Frequency Comb

Femtosecond Optical Frequency Comb:
Principle, Operation, and Applications

Kluwer Academic Publishers / Springer
Norwell, MA
Jun Ye and Steven T. Cundiff

Editors
PREFACE
Over the last few years, there has been a convergence between the fields
of ultrafast science, nonlinear optics, optical frequency metrology, and
precision laser spectroscopy. These fields have been developing largely
independently since the birth of the laser, reaching remarkable levels of
performance. On the ultrafast frontier, pulses of only a few cycles long have
been produced, while in optical spectroscopy, the precision and resolution
have reached one part in 1014. Although these two achievements appear to be
completely disconnected, advances in nonlinear optics provided the essential
link between them. The resulting convergence has enabled unprecedented
advances in the control of the electric field of the pulses produced by
femtosecond mode-locked lasers. The corresponding spectrum consists of a
comb of sharp spectral lines with well-defined frequencies. These new
techniques and capabilities are generally known as “femtosecond comb
technology.” They have had dramatic impact on the diverse fields of
precision measurement and extreme nonlinear optical physics.
The historical background for these developments is provided in the
Foreword by two of the pioneers of laser spectroscopy, John Hall and
Theodor Hänsch. Indeed the developments described in this book were
foreshadowed by Hänsch’s early work in the 1970s when he used
picosecond pulses to demonstrate the connection between the time and
frequency domains in laser spectroscopy. This work complemented the
advances in precision laser stabilization developed by Hall. The parallel
efforts on mode-locked lasers by Charles Shank, Erich Ippen, and others laid
the groundwork for the development in the 1990s by Wilson Sibbett of Kerr-
lens mode locking, the instantaneous nature of which yields sub-10 fs pulses
directly from laser oscillators that correspond to strong phase-locking of the
comb components across a broad optical spectrum. The synergy between
precision spectroscopy and ultrafast lasers was catalyzed by the development
of novel optical fiber with high nonlinearity and controlled dispersion.
In Chapter 1 we provide an introductory description of mode-locked
lasers, the connection between time and frequency descriptions of their
ii Preface
output, the physical origins of the electric field dynamics, and an overview
of applications of femtosecond comb technology. Chapter 2 by Ippen,
Kärtner and Cundiff discusses the development of ultrashort lasers,
particularly focusing on how to achieve an octave-spanning spectrum and
pulse dynamics that are relevant to the stability and control of the comb.
Chapter 3 by Bartels describes in detail high-repetition-rate ring oscillators
for precision frequency metrology. Chapter 4 by Gaeta and Windeler
provides in-depth discussions relating to the physics of bandwidth generation
and the underlying noise process during pulse propagation through
microstructure fibers. Certain aspects of comb dynamics and stability are
presented in Chapters 5 by Steinmeyer and Keller. An attractive approach
presented in Chapter 6 by Kobayashi makes use of optical parametric
generation to produce high-peak-power, femtosecond pulses in the IR
spectral domain. A detailed review of the traditional harmonic-based
frequency chain is provided in Chapter 8 by Schnatz, Stenger, Lipphardt,
Haverkamp, and Weiss, while the new epoch of absolute optical frequency
measurement using femtosecond comb technology is reviewed in Chapter 7
by Udem, Zimmermann, Holzwarth, Fischer, Kolachevsky, and Hänsch.
Chapter 9 by Diddams, Ye, and Hollberg provides an account of the current
state-of-the-art performance and characterization of femtosecond comb
systems used for optical frequency measurement, synthesis, and optical
atomic clocks. Chapter 10 by Baltuška, Paulus, Lindner, Kienberger, and
Krausz provides a thorough discussion of the generation of the high-intensity
pulses needed to access the regime of extreme nonlinear optics and a review
of the results obtained for above-threshold ionization. Control of high-
harmonic generation is addressed in Chapter 11 by Gibson, Christov,
Murnane and Kapteyn. Stabilization of mode-locked lasers and their
applications to ultrasensitive sensors are discussed in Chapter 12 by Diels,
Jones, and Arissian.
The rapid progress during the last 5–6 years has been breathtaking and
has made a tremendous impact on both science and technology. We foresee
an undiminished potential for similar advances in the near future. We hope
that the readers of this book will share our enthusiasm and benefit from the
material presented in this book.
The editors thank all of the chapter authors for their contributions. The
efforts of Julie Phillips and Lynn Hogan in the JILA Scientific Reports
Office are also gratefully acknowledged.
Boulder
September, 2004 Jun Ye and Steven T. Cundiff
CONTENTS
Preface.......................................................................................................i
Contents...................................................................................................iii
History of Optical Comb Development ..................................................1
Introduction ..........................................................................................12
1. Time- and Frequency-Domain Pictures of a Mode-Locked
Laser ..........................................................................................14
1.1 Introduction to mode-locked lasers......................................14
1.2 Frequency spectrum of a mode-locked laser ........................15
1.3 Determining absolute optical frequencies with octave-
spanning spectra..................................................................17
1.4 Femtosecond optical-frequency comb generator ..................19
1.5 Time- and frequency-domain characterizations of f0 ............22
2. Precision Optical Frequency Metrology Using
Femtosecond-Optical-Frequency Combs.....................................24
2.1 Measurement of absolute optical frequency .........................24
2.2 Optical atomic clocks..........................................................27
2.3 Optical frequency synthesizer..............................................29
3. Atomic and Molecular Spectroscopy...........................................30
3.1 Precise, simultaneous determination of global atomic
structure and transition dynamics ........................................30
3.2 I2 hyperfine interactions, optical frequency standards,
and clocks ...........................................................................33
4. Carrier-Envelope Phase Coherence and Time-Domain
Applications ...............................................................................38
4.1 Timing synchronization of mode-locked lasers....................39
4.2 Phase lock between separate mode-locked lasers .................40
4.3 Extending phase-coherent femtosecond combs to the
mid-IR spectral region.........................................................41
4.4 Femtosecond lasers and external optical cavities .................42
iv Contents
4.5 Coherent control via quantum interference between

one- and two-photon absorption ..........................................46
4.6 Extreme nonlinear optics.....................................................48
5. Summary....................................................................................48
Femtosecond Laser Development .........................................................54

1. Introduction................................................................................54
2. Pulse Dynamics ..........................................................................57
3. Octave-Spanning Lasers .............................................................63
3.1 Octave-spanning laser using prisms.....................................64
3.1.1 Design .....................................................................65
3.1.2 Carrier-envelope phase stabilization.........................66
3.1.3 Frequency-dependent spatial mode ..........................67
3.2 Prismless octave-spanning laser ..........................................70
3.2.1 Design .....................................................................70
3.2.2 Carrier-envelope phase stabilization.........................72
4. Carrier-Envelope Phase Dynamics..............................................74
5. Summary....................................................................................75
Gigahertz Femtosecond Lasers.............................................................78

1. Introduction................................................................................78
2. High-repetition-Rate Ring Oscillators .........................................80
2.1 Design criteria and basic resonator layout............................80
2.2 Standard Ti:sapphire lasers for 0.3–3.5 ghz repetition
rate .....................................................................................83
2.3 Cr:forsterite oscillator at 433 mhz — extension to
telecommunication wavelengths..........................................86
3. Broadband Ti:sapphire Oscillator................................................88
3.1 How does it work? ..............................................................89
3.2 Application in frequency metrology and optical clocks........91
4. Conclusion .................................................................................93
Microstructure Fiber and White-Light Generation.............................97

1. Introduction................................................................................97
2. Microstructure Fiber Fabrication.................................................98
2.1 Preform fabrication .............................................................98
2.2 Fiber draw ........................................................................100
3. Microstructure Fiber Types.......................................................100
4. Linear Optical Properties of Microstructure Fiber .....................102
5. Supercontinuum Generation .....................................................104
5.1 Nonlinear envelope equation .............................................104
5.2 Spectral superbroadening ..................................................105
Contents v
5.3 Continuum instability and noise ........................................107

6. Conclusions..............................................................................109
Optical Comb Dynamics and Stabilization ........................................112

1. Introduction..............................................................................112
2. Comb Parameters and Their Connection to Intracavity
Dispersion ................................................................................113
2.1 Carrier-envelope-offset phase and frequency in the time
domain..............................................................................113
2.2 The frequency comb and its dynamics ...............................115
3. Measurement of the CEO Frequency ........................................ 117
3.1 Heterodyning different laser harmonics .............................117
3.2 Transfer oscillators and interval bisection..........................119
4. CEO Phase Noise .....................................................................120
4.1 Noise densities and rms phase jitter ...................................121
4.2 CEO-phase noise of mode-locked oscillators.....................121
4.3 Physical mechanisms behind CEO fluctuations .................123
4.4 Amplitude-to-phase conversion effects..............................124
5. Stabilization of the CEO Frequency.......................................... 126
5.1 Controlling the CEO frequency of a laser oscillator ...........126
5.2 Performance of CEO phase locks ......................................127
5.3 Limitations of CEO control ...............................................129
6. Summary..................................................................................130
Femtosecond Noncollinear Parametric Amplification and

Carrier-Envelope Phase Control ...................................................133
1. Introduction..............................................................................133
2. Advances of Noncollinear-Phase-Matched Optical
Parametric Conversion..............................................................135
3. Principle of Parametric Amplification.......................................138
3.1 Noncollinear-optical-parametric amplification (NOPA).....140
4. Signal-Wavelength-Insensitive Phase Matching........................144
5. Group-Velocity Matching in β-BaB2O4 .....................................146
6. Femtosecond NOPA Based on β-BaB2O4..................................150
6.1 Broadband amplification of a single-filament
continuum.........................................................................150
6.2 Output properties ..............................................................152
7. Limitation of Pulse Width because of Pulse-Front Mismatch ....155
7.1 Tilted pump geometry for pulse front matching .................157
8. Chirp Property of Signal ........................................................... 159
8.1 Compression to the sub-5 fs regime...................................160
9. Second-Generation Noncollinear Parametric Amplifier............. 166
vi Contents
10. Conclusions and Outlook..........................................................170
Optical Frequency Measurement .......................................................176

1. Frequency Combs..................................................................... 176
2. The Cesium D1 Line and the Fine Structure Constant................179
3. Optical Synthesizers .................................................................181
4. Octave-Wide Frequency Combs ...............................................183
5. Application to Hydrogen ..........................................................184
6. The First Optical Synthesizer....................................................185
7. The Hydrogen Spectrometer .....................................................187
8. The 1S–2S Transition Frequency ..............................................190
9. Checking for Slow Drifts of a Natural Constant ........................192
Optical Frequency Measurement Using Frequency

Multiplication and Frequency Combs ...........................................198
1. Introduction..............................................................................198
2. Frequency Measurements by Repeated Harmonic Mixing
(Frequency Chains) ..................................................................200
3. Frequency Interval Division Approach......................................202
4. Optical Frequency Measurement Using Femtosecond Lasers ....204
5. Optical Frequency Measurements .............................................209
5.1 Ca frequency measurement ...............................................210
5.2 Yb+ frequency measurement..............................................212
6. Test on the Precision of Frequency Measurement with
Frequency Combs..................................................................... 213
6.1 Transfer technique ............................................................213
6.2 Frequency ratio .................................................................216
7. Summary..................................................................................220
Femtosecond Lasers for Optical Clocks and Low Noise

Frequency Synthesis.......................................................................225
1. Introduction..............................................................................225
1.1 Basic components of optical clocks ...................................227
1.2 Uses of optical atomic clocks ............................................229
1.3 A brief history of the development of optical clocks ..........230
2. The Atomic Reference..............................................................231
2.1 Single ion references .........................................................233
2.2 Neutral atom references ....................................................234
2.3 Molecular references.........................................................236
2.4 Local oscillator requirements ............................................238
3. Femtosecond Laser-Based Optical Frequency Synthesizers.......238
Contents vii
3.1 Considerations in designing a femtosecond comb for

use in an optical clock .......................................................240
3.2 Frequency synthesis with a femtosecond laser ...................243
3.2.1 methods of control .................................................244
3.3 Testing the synthesizer ......................................................249
3.4 Alternatives to Ti:sapphire ................................................251
4. Signal Transmission and Cross-Linking....................................253
Generation And Measurement of Intense Phase-Controlled Few-

Cycle Laser Pulses..........................................................................263
1. Introduction..............................................................................264
2. Carrier-Envelope Phase of a Mode-Locked Pulse Train and
a Single Attosecond Pulse.........................................................268
3. Measurement of Phase Variations .............................................270
3.1 Detecting carrier-envelope drift of oscillator pulses ...........270
3.2 Detecting carrier-envelope drift of amplified pulses...........271
3.3 Measuring the phase difference by linear interferometry....273
4. Phase Jitter of the White-Light Continuum ...............................276
4.1 Phase lock between the input pulse and the white-light
continuum.........................................................................276
4.2 Phase noise resulting from intensity fluctuations ...............278
5. Concepts of Phase-Controlled Amplification ............................280
6. Phase-Stabilized 5 fs, 0.1 TW-Amplified System......................284
7. Control of Light Field Oscillations............................................288
8. Carrier-Envelope-Phase Measurement without Ambiguity ........296
9. Gouy Phase Shift for Few-Cycle Laser Pulses...........................302
10. Conclusions and Outlook..........................................................307
Quantum Control of High-Order Harmonic Generation ..................314

1. The Physics of High-Order-Harmonic Generation.....................315
2. “Single-Atom” Effects in High-Order-Harmonic
Generation: Manipulation and Coherent Control .......................317
3. Phase Matching of High-Harmonic Generation .........................319
4. Quasi-Phase Matching of High-Harmonic Generation...............323
5. Conclusion ...............................................................................329
Applications of Ultrafast Lasers .........................................................333

1. Mode Locking ..........................................................................334
2. Group and Phase Velocities in Ring Lasers...............................337
3. Ring Lasers as Sensors .............................................................338
3.1 Case (1): nonreciprocal effects ..........................................340
viii Contents
3.2 Case (2): reciprocal effects that can be synchronized to

the cavity repetition rate....................................................341
3.3 Case (3): reciprocal effects, slow motions .........................342
4. Intracavity-Pumped Optical Parametric Oscillator as a
Mode-Locked Ring Laser .........................................................343
5. Stabilization of the Frequency Comb ........................................ 345
5.1 Locking femtosecond pulses to stable reference cavities ....346
5.2 Characterization of femtosecond comb stability.................348
6. Dispersion Measurement Applications ......................................349
6.1 Cavity characterization......................................................349
6.2 Atmospheric dispersion.....................................................351
7. Ring Laser Stabilization............................................................352
Author addresses...................................................................................355
Index ....................................................................................................359
Foreword
HISTORY OF OPTICAL COMB DEVELOPMENT
John L. Hall,1 and Theodor W. Hänsch,2

1
JILA, National Institute of Standards and Technology and University of Colorado
2
Max-Planck-Institut für Quantenoptik
In the past five years, progress in laser stabilization, optical frequency

measurement, femtosecond laser development and stabilization, nonlinear
optics, and related topics has been stunning and unexpected. The excitement
surrounding the rapid evolution in these fields since 1999 gives us a hint of
what it must have been like after 1927 when the first ideas of quantum
mechanics were being introduced. With laser optics, however, the explosion
of knowledge is based upon years of detailed, painstaking research in the
independent fields of laser stabilization, ultrafast laser development, and
highly nonlinear optics [1]. The coalescence of these fields has provided five
years of almost unprecedented discovery while, at the same time, a new
millennium in metrology has generated advances of fundamental value in the
contributing fields and in their spin-offs. To give just two examples: (1) the
precise synchronization of picosecond and femtosecond lasers [2] allows
nonlinear surface probing at the single-molecule level using coherent anti-
Stokes Raman microscopy and spectroscopy [3] and (2) the use of stabilized
comb pulses allows time and frequency dissemination [4] over extended
distances via fiber optics. The latter offers a new capability to metrologists
and researchers interested in developing new accelerators and large-array
radio telescopes.
Here we offer our perspective on how we got started on this incredible
journey of discovery and why it is occurring at this time. The rest of this
timely book highlights key technical advances from the perspective of the
researchers who made them. Consequently, this book should be of interest to
students, practitioners in this rapidly evolving field, and physicists, chemists,
and biologists whose research will be enhanced by the insights and
discoveries recounted here.
2 Foreword
The historical development of pulsed and continuous wave (cw) lasers

diverged right from the beginning. The solid-state ruby laser of Ted Maiman
involved kilojoule discharges into flash lamps and repetition rates from zero
up to once per minute. (The laser’s high ion density implied that a serious
power level would be involved.) In contrast, the gas laser of Ali Javan had
much lower gain margins and required temporal stability to allow its
incremental approach to the laser threshold. From the early 1960s until about
1990, the pulse and cw laser communities continued to diverge. The cw laser
folks liked to enhance the stability of their lasers, because stability was the
main good feature they had. Certainly it wasn’t the ability to burn holes
through Gillette razor blades, a popular specification for pulsed lasers of the
day. Over the years, the cw-laser teams learned to frequency stabilize their
milliwatt-scale lasers, eventually reaching into the subhertz domain. Some
members of the pulse-laser community escalated to kilojoule pulse energies
delivered in nanoseconds for fusion target compression and even larger
discharges to be delivered to potentially hostile moving targets.
On the other hand, the university research community wanted pulse
lasers with high repetition rates that would enhance nonlinear responses,
enable signal averaging over many pulses, and reduce the destructive impact
of the probing radiation on the probed system. Thus began the search for
laser media that were quiet and calm enough to lead to repeatable pulses,
even at high repetition rate. First came the mode-locked He-Ne laser
followed by actively mode-locked Argon lasers, which were developed
commercially to give nanosecond pulses at 100 MHz rates. By the time
synchronously pumped mode-locked picosecond dye lasers were introduced
in the mid-1970s, the comb approach to frequency measurement was
probably already inevitable, albeit more than 20 years in the future.
A driving force in the development of laser (and optical comb)
technologies was the desire to learn more about the physical world. For
instance, Professor Hänsch’s group at Stanford (and later in Garching)
focused on learning about the hydrogen spectrum, comparing hydrogen with
deuterium to see the isotope shifts, accurately determining energies
associated with various principal quantum numbers, and so forth. From the
beginning, they realized that certain physical parameters, such as the Lamb
shift or proton-size changes, could be isolated by using their quantum-
number-scaling dependence and thus probed in the optical regime.
At Stanford, Hänsch’s team demonstrated one of the first mode-locked
“femtosecond” dye lasers (with a pulse length of less than one picosecond)
in 1977 [5]. Their progress in laser research was closely connected to the
invention of spectroscopic techniques in the sub-Doppler regime as part of
their work on precision spectroscopy of the simplest atomic system,
hydrogen. The group’s new tunable dye-laser pulses were soon used in a
HISTORY OF OPTICAL COMB DEVELOPMENT 3
landmark experiment to demonstrate Fourier’s reality to us: stable pulse

trains — evenly-spaced in time — represented as stable combs of frequency
components. We note that in the late 1970s, Veniamin Chebotayev in
Novosibirsk had also begun thinking about stable, repetitive laser pulse
trains. Even though the early experiments could not yet verify the precisely
equal spacing of the comb frequencies, Eckstein, Ferguson, and Hänsch [6]
used a comb structure to measure some fine and hyperfine frequency
intervals in atomic sodium in 1978. This classic paper would help define the
future impact of ultrafast lasers on precision measurement. Then, as the work
at Garching progressed over the years, one dream became clear and insistent:
if only we could measure optical frequencies directly and accurately!
This vision led the Garching team to flirt with extended chains of
synchronized frequency sources. In the late 1980s, their new idea [7] was to
deal with a series of increasing frequency differences between tunable diode
laser sources. In this way, they hoped to avoid one of the major headaches of
traditional chains — the step-wise increase of absolute frequencies
throughout the chain that obliges one to develop a different laser technology
at many points to cover the 105 frequency ratio from the microwave regime
to the visible. A number of laser-diode frequency-interval-divider stages
were built. A four-stage system was used for one of the hydrogen frequency
measurements.
The comb idea jumped ahead with the demonstration of intracavity
modulator-based spectral comb generators by Kourogi et al. [8]. In these
experiments, pulses were modulated onto stable cw laser beams. Researchers
in both Garching [9] and Boulder [10] recognized the utility of these devices
and launched frequency-measurement programs using their few terahertz
width modulator-based optical combs to bridge the awkward frequency gaps.
In the meantime, the ultrafast laser community continued to work toward
improved pulse train stability and shorter pulses. Researchers developed
several solid-state lasers, such as the cw-pumped mode-locked Nd:YAG,
that were attractive for their stability. However, pulse durations below 30 ps
were difficult to obtain because of the gain-bandwidth limitation of the laser
material itself. Eventually the “intelligent and beautiful princess” — the
titanium-doped sapphire laser system — was introduced and developed,
along with the important discovery of Kerr-lens mode locking by Wilson
Sibbett in St. Andrews [11]. These inventions changed femtosecond lasers
from delicate contraptions to simple and reliable devices. Soon afterwards,
commercial Ti:sapphire femtosecond lasers became readily available and
they offered sub-100 fs pulses by the early 1990s. By 1994, the Garching
group had acquired a Coherent Mira* laser for frequency metrology
experiments. This laser and other similar devices opened the door to solving
an array of challenging problems.
4 Foreword
The factors that limit the shortness of the generated laser pulses arise
from two issues: (1) finite gain-bandwidth product (which is not a problem
for Ti:sapphire) and (2) intracavity dispersion. A short pulse can be viewed
as the superposition of many cw phase-locked modes, all of which oscillate
at their own cavity-defined frequencies. For the pulse train to be stable in
time, the modes must have a common frequency separation. Because of
dispersion in the sapphire, intracavity air, and mirror coatings, these cavity
frequencies are generally not exactly evenly spaced. This is particularly true
as the spectral bandwidth dramatically increases. So even with the ~30%
bandwidth of Ti:sapphire, further shortening of sub-100 fs pulses proved
difficult until Asaki et al. [12] employed a sufficiently general analysis of the
pulse laser cavity. This model included the index of refraction characteristics
of the intracavity dispersion-compensating prisms, the resulting color
influence on the refraction direction, associated cavity path lengths, and
some modeling of the air and laser crystal dispersion. Space-time focusing of
the light bullet in the laser crystal was another important consideration.
In the early 1990s, when most of researchers were working feverishly to
shorten pulse widths, some dreamers began to think of pulses so short that
their Fourier representation would span from radio frequencies up into the
visible domain. This idea seemed like science fiction to one of the authors
(JLH) and a likely possibility to the other (TWH). Of course, spectral self-
broadening was well known. By focusing powerful amplified pulses into
water or some solids, one could basically generate a white-light continuum.
At elevated pulse energy levels, one expected serious disruption of the
calmness of the intermolecular bonds; consequently, one would not expect to
find a phase-stable repetition of the generated white light. Perhaps a glass
sample could melt and recrystallize at a 100 MHz rate, emitting a similar
thermal spectrum on every heating cycle. However, to form a coherent
optical comb in the forward direction, the timing would need to be stable to
~1 radian — at the visible frequency! Few believed that this thermal process
would be stable at the 0.3 fs level needed. Rather, it seemed clear that a more
gentle process would be required, in which somewhat less-powerful laser
pulses would strongly distort some atomic wave functions but not disrupt
chemical bonds. Atomic frequencies are so high that when the pulse is gone,
calmness can return; the next pulse would be able to generate just the same
effect on the system. In this case, the phase-coherence of the source pulses
could insure that the nonlinearly generated frequencies would be mutually
coherent pulse-to-pulse.
Researchers in both Garching and Boulder set out to learn about this
subject. In Europe, an amplified pulse was split into two parts and focused
onto two separate spots in a CaF2 crystal. The white light produced in each
spot in the CaF2 plate interfered with each other in a geometry that allowed
the time delays to be equalized. Miraculously, stable white interference

fringes could be seen by eye [13]. In this way, the Garching group realized
that the phase of the nonlinearly generated light was stable enough to form
an optical comb! This experiment led to Hänsch’s detailed six-page
proposal, dated March 30, 1997, for an octave-spanning self-referenced
universal optical-frequency comb synthesizer. Following this proposal,
developments in the art of femtosecond-laser frequency comb generation
began to appear quickly.
The next year (1998) saw the first crucial test of a Kerr-lens mode-locked
Ti:sapphire laser in Garching [14]. This experiment clearly proved the
viability of femtosecond-laser frequency comb synthesizers. Before this
experiment, some researchers had argued that the laser comb spectrum
would be completely washed out by phase noise. Although some
publications were held back due to the restrictions of German patent law, by
2000 the Garching team reported its first absolute frequency measurement
with a comb (made in 1999) [15]. In a direct comparison with the
transportable cesium fountain clock of the Bureau National de Métrologie –
Systèmes de Référence Temps Espace, they measured the frequency of the
hydrogen 1S–2S two-photon resonance with an uncertainty of 1.9 x 10-14,
which is more than an order of magnitude more accurate than any previous
optical frequency measurement. This experiment firmly established the
viability of optical frequency metrology with femtosecond laser frequency
combs. It also electrified the frequency metrology community. The
convenience and simplicity of octave-spanning frequency combs added to
the attraction of this new approach.
In the meantime, researchers in Boulder were studying the spectral
expansion associated with pulse propagation in an optical fiber as a possible
replacement for their modulator-based comb generator. They showed that a
cw test beam in the fiber developed a comb structure on it because of the co-
propagating femtosecond pulses. However, the new comb only contained
information about the repetition rate of the pulses, not about their optical
frequency [16]. Of course, in retrospect, it is clear that this “cross-
modulation” result (as opposed to “four-wave mixing”) was preordained
because of the huge difference in the phase velocities of the two colors. The
interesting frequency-coupled wavelets never got a chance to coherently
build up along the fiber’s length. This experiment also showed that
spectrally narrow features would be generated across a broad spectral range
even when “pounding” on a fiber with powerful femtosecond pulses.
A classic 1999 paper [17] offered a complete description of issues and
techniques for comb-based optical frequency measurement, including the
famous carrier-envelope phase-slipping issue [18]. The Garching team made
their first self-contained rf-to-optical frequency comparison [19] using a
6 Foreword
comb configuration that had a stabilized carrier-envelope-offset frequency.

Since the comb spectrum did not yet span an octave, some interval-divider
stages with auxiliary lasers had to be used for self-referencing.
Soon afterwards, the world was turned upside down by Jinendra Ranka,
Robert Windeler, and Andrew Stentz. During the 1999 CLEO postdeadline
session, they announced that they had demonstrated that white light could be
produced in a revolutionary way by using an internally structured fiber
incorporating a number of air holes [20]. They showed pictures in which the
input dark red pulse gradually transformed itself into green and blue, with
expansion into the IR direction occurring as well. Soon it was discovered
that supercontinuum could also be obtained in tapered fibers [21].
The central rod in the fiber preform had been surrounded by a number of
hollow tubes. When drawn down to fiber scale, the inner “core” was
surrounded by air that presented a vastly larger index contrast than found in
normal fiber (where the contrast may be ~0.01). Consequently, a single-
mode microstructure fiber would have a much smaller diameter for a given
wavelength, for example 1.7 instead of 5 micrometers for 800 nm light. This
meant that a laser beam focused into this fiber would have a ~tenfold higher
intensity and generate ~tenfold larger self-induced phase shifts during the
pulse because of the Kerr effect, which is analogous to the quadratic Stark
effect in atoms. Suddenly, researchers learned that white light could be
generated by normal femtosecond oscillators without expensive and low-
repetition-rate amplifiers!
Researchers also realized that the special fiber could help with the phase
matching needed for a big coupling to accrue. In the “holey” fiber, a number
of fiber parameters could be designer variables: the basic glass and its
dispersion, the core size, the fractional angular coverage of supporting web,
the size, and number of surrounding air holes. By design, it proved possible
to make a strong cancellation between dispersion of the fiber core material
and the dispersion associated with the geometric structure. In this way an
input pulse could propagate vastly longer distances — millimeters rather
than micrometers — before its peak intensity was diminished by the
different propagation speeds of its spectral components.
Knowing that the pulses were so gentle the fiber was not damaged and
that all the frequency components would be cross-coupled together by the
nearly constant propagation speed, we predicted the output of the fiber
would be a spectral comb of coherent frequencies. Of course, for precision
metrology, this idea would have to be tested.
The first question was: how can we get some “Magic Rainbow Fiber?”
Our combined approaches to friends, colleagues, and administrators at Bell
Labs all came to nothing, apparently because of the lawyers there. Luckily
for the JILA team, its most recently recruited colleague, Steve Cundiff, had
been in a nearby group in the same part of Bell Labs. By some unknown
means, the JILA team came up with a sample of “Magic Fiber” to test by
November 1999. The Garching group teamed up with the powerful fiber
group of P. St. J. Russell at the University of Bath (UK), which had been
working with both microstructure and tapered fibers. New results began
immediately rolling in [21, 22], and the publication competition began in
earnest! (Since then, many alternatives to the Magic Fiber approach have
emerged, including supercontinuum generation in tapered fibers [21],
octave-spanning laser oscillators without the use of external fibers, mode-
locked fiber lasers with some highly nonlinear ordinary fiber for spectral
broadening, and schemes incorporating difference-frequency generation to
determine the carrier-envelope-offset frequency with combs spanning less
than an octave.)
At this point, the Boulder team had a significant advantage. They had
already worked on laser stabilization and optical frequency standards for
many years because of JILA’s affiliation with the Boulder campus of the
National Institute of Standards and Technology (NIST). Indeed, the authors
of this foreword first met in Novosibirsk in 1969 at a conference organized
by the late Veniamin Chebotayev on the topic of stabilized lasers. There,
JLH presented his progress with a methane-stabilized He-Ne laser.
While other researchers were busy improving cw and pulsed lasers,
national metrology and standards laboratories around the world had been
trying to verify the frequencies of their “as-maintained” national wavelength
standards. The first such measurements occurred in Boulder and led, in
1972, to the measurement of the frequency and wavelength of the methane
standard. This measurement, in turn, led to a new and definitive value for the
speed of light. Other national laboratories joined in, and a long discussion
ensued about the philosophical and practical issues associated with
calculating meters from the frequency of light rather than simply adopting
new wavelength standards as they became available. Within 10 years,
national laboratories in Canada, the United Kingdom, Japan, Germany, and
Gaithersburg, Maryland, had confirmed parts of the Boulder work and the
frequency of the He-Ne iodine-stabilized red laser had been determined. In
1983, the meter was redefined based on the speed of light.
The reproducibility of most of the reference lasers developed during this
era was so good that their imperfections had little practical consequence for
length metrology. Still, the optical frequency standards business continued to
develop as researchers sought better designs and new reference transitions.
More importantly, each nation wanted to confirm its own standards at the
highest level.
Both NIST and PTB built very good systems to measure the calcium
intercombination transition at 657 nm, for example. In addition, the PTB
8 Foreword
team built up one of the best rf-to-optical harmonic frequency “chains,”

which took advantage of several decades of work on system components.
While PTB’s “traditional” frequency chain worked well, its sheer
complexity was daunting. The German laboratory’s state-of-the-art
measurement [23] of the Ca frequency was published in 1996 and had a
frequency uncertainty of ~ 430 hertz, arising from both the Ca standard and
the measurement scheme. (By 2003 using femtosecond comb techniques,
both NIST and PTB were reporting uncertainties for this frequency in the
~10 hertz range, limited mostly by interesting spectroscopic issues with the
optical Ca standard.)
Its experience with optical frequency standards allowed the Boulder team
at JILA to take Ranka et al.’s groundbreaking discovery of the properties of
Magic Fiber and run with it. The Boulder team was the first to measure (and
control) the carrier-envelope-offset frequency with a ν-to-2ν self-referenced
comb [24]. With this method, they determined the frequency offset of the
comb lines from the positions of harmonics of the repetition rate. Some
known optical frequencies were confirmed. Within a few months, the group
was also attempting to generate femtosecond pulses of controlled shape, in
which the pulse-to-pulse carrier-envelope-offset phase was under the
experimenter’s control.
The absolute frequency measurements in the pre-comb epoch (prior to
1999) were suddenly such a bother! The simplicity and efficiency of the new
comb-based measurements attracted wide interest and provided a huge boost
to the optical frequency metrology and standards field. Some parts of a total
optical frequency metrology system even became available commercially
(see, for example, http://www.menlosystems.com/*). Five years into the
optical frequency comb epoch, the once-independent laser research
subcommunities are now extremely interdependent, at least as viewed from
the perspective of results. Of course, each advance has been primarily an
independent step, attractive in its own context. Which market planner could
have organized this beauty?
One question remains, however: how can we be sure of the frequency comb
results? The accuracy of comb techniques has been the subject of a number of
tests, but no problems have turned up so far. JILA has used the comb
technique to measure “known” optical standards. Rather than discovering
limitations of the comb technique, we have typically found that these
measurements refine our knowledge of the “known” standard [25]. Thus the
obvious method of testing the comb by measuring a physical standard by both
methods is not successful because of limitations of the standards themselves.
Using another approach, the Garching team compared an octave-spanning
frequency-comb synthesizer [26] with the more complex frequency synthesizer
used in their 1999 hydrogen-frequency measurement. By starting with a
common 10 MHz rf reference and comparing comb lines near 350 THz, they
verified agreement within a few parts in 1016; the precision of the experiment
was probably limited by Doppler shifts due to air pressure changes or thermal
expansion of the optical tables.
The Garching group also made additional accuracy tests on the comb-
spacing uniformity using one stage of a frequency-interval divider [14]. The
interval between comb lines near the edges of the “white” spectrum could be
divided to find the “center” by two methods. Or, an edge line could be
combined with one near the middle to seek a dispersive effect. No problems
could be found. To really press to the testing limits, comb vs comb tests —
using different comb frequencies, materials, or whatever we think could be
important — are probably necessary. By late 2004, tests still had discovered no
problems, and the accuracy in the context of frequency combs had reached
10-21 [27].
As we begin the next era in optical comb research, with each group
measuring specific frequencies with femtosecond comb techniques, the
realization of the cesium standard frequency is likely to be the first weak
point. With a day’s averaging, the GPS system can help us know our local
frequency standard’s average performance, but it takes about a day to deliver
an accuracy ~1 x 10-14; achieving this accuracy requires us to know the
timing comparisons separately with each of the satellites used in the test. To
test optical frequency combs against the highest traditional standard, modern
fountain cesium clocks are usually available either by fiber link [4] or by
physical transport.
Femtosecond combs are now ready to accurately measure any desired
frequency such as those of some isolated hydrogen atoms at rest in a field-
free vacuum, single Hg+ or Yb+ ions in an ion trap, or a million cold Sr
atoms trapped in an optical lattice/trap. In addition, frequency comb
techniques are having an impact on ultrafast physics. By making it possible
to produce few-cycle pulses with a stabilized carrier-envelope phase, these
tools are leading to the discovery of novel phenomena in nonlinear light-
matter interactions. For example, by 2004 amplified phase-stabilized pulses
had been used to produce controlled bursts of soft x-rays with time durations
in the attosecond range [28].
So where do we go from here?
In contrast to the digital security of frequency measurement, in the final
accuracy-defining step, it usually comes out that spectroscopic line-shape
issues are what limit our results. How well do we know the connection
between the center of the observed resonance line and the desired physical
quantity? Resonance frequencies can be shifted by fields, laser intensity,
Doppler shifts, and ... Careful treatment of such issues is part of what makes
our field so fun. And, of course, many interesting measurements can be
10 Foreword
designed so they measure differences or temporal changes and hence are

better isolated from such limitations. But as our experiments improve, the
line-center question will continue to reappear, ideally at an increasingly
reduced sensitivity.
In summary, optical frequency combs have given us some comfortable
metrological headroom for pushing ideas for new optical frequency
standards and for measuring interesting physical constants. And, we feel
confident that the new metrology based on integer arithmetic and
femtosecond combs will be sufficient to reveal shortcomings in our
spectroscopy ideas and implementations.
*Use of product name for technical information only and does not constitute an
endorsement by NIST.
REFERENCES
[1] J. L. Hall, IEEE J. Sel. Top. Quantum Electron. 6, 1136-1144 (2000).
[2] R. K. Shelton, L. S. Ma, H. C. Kapteyn, M. M. Murnane, J. L. Hall, and J. Ye,
Science 293, 1286-1289 (2001).
[3] E. O. Potma, D. J. Jones, J. X. Cheng, X. S. Xie, and J. Ye, Opt. Lett. 27,
1168-1170 (2002).
[4] J. Ye, J. L. Peng, R. J. Jones, K. W. Holman, J. L. Hall, D. J. Jones, S. A.
Diddams, J. Kitching, S. Bize, J. C. Bergquist, L. W. Hollberg, L. Robertsson,
and L. S. Ma, J. Opt. Soc. Am. B 20, 1459-1467 (2003).
[5] A. I. Ferguson, J. N. Eckstein, and T. W. Hänsch, Appl. Phys. Lett. 49, 5389-
5391 (1978).
[6] J. N. Eckstein, A. I. Ferguson, and T. W. Hänsch, Phys. Rev. Lett. 40, 847-850
(1978).
[7] H. R. Telle, D. Meschede, and T. W. Hansch, Opt. Lett. 15, 532-534 (1990).
[8] M. Kourogi, K. Nakagawa, and M. Ohtsu, IEEE J. Quantum Electron. 29,
2693-2701 (1993).
[9] A. Huber, T. Udem, B. Gross, J. Reichert, M. Kourogi, K. Pachucki, M.
Weitz, and T. W. Hänsch, Phys. Rev. Lett. 80, 468-471 (1998).
[10] J. L. Hall, L. S. Ma, M. Taubman, B. Tiemann, F. L. Hong, O. Pfister, and J.
Ye, IEEE Trans. Instrum. Meas. 48, 583-586 (1999).
[11] D. E. Spence, P. N. Kean, and W. Sibbett, Opt. Lett. 16, 42-44 (1991).
[12] M. T. Asaki, C. P. Huang, D. Garvey, J. P. Zhou, H. C. Kapteyn, and M. M.
Murnane, Opt. Lett. 18, 977-979 (1993).
[13] M. Bellini and T. W. Hänsch, Opt. Lett. 25, 1049-1151 (2000).
[14] T. Udem, J. Reichert, R. Holzwarth, and T. W. Hänsch, Opt. Lett. 24, 881-883
(1999).
[15] M. Niering, R. Holzwarth, J. Reichert, P. Pokasov, T. Udem, M. Weitz, T. W.
Hänsch, P. Lemonde, G. Santarelli, M. Abgrall, P. Laurent, C. Salomon, and
A. Clairon, Phys. Rev. Lett. 84, 5496-5499 (2000).
[16] D. J. Jones, S. A. Diddams, M. S. Taubman, S. T. Cundiff, L. S. Ma, and J. L.

Hall, Opt. Lett. 25, 308-310 (2000).
[17] J. Reichert, R. Holzwarth, T. Udem, and T. W. Hänsch, Opt. Commun. 172,
59-68 (1999).
[18] J. Eckstein, Ph.D Thesis, Stanford University (1978).
[19] J. Reichert, M. Niering, R. Holzwarth, M. Weitz, T. Udem, and T. W. Hänsch,
Phys. Rev. Lett. 84, 3232-3235 (2000).
[20] J. K. Ranka, R. S. Windeler, and A. J. Stentz, Opt. Lett. 25, 25-27 (2000).
[21] T. A. Birks, W. J. Wadsworth, and P. S. Russell, Opt. Lett. 25, 1415-1417
(2000).
[22] J. K. Ranka and R. S. Windeler, in Opt. Photonics News, 2000), Vol. 11, p.
20-25; J. K. Ranka, R. S. Windeler, and A. J. Stentz, Opt. Lett. 25, 796-798
(2000).
[23] H. Schnatz, B. Lipphardt, J. Helmcke, F. Riehle, and G. Zinner, Phys. Rev.
Lett. 76, 18-21 (1996).
[24] D. J. Jones, S. A. Diddams, J. K. Ranka, A. Stentz, R. S. Windeler, J. L. Hall,
and S. T. Cundiff, Science 288, 635-639 (2000); S. A. Diddams, D. J. Jones, J.
Ye, S. T. Cundiff, J. L. Hall, J. K. Ranka, R. S. Windeler, R. Holzwarth, T.
Udem, and T. W. Hänsch, Phys. Rev. Lett. 84, 5102-5105 (2000).
[25] J. Ye, T. H. Yoon, J. L. Hall, A. A. Madej, J. E. Bernard, K. J. Siemsen, L.
Marmet, J.-M. Chartier, and A. Chartier, Phys. Rev. Lett. 85, 3797 (2000).
[26] R. Holzwarth, T. Udem, T. W. Hänsch, J. C. Knight, W. J. Wadsworth, and P.
S. J. Russell, Phys. Rev. Lett. 85, 2264-2267 (2000).
[27] M. Zimmermann, C. Gohle, R. Holzwarth, T. Udem, and T. W. Hänsch, Opt.
Lett. 29, 310-312 (2004).
[28] R. Kienberger, E. Goulielmakis, M. Uiberacker, A. Baltuska, V. Yakovlev, F.
Bammer, A. Scrinzi, Th.Westerwalbesloh, U. Kleineberg, U. Heinzmann, M.
Drescher, and K. F., Nature 427, 817-821 (2004).
Chapter 1
INTRODUCTION
Optical Frequency Combs and their Applications
Jun Ye and Steven T. Cundiff

JILA, National Institute of Standards and Technology and the University of Colorado
Abstract: Recently there has been a remarkable synergy between the technology of
precision laser stabilization and mode-locked ultrafast lasers. This has resulted
in control of the frequency spectrum, which consists of a regular “comb” of
sharp lines, produced by mode-locked lasers. Such a controlled mode-locked
laser is a “femtosecond optical frequency comb generator.” For a sufficiently
broad comb, it is possible to determine the absolute frequencies of all of the
comb lines. This ability has revolutionized optical frequency metrology and
synthesis. It has also served as the basis for the recent demonstrations of
atomic clocks that utilize an optical frequency transition. In addition, it is
having an impact on time-domain applications, including phase-sensitive
extreme nonlinear optics and pulse manipulation and control. In this chapter,
we first review the frequency-domain description of a mode-locked laser and
the connection between the pulse phase and the frequency spectrum to provide
a basis for understanding how the absolute frequencies can be determined and
controlled. Using this understanding, applications in optical frequency
metrology, optical atomic clocks, and precision spectroscopy are discussed.
Next, we discuss applications of the carrier-envelope phase coherence in time-
domain experiments. This chapter serves as a broad introduction and summary
for all subsequent chapters that present detailed discussions of specific topics.
Key words: frequency comb, carrier-envelope phase, spectroscopy, optical frequency

metrology, optical atomic clock, quantum coherence
Mode-locked lasers generate “ultrashort” optical pulses by establishing a

fixed phase relationship across a broad spectrum of frequencies. Progress in
the technology of mode-locked lasers has resulted in the generation of
optical pulses that are only 5 femtoseconds (fs) in duration [1], which
corresponds to less than 2 cycles of the laser light. Although “mode locking”
1. INTRODUCTION 13
is a frequency-domain concept, mode-locked lasers and their applications are

typically discussed in the time domain. Recently, a paradigm shift in the
field of ultrafast optics has been brought about by switching to a frequency-
domain treatment of the lasers and the pulse trains that they generate.
Understanding mode-locked lasers in the frequency domain has allowed the
extensive tools of frequency-domain laser stabilization to be employed with
dramatic results.
The central concept of these advances is that the pulse train generated by
a mode-locked laser has a frequency spectrum that consists of a discrete,
regularly spaced series of sharp lines, known as an optical frequency comb.
As described below, if the comb spectrum is sufficiently broad, it is possible
to directly measure the two radio frequencies (rf) that describe the comb.
This fact has had an immediate impact in the field of optical frequency
metrology/synthesis [2-5] and has enabled the recent demonstration of
optical atomic clocks [6]. Because the comb spectrum can be related to
phase evolution in the pulse train [2, 7], these results also promise important
advances in ultrafast science, specifically in extreme nonlinear optics [8] and
coherent control. In addition, the union of time- and frequency-domain
techniques has yielded remarkable results in pulse synthesis [9].
The idea that a regularly spaced train of pulses corresponds to a comb in
the frequency domain and can thereby excite narrow resonances was realized
more than 20 years ago [10]. Teets et al. [11] used a train of pulses generated
externally to the laser. However, it was quickly realized that mode-locked
lasers were superior as demonstrated in a measurement of the sodium
hyperfine splitting using a picosecond laser [12]. Some of the concepts being
developed today were described in these early papers, but the technology
was insufficient to demonstrate them at the time. Advances in mode-locked
laser technology, specifically the Kerr-lens–mode-locked Ti:sapphire laser,
renewed interest in this area [13-15]. The observation of supercontinuum
generation from nanojoule pulses in microstructure fiber [16] led to the
recent, sudden explosion in activity.
In this chapter, we first discuss the time- and frequency-domain pictures
of mode-locked lasers and the intrinsic connections between the evolution of
the optical phase of the pulses and the frequency shift in the comb spectrum.
This discussion provides the background for understanding how the absolute
frequencies of the comb lines are determined and how the frequency-
domain-based control techniques are used to exert time-domain effects. The
ease and precision with which different parts of the electromagnetic
spectrum can be connected has led to wide-ranging applications in precision
optical frequency metrology including rf-based absolute optical frequency
measurement, optical frequency synthesis and distribution, and rf-clock
signal generation from a precisely stabilized optical clock oscillator. These
14 Chapter 1
new capabilities have brought novel approaches to precision atomic and

molecular spectroscopy and sensitive measurement instrumentation. For
time-domain applications, precise control of comb frequency and phase has
resulted in tight synchronization and phase lock among independent mode-
locked laser sources, connecting various parts of optical spectra and
synthesizing optical waveforms. The precise control of the carrier-envelope
phase has led to new experimental capabilities in ultrafast science, including
extreme nonlinear optical phenomena and coherent control. For example,
interference among multiple photon-order excitation pathways can now be
manipulated and controlled via a single mode-locked laser. The intrinsic
ultrafast time scale associated with the sub-optical-cycle phase control has
enabled ultrafast science to advance into the attosecond domain with a
remarkable level of experimental command and precision.
1. TIME- AND FREQUENCY-DOMAIN PICTURES

OF A MODE-LOCKED LASER
Understanding the connection between the time-domain and frequency-

domain descriptions of a mode-locked laser and the pulse train that it emits
is crucial. In this section, we first briefly introduce mode-locked lasers to
provide the necessary background. Then we discuss how, given a spectrum
that spans an octave, to determine the frequency spectrum of the pulse train
emitted by a mode-locked laser and how the absolute frequencies of the
comb spectrum can be determined. Finally, we present a prototype
femtosecond comb generator along with some relevant characterization
techniques.
A key concept in this discussion will be the carrier-envelope phase. This
phase is based on the decomposition of an ultrashort pulse into an envelope
function, Ê (t ) , that is superimposed on a continuous carrier wave with
frequency ωc, so that the electric field of the pulse is written E (t ) = Eˆ (t )e iω ct .
The carrier-envelope phase, φce, is the phase shift between the peak of the
envelope and the closest peak of the carrier wave. In any dispersive material,
the difference between group and phase velocities will cause φce to evolve as
the pulse propagates.
1.1 Introduction to mode-locked lasers
Mode-locked lasers generate short optical pulses by establishing a fixed-

phase relationship between all of the lasing longitudinal modes (for a
textbook level discussion, see [17]). Mode locking requires a mechanism
1. INTRODUCTION 15
that results in a higher net gain for short pulses as compared to continuous
wave (cw) operation. This mechanism can be either an active element or
implemented passively with saturable absorption (real or effective). Passive
mode locking yields the shortest pulses because, up to a limit, the self-
adjusting mechanism becomes more effective as the pulse shortens [18].
Real saturable absorption occurs in a material with a finite number of
absorbers, for example, a dye or semiconductor. The shortness of the pulses
is limited by the finite lifetime of the excited state. Effective saturable
absorption typically uses the nonlinear index of refraction of some material
together with spatial effects or interference to produce higher net gain for
shorter pulses. The ultimate limit on minimum pulse duration in such a
mode-locked laser is due to the interplay between the mode-locking
mechanism, group-velocity dispersion (GVD), and net gain bandwidth [18].
Chapter 2 discusses the development of ultrashort lasers, particularly
focusing on how to achieve an octave-spanning spectrum and pulse
dynamics that are relevant to the stability and control of the comb. Further
aspects of the comb dynamics and stability are presented in Chapters 5, 7, 9,
10, and 12.
Currently, the generation of ultrashort optical pulses is dominated by the
Kerr-lens–mode-locked Ti:sapphire (KLM Ti:sapphire) laser because of its
excellent performance and relative simplicity. Kerr-lens mode locking is
based on a combination of self-focusing in the Ti:sapphire crystal and an
aperture that selects the spatial mode corresponding to the presence of self-
focusing. The Ti:sapphire crystal is pumped by green light from either an
Ar+-ion laser or a diode-pumped–solid-state (DPSS) laser, which provides
far superior performance in terms of laser stability and noise. The
Ti:sapphire crystal provides gain and serves as the nonlinear material for
mode locking. Prisms or dispersion-compensating mirrors compensate the
GVD in the gain crystal [19]. Chapter 3 describes in detail a high repetition-
rate ring oscillator system. Since the discovery of KLM [20], the pulse width
obtained directly from mode-locked lasers has been shortened by
approximately an order of magnitude by first optimizing the intracavity
dispersion [21] and then using dispersion-compensating mirrors [1], yielding
pulses that are less than 6 fs in duration, i.e., less than two optical cycles.
Recently, output spectra that span an octave (factor of two in frequency)
have been obtained directly from a mode-locked laser [22], which is an
important accomplishment for phase stabilization.
1.2 Frequency spectrum of a mode-locked laser
To understand how frequency-domain techniques can be used to control

mode-locked lasers, we must first connect the time- and frequency-domain
16 Chapter 1
descriptions [23]. To start, we ignore the carrier-envelope phase and assume

identical pulses, i.e., φce is a constant. If we just consider a single pulse, it
will have a power spectrum that is the Fourier transform of its envelope
function and is centered at the optical frequency of its carrier. Generally, for
any pulse shape, the frequency width of the spectrum will be inversely
proportional to the temporal width of the envelope. For a train of identical
pulses, separated by a fixed interval, the spectrum can easily be obtained by
a Fourier series expansion, yielding a comb of regularly spaced frequencies,
where the comb spacing is inversely proportional to the time between pulses,
i.e., it is the repetition rate (fr) of the laser that is producing the pulses. The
Fourier relationship between time and frequency resolution guarantees that
any spectrometer with sufficient spectral resolution to distinguish the
individual comb lines cannot have enough temporal resolution to separate
successive pulses. Therefore, the successive pulses interfere with each other
inside the spectrometer and the comb spectrum occurs because there are
certain discrete frequencies at which the interference is constructive. Using
the result from Fourier analysis that a shift in time corresponds to a linear
phase change with frequency, we can readily see that the constructive
interference occurs at n fr, where n is an integer.
When φce is evolving with time, such that from pulse to pulse (at a time
separation of T = 1/ fr) there is a phase increment of ∆φce, then in the spectral
domain, a rigid shift will occur for the frequencies at which the pulses add
constructively. This shift is easily determined to be (1/2π) ∆φce /T. Thus the
optical frequencies, νn, of the comb lines are νn = nfr + f0, where n is a large
integer of the order of 106 that indexes the comb line, and f0 is the comb
offset due to the pulse-to-pulse phase shift. The comb offset is connected to
the pulse-to-pulse phase shift by f 0 = (1 2π ) f r ∆φce . The relationship
between time- and frequency-domain pictures is summarized in Figure 1-1.
The pulse-to-pulse change in the phase for the train of pulses emitted by a
mode-locked laser occurs because the phase and group velocities inside the
cavity are different. The pulse-to-pulse change in the phase for the train of
pulses emitted by a mode-locked laser can be expressed in terms of the
average phase (vp) and group (vg) velocities inside the cavity.
( )
Specifically, ∆φ ce = 1 v g − 1 v p l cω c , where lc is the round-trip length of
the laser cavity and ωc is the “carrier” frequency.
1. INTRODUCTION 17

aTime Domain ∆φFH ∆φFH
Et τ
T = 1/fr

Iν

bFrequency Domain
fr
f0 1/τ
0 νn = nfr+f0 ν
Figure 1-1. Summary of the time-frequency correspondence for a pulse train with evolving
carrier-envelope phase.
1.3 Determining absolute optical frequencies with

octave-spanning spectra
Armed with the understanding of the frequency spectrum of a mode-

locked laser, we now turn to the question of measuring the absolute
frequencies of comb lines. For a frequency measurement to be absolute, it
must be referenced to the hyperfine transition of 133Cs that defines the
second. From the relations listed above, we see that determining the absolute
optical frequencies of the femtosecond comb requires two rf measurements,
that of fr and f0. Measurement of fr is straightforward: we simply detect the
pulse train's repetition rate (from tens of megahertz to several gigahertz) with
a fast photodiode. On the other hand, measurement of f0 is more involved as
the pulse-to-pulse–carrier-envelope phase shift requires interferometric
measurement, whether it is carried out in the time domain or in the
frequency domain. When the optical spectrum spans an octave in frequency,
i.e., the highest frequencies are a factor of two larger than the lowest
frequencies, measurement of f0 is greatly simplified. If we use a second
harmonic crystal to frequency double a comb line, with index n, from the
low-frequency portion of the spectrum, it will have approximately the same
18 Chapter 1
frequency as the comb line on the high-frequency side of the spectrum with
index 2n. Measuring the heterodyne beat between these two families of
optical comb lines yields a difference,
2ν n − ν 2n = 2(nf r + f 0 ) − (2nf r + f 0 ) = f 0 , which is just the offset frequency.
Thus an octave-spanning spectrum enables a direct measurement of f0.
However, an octave-spanning spectrum is not required; it just represents the
simplest approach. We designate this scheme, shown in Figure 1-2(a), as
“self-referencing” since it uses only the output of the mode-locked laser.
(a) f0
I(ν) fr
ν
0
νn=nfr - f0 x2 2νn=2nfr -2 f0 ν2n=2nfr - f0
f0
(b) νstandard 2νstandard
f0
fr
ν
0
fbeat1 fbeat2
f0 = fbeat2 – 2fbeat1
Figure 1-2. Two equivalent schemes for the measurement of f0 using an octave-spanning
optical frequency comb. In the self-referencing approach, shown in (a), frequency doubling
and comparison are accomplished with the comb itself. In the second approach shown in (b),
the fundamental frequency (νstandard) and its second harmonic of a cw optical standard are
used to determine f0. These two basic schemes are employed for absolute optical frequency
measurement and implementation of optical atomic clocks.
Self-referencing is not the only means of determining the absolute optical

frequencies given an octave-spanning spectrum. For example, the absolute
optical frequency of a cw laser can be determined if its frequency lies close
to comb line n in the low-frequency portion of the femtosecond comb
1. INTRODUCTION 19
spectrum. Then the second harmonic of the cw laser will be positioned close
to the comb line 2n. Measurement of the heterodyne beat between the cw-
laser frequency, νs, and the comb line n gives f beat1 = ν s − (nf r + f 0 ) and
between the second harmonic of the cw laser and comb line 2n gives
f beat 2 = 2ν s − (2nf r + f 0 ) . Mixing the beats with appropriate weighting
factors gives f beat 2 − 2 f beat1 = 2ν s − (2nf r + f 0 ) − (2ν s − 2(nf r + f 0 )) = f 0 ,
which represents the second detection scheme shown in Figure 1-2(b).
Another interesting fact is that by mixing the two beat signals, one
establishes a direct link between the optical and rf frequencies (νs and fr) as
in f beat 2 − f beat1 = 2ν s − (2nf r + f 0 ) − ν s + (nf r + f 0 ) = ν s − nf r .
1.4 Femtosecond optical-frequency comb generator
A frequency comb generator produces a spectrum that consists of a series

of equally spaced sharp lines with known frequencies. Microwave comb
generators are commercially available. Optical-frequency comb generators
have been constructed by injecting a single-frequency cw laser into a high-
quality optical cavity that contains an electro-optic modulator [24].
Typically, comb bandwidths of a few terahertz have been achieved using this
method [25]. If the absolute optical frequency of the cw laser is known, then
the resulting comb can be used to directly measure nearby frequencies [26].
Alternatively, without needing to know the absolute frequency of the cw
laser, a comb generator can be used to span a frequency gap in a more
complex frequency-measurement chain [27, 28].
Based on our discussion above, it is clear that a mode-locked laser also
generates an optical frequency comb. However, there is no equivalent to the
cw laser, which can provide a priori knowledge about the absolute
frequencies of the comb. Thus, the first applications of mode-locked lasers
were based on the use of precisely determined spanning intervals in more
complex chains [14] or between a known and unknown frequency [15].
The discussion in Section 1.3 shows that knowledge of the absolute
frequencies of the comb generated by a mode-locked laser is easily obtained
if it generates a spectrum that spans an octave. A Fourier-transform-limited
pulse with a full-width–half-maximum (FWHM) bandwidth of an octave
centered at 800 nm would only be a single optical cycle in duration. Such
short pulses have not been achieved; the shortest pulses generated by a
mode-locked oscillator (i.e., not including external amplification and
broadening) are just under two cycles in duration [1, 22]. Fortunately,
neither a transform-limited pulse nor a FWHM of an octave is needed to
implement the methods for obtaining the absolute frequencies. The pulse
width is unimportant as the methods are purely frequency-domain
20 Chapter 1
techniques. Experimentally, it has been found that even if the power at the
octave-spanning points is 40 dB below the peak, it is still possible to observe
strong ν-to-2ν heterodyne beats.
Since Ti:sapphire, which has the broadest gain bandwidth of all known
laser media, does not support an octave-spanning spectrum, additional
spectral content must be generated. This is accomplished by using self-phase
modulation, which is based on a temporal variation in the index of refraction
because of a combination of a short optical pulse and an intensity-dependent
index of refraction [29]. The broadening can be achieved outside the laser
cavity by using optical fiber or internally by creating coincident secondary
time and space foci [22]. The latter technique requires carefully designed
mirrors [30] and is challenging to implement; therefore the former is
significantly more common. (For more detailed discussions, see Chapter 2.)
Recent results have shown that additional spectral bandwidth can be
obtained by minor changes in the cavity configuration of a high-repetition-
rate laser, although this technique has not yet yielded sufficient intensity at
the octave points for observation of ν-to-2ν beats. (See Reference [31] and
related discussions in Chapter 3.)
The amount of spectral broadening that can be obtained in ordinary
optical fiber is limited, primarily because temporal spreading of the pulse,
due to GVD in the fiber, reduces the peak intensity. Using a low-repetition-
rate laser (to raise the pulse energy) an octave-spanning spectrum has been
obtained with ordinary fiber [7]. The discovery [16] that microstructure fiber
can have zero group-velocity dispersion within the emission spectrum of a
Ti:sapphire laser eliminated this difficulty and led to rapid progress in the
field of femtosecond-optical-frequency combs by allowing broadband-
continuum generation with only nanojoule-pulse energies.
Microstructure fiber uses air holes surrounding a fused silica core to
obtain the index-of-refraction contrast needed for waveguiding. This method
results in a much larger index contrast than can be obtained using doping.
The large index contrast has two consequences: (1) the ability to generate a
zero in the GVD at visible or near-infrared wavelengths and (2) the
possibility of using a much smaller core size. The first consequence means
that the pulse does not spread temporally and hence maintains its high peak
power. In addition, it results in phase matching between the generated
spectral components. The second outcome greatly increases the light
intensity in the core, thereby enhancing nonlinear effects. Chapter 4 provides
in-depth discussions relating to the physics of bandwidth generation and the
underlying noise process during pulse propagation through these relatively
novel microstructure fibers.
We have now introduced all of the concepts and components needed to
construct a femtosecond-optical-frequency comb generator that produces
1. INTRODUCTION 21
known absolute frequencies. There are several possible implementations,

one of which we present in Figure 1-3.
output
microstructure
fiber
KLM Ti:sapphire Laser ν−2ν interferometer
AOM
Pump
Laser Mod.
· n/16 SHX
fr
Micro-
wave servo counter
Clock
f0 photo-
counter
servo diode
Figure 1-3. Schematic of a femtosecond comb generator. AOM: acousto-optic modulator;

SHX: second-harmonic-generation crystal.
The heart of the comb generator is a KLM Ti:sapphire laser. A small

portion of the output is detected using a high-speed photodiode to measure
the repetition rate. Greater precision is obtained by measuring a large
harmonic of the repetition rate rather than the fundamental. Ultimately, in
implementing an optical clockwork with a frequency comb (cf. Chapter 9),
the relevant information regarding the repetition frequency is collected in the
optical domain with a gain factor of nearly a million for enhanced
measurement precision. A servo loop controls the repetition rate of the laser
by comparing this signal to a microwave clock or an optical-frequency
standard.
The output of the KLM Ti:sapphire laser is launched into a length of
microstructure fiber. Using the minimum possible amount of spectral
broadening in the microstructure fiber works best. For this reason, all
metrology experiments start with KLM Ti:sapphire lasers that produce pulse
widths of 30 fs or less. This strategy generally results in an octave-spanning
spectrum for modest pulse energies and short lengths of microstructure fiber.
The output of the microstructure fiber is split into two parts. One part serves
as the useful output of the comb generator, while the other part is used in an
ν-to-2ν interferometer to measure f0.
The input to the ν-to-2ν interferometer is divided into long and short
wavelength portions by a dichroic beam splitter. The long wavelength
portion is frequency doubled by a second-harmonic crystal. The beams in the
two arms of the interferometer, which now have the same spectral
components, are recombined and detected with a photodiode. The lengths of
22 Chapter 1
the two arms must be matched in order to achieve temporal overlap,

including compensation for GVD in the microstructure fiber.
The detected signal from the ν-to-2ν interferometer contains a forest of
signals including multiples of fr and ν-to-2ν beat-note signals spaced above
and below each repetition-rate signal by f0. One of the beat notes must be
chosen and isolated for counting and stabilizing the laser. If the signal-to-
noise ratio is sufficiently large, an appropriate rf-bandpass filter is usually
sufficient to process the signal without cycle slips, otherwise regeneration
with a tracking oscillator can be employed.
The final step is to close the loop to stabilize f0. This requires a “knob” on
the laser that can be used to adjust f0, which is determined by the difference
between the intracavity group and phase velocities. One common method for
adjusting f0 is to swivel the end mirror in the arm of the laser cavity that
contains the prism sequence [32]. Since the spectrum is spatially dispersed
on this mirror, a small swivel produces a linear phase delay with frequency,
which is equivalent to a group delay. An alternative method of controlling f0
is to modulate the pump power [7, 33]. Empirically, this clearly causes a
change in f0 [34]. However, the details are somewhat unclear, as there are
likely contributions from the nonlinear phase, spectral shifts, and intensity
dependence in the group velocity [35]. Each method has advantages and
disadvantages with respect to servo speed and impact on amplitude noise.
1.5 Time- and frequency-domain characterizations of f0
Carrier-envelope phase coherence is critical for all of the time-domain

processes discussed in the remainder of this section. Physically, the carrier-
envelope phase coherence simply reflects how well we can tell what the
carrier-envelope phase is of a given pulse in the train if we know the phase
of an earlier pulse. Knowing the carrier-envelope phase of a given pulse is
important, however, for coherent pulse synthesis because we need to
maintain carrier-envelope coherence between the two lasers. For
experiments sensitive to φce, it is difficult to determine how φce affects the
outcome if φce is varying wildly during the measurement.
The connection between time and frequency can be illustrated by
measuring the cross-correlation between successive pulses emitted by the
comb generator. A cross-correlator is based on the interferometric
autocorrelator commonly used to measure ultrashort pulses [17], with the
modification that one arm of the interferometer is longer than the other by a
multiple of the cavity round-trip time [34]. The asymmetric arm lengths
allow appropriate delays to be introduced so that the pulse-to-pulse phase
shift ∆φce can be measured. An autocorrelation curve is always symmetric.
However, in an interferometric cross-correlation only the envelope is
1. INTRODUCTION 23
symmetric, and the shift of the fringes from the peak of the envelope is due
to the pulse-to-pulse phase shift ∆φce. By stabilizing f0 at different frequency
offsets, differing values of ∆φce are obtained, as discussed in Section 1.2.
Another technique, known as spectral interferometry, can also be employed
to determine the values of ∆φce and the related f0 if the interfering pulses are
separated by at least one cavity round-trip time. This technique also yields
additional spectral-phase information. For single-shot experiments, a
variation of the spectral-interferometry technique that involves interference
between the fundamental and its second harmonic spectra, which overlap if
the fundamental optical bandwidth exceeds one octave, becomes the most
effective approach to measuring ∆φce [36].
Although cross-correlation measurements can demonstrate some degree
of phase coherence, they are actually quite insensitive to phase fluctuation
because they measure the change between pulses that are separated by only a
few round-trip times because of the practical size limitation of the
interferometer. Instead, frequency-domain measurements of the frequency-
noise spectrum of f0 are much more sensitive since they can monitor the
frequency/phase evolution of f0 over much longer time intervals. Typically, f0
is determined from the self-referencing technique discussed earlier, from
which one can make a close examination of the frequency- and phase-noise
power spectral density associated with f0.
Given a measurement of the frequency-noise power spectral density of
f0, sνf 0 , the accumulated root-mean-square fluctuations of φce are given by
∞
sν ( f )df for an observation time τobs [37]. Once a
1 f0
∆φ ce
rms
= 2 ∫
τ obs f2
1 2πτ obs
femtosecond laser has been properly phase stabilized, experimental results

have indicated that the linewidth of f0 is actually limited by the measurement
time of a few thousand seconds [38]. The standard definition for coherence
time is the τobs at which 1 radian of phase fluctuations have accumulated.
Performing such an integration reveals that the coherence time can reach
beyond 1000 s.
We would like to emphasize that carrier-envelope phase coherence is not
the same as optical coherence. A process that shifts the position the pulse
without changing the φce of the pulse destroys the optical coherence but does
not affect the carrier-envelope phase coherence.
24 Chapter 1
2. PRECISION OPTICAL FREQUENCY

METROLOGY USING FEMTOSECOND-
OPTICAL-FREQUENCY COMBS
It is useful to review the historical development of optical frequency

metrology and its perspectives in precision spectroscopy, clock-signal
generation, and frequency synthesis. The outstanding spectral properties of
optical frequency standards offer unprecedented resolution and precision and
potentially the highest accuracy for physical measurements [39]. Even at the
cost of extraordinary complexity and remarkable resources, researchers have
explored and constructed optical-frequency synthesis chains that span the
vast frequency gap between the optical and microwave spectral regions (see
for example References [40, 41]). Previous efforts concentrated on
measurement of a few discrete optical lines that are chosen to be optical
frequency standards. Even limited frequency-measurement capabilities
brought a number of significant advances in fundamental physics including:
(1) the determination of the speed of light [40, 42] and the linkage between
the fundamental physical units of length and time; (2) the refinement of the
Rydberg constant and the Lamb shift [43] predicted by quantum
electrodynamics; (3) the competitive measurement of various fundamental
constants such as the fine structure constant [14, 44] and the ratio of proton-
to-electron mass [45]; and (4) a test of relativity theory [46]. However, until
recently it was deemed an overwhelming challenge to synthesize arbitrary,
absolute optical frequencies. The significance of the wide-bandwidth optical
comb lies in the fact that it has substantially reduced this challenge. A
detailed review of the traditional harmonic-based frequency chain is
provided in Chapter 8, while the new epoch of absolute optical frequency
measurement using femtosecond comb technology is reviewed in Chapter 7.
2.1 Measurement of absolute optical frequency
The dramatic simplification of a complex optical frequency chain to a

single mode-locked laser has facilitated optical frequency measurement. An
important aspect of this new technology is its high degree of reliability and
precision together with a lack of systematic errors. For example, recent tests
have shown that the repetition rate of a mode-locked laser equals the mode
spacing to within the measurement uncertainty of 10-16 [13]. The uniformity
of the comb’s mode spacing has also been verified to a level below 10-17
[13], even after spectral broadening in fiber. Comparison between two
separate femtosecond comb systems, both linked to a common reference
source (microwave or optical), allows one to examine the intrinsic accuracy
1. INTRODUCTION 25
of a femtosecond comb-based–frequency-measurement system at a level of a

few parts in 1016, with no measurable systematic effects [47]. Chapter 9
provides an account of the current state-of-the-art performance and
characterization of femtosecond comb systems used for optical frequency
measurement and synthesis.
Another confirmation of femtosecond comb accuracy was provided by
comparison of the measured frequency of a He-Ne laser frequency stabilized
on a molecular iodine transition. Using an elaborate scheme of transfer
standards and crosschecks, we were able to reliably compare the absolute
frequency of the laser measured with a femtosecond comb in JILA and that
measured by a traditional harmonic-optical-frequency synthesis chain
located at the National Research Council Canada [48]. The difference
between the two measurements is below 1.6 × 10-12. While the accuracy of
this test has fewer digits, it is comforting to find such agreement in a direct
comparison of the two synthesis methods at two national laboratories.
As mentioned in Sections 1.3 and 1.4, one simple use of a femtosecond
comb is to stabilize one degree of freedom, namely fr, and use the comb to
span a frequency difference between two cw optical frequencies to be
measured. A typical approach is to stabilize fr directly to a known
microwave standard. Experimental observation has clearly confirmed that
the actual limitation in precision with femtosecond comb-based
measurements is the quality of the rf-reference sources [49]. For example,
commercial cesium clocks have a stability ~5 × 10-12/τ1/2 and can be
calibrated to an accuracy ~1 x 10-14. For absolute measurement of a single
optical frequency, one also needs to measure f0 in addition to fr , and then
count the heterodyne beat between one of the comb lines and the cw optical
frequency. The comb mode order can be determined by varying fr, for
example. Telle et al. have also introduced a clever method of using the comb
from a mode-locked laser to measure the ratio of two optical frequencies
[50]. By a judicious choice of mixing frequencies, they show it is possible to
transfer the frequency stability of one optical source to a different optical
frequency without additional noise from the comb. The advantage of this
frequency-transfer method is that the frequency noise introduced by the
mode-locked laser is taken out of the measurement process. Thus it is
possible to compare two narrow optical atomic lines without requiring the
optical frequency comb to be highly stabilized. In a demonstration
experiment, Telle et al. obtained narrow optical beat signals in the terahertz
range that were actually the transferred beat notes between three optical
frequency standards [50]. This frequency-transfer method and the use of a
free-running frequency comb have limitations but can be useful tools for
measuring frequency ratios from rf to the visible.
26 Chapter 1
There has been an explosion of absolute frequency measurements using

femtosecond comb methods since 2000. Not surprisingly, the most accurate
results come from optical standards that are based on transitions with
extraordinary quality factors. One of the most precise measurements is the
determination of the Hydrogen 1S–2S transition frequency at the Max-
Planck Institute for Quantum Optics (MPQ) in Garching [51], as discussed
in Chapter 7. The MPQ team took advantage of the high stability of a
transportable cesium fountain clock as the reference for their femtosecond
comb. Their result was quoted at the 1.8 × 10-14 accuracy level. Some
national laboratories have both trapped-ion teams and primary frequency-
standard teams that form a powerful collaboration. After investing many
years of effort on the Hg+ trap [52], a team at the National Institute of
Standards and Technology (NIST) in Boulder, Colorado, can now determine
the frequency of an electric quadrupole transition of the Hg+ ion to <1 × 10-14
[5]. Yb+ was measured with similar accuracy at the Physikalish-Technische
Bundesanstalt (PTB) in Braunschweig [53]. Excellent results are also
obtained for frequency determination of a spin-forbidden intercombination
line in cold calcium atoms at both NIST and PTB [4, 5]. Indeed, with this
new increase in measurement precision, testing fundamental physical
postulates or determining constants at the next decimal place are again
attracting great interest [54].
Another direction to explore is the “everyman’s frequency measurement
system” where one can consider tradeoffs such as a ~tenfold lower accuracy
for a ~103 scale reduction for the apparatus. Femtosecond combs based on
compact and energy efficient mode-locked lasers (such as fiber lasers) and
super-continuum generation fibers can offer portable versions of a
frequency-measurement device. Cell-based optical frequency standards, such
as a solid-state laser stabilized on sub-Doppler transitions of molecular
iodine, already offer a competitive stability near or below 1 × 10 -14 when
averaged over 10 to 1000 s [55, 56]. The long-term reproducibility of such a
molecular standard over a period of more than three years was limited to
about 3 × 10-13. Better stability and reproducibility are expected from an
improved iodine spectrometer [57, 58].
Cell-based optical standards play an essential role in length metrology
[59]. For different national laboratories to establish a common basis for a
length standard, an assortment of wavelength-reference lasers realized
separately at each laboratory need to be regularly intercompared. The
traditional practice has been to hold regular conventions, organized by the
Bureau International des Poids et Mesures (BIPM), where stabilized lasers
from different national laboratories were gathered at the same physical
location and directly compared. A more economic and precise approach to
carry out this task can now be accomplished by local calibrations of the
1. INTRODUCTION 27
length-standard lasers with GPS systems to an accuracy ~1 × 10-14. Another

rather pleasant outcome of the absolute frequency-calibration process is the
associated capability of unambiguous testing of the manufacturing process of
reference cells and their long-term variations [56].
2.2 Optical atomic clocks
With the advent of wide-bandwidth–optical-comb technology, it is now

possible to transfer the stability of the highest-quality optical frequency
standards across vast frequency gaps to other optical spectral regions.
Furthermore, the comb technology has also established a straightforward
possibility to transfer the optical stability down to the rf domain and vice
versa. One can now realize a network of microwave and optical frequencies
at a level of stability and reproducibility that surpasses the properties of
basically all commercially available frequency sources ─ and at a reasonable
cost. Easy access to the resolution and stability offered by optical standards
will greatly facilitate the application of frequency metrology both to
precision experiments for fundamental physics and to practical devices.
As elaborated in Chapter 9, recent experimental demonstrations support
the concept that, in the future, the most stable and accurate frequency
standards will be based on optical transitions. The advantage of an optical
frequency standard over a traditional microwave standard is apparent if we
examine the frequency stability of an atomic clock. Resonance natural
widths, ∆ν, in the few kilohertz to the subhertz domain are available by
selection of an atomic transition with a natural decay time, τ0, in the 100 µs
to 1 s domain. In principle, one could obtain ~1/(2τ0) interactions per second
with approximately twofold broadening of the resonance linewidth by the
interrogation process. So, if we collect all the available information-bearing
photons for a single measurement, a signal-to-noise ratio (SNR) of ~ N
should be available, where N is the number of participating particles.
Normalizing to a standard 1 s measurement time produces SNR ~ N ×
1 (2τ 0 ) . An optimum-frequency control system could find the center of
the resonance with a precision of ~1/SNR in 1 s. Taking the resonance
linewidth into account leads to a frequency uncertainty δν (at 1 s) ~∆ν/SNR
= (2/Nτ0)1/2. Assuming the Ramsey separated-field method and in the case
where the interrogation time, TR, is shorter than the actual lifetime of the
transition under study, the fractional frequency (in-)stability is
δν 1
σ y (τ ) = = . In this expression, ω0 (=2πν0) is the clock
ν 0 ω 0 NTRτ
transition frequency and τ (τ > TR) is the total averaging time. Clearly, higher
28 Chapter 1
stability is most easily attained if we can increase ω0 by changing ω0 from a

microwave to an optical frequency.
To create an optical atomic clock, one uses an optical frequency standard
(νcw) to stabilize fr of a femtosecond comb, thus transferring the optical-
phase information to the microwave domain. Since the comb system has two
degrees of freedom, fr and f0, one has to ensure that a direct and
unambiguous phase relation between νcw and fr is established. This implies
that either f0 is strictly known, for example, via tight phase stabilization to a
known reference, or the variable f0 is somehow eliminated from the optical
heterodyne beat that connects a comb component to that of νcw, such as in
the example shown at the end of Section 1.3.
Figure 1-2 and the related discussions presented in Section 1.3 provide
basic guidelines for implementation of these schemes. The first approach, as
shown in Figure 1-2 (a), uses the self-referencing technique to recover f0,
which is then stabilized with respect to either fr or an auxiliary stable rf
source. Stabilization of f0 to a few millihertz is more than adequate, as it
yields a fractional frequency noise of < 10-17 for an optical carrier. A
heterodyne beat between one of the comb components and the cw laser (νcw),
which acts as the optical frequency standard, reveals fluctuations in fr. After
appropriate processing, this error signal is used to stabilize the phase of fr
coherently to νcw, thereby producing a clock signal output in the rf domain
derived from νcw. The second approach, as shown in Figure 1-2(b), uses two
beat signals between a stabilized cw laser (νcw) and its second harmonic
(2νcw) against two respective comb components in the corresponding spectral
domains. One immediately sees that we are taking the same advantage of the
octave bandwidth of the femtosecond comb. Through appropriate electronic
mixing of the two beats, one can derive two servo-control error signals
associated with f0 and fr, respectively, or one can simply eliminate the
variable f0 from the equation.
One of the JILA optical frequency standards is a diode-pumped solid
state Nd:YAG laser (νcw) with its second harmonic (2νcw) locked on a
hyperfine component of an iodine transition (R(56) 32-0, a10) near 532 nm;
this system offers an (in-)stability of 4 × 10-14 at 1 s [27, 55]. This optical
frequency standard presents an ideal case for the scheme shown in Figure 1-
2(b), since both νcw and 2νcw are located within the spectrum of a
Ti:sapphire-based femtosecond comb.
With the tracking of the comb system exceeding the stability of the
current optical frequency standards, we expect the stability of the derived
clock signal to be basically that of the optical standard. To characterize the
system, the optical clock signal is compared against other well-established
microwave/rf frequency standards such as cesium atomic clocks or hydrogen
masers. The comparison typically involves a heterodyne-beat experiment
1. INTRODUCTION 29
between the two signal sources with a frequency counter recording the
resultant beat-frequency fluctuations over a period of time. There has been
good progress in the process of extracting high-stability rf-output signals
from an optical clock at a level that approaches the optical stability. Further
work along this line is continuing. We also still face some technical
challenges in making an optical clock a reliable scientific device. Technical
developments necessary for an advanced optical atomic clock include: (1)
highly accurate, cold-atom optical frequency standards and portable, high-
stability optical frequency standards (see Chapters 8, 9); (2) development of
ultrastable optical local oscillators suitable for the most demanding
spectroscopic tasks (see Chapter 9); (3) stabilization and control of wide-
bandwidth optical combs, including exploration of novel generation and
detection techniques and approaches to reduce noise (see Chapters 5, 7, 9);
(4) reliable, stable, and compact ultrafast laser technology for practical
implementation of optical clocks (see Chapters 2, 3, 4, 6); (5) development
and use of femtosecond combs for intercomparison of optical frequency
standards and the cesium primary standard (see Chapters 7, 8, 9); and (6)
development of frequency- and time-transfer methods over extended fiber
optic links that can support the next generation of atomic frequency
standards (see Chapter 9).
2.3 Optical frequency synthesizer
As will be discussed in Section 4, a future goal of ultrafast technology is

to demonstrate arbitrary pulse synthesis in the time domain, including the
capability of phase-coherent stitching of distinct optical bandwidths. This
time-domain capability would complement the goal of constructing an
optical frequency synthesizer that allows access in the frequency domain to
any optical-spectral feature with a well-defined optical carrier wave. Such a
capability would greatly simplify precision-laser spectroscopy.
With the development of an optical comb, we have now established an
optical frequency grid with lines repeating every repetition frequency (100
MHz–1 GHz) over an octave optical bandwidth and with every line stable at
or below the 1-hertz level. This capability creates the basic infrastructure to
construct a highly stable frequency synthesizer for both rf and optical
spectral domains and, in principle, anything in between. An rf (optical)
reference signal can be used to derive a phase-coherent secondary signal in
any part of either the rf or the optical domains. Chapter 9 presents a more
detailed discussion about transfer noise and accuracy relevant to this
frequency distribution and synthesis process. In a traditional rf synthesizer,
the output is a single-frequency rf-“delta” function (of reasonable power)
that can be tuned to any desired frequency on demand. Realization of such a
30 Chapter 1
frequency synthesizer in the optical domain requires a wide-tunable cw laser

or an array of them that would collectively cover most of the visible
spectrum. The frequency of the cw lasers would be controlled by the
underlying optical frequency comb and therefore be directly related to the
absolute time/frequency standard in a phase-coherent fashion. Setting of the
optical frequency will be accomplished via computer control. Such a system
(with continuous, precise frequency tuning and arbitrary frequency setting
on demand) has indeed already been demonstrated [60].
3. ATOMIC AND MOLECULAR SPECTROSCOPY
The advent of precision femtosecond optical combs brings a new set of

tools for precision atomic and molecular spectroscopy and allows us to
explore spectroscopy in a broader sense. For example, ultrafast lasers are
now being used not only for time-resolved spectroscopy on fast dynamics,
but also for precision spectroscopy on structural information. Indeed,
coherent control of dynamics and precision measurement are merging into a
joint venture.
The ability to make absolute optical frequency measurements in both
visible and infrared (IR) spectral regions adds a new meaning to the term
“precision molecular spectroscopy.” The understanding of molecular
structure and dynamics often involves detailed spectral analysis over a broad
wavelength range. Such a task can now be accomplished accurately and
uniformly across all relevant spectral windows, allowing precise
investigations of minute changes in the molecular structure over a large
dynamic range. For example, absolute frequency measurement of vibration-
overtone transitions and other related resonances (such as hyperfine
splitting) can reveal precise information about the molecular potential-
energy surface and relevant perturbation effects. We have pursued such a
study with iodine molecules, as we discuss in Section 3.2.
3.1 Precise, simultaneous determination of global atomic

structure and transition dynamics
The investigation of two-photon transitions in laser-cooled 87Rb atoms

using a precisely stabilized femtosecond comb represents the first example
of direct femtosecond comb spectroscopy. Phase coherence among the
successive pulses interacting with a cold atomic sample is similar to the
approach of Ramsey interference for precision atomic spectroscopy.
However, the difference here is that the bandwidth associated with the
1. INTRODUCTION 31
femtosecond pulse is so broad that one is able to explore the structure of a

large number of atomic states simultaneously, while at the same time, one
could study coherent and incoherent population transfer in a multilevel
system. In short, one can simultaneously explore the global structure and
dynamics of an atomic system. The multipulse interference in the time
domain gives an interesting variation and generalization of the two-pulse-
based, temporal coherent control of the excited state wave packet.
From the frequency domain perspective, spectroscopic resolution and
precision will not be compromised by the use of ultrafast pulses, since they
are associated with a phase-stabilized, wide-bandwidth femtosecond comb.
Phase coherence among various transition pathways through different
intermediate states produces multipath quantum-interference effects on the
resonantly enhanced two-photon-transition probability in cold 87Rb atoms.
The two-photon-transition spectrum can be analyzed in terms of the pulse
repetition rate (fr) and the carrier-envelope-offset frequency (f0). Both can be
stabilized to high precision. With a set of measurements taken at a few
different combinations of fr and f0, one can essentially derive all relevant
atomic-energy-level positions in absolute terms.
Doppler-free–two-photon spectroscopy is usually carried out with two
equal-frequency cw laser beams propagating in opposite directions. The two-
photon-transition rate can be resonantly enhanced via the intermediate states
with two different laser frequencies [61] or accelerated atomic beams [62].
High-resolution–two-photon spectroscopy using pulsed picosecond light has
also been demonstrated [12], with the recent extension to cold atoms [63]. A
unique feature of the wide-bandwidth optical comb allows all relevant
intermediate states to resonantly participate in the two-photon excitation
process. This participation, in turn, permits phase coherence among different
comb components to induce a stronger transition rate through quantum
interference. The resonant interaction with the intermediate states also makes
it possible to explore population-transfer dynamics and the mechanical
consequences of light-atom interactions. Following the initial proposal [64]
and the subsequent theoretical investigations, we are exploring this novel,
high-resolution spectroscopy using a femtosecond laser.
Figure 1-4 shows the relevant 87Rb energy levels involved in the two-
photon transition from the ground state 5S1/2 to the excited state 5D3/2. The
dipole-allowed intermediate states, 5P3/2 and 5P1/2, are located ~2 and 17 nm
below the virtual level, respectively. Also shown is a regularly spaced comb
of optical frequencies around 800 nm. The experimental bandwidth of the
comb (emitted from a 10 fs, 100 MHz repetition-rate, mode-locked
Ti:sapphire laser) is ~50 nm. Adjustment of fr and f0 allows the comb
components to line up with the corresponding hyperfine states of 5P3/2 and
5P1/2 to resonantly enhance the two-photon transition. The frequency-domain
32 Chapter 1
analysis is complemented by the time-domain-multipulse Ramsey

interference picture, also illustrated in Figure 1-4, where the relevant
quantities for interaction are τ = 1/fr and ∆φce. Both the frequency-domain
and the time-domain analyses produce the same result on the two-photon-
transition spectra when one assumes a static distribution among the relevant
atomic states. However, to follow the time evolution of the system, it is
necessary to explore the interaction dynamics from one pulse to the next,
taking into account both the atomic coherence and the optical coherence.
The general Liouville equation for the density-matrix components of the
atomic states, along with phenomenological decay terms, are used to derive a
set of Bloch equations describing the evolution of all relevant levels
associated with the ground, excited, and intermediate states.
F” =
1
5D5/2 2
3 5D δSD
5D3/2 4
3
Φ-coherent 2
comb 1
0 De 6P
νPD = ca
yt
m fr + f0 o5
F’ = S
fr 3
2
5P3/2 1
0
5P1/2 5P
2 δSP
1
νSP =
f0 F= n fr + f0
…
2
5S1/2 1 5S
∆φce 2∆φce
E(t)
ωc
Cold Rb
t
τ = 1/fr ∆φce = 2π (f0/fr)
Figure 1-4. Top: Schematic of the relevant energy levels of the 87Rb atom and the frequency-
domain perspective of the atom-light interaction. Bottom: Time-domain picture showing a
sequence of mode-locked pulses, with the relevant interaction parameters in fr and ∆φce. The
inset at right shows the relevant “three-level” model used for construction of the Bloch
equations to solve for population-transfer dynamics. An example for the on-resonance,
stepwise transition is shown with relevant detunings at δSP and δSD for the pair of comb modes
that make the dominant contribution to the transition probability amplitude.
1. INTRODUCTION 33
Experimental two-photon spectra resonantly enhanced by the

intermediate states confirm the theoretical model where the effect of
population transfer becomes non-negligible as the number of interacting
pulses increases. Not surprisingly, the most dominant transition pathway
when a large number of pulses is involved is 5S1/2 (F=2) → 5P3/2 (F=3) →
5D5/2 (F=4), which represents a closed transition. If necessary, an appropriate
repumping scheme, sometimes involving the probing femtosecond laser
itself, can be employed to allow investigation of various transition pathways
at longer time scales. This incoherent population-redistribution process is
complemented by coherent-accumulation effects observed at shorter time
scales. The relatively long lifetime associated with the atomic D-state allows
phase-coherent accumulation of the transition amplitudes stimulated from
successive pulses, resulting in a significant enhancement of the desired D-
state population. Only a few comb components are primarily responsible for
the on-resonance, stepwise excitation since the linewidth (~6 MHz) of the
intermediate P3/2 and P1/2 states and their hyperfine sublevels is much smaller
than the comb-frequency spacing (fr = 100 MHz). Another interesting fact is
that mechanical effects of the light on the cold atomic sample are clearly
visible and can be controlled via various settings of fr and f0, which control
the detuning δSP of the corresponding comb modes from the intermediate
states. Although the laser spectrum spans about 50 nm, the resultant
spectroscopy resolution approaches the limit of the atomic natural linewidth
of 660 kHz associated with the D-state lifetime. This level of resolution is a
result of careful control of the comb parameters, the use of ultracold atoms,
and a good understanding of the light’s mechanical effects. The work on
simple two-photon-transition dynamics provides a solid link between the
time-domain picture of the carrier-envelope phase and the frequency-domain
picture of fr and f0. One practical consequence of these results is that we can
now directly control both degrees of freedom for the femtosecond comb by a
transition in cold atoms. In future experiments, the exquisite control in the
time-domain phase coherence will be complemented by spectral-domain
amplitude and phase manipulations, leading to even greater control and
spectroscopic investigative power.
3.2 I2 hyperfine interactions, optical frequency

standards, and clocks
With the development of an optical frequency synthesizer, namely a

tunable, cw Ti:sapphire laser guided by and referenced to an accurate
femtosecond comb [Figure 1-5(a)], we have performed high-resolution and
high-precision measurement of hyperfine interactions of the first excited
electronic state (B) of I2 over an extensive range of vibrational and rotational
34 Chapter 1
quantum numbers towards the dissociation limit [65]. Experimental data

demonstrate systematic variations in the hyperfine parameters that confirm
calculations based on ab initio molecular potential-energy curves and
electronic wave functions derived from a separated-atomic-basis set. We
have accurately determined the state-dependent quantitative changes of
hyperfine interactions caused by perturbations from other electronic states
and identified the respective perturbing states. Our work on I2 near the
dissociation limit is also motivated by the desire to improve cell-based,
portable optical frequency standards. Indeed, I2-stabilized lasers have already
demonstrated high stability (< 5 x 10-14 at 1 s averaging time) and have
served well for optical atomic clocks.
νstandard
(a)
fr f0
Transfer stability
~ 10-16
< 10 nW per comb
f
0
νcw Single mode,

> mW Total search time: 1 minute
(b) 2P
1/2 + 2P3/2
B
E - energy
2P + 2P3/2
1 3/2
u
X
R
R - internuclear distance
Figure 1-5. (a) Random access and precise stabilization of a cw laser (νcw) using the
frequency reference grid provided by the femtosecond comb stabilized by optical (νstandard)
and rf standards. (b) The ground state and the first excited state of I2 with their associated
dissociation limits. The transition linewidth narrows when the excited state approaches the
dissociation limit.
The hyperfine structure of I2 rovibrational levels includes four

contributions: (1) nuclear electric quadrupole (eqQ), (2) spin-rotation (C),
(3) tensorial spin-spin (d), and (4) scalar spin-spin (δ) interactions.
1. INTRODUCTION 35
Agreement between experiment and theory using the four-term effective

hyperfine Hamiltonian is at the kilohertz level for a few selected transitions.
For the first excited electronic state B with the 2P3/2 + 2P1/2 dissociation limit,
our goal is to perform a systematic, high-precision investigation of hyperfine
interactions over an extensive range of rovibrational quantum numbers
coupled with a large range of internuclear separations. Such a study has
allowed us to understand the rovibrational dependence of the hyperfine
interactions (and the dependence on internuclear distance) based on ab initio
molecular potential-energy curves and the associated electronic wave
functions. Analysis of various perturbation effects leads to precise
determination of molecular structure over a large dynamic range [66].
Prior studies have concentrated on a few, isolated rovibrational levels for
the high vibrational levels ν' = 40–82 in the B state. For vibrational levels
below ν' = 43, only functional forms of the state-dependent variations of the
hyperfine interactions have been investigated from empirical data.
Combining absolute optical frequency metrology with high-resolution and
broad-wavelength-coverage laser spectroscopy, we have measured ~80
rovibrational transitions with the upper vibrational levels (from ν' = 42 up to
ν' = 70) stretching from a closely bonded molecular basis to a separated-
atomic basis appropriate for the 2P3/2 + 2P1/2 dissociation limit and providing
kilohertz-level line accuracies for most hyperfine components. The study is
performed in the wavelength region of 530–498 nm. Measurements
performed on a large set of rovibrational quantum numbers provide
systematic information on state-dependent variations in the hyperfine
interactions caused by perturbation from other nearby states. Figure 1-5(b)
shows a simple schematic of the ground and the first excited electronic states
of I2 and their relevant dissociation limits. There is a clear trend of linewidth
narrowing with decreasing transition wavelength [57]. However, this
tendency is complicated by variations in linewidths among different
rotational or hyperfine components when the transitions approach the
predissociation region. The initial linewidth narrowing at shorter wavelength
may indicate, among other interesting effects, that the Franck-Condon factor
in the transition probability is reduced when the excited state reaches a
higher vibration level. As the excited state approaches the dissociation
threshold, the limit on lifetime imposed by predissociation and other effects
will need to be taken into consideration.
Figure 1-6 illustrates the systematic rovibrational dependences for all
four hyperfine parameters. Each solid line is a fit of the experimental data
for rotational dependence belonging to a single vibrational level (ν'). In
general, all hyperfine parameters have a monotonic dependence on both
rotational and vibrational quantum numbers except for the levels in the
vicinity of ν' = 57–59. However, the ν-dependence of eqQB reverses its trend
36 Chapter 1
after ν' = 60. For the sake of figure clarity, the eqQB data for ν' > 60 are not
shown. Another important observation is that for levels of ν' = 57–59, all
hyperfine parameters except for CB bear abnormal J-dependences due to
perturbations from a 1g state through accidental rotational resonances.
Combining data from this work and the literature [67], investigations of
the hyperfine spectra now cover the majority of the vibrational levels (3 ≤ v'
≤ 82) in the B state. Therefore, it is now possible and useful to explore the
global trend of these hyperfine parameters in the B state. Suppressing the
rotational dependence, hyperfine parameters as functions of pure vibrational
energy E(v') are found to increase rapidly when molecules approach the
dissociation limit, which is a result of the increasingly strong perturbations
from other high-lying electronic states sharing the same dissociation limit
with the B state. While the variation of CB is smooth over the whole range,
eqQB , dB , and δB all have local irregularities at three positions: (1) ν' = 5
where the B":1u state crosses nearby, (2) around ν' = 57–59 (see discussions
above), and from ν' = 76–78, because of the same 1g state.
To examine these hyperfine parameters in terms of the internuclear
separation R, the vibrational average of the hyperfine parameters is removed
by inverting the expression O (v ′, J ′) = v ′J ′ O(R ) v ′J ′ , where O(v',J') denotes
one of the four hyperfine parameters. eqQB , CB , dB , and δB against R-
centroid are evaluated from v′J ′ R v′J ′ (with v ′J ′ properly normalized).
Consistent with CB’s smooth variation, the interpolation function CB (R) has
small residual errors (within ±0.03, relative) for the entire range from v' = 3–
70. In contrast, the large residual errors in the interpolation of eqQB , dB, and
δB for v' ≥ 56 reflect abnormal variations observed around v' = 57 and 59,
restricting a reliable interpolation only to levels of v' < 56. In the region of R
< 5 Å, valuable information can be readily extracted from eqQB to assist the
investigation of I2’s electronic structure. Unlike the other three hyperfine
parameters whose major parts originate from perturbations at nearly all
possible values of R, a significant part of eqQB is due to the interaction
between the nuclear quadrupole moment Q and the local electric-field
gradient q(R) generated by the surrounding charge distribution of a largely
B-state character. Thus, for R < 5 Å, where perturbations from other
electronic states are negligible, the vibration-removed interpolation function
eqQB (R), coupled with a priori information on q(R), can be used to
determine the I2 nuclear quadrupole moment or serve as a benchmark for
molecular ab initio calculations of the electronic structure at various values
of R.
Precision measurements on B–X hyperfine spectra provide an alternative,
and yet effective, way to investigate the potential energy curves (PECs)
sharing the same dissociation limit with the B state and the associated
1. INTRODUCTION 37
electronic wave functions. To demonstrate this method, we perform

calculations of eqQB, CB, dB, and δB based on the available PECs and
electronic wave functions derived from a separated atomic-basis set. For
both vibrational and rotational dependences, the ab initio calculation results
agree very well with the experimental data for ν' ≥ 42 (R centroid ≥ 3.9 Å).
In short, we have extended the range of separated atomic-basis calculations
from levels near the dissociation limit to low vibrational levels (ν' = 5) and
have found very good agreement with the experimental data on both
vibrational and rotational dependences.
42 (a) 70 (b)
2 69 65
43 63
-560
61
eqQ B (MHz)
60
45 59
CB (MHz)
1 57
-564
47 55
50 6 53
49 5
51 51
-568 4 49
53 50 47
3 45
57 55
43
-572 2
59 60 42
1 70 70
8 69 (c) (d)
69 65
7 63
6
59 1 61
65 60 60
5 63
59 57
-dB (MHz)
δB (MHz)
4 55
61 57 55
3 53
53 51
51
0.1
49
2 50 49 50
47 47
45 45
43 0.01 42
0.1
42 42
8
3 3
0 5 10x10 0 5 10x10
J' (J' +1) J' (J' +1)
Figure 1-6. Rovibrational dependence of the B-state hyperfine parameters: (a) eqQB , (b) CB,
(c) dB , and (d) δB. Graphs (b), (c), and (d) are semilog plots and the vertical scale of (c) has
been inverted. Each solid line is a fit for J-dependence for each vibrational level (ν' indicated
in the figure). Experimental data in squares and open circles show abnormal variations of
eqQB , dB , and δB around ν' = 57 and 59.
38 Chapter 1
4. CARRIER-ENVELOPE PHASE COHERENCE

AND TIME-DOMAIN APPLICATIONS
Prior to the development of femtosecond comb technology, mode-locked

lasers were used almost exclusively for time-domain experiments. Although
femtosecond comb technology has primarily impacted the frequency-domain
applications described earlier, it is having an effect on time-domain
experiments and promises to bring about just as dramatic advances in the
time domain as it has in optical frequency metrology and optical clocks.
Indeed, it is fascinating to blur the boundary between traditional cw
precision spectroscopy and ultrafast phenomena. The time-domain
applications put stringent requirements on carrier-envelope phase coherence.
Stabilization of the “absolute” carrier-envelope phase at a level of tens of
milliradians has been demonstrated. This phase coherence is maintained over
an experimental period exceeding many minutes, laying the groundwork for
synthesizing electric fields with known amplitude and phase at optical
frequencies. Working with two independent femtosecond lasers operating at
different wavelength regions, we can synchronize the two pulse trains at the
femtosecond level and also phase lock the carrier frequencies. This
procedure establishes phase coherence between the two lasers. By coherently
stitching optical bandwidths together, a “synthesized” pulse has been
generated. With the same pair of Ti:sapphire mode-locked lasers, we can
generate widely tunable femtosecond pulses in the mid and far IR using
difference-frequency generation. The flexibility of this new experimental
approach is evidenced by the capability for rapid and programmable
switching and modulation of the wavelength and amplitude of the generated
IR pulses. A fully developed capability for producing phase-coherent visible
and IR pulses over a broad spectral bandwidth, coupled with arbitrary
control of amplitude and pulse shape, represent the ultimate instrumentation
for coherent control of molecular systems. A pulse train with good carrier-
envelope-phase coherence is also very promising for experiments that are
sensitive to φce, i.e., the “absolute” pulse phase. This sensitivity can be
manifested in “extreme” nonlinear optics experiments or coherent control.
The capability of precisely controlling pulse timing and the carrier-
envelope phase allows one to manipulate pulses using novel techniques and
achieve unprecedented levels of flexibility and precision. For example, the
simultaneous control of timing jitter and carrier-envelope phase can be used
to phase coherently superpose a collection of successive pulses from a mode-
locked laser. By stabilizing the two degrees of freedom of a pulse train to an
optical cavity acting as a coherent delay, constructive interference of
sequential pulses will be built up until a cavity dumper switches out the
“amplified” pulse [68]. Such a passive-pulse ‘amplifier,” along with the
1. INTRODUCTION 39
synchronization technique we developed for pulse synthesis, have made a

strong impact on the field of nonlinear optical spectroscopy and imaging of
biomolecular systems, offering significant improvements in experimental
sensitivity and spatial resolution [69]. With enhanced detection sensitivity
comes the capability of tracking real-time biological dynamics. An ultrafast
laser locked to a high-stability cavity is also expected to demonstrate
extremely low pulse jitter and carrier-envelope-phase noise, features that
will be particularly attractive for time-domain experiments. In addition, we
are exploring the use of pulse-cavity interactions to obtain a high sensitivity
in intracavity spectroscopy (linear and nonlinear) with wide spectral
coverage and to enhance nonlinear interaction strengths for high-efficiency,
nonlinear optical experiments.
The carrier-envelope phase can also be used for deterministic control in
the low-field regime because of interference between reaction channels in
chemical systems and population transfer between quantum states [70, 71].
This interference had been demonstrated using separate phase-related pulses
in atoms [72, 73], molecules [74], and in direct-gap semiconductors [75, 76].
Recently, it was demonstrated in a direct-gap semiconductor using φce
stabilized pulses [77].
4.1 Timing synchronization of mode-locked lasers
To establish phase coherence among independent ultrafast lasers, it is

necessary to first achieve a level of synchronization among these lasers so
that the remaining timing jitter is less than the oscillation period of the
optical carrier, namely 2.7 fs for lasers centered around 800 nm. In the push
for greater stability and precision of femtosecond optical combs, a number of
effective techniques for ultralow-jitter timing synchronization have emerged.
They include an all-electronic approach for active stabilization of repetition
rates [78], cross-phase modulation to passively synchronize two mode-
locked lasers that share the same intracavity gain medium [79], the linkage
of lasers’ repetition rates to the same optical standard [80], and optical cross-
correlation between the pulse trains to be synchronized [81].
Detecting timing jitter should be carried out at a high harmonic of fr to
attain much-enhanced detection sensitivity. The harmonic order can range
from 100 to 106. This approach has enabled tight synchronization between
two independent mode-locked Ti:sapphire lasers with a residual-rms timing
jitter on the order of 1 fs or less, integrated over a bandwidth of a few
megahertz. Of course the enhanced detection sensitivity comes with a price
of reduced dynamic range for the change in the nominal value of fr. This
problem can be alleviated by invoking another control loop that works with
lower harmonics of fr. The low-frequency control loop can be used to
40 Chapter 1
achieve a desired timing offset between the two pulse trains. The high
stability loop can then be activated to achieve the ultimate level of
synchronization at the preset value of timing offset. For example, in the all-
electronic implementation of laser synchronization, we use two phase-locked
loops (PLLs). The low-resolution PLL compares and locks the fundamental
repetition frequencies (100 MHz) of the lasers. The second, high-resolution,
PLL compares the phase of the 140th harmonic of the two repetition
frequencies at 14 GHz. A transition of control from the first PLL to the
second PLL is sufficiently smooth to allow synchronization at the
femtosecond level for any timing offsets within the entire dynamic range,
e.g., one pulse period of 10 ns [82]. The synchronization lock can be
maintained for several hours.
The ability to synchronize a passively mode-locked laser to an external
reference or to a second laser has many applications. Previous work in
electronic synchronization of two mode-locked Ti:sapphire lasers
demonstrated timing jitter of, at best, a few hundred femtoseconds.
Therefore the present level of synchronization would make it possible to take
full advantage of this time resolution for applications such as high-power
sum- and difference-frequency mixing [83], novel pulse generation and
shaping [9], new generations of laser/accelerator-based light sources, or
experiments requiring synchronized laser light and x-rays or electron beams
from synchrotrons [84]. Indeed, accurate timing of high-intensity fields is
essential for several important schemes in quantum coherent control and
extreme nonlinear optics such as efficient x-ray generation. Two recent
applications that have been developed in our laboratories include tunable,
subpicosecond pulse generation in the IR [85] and coherent anti–Stokes-
Raman scattering (CARS) microscopy with two tightly synchronized
picosecond lasers [69]. The flexibility and general applicability of the two-
laser-synchronization approach are clearly demonstrated in the
straightforward generation of programmable light sources for these
applications.
4.2 Phase lock between separate mode-locked lasers
Phase locking of separate femtosecond lasers requires a step beyond

synchronization of the two pulse trains. One needs effective detection and
stabilization of the phase difference between the two optical carrier waves
underlying the pulse envelopes [86]. After synchronization matches the
repetition rates (fr 1 = fr 2), phase locking requires that the spectral combs of
the individual lasers are maintained exactly coincident in the region of
spectral overlap so that the two sets of optical frequency combs form a
continuous and phase-coherent entity. We detect a coherent heterodyne-beat
1. INTRODUCTION 41
signal between the corresponding comb components of the two mode-locked

lasers. Such heterodyne detection yields information related to the difference
in the offset frequencies of the two lasers, δf0 = f01 – f02, which can then be
controlled. By phase locking δf0 to a frequency of a mean zero value, we
effectively achieve (∆φce1 – ∆φce2) = 0, leading to two pulse trains that have
nearly identical phase evolution. When stabilized, fluctuations associated
with the recorded beat-frequency signal (δf0) can be suppressed to just a few
millihertz with an averaging time of 1 s.
The established phase coherence between the two mode-locked lasers can
also be revealed via a direct time-domain analysis. For example, spectral
interferometry analysis of the joint spectra of the two pulses produces
interference fringes that correspond to phase coherence between the two
pulse trains persisting over the measurement time period. A cross-correlation
measurement between the two pulse trains also manifests phase coherence in
the display of persistent fringe patterns. A more powerful and
straightforward demonstration of the “coherently synthesized” aspect of the
combined pulse is through a second-order autocorrelation measurement of
the combined pulse. For this measurement, the two pulse trains are
maximally overlapped in the time domain before the autocorrelator. When
the two femtosecond lasers are phase locked, autocorrelation reveals a clean
pulse that is often shorter in apparent duration and larger in amplitude than
the individual original pulses. A successful implementation of coherent light
synthesis has therefore become reality: the coherent combination of output
from more than one laser where the combined output can be viewed as a
coherent femtosecond pulse being emitted from a single source [9].
4.3 Extending phase-coherent femtosecond combs to the

mid-IR spectral region
Being able to combine the characteristics of two or more pulsed lasers

working at different wavelengths provides a more flexible approach to
coherent control. The capability of synchronizing the repetition rates and
phase locking the carrier frequencies of two mode-locked lasers opens up
many applications. This capability may be particularly important in the
generation of tunable femtosecond sources in other previously unreachable
spectral regions. For example, two stabilized mode-locked Ti:sapphire lasers
can be employed to enable both sum- (SFG) and difference-frequency
generation (DFG). The DFG signal produced by a GaSe crystal can be tuned
from 6 microns to any longer wavelength region with a high repetition rate
(the same as the original laser's) and a reasonable average power (tens of
microwatts). Arbitrary amplitude-waveform generation and rapid
wavelength switching in these nonlinear signals are simple to implement.
42 Chapter 1
The ultimate goal of this work is to make an optical-waveform synthesizer

that can create an arbitrary optical pulse on demand and use the novel source
to study and control molecular motion. For frequency metrology and
precision molecular spectroscopy in the IR region, the DFG approach
produces an absolute frequency-calibrated-IR comb when the two
Ti:sapphire lasers are synchronized and share a common offset frequency f0.
Another attractive approach presented in Chapter 6 makes use of optical
parametric generation to produce high-peak-power femtosecond pulses in
the IR spectral domain. This scheme presents an attractive property of
automatic cancellation of the carrier-envelope-phase shift in IR pulses.
One of the important spectral regions is 1.5 µm, where compact, reliable,
and efficient mode-locked lasers exist. Here there is rich information on
molecular spectra. Frequency reference grids in this spectral window could
find applications in dense-wavelength–division-multiplexed (DWDM)
communications systems, photonic samplers in high-speed–analog-to-digital
conversion, and distribution of optical frequency standards over optical fiber
networks. The synchronization and phase-locking approach for Ti:sapphire
lasers can be extended to cover mode-locked lasers at 1.5 µm [87]. We have
indeed achieved tight synchronization and coherent phase locking between
the repetition rates and the optical carriers, respectively, for the 1.5 µm
mode-locked laser sources and a Ti:sapphire-based femtosecond frequency
comb, which is used as the clockwork for an optical atomic clock [88].
To achieve phase locking between the optical carriers of the Ti:sapphire
and the 1.5 µm mode-locked lasers, one requires spectral overlap between
the two combs. The wide-bandwidth optical frequency comb generated by
the mode-locked femtosecond Ti:sapphire laser is phase locked to a highly
stable, visible optical frequency standard. The optical comb of the 1.5 µ m
source is frequency doubled and compared against the Ti:sapphire comb at a
mutually accessible spectral region to generate a heterodyne beat between
the two combs. Under simultaneous control of synchronization and phase
locking, the 1.5 µm laser faithfully duplicates both the repetition frequency
and the carrier phase of the Ti:sapphire laser. Therefore the optical clock
information has now been transferred onto the 1.5 µm mode-locked laser,
which can be used in a fiber network to transfer highly stable clock signals
to remote ends [89]. This topic is discussed in Chapter 9.
4.4 Femtosecond lasers and external optical cavities
The combination of ultrashort pulse trains and optical cavities will open
doors for a variety of exciting experiments. Such experiments will require an
understanding of intricate pulse-cavity interactions and the development of
techniques to efficiently couple ultrashort pulses into a high-finesse optical
1. INTRODUCTION 43
cavity and coherently store them in it. An immediate impact will be on the
precision stabilization of ultrafast lasers (see more detailed discussions in
Chapter 12). Similar to the state-of-the-art stabilization of cw lasers, a
cavity-stabilized ultrafast laser is expected to demonstrate superior short-
term stability of both the pulse-repetition frequency and the carrier-envelope
phase [90]. The improved stability is beneficial in particular for time-domain
applications where the signal-processing bandwidth is necessarily large.
Another attractive application lies in broadband and ultrasensitive
spectroscopy. The use of high-finesse cavities has played a decisive role in
enhancing sensitivity and precision in atomic and molecular spectroscopy.
We expect a dramatic advancement in the efficiency of intracavity
spectroscopy by exploiting the application of ultrashort pulses. In other
words, high detection sensitivity is achievable uniformly across the broad
spectrum of the pulse. Applying cavity-stabilization techniques to
femtosecond lasers, the comb structure of the probe laser can be precisely
matched to the resonance modes of an empty cavity, allowing an efficient
energy coupling for a spectroscopic probe. Molecular samples located inside
the high-finesse cavity will have a strong impact on the dispersive properties
of the cavity. In fact, it is this dispersion-related–cavity-pulling effect that
will aid our sensitive detection process when we analyze the light
transmitted through the cavity. Preliminary data on spectrally resolved, time-
domain–ring-down measurement for intracavity loss over the entire
femtosecond-laser bandwidth are already quite promising.
To develop sources for ultrafast nonlinear spectroscopy, a properly
designed, dispersion-compensated cavity housing a nonlinear crystal will
provide efficient nonlinear optical frequency conversion of ultrashort optical
pulses at spectral regions where no active gain medium exists. Furthermore,
by simultaneously locking two independent mode-locked lasers to the same
optical cavity, efficient SFG and/or DFG can be produced over a large range
of wavelengths. Similarly, a passive cavity can be used to explore coherent
superposition of ultrashort pulses, with cavity stabilization providing the
means to phase coherently superpose a collection of successive pulses from a
mode-locked laser. The coherently enhanced pulse stored in the cavity can
be switched out using a cavity-dumping element (such as a Bragg cell),
resulting in a single phase-coherent, amplified pulse. The use of a passive
cavity also offers the unique ability to effectively amplify pulses in spectral
regions where no suitable gain medium exists such as for IR pulses from
difference-frequency mixing or the ultraviolet light from harmonic
generation. Unlike actively dumped laser systems, the pulse energy is not
limited by the saturation of a gain medium or the requirement for a saturable
absorber for mode locking. Instead, the linear response of the passive cavity
allows the pulse energy to build up inside the cavity until limited by cavity
44 Chapter 1
loss and/or dispersive pulse spreading. Therefore storage and amplification

of ultrashort pulses in the femtosecond regime require precise control of the
reflected spectral phase of the resonator mirrors and the optical loss of the
resonator. While the reflected group delay of the mirrors only changes the
effective length of the resonator, the group-delay dispersion (GDD) and
higher-order derivatives of the group delay with respect to frequency affect
the pulse shape. The net cavity GDD over the bandwidth of the pulse needs
to be minimized to maintain the shape of the resonant pulse and allow for the
coherent addition of energy from subsequent pulses.
We have applied the coherent pulse-stacking technique to both
picosecond [91] and femtosecond pulses [92]. Initial studies have already
demonstrated amplification of picosecond pulses of more than thirtyfold at
repetition rates of 253 kHz, yielding pulse energies greater than 150 nJ. With
significant room left for optimization of the cavity finesse (current value of
~350, limited by the cavity input-coupling mirror), we expect that
amplifications greater than a hundredfold are feasible, which would bring
pulse energies into the µJ range. While the use of picosecond pulses allows
us to separate out complications arising from intracavity dispersion, for sub-
100 fs pulses, dispersive phase shifts in the cavity mirrors become an
important topic. Preliminary results in enhancing low individual pulse
energies for sub-50 fs pulses illustrate the importance of GDD control. The
external enhancement cavity incorporated specially designed negative GDD
low-loss mirrors to simultaneously compensate for the Bragg cell's 3 mm of
fused silica and provide a high finesse. The input-coupling mirror
transmission is ~0.8%, with a measured cavity finesse of 440. An intracavity
energy build-up ratio of ~130 is expected, leading to single pulse
amplifications of approximately 52 for the current setup, given the 40%
dumping efficiency of our Bragg cell. The negative GDD mirrors are
designed to only partially compensate for the total cavity dispersion. The
remaining cavity GDD is estimated at +20 to +30 fs2. Controlling the
intracavity pressure allows fine tuning of the net cavity GDD to zero.
Experimental results are in good agreement with independent numerical
calculations. The input pulses of 47 fs duration are experimentally enhanced
by a factor of ~120 inside the passive cavity, with the output pulses
broadened only to ~49 fs. These results are shown in Figure 1-7.
1. INTRODUCTION 45
T = 1/fr tin (Optional) cavity

dumper
Mode-locked laser
2
 F 
Cavity enhancement: N = 4tin   τroundtrip = k T or (1/k) T; k: integer
 2π 
1.2
(a) enhanced (112 x buildup)
FROG trace input spectrum (b)
Amplitude (a.u.)
Amplitude (linear scale)

1.0
0.8
of output pulse
0.8
49 fs FWHM
0.6
0.4
0.4
0.2
0.0
0.0
-50 0 50 100 150 780 800 820 840

Time (fs) Wavelength (nm)
Figure 1-7. Top: Illustration of the principle of coherent-pulse amplification with the aid of an
optical cavity, showing the matching of the pulse-repetition period with the cavity round-trip
time. The intracavity pulse is switched out when sufficient energy is built up in the cavity.
The build-up factor is given ideally by the cavity finesse, with appropriate intracavity
dispersion compensation. Bottom: Coherent evolution of a 50 fs pulse inside the cavity. The
left panel displays a FROG measurement of the amplified pulse switched out of the passive
cavity shown at FWHM of 49 fs, slightly wider than the input pulse width of 47 fs. The right
panel presents a comparison of the input and output pulse spectra, showing no significant
distortion to the input pulse spectrum after the intracavity power is built up by a factor of 112.
An important application of these advanced pulse-control technologies is

in the field of nonlinear optical spectroscopy and nanoscale imaging. For
example, using two tightly synchronized picosecond lasers, one can achieve
significant improvements in experimental sensitivity and spatial resolutions
for CARS microscopy. Vibrational imaging based on CARS spectroscopy is
a powerful method for acquisition of chemically selective maps of biological
samples. In CARS microscopy, pulsed-pump and Stokes beams are focused
tightly to a single focal spot in the sample to achieve a high spatial
resolution. The third-order nonlinear interaction produces a signal photon
that is blue-shifted (anti-Stokes signal) with respect to the incident beams.
Strong CARS signals are obtained whenever the frequency difference
between the pump and Stokes coincides with a Raman-active vibrational
mode, which gives rise to the molecule-specific vibrational contrast in the
image. Recent studies and technological improvements have demonstrated
the capability of CARS microscopy to attain high-resolution vibrational
images of unstained living cells [93]. Practical applications of the CARS
46 Chapter 1
microscopy technique require pulsed light sources, because optimized peak

powers help boost the nonlinear signal. Pulses with temporal widths of 1–2
picoseconds (ps) are used to match to the vibration bandwidths to optimize
the CARS signal with minimized nonresonant background and
compromising spectral resolution [94]. An important technical challenge is
to achieve tight synchronization between two mode-locked lasers that
produce the frequency difference that matches the vibrational resonance
[95]. Another important consideration is that while the repetition rate of the
pulse train needs to be low enough to avoid thermal damage to the cell due
to a high average power, the peak power of the pulses needs to be reasonably
high to improve the nonlinear signal strength. The technologies of pulse
synchronization and coherent pulse stacking therefore become ideal tools for
carrying out this task of combining spectroscopy with microscopy.
4.5 Coherent control via quantum interference between

one- and two-photon absorption
Control of molecular reactions is a central goal of chemistry. The

development of the laser led to the proposal that light fields could be used to
control reaction pathways; for recent reviews, see [70] and [96]. Many
techniques are sensitive to the phase of the applied fields and thus dubbed
“coherent control.” To date, only the relative phase between two laser fields,
or the relative internal phase of a femtosecond pulse (i.e., its chirp) has been
demonstrated to have physical impacts. Some new schemes explore
interference between pathways involving n-photons and m-photons. When n
and m have opposite parity, a dependence on φce will occur for excitation by
single ultrashort pulse. Early studies demonstrated the interference
phenomenon by using a pair of phase-controlled pulses to ionize rubidium
[72] and control electrical currents in bulk semiconductors [75]. In both
cases, there is a connection between spatial direction and the relative phase.
The latter system has recently been shown to be sensitive to φce [77].
In a semiconductor, the phase sensitivity can be understood by
considering the k dependence of the transition amplitudes. In GaAs, when
the phase is such to produce constructive interference between one- and two-
photon absorption at –k, it produces, to good approximation, destructive
interference at k. This effect is schematically shown in Figure 1-8. For an
octave-spanning pulse, the frequency components for one-photon absorption
come from the high-energy tail of the spectrum, while for the two-photon
absorption, they come from the low-frequency tail. Thus φce determines their
relative phase in a manner analogous to the beat signal that occurs in a
standard ν-to-2ν interferometer. The noninterfering carrier populations do
not contribute to the signal, which makes it intrinsically balanced and thus
1. INTRODUCTION 47
insensitive to amplitude fluctuations. This insensitivity provides an

advantage over the standard ν-to-2ν interferometer, which relies on an
ordinary photodiode for conversion of the light to an electrical signal.
Conduction
Band
2ν Phase-Coherent
Harmonically Related
ν Pulses
2ν E
ν
k Phase-Coherent
ν Octave-Spanning
Pulse
0
Valence
Band
Figure 1-8. Conceptual diagram of quantum interference between one- and two-photon
absorption in a direct-gap semiconductor. The two interfering absorption pathways are driven
by the spectral wings of a single octave-spanning pulse. An imbalance in the otherwise
symmetric carrier-population distribution in momentum space (represented by ovals) occurs
due to interference and results in a net current. The resulting photocurrent is sensitive to φce.
The experimental demonstration of quantum interference control of the

injected photocurrent in a semiconductor is shown in Figure 1-9 [77]. The
carrier-envelope evolution of the laser was stabilized so that f0 = 2400 Hz
using a standard ν-to-2ν interferometer in a servo loop. A portion of the
resulting phase-stabilized pulse train was then used to illuminate a low-
temperature grown GaAs sample with gold electrodes. The electrodes are
used to collect the resulting photocurrent, which also oscillates at f0.
10 dB
2000 2200 2400 2600

frequency (Hz)
Figure 1-9. Spectrum of the measured quantum interference signal in a semiconductor. The
dotted line is the background (no light on the sample).
These results provide an interesting route to a simple, solid-state detector

of the carrier-envelope phase. The first step is to use it simply to stabilize f0.
48 Chapter 1
However this requires a significant improvement in signal-to-noise ratio and

bandwidth. This technique does not suffer from the arbitrary offsets that
plague detection using a standard ν-to-2ν interferometer and thus may allow
measurement of the “absolute” φce [97]. However, shifts due to dispersion in
the GaAs detector still need to be considered.
4.6 Extreme nonlinear optics
The expression “extreme nonlinear optics” refers to experiments where

the optical pulses are so intense that the electric field of the pulse is more
relevant than the intensity profile, as is the case in the “perturbative” regime
[8]. In this regime, the electric field can be strong enough to distort the
potential energy well for an electron bound to an atom so much that
ionization occurs via tunneling. Such tunneling ionization, also known as
“above-threshold ionization,” typically displays a threshold with respect to
the electric field of the pulse. This effect causes tunneling ionization to
depend both on the phase of sufficiently short pulses and whether the
threshold is close to the maximum field in the pulse. Indirect evidence for
such a phase dependence has been observed in high-harmonic generation
[98], which occurs when the ionized electrons slam back into the ion. More
recently, measurements of the photoelectron yield in opposite directions
have provided evidence for a phase dependence in above-threshold
ionization [99]. Chapter 10 provides a thorough discussion of the generation
of the high-intensity pulses needed to access the regime of extreme nonlinear
optics together with a review of the results obtained for above-threshold
ionization. Control of high-harmonic generation is addressed in Chapter 11.
5. SUMMARY
Recent developments in femtosecond comb generators have enabled

breakthroughs in optical frequency metrology, optical frequency synthesis,
and optical atomic clocks. Femtosecond combs have been built, or are being
built, for these applications around the world. Although not large on an
absolute scale, the number is a large multiple of the total number of
frequency multiplication chains ever built. Indeed, here in Boulder,
Colorado, between NIST and JILA, there are more femtosecond comb-based
frequency chains than were operating in the world prior to 1999! Thus, we
are confident that precision, absolute optical frequency metrology and
synthesis are becoming common laboratory tools.
The time-domain applications engendered by femtosecond combs are just
being realized. They also promise very exciting results in the near future.
1. INTRODUCTION 49
ACKNOWLEDGEMENTS
The authors would like to acknowledge the many contributors to the

work on femtosecond combs at JILA. Special acknowledgement goes to Jan
Hall for building the foundation for precision-laser stabilization at JILA over
the last 40 years. Others include D. J. Jones, R. J. Jones, T. M. Fortier, K.
Holman, S. Foreman, L. Chen, M. Stowe, A. Marian, P. A. Roos, K. Moll,
M. Thorpe, S. A. Diddams, L.-S. Ma, D. Hudson, R. Shelton, L. Hollberg,
H. C. Kapteyn, M. M. Murnane, T. W. Hänsch, and T. Udem. Our research
is funded by NIST, NSF, ONR, NASA, AFOSR, and DARPA. Both authors
are staff members of the NIST Quantum Physics Division. Ye’s email is
[email protected] and Cundiff’s email is [email protected].
REFERENCES
[1] U. Morgner, F. X. Kärtner, S. H. Cho, Y. Chen, H. A. Haus, J. G. Fujimoto,
E. P. Ippen, V. Scheuer, G. Angelow, and T. Tschudi, Opt. Lett. 24, 411-
413 (1999); D. H. Sutter, G. Steinmeyer, L. Gallmann, N. Matuschek, F.
Morier-Genoud, U. Keller, V. Scheuer, G. Angelow, and T. Tschudi, Opt.
Lett. 24, 631-633 (1999).
[2] D. J. Jones, S. A. Diddams, J. K. Ranka, A. Stentz, R. S. Windeler, J. L.
Hall, and S. T. Cundiff, Science 288, 635-639 (2000).
[3] S. A. Diddams, D. J. Jones, J. Ye, S. T. Cundiff, J. L. Hall, J. K. Ranka, R.
S. Windeler, R. Holzwarth, T. Udem, and T. W. Hänsch, Phys. Rev. Lett.
84, 5102-5105 (2000); R. Holzwarth, T. Udem, T. W. Hänsch, J. C.
Knight, W. J. Wadsworth, and P. S. J. Russell, Phys. Rev. Lett. 85, 2264-
2267 (2000); S. T. Cundiff, J. Ye, and J. L. Hall, Rev. Sci. Instrum. 72,
3746-3771 (2001).
[4] J. Stenger, T. Binnewies, G. Wilpers, F. Riehle, H. R. Telle, J. K. Ranka, R.
S. Windeler, and A. J. Stentz, Phys. Rev. A 63, 021802 (2001).
[5] T. Udem, S. A. Diddams, K. R. Vogel, C. W. Oates, E. A. Curtis, W. D.
Lee, W. M. Itano, R. E. Drullinger, J. C. Bergquist, and L. Hollberg, Phys.
Rev. Lett. 86, 4996-4999 (2001).
[6] S. A. Diddams, T. Udem, J. C. Bergquist, E. A. Curtis, R. E. Drullinger, L.
Hollberg, W. M. Itano, W. D. Lee, C. W. Oates, K. R. Vogel, and D. J.
Wineland, Science 293, 825-828 (2001); J. Ye, L. S. Ma, and J. L. Hall,
Phys. Rev. Lett. 87, 270801 (2001).
[7] A. Apolonski, A. Poppe, G. Tempea, C. Spielmann, T. Udem, R.
Holzwarth, T. W. Hänsch, and F. Krausz, Phys. Rev. Lett. 85, 740-743
(2000).
[8] T. Brabec and F. Krausz, Rev. Mod. Phys. 72, 545-591 (2000).
[9] R. K. Shelton, L. S. Ma, H. C. Kapteyn, M. M. Murnane, J. L. Hall, and J.
Ye, Science 293, 1286-1289 (2001).
50 Chapter 1
[10] T. W. Hänsch, in Tunable Lasers and Applications, edited by A.

Mooradain, T. Jaeger and P. Stokseth (Springer-Verlag, Berlin, 1976); E.
V. Baklanov and V. P. Chebotaev, Sov. J. Quantum Electr. 7, 1252-1255
(1977).
[11] R. Teets, J. Eckstein, and T. W. Hänsch, Phys. Rev. Lett. 38, 760-764
(1977).
[12] J. N. Eckstein, A. I. Ferguson, and T. W. Hänsch, Phys. Rev. Lett. 40, 847-
850 (1978).
[13] T. Udem, J. Reichert, R. Holzwarth, and T. W. Hänsch, Opt. Lett. 24, 881-
883 (1999).
[14] T. Udem, J. Reichert, R. Holzwarth, and T. W. Hänsch, Phys. Rev. Lett. 82,
3568-3571 (1999).
[15] S. A. Diddams, D. J. Jones, L. S. Ma, S. T. Cundiff, and J. L. Hall, Opt.
Lett. 25, 186-188 (2000).
[17] J.-C. Diels and W. Rudolph, Ultrashort Laser Pulse Phenomena :
Fundamentals, Techniques, and Applications on a Femtosecond Timescale
(Academic Press, San Diego, 1996).
[18] E. P. Ippen, Appl. Phys. B 58, 159-170 (1994).
[19] R. L. Fork, O. E. Martinez, and J. P. Gordon, Opt. Lett. 9, 150-152 (1984).
[20] D. E. Spence, P. N. Kean, and W. Sibbett, Opt. Lett. 16, 42-44 (1991); D.
K. Negus, L. Spinelli, N. Goldblatt, and G. Feugnet, in Advanced Solid-
State Lasers (OSA, 1991), Vol. 10.
[21] M. T. Asaki, C. P. Huang, D. Garvey, J. P. Zhou, H. C. Kapteyn, and M.
M. Murnane, Opt. Lett. 18, 977-979 (1993).
[22] R. Ell, U. Morgner, F. X. Kärtner, J. G. Fujimoto, E. P. Ippen, V. Scheuer,
G. Angelow, T. Tschudi, M. J. Lederer, A. Boiko, and B. Luther-Davies,
Opt. Lett. 26, 373-375 (2001).
[23] S. T. Cundiff, J. Phys. D. 35, R43 (2002).
2693-2701 (1993).
[25] J. Ye, L. S. Ma, T. Day, and J. L. Hall, Opt. Lett. 22, 301-303 (1997).
[26] K. Nakagawa, M. deLabachelerie, Y. Awaji, and M. Kourogi, J. Opt. Soc.
Am. B 13, 2708-2714 (1996).
[27] J. L. Hall, L. S. Ma, M. Taubman, B. Tiemann, F. L. Hong, O. Pfister, and
J. Ye, IEEE Trans. Instrum. Meas. 48, 583-586 (1999).
[28] T. Udem, J. Reichert, T. W. Hänsch, and M. Kourogi, Phys. Rev. A 62,
1801-1804 (2000).
[29] G. P. Agrawal, Nonlinear Fiber Optics (Academic Press, San Diego, 1995).
[30] F. X. Kärtner, U. Morgner, R. Ell, T. Schibli, J. G. Fujimoto, E. P. Ippen,
V. Scheuer, G. Angelow, and T. Tschudi, J. Opt. Soc. Am. B 18, 882-885
(2001).
[31] A. Bartels and H. Kurz, in XIII International Conference on Ultrafast
Phenomena, edited by M. M. Murnane, N. F. Scherer, D. J. D. Miller and
A. M. Weiner (OSA, Vancouver, BC, 2002), Vol. 72, p. 173-174.
1. INTRODUCTION 51
[32] J. Reichert, R. Holzwarth, T. Udem, and T. W. Hänsch, Opt. Commun.

172, 59-68 (1999).
[33] A. Poppe, R. Holzwarth, A. Apolonski, G. Tempea, C. Spielmann, T. W.
Hänsch, and F. Krausz, Appl. Phys. B 72, 977-977 (2001).
[34] L. Xu, C. Spielmann, A. Poppe, T. Brabec, F. Krausz, and T. W. Hänsch,
Opt. Lett. 21, 2008-2010 (1996).
[35] H. A. Haus and E. P. Ippen, Opt. Lett. 26, 1654-1656 (2001).
[36] P. Dietrich, F. Krausz, and P. B. Corkum, Opt. Lett. 25, 16-18 (2000).
[37] J. Ye, S. T. Cundiff, S. Foreman, T. M. Fortier, J. L. Hall, K. W. Holman,
D. J. Jones, J. D. Jost, H. C. Kapteyn, K. Leeuwen, L. S. Ma, M. M.
Murnane, J. L. Peng, and R. K. Shelton, Appl. Phys. B 74, S27-S34 (2002).
[38] T. M. Fortier, D. J. Jones, J. Ye, S. T. Cundiff, and R. S. Windeler, Opt.
Lett. 27, 1436-1438 (2002).
[39] J. C. Bergquist ed., Proceedings of the Fifth Symposium on Frequency
Standards and Metrology (Singapore, Woods Hole, Massachusetts, 1996).
[40] K. M. Evenson, J. S. Wells, F. R. Petersen, B. L. Danielson, and G. W.
Day, Appl. Phys. Lett. 22, 192 (1973).
[41] A. Clairon, B. Dahmani, A. Filimon, and J. Rutman, IEEE Trans. Instrum.
Meas. 34, 265-268 (1985).
[42] D. A. Jennings, K. M. Evenson, and D. J. E. Knight, Proc. IEEE 74, 168-
179 (1986); R. L. Barger and J. L. Hall, Appl. Phys. Lett. 22, 196-199
(1973).
[43] T. Udem, A. Huber, B. Gross, J. Reichert, M. Prevedelli, M. Weitz, and T.
W. Hänsch, Phys. Rev. Lett. 79, 2646-2649 (1997); C. Schwob, L.
Jozefowski, B. de Beauvoir, L. Hilico, F. Nez, L. Julien, F. Biraben, O.
Acef, and A. Clairon, Phys. Rev. Lett. 82, 4960-4963 (1999).
[44] A. Peters, K. Y. Chung, B. Young, J. Hensley, and S. Chu, Philos. Trans.
R. Soc. Lond. Ser. A 355, 2223-2233 (1997).
[45] B. de Beauvoir, C. Schwob, O. Acef, L. Jozefowski, L. Hilico, F. Nez, L.
Julien, A. Clairon, and F. Biraben, Euro. Phys. J. D 12, 61-93 (2000).
[46] D. Hils and J. L. Hall, Phys. Rev. Lett. 64, 1697-1700 (1990).
[47] S. A. Diddams, L. Hollberg, L. S. Ma, and L. Robertsson, Opt. Lett. 27, 58-
60 (2002).
[49] J. Ye, J. L. Hall, and S. A. Diddams, Opt. Lett. 25, 1675 (2000).
[50] J. Stenger, H. Schnatz, C. Tamm, and H. R. Telle, Phys. Rev. Lett. 88,
073601 (2002).
[51] M. Niering, R. Holzwarth, J. Reichert, P. Pokasov, T. Udem, M. Weitz, T.
W. Hänsch, P. Lemonde, G. Santarelli, M. Abgrall, P. Laurent, C. Salomon,
and A. Clairon, Phys. Rev. Lett. 84, 5496-5499 (2000).
[52] B. C. Young, F. C. Cruz, W. M. Itano, and J. C. Bergquist, Phys. Rev. Lett.
82, 3799-3802 (1999).
[53] J. Stenger, C. Tamm, N. Haverkamp, S. Weyers, and H. R. Telle, Opt. Lett.
26, 1589-1591 (2001).
52 Chapter 1
[54] S. G. Karshenboim, Laser Physics 11, 1083-1087 (2001).

[55] J. Ye, L. Robertsson, S. Picard, L. S. Ma, and J. L. Hall, IEEE Trans.
Instrum. Meas. 48, 544-549 (1999).
[56] R. J. Jones, W. Y. Cheng, K. W. Holman, L.-S. Chen, J. L. Hall, and J. Ye,
Appl. Phys. B 74, 597 (2002).
[57] W. Y. Cheng, L. S. Chen, T. H. Yoon, J. L. Hall, and J. Ye, Opt. Lett. 27,
571-573 (2002).
[58] C. Ishibashi, J. Ye, and J. L. Hall, in SPIE - Methods for ultrasensitive
detection II, edited by C. W. Wilkerson (SPIE, San Jose, 2002), Vol. SPIE
4634, p. 58.
[59] T. J. Quinn, Metrologia 36, 211-244 (1999).
[60] J. D. Jost, J. L. Hall, and J. Ye, Opt. Expr. 10, 515 (2002).
[61] J. E. Bjorkholm and P. F. Liao, Phys. Rev. Lett. 33, 128-131 (1974).
[62] O. Poulsen and N. I. Winstrup, Phys. Rev. Lett. 47, 1522-1525 (1981).
[63] M. J. Snadden, A. S. Bell, E. Riis, and A. I. Ferguson, Opt. Commun. 125,
70-76 (1996).
[64] T. H. Yoon, A. Marian, J. L. Hall, and J. Ye, Phys. Rev. A 63, 011402
(2000).
[65] L. S. Chen and J. Ye, Chem. Phys. Lett. 381, 777-783 (2003).
[66] L. S. Chen, W. Y. Cheng, and J. Ye, J. Opt. Soc. Am. B 21, 820-832
(2004).
[67] B. Bodermann, H. Knockel, and E. Tiemann, Eur. Phys. J. D 19, 31 (2002).
[68] R. J. Jones and J. Ye, Opt. Lett. 27, 1848-1850 (2002).
[69] E. O. Potma, D. J. Jones, J. X. Cheng, X. S. Xie, and J. Ye, Opt. Lett. 27,
1168-1170 (2002).
[70] P. Brumer and M. Shapiro, Chem. Phys. Lett. 126, 541-546 (1986).
[71] M. Shapiro and P. Brumer, Advances in Atomic Molecular, and Optical
Physics, Vol. 42 42, 287-345 (2000).
[72] Y.-Y. Yin, C. Chen, D. S. Elliot, and A. V. Smith, Phys. Rev. Lett. 69,
2353 (1992).
[73] D. Z. Anderson, N. B. Baranova, C. Green, and B. Y. Zeldovich, Zhurnal
Eksperimentalnoi I Teoreticheskoi Fiziki 102, 397-405 (1992).
[74] B. Sheehy, B. Walker, and L. F. Dimauro, Phys. Rev. Lett. 74, 4799-4802
(1995).
[75] A. Haché, Y. Kostoulas, R. Atanasov, J. L. P. Hughes, J. E. Sipe, and H. M.
van Driel, Phys. Rev. Lett. 78, 306-309 (1997).
[76] A. Haché, J. E. Sipe, and H. M. van Driel, IEEE Journal of Quantum
Electronics 34, 1144-1154 (1998).
[77] T. M. Fortier, P. A. Roos, D. J. Jones, S. T. Cundiff, R. D. R. Bhat, and J.
E. Sipe, Phys. Rev. Lett. 92, 147403 (2004).
[78] L.-S. Ma, R. K. Shelton, H. C. Kapteyn, M. M. Murnane, and J. Ye, Phys.
Rev. A 64, 021802 (2001).
[79] A. Leitenstorfer, C. Furst, and A. Laubereau, Opt. Lett. 20, 916-918
(1995); Z. Y. Wei, Y. Kobayashi, Z. G. Zhang, and K. Torizuka, Opt. Lett.
26, 1806-1808 (2001).
1. INTRODUCTION 53
[80] A. Bartels, S. A. Diddams, T. M. Ramond, and L. Hollberg, Opt. Lett. 28,

663-665 (2003).
[81] T. R. Schibli, J. Kim, O. Kuzucu, J. T. Gopinath, S. N. Tandon, G. S.
Petrich, L. A. Kolodziejski, J. G. Fujimoto, E. P. Ippen, and F. X. Kaertner,
Opt. Lett. 28, 947-949 (2003).
[82] R. K. Shelton, S. M. Foreman, L. S. Ma, J. L. Hall, H. C. Kapteyn, M. M.
Murnane, M. Notcutt, and J. Ye, Opt. Lett. 27, 312-314 (2002).
[83] R. A. Kaindl, M. Wurm, K. Reimann, P. Hamm, A. M. Weiner, and M.
Woerner, J. Opt. Soc. Am. B 17, 2086-2094 (2000).
[84] R. W. Schoenlein, W. P. Leemans, A. H. Chin, P. Volfbeyn, T. E. Glover,
P. Balling, M. Zolotorev, K. J. Kim, S. Chattopadhyay, and C. V. Shank,
Science 274, 236-238 (1996).
[85] S. M. Foreman, D. J. Jones, and J. Ye, Opt. Lett. 28, 370-372 (2003).
Ye, J. Mod. Opt. 49, 401-409 (2002).
[87] J. Rauschenberger, T. M. Fortier, D. J. Jones, J. Ye, and S. T. Cundiff, Opt.
Expr. 10, 1404-1410 (2002); D. J. Jones, K. W. Holman, M. Notcutt, J. Ye,
J. Chandalia, L. A. Jiang, E. P. Ippen, and H. Yokoyama, Opt. Lett. 28,
813-815 (2003).
[88] K. W. Holman, D. J. Jones, J. Ye, and E. P. Ippen, Opt. Lett. 28, 2405-2407
(2003).
[89] K. W. Holman, D. J. Jones, D. D. Hudson, and J. Ye, Opt. Lett. 29, 1554-
1556 (2004).
[90] R. J. Jones, I. Thomann, and J. Ye, Phys. Rev. A 69 (2004).
[91] E. O. Potma, C. Evans, X. S. Xie, R. J. Jones, and J. Ye, Opt. Lett. 28,
1835-1837 (2003).
[92] R. J. Jones and J. Ye, Opt. Lett., in press (2004).
[93] A. Volkmer, J. X. Cheng, and X. S. Xie, Phys. Rev. Lett. 87, 023901
(2001).
[94] L. D. Book, J. X. Cheng, A. Volkmer, and X. S. Xie, Biophys. J. 80, 171A-
171A (2001).
[95] D. J. Jones, E. O. Potma, J. X. Cheng, B. Burfeindt, Y. Pang, J. Ye, and X.
S. Xie, Rev. Sci. Instrum. 73, 2843-2848 (2002).
[96] H. Rabitz, R. de Vivie-Riedle, M. Motzkus, and K. Kompa, Science 288,
824-828 (2000).
[97] E. O. Potma, X. S. Xie, L. Muntean, J. Preusser, D. Jones, J. Ye, S. R.
Leone, W. D. Hinsberg, and W. Schade, J. Phys. Chem. B 108, 1296-1301
(2004).
[98] C. G. Durfee, A. R. Rundquist, S. Backus, C. Herne, M. M. Murnane, and
H. C. Kapteyn, Phys. Rev. Lett. 83, 2187-2190 (1999); C. Spielmann, N.
H. Burnett, S. Sartania, R. Koppitsch, M. Schnurer, C. Kan, M. Lenzner, P.
Wobrauschek, and F. Krausz, Science 278, 661-664 (1997).
[99] G. G. Paulus, F. Grasbon, H. Walther, P. Villoresi, M. Nisoli, S. Stagira, E.
Priori, and S. De Silvestri, Nature 414, 182-184 (2001).
Chapter 2
FEMTOSECOND LASER DEVELOPMENT
Franz X. Kärtner1, Erich P. Ippen1 and Steven T. Cundiff2

1
Research Laboratory of Electronics, Massachusetts Institue of Technology
2
Abstract: Advances in ultrashort-pulse laser technology have led to few-cycle-duration
optical pulses, with octave-spanning spectra, generated directly from laser
oscillators at high repetition rates. These new laser systems promise to be
reliable sources of carrier-envelope–phase-stabilized femtosecond pulse trains
and highly stable laser frequency combs. In this chapter, we review the pulse
and carrier-envelope phase dynamics of these laser sources and discuss
different laser designs based on prismless cavities and cavities including prism
pairs for dispersion compensation.
Key words: mode-locked lasers, carrier-envelope phase, frequency comb
1. INTRODUCTION
Remarkable progress in the generation of femtosecond pulses with solid-

state lasers has followed from the discovery of self-mode-locking in a
Ti:sapphire laser by the Sibbett group in 1991 [1]. Explained as a
consequence of self-focusing inside the laser [2], this self-mode-locking
behavior has become known as Kerr-lens mode locking (KLM). It is now the
basis for femtosecond pulse generation in a wide variety of other solid-state
laser systems as well. As shown in Figure 2-1, KLM has resulted in the
shortest pulses, superceding previous marks set using dye lasers,
amplification, and spectral broadening.
Self-focusing in the KLM laser in the presence of aperturing due either to
the size of the gain spot or some other physically introduced aperture can
cause pulse shortening in precisely the manner described by "fast saturable
absorber" mode-locking theory [3]. In addition, simultaneously occurring
self-phase modulation (SPM), in the presence of group-velocity dispersion
(GVD) in the resonator, provides a strong soliton like shaping to the pulses.
This latter process can, in fact, be the dominant pulse-shaping effect;
2. FEMTOSECOND LASER DEVELOPMENT 55
however, the KLM remains necessary to suppress the growth of noise

between the pulses, thereby stabilizing the mode locking. An important
variant of soliton shaping occurs when the GVD alternates between positive
and negative values as the pulse propagates. The result can be the formation
of "dispersion-managed solitons" even when the average GVD in the
resonator is zero or net normal [4]. Such dispersion management has found
important application in "stretched-pulse" fiber lasers [5] and may also play
a role in very short pulse Ti:sapphire lasers [6].
10ps
Nd:glass
Nd:YLF
Nd:YAG
S-P Dye Diode

Dye
SHORTEST PULSE DURATION
1ps
SOLID-STATE
CW Dye SYSTEMS
Color
Center
100fs
CPM Dye Er:fiber
DYE LASER Cr:YAG
Nd:fiber
TECHNOLOGY
Cr:forsterite
10fs
Cr:LiS(G)AF
w/Compression Ti:sapphire
1965 1970 1975 1980 1985 1990 1995 2000 2005

YEAR
Figure 2-1. History of ultrashort pulse duration.
Ultrashort pulse durations have been reduced to well below 10 fs by a

variety of researchers including the university groups at Washington State
University [7], the Technical Universiy of Vienna [8], ETH-Zurich [9], the
Massachusetts Institute of Technology (MIT) [10], and the University of
Karlsruhe [11]. The Washington State group, responsible for a number of
successful early cavity designs, first documented the limitations to pulse
shortening by higher-order dispersion. The TU-Vienna group introduced
chirped mirrors for dispersion compensation, an approach that was further
refined at ETH-Zurich with the concept of a double-chirped mirror design.
Dispersion compensation over one octave, ultimately achieved by the use of
double-chirped mirror pairs at MIT and the University of Karlsruhe, finally
56 Chapter 2
led to the generation of 5 fs pulses containing less than two optical cycles.
The wings of the spectrum of these pulses extend over more than one octave,
an achievement that has important implications for the control of the electric
field waveform of the pulse. It had been recognized earlier by Xu et al.[12]
that for few-cycle pulses, the maximum electric field in a pulse
quantitatively depends on the phase relationship between the carrier wave
and the maximum of the envelope (see Figure 2-1).
Nonlinear optical processes, which depend on powers of the electric field
amplitude, must therefore also depend on the carrier-envelope phase for very
short pulses. This dependence is visualized in Figure 2-2, which shows
sketches of spectra resulting from instantaneous χ (2) and χ (3) processes
produced by a chirp-free pulse with a rectangular-shaped spectrum spanning
an octave. For a one-octave pulse, the spectra of the second-harmonic and
optical-rectification terms overlap with the fundamental spectrum.
Interference in the overlap regions will be constructive or destructive
depending on the carrier-envelope phase. Use of this interference for carrier-
envelope phase control is called the ν-to-2ν self-heterodyne technique. If
higher-order processes are employed, such as a χ (3) nonlinearity, there are
similar interference terms between the χ (2) and χ (3) terms. Thus a phase-
dependent interference is already possible even if the pulse spans only 2/3 of
an octave. All of these interferences have been used to detect and control the
carrier-envelope phase.
Fundamental
eiφ
O
OR SHG χ(2)
e2iφO
eiφ O
SPM
THG χ(3)
e3iφO
0 ω0 2ω0 3ω0 ω
Figure 2-2. Top: Fundamental spectrum of an octave-spanning pulse centered at ω0; Below:
Spectra generated by instantaneous second- and third-order nonlinearities due to optical
rectification (OR), second-harmonic generation (SHG), self-phase modulation (SPM), and
third-harmonic generation (THG).
A special case, pointed out by Baltuška et al. [13], is that optical

rectification or difference-frequency mixing of an optical pulse directly
produces a phase-independent IR pulse. Currently, control of the carrier-
envelope phase for a variety of applications is one of the strongest

motivations for the development of ultrabroadband lasers.
This chapter reviews progress made in phase-stabilized–octave-spanning
lasers, starting with a discussion of the dispersion-managed pulse dynamics
in sub-10 fs Ti:sapphire lasers in Section 2. In Section 3, the experimental
realization of systems using combinations of prism pairs and chirped mirrors
for dispersion compensation and prismless systems are described. Section 4
discusses the carrier-envelope phase dynamics of some sources. The results
achieved are summarized in Section 5.
2. PULSE DYNAMICS
The generation of ultrashort pulses from Ti:sapphire lasers has

progressed rapidly over the past decade and led to the generation of pulses as
short as 5 fs directly from the laser. The spectra generated by these lasers are
not of simple shape [7, 11, 14], and various models to explain this
complexity have been presented. Christov et al. [15] have shown by
numerical simulations that, when the second-order group-velocity dispersion
is balanced, fourth-order dispersion with gain bandwidth filtering and KLM
action can lead to steady-state pulse generation. Extensive three-dimensional
simulations have been carried out [15-17]. However, so far, a fully three-
dimensional numerical simulation of a KLM laser is still too time consuming
for computation of the steady-state solution. Also, the solution is known
from experimental observation to be very sensitive to changes in the self-
focusing dynamics caused by slight parameter variations. This sensitivity
adds to the difficulty of obtaining meaningful information from steady-state
simulations that require many hours of computer time.
A schematic setup of current 5 fs Ti:sapphire lasers using only chirped
mirrors for dispersion compensation is shown in Figure 2-3(a). For insight
into the dynamics of this laser, we start by considering only the temporal
dynamics, which are modeled by the nonlinear Schrödinger equation
∂ ∂2
a = −iD (z ) 2 a + iδ (z ) a a ,
2
(1)
∂z ∂t
describing propagation of the envelope function a in a medium with

distributed dispersion, D(z), and self-phase modulation, δ(z). The
distribution of dispersion and nonlinearity in the laser is sketched in Figure
2-3(b). It was first pointed out by Spielmann et al. that for very-short-pulse
Ti:sapphire lasers, large changes in the pulse occur within one round trip and
58 Chapter 2
that the ordering of the pulse-shaping elements within the cavity has a major
effect on the pulse formation [18]. The discrete action of linear dispersion in
the arms of the laser resonator and the discrete, but simultaneous, action of
positive SPM and positive GDD in the laser crystal cannot be neglected any
longer. The importance of the strong variations in dispersion was first
discovered in a fiber laser and called stretched-pulse mode locking [19]. The
positive dispersion in the Er-doped fiber section of a fiber ring laser was
balanced by a negative-dispersion passive fiber. The pulse circulating in the
ring was stretched and compressed by as much as a factor of 20 in one round
trip. One consequence of this behavior was a dramatic decrease of the
nonlinearity and thus increased stability against the SPM-induced
instabilities. Also, the formation of Kelly sidebands [20] occurring for large
nonlinear phase shifts per round trip in conventional soliton lasers was
greatly suppressed. The energy of the output pulses could be increased a
hundredfold [21]. These important consequences can explain the high power
densities that can be achieved in sub-10 fs KLM lasers without undergoing
catastrophic self-focusing in the laser crystal.
a) Silver
Double-Chirped Mirror I
Mirror 1mm BaF2
Laser crystal:
2mm Ti:Al 2O3
PUMP
L = 20 cm
Double-Chirped Mirror II
OC 2
BaF2 - wedges
Base Length = 30cm for 82 MHz Laser
b)
D D
- D, SPM -
2 2
Figure 2-3. (a) Schematic of a Kerr-lens–mode-locked Ti:sapphire laser using double-chirped

mirrors and BaF2 for dispersion compensation. (b) Distribution of dispersion and nonlinearity
inside the laser cavity.
Here, we consider the impact of the discrete action of dispersion and

nonlinearity, which become important in the sub-10 fs range, in Ti:sapphire
lasers on the spectral shape of the laser pulses. The dispersion distribution
shown in Figure 2-3(b) suggests an analogy with pulse propagation along a
dispersion-managed fiber transmission link [4]. A system with sufficient
variation of dispersion can support waves that have been called nonlinear
Bloch waves [22]. One can show that the Kerr nonlinearity produces a self-
consistent nonlinear scattering potential that permits formation of a periodic
solution with a simple phase factor in a system with zero net dispersion. It
has been shown further that nonlinear propagation along dispersion-managed
fiber near zero net GDD possesses a narrower spectrum in the segment of
positive dispersion than in the segment of negative dispersion [4, 22]. Thus,
the effect of negative dispersion is greater than that of the positive
dispersion, and the pulse sees an effective net negative dispersion. This
effective negative dispersion can balance the Kerr-induced phase, leading to
steady-state pulses at zero net dispersion. This is true even when there is no
nonlinearity in the negative dispersion segment. The pulses are analogous to
solitons in that they are self-consistent solutions of the Hamiltonian
(lossless) problem as are conventional solitons. But, they are not secant
hyperbolic in shape. Figure 2-4 shows a numerical simulation of a self-
consistent solution of the Hamiltonian pulse-propagation problem in a linear
medium of negative dispersion and subsequent propagation in a nonlinear
medium of positive dispersion with positive self-phase modulation following
Equation (1).
The dispersion map D(z) is shown as an inset in Figure 2-5. The
dispersion coefficient D(z) and the nonlinear coefficient δ(z) are defined per
unit length. In Figure 2-4, the steady-state intensity profiles are shown at the
center of the negative dispersion segment over 1000 round trips. In addition,
we can include in the model the saturable gain, Lorentzian gain filtering, and
KLM modeled by a fast saturable absorber. Figure 2-5 shows the behavior in
one period (one round trip through the resonator) including these effects.
( )
The response of the absorber is q(a ) = q0 1 + a PA , with q0 = 0.01/mm
2
and PA = 1 MW. The bandwidth-limited gain is modeled by a Lorentzian

gain profile with bandwidth Ω g = 2π × 43 THz. The filtering and saturable
absorber reduce the spectral and temporal side lobes of the Hamiltonian
problem, respectively, as can be inferred from Figures 2-4 and 2-5. The
steady-state pulse formation can be understood in the following way: By
symmetry, the pulses are chirp free in the middle of the dispersion cells. A
chirp-free pulse starting in the center of the gain crystal, i.e., nonlinear
segment, is spectrally broadened by the SPM and disperses in time because
of the GVD, which generates a mostly linear chirp over the pulse.
60 Chapter 2
Figure 2-4. Simulation of the Hamiltonian problem. Temporal (a) and spectral (b) intensity
profiles at the center of the negatively dispersive segment are shown for successive round
trips. The total extent in 1000 round trips. D = D (± ) = ±60 fs/mm for the segment of crystal
length L = 2 mm, τ FWHM = 5.5 fs, δ = 0 for D < 0, δ = 1 (MW-mm) for D > 0 .
Figure 2-5. Pulse shaping in one round trip. The negative segment has no nonlinearity.
After the pulse leaves the crystal, it experiences GVD in the arms of the
laser resonator. The positively chirped pulse is compressed to its transform
limit at the end of each arm, where output couplers can be placed.
Propagation back towards the crystal imposes a negative chirp, generating
the time-reversed solution of the nonlinear Schrödinger Equation (1).
Therefore, subsequent propagation in the nonlinear crystal compresses the
pulse spectrally and temporally to its initial shape in the center of the crystal.
The spectrum is narrower in the crystal than in the negative-dispersion
sections, because it is negatively prechirped before it enters the SPM section.
Spectral spreading occurs again only after the pulse has been compressed.
This result further explains that in a laser with a linear cavity, for which the
negative dispersion is located in only one arm of the laser resonator (i.e., in
the prism pair and no use of chirped mirrors), the spectrum is widest in the
arm that contains the negative dispersion [18]. In a laser with a linear cavity,
for which the negative dispersion is equally distributed in both arms of the
cavity, the pulse runs through the dispersion map twice per round trip. The
pulse is short at each end of the cavity, and, most importantly, the pulses are
identical in all passes through the crystal; thus they can exploit the full KLM
action twice per round trip [23]. For this reason, a symmetric dispersion
distribution may lead to an effective saturable absorption that is twice as
strong as an asymmetric dispersion distribution and that can produce
substantially shorter pulses. Furthermore, the dispersion swing between the
negative and positive dispersion sections is only half as large, which allows
62 Chapter 2
for shorter dispersion-managed solitons operating at the same average power

level.
Figure 2-6. Energy of the pulse in the lossless dispersion-managed system with stretching
S = LD τ FWHM
2
or, for a fixed crystal length L and pulse width as parameters;
D = 60 fs2/mm for Ti:sapphire at 800 mm.
A major difference from conventional soliton formation is illustrated in

Figure 2-6. The figure shows the parameter ranges for a dispersion-managed
Hamiltonian system (no gain, no loss, no filtering) that is unbalanced by the
net dispersion that serves as the abscissa of the figure. Each curve gives the
locus of energy vs net cavity dispersion for a stretching ratio S = LD τ FWHM
2
(or pulse width with fixed crystal length L). One can see that for a pulse
width longer than 8 fs with crystal length L = 2 mm, no solution of finite
energy exists in the dispersion-managed system for zero or positive net
dispersion. Pulses of durations longer than 8 fs require net negative
dispersion. Hence, one can reach the ultrashort operation at zero net
dispersion only by first providing the system with negative dispersion. At the
same energy, one can form a shorter pulse by reducing the net dispersion,
provided that the 8 fs threshold has been passed. For a fixed-dispersion
swing ± D, the stretching increases quadratically with the spectral width or
the inverse pulse width. Long pulses with no stretching have a sech shape.
For stretching ratios of 3–10, the pulses are Gaussian shaped. For even larger
stretching ratios, the pulse spectra become increasingly more flat topped, as
shown in Figure 2-4.
These dynamics explain why there are steady-state pulses even at zero
and slightly positive linear intracavity dispersion as confirmed in the
experiment. These pulses would, of course, be unstable in the presence of
gain filtering due to the finite gain bandwidth of Ti:sapphire and, therefore,
there is a need for strong-saturable-absorber action to keep the pulse stable
against continuum radiation exploiting the peak of the gain. This saturable-
absorber action is due to the Kerr-lens effect and can be optimized by
various techniques [23, 24], which shall not be repeated here. Also, the
spectral shape of the output pulses is greatly distorted because of the finite
bandwidth of the output coupler, which can greatly enhance the wings of the
laser spectrum, as we will see in the next section.
3. OCTAVE-SPANNING LASERS
Although an octave of spectrum can be generated by spectral broadening

in microstructure fiber, there are advantages to using a laser that generates an
octave directly. Since there is no medium with a gain spectrum that spans an
octave (for Ti:sapphire, it approaches a half-octave), an octave-spanning
laser clearly must rely on nonlinear broadening inside the laser cavity. The
first demonstration of an octave-spanning laser used a second time-and-
space focus in a separate glass element to enhance the nonlinearity [11].
More recently, there have been several demonstrations of Kerr-lens–mode-
locked Ti:sapphire lasers that use the Ti:sapphire alone as the nonlinear
medium. In this section, we discuss an implementation that uses intracavity
prisms for dispersion compensation [25] and one that only uses dispersion-
compensating mirrors [26].
The primary advantage of using an octave-spanning laser is that, by
eliminating the microstructure fiber [27], issues of coupling and damage are
also eliminated. Since the microstructure fiber has a very small core, the
coupling is very alignment sensitive and tends to degrade with time. This
poses a limit of approximately 10 hours on the duration of experiments.
Although long enough for many applications, a proper clock needs to be able
to run for days or longer. Damage to the microstructure fiber also limits the
durations of experiments. However, there is significant variability as to how
long it takes for damage to occur. Under some conditions, the fiber only lasts
for hours, while under others it seems to last for weeks.
For time-domain experiments that are affected by carrier-envelope phase,
it is important to maintain a short (transform-limited) pulse. The large
higher-order dispersion in the fiber coupled with high nonlinearity
essentially makes this impossible, whereas with an octave-spanning laser, it
is possible.
With these advantages also come some disadvantages. The very high
intracavity peak powers needed to obtain sufficient nonlinear spectral
broadening makes these lasers susceptible to damaging the Ti:sapphire
64 Chapter 2
crystal. This damage takes some time to occur but can limit the duration of
experiments. In addition, the output spatial mode exhibits strong frequency
dependence, which can also impact experiments.
Intracavity continuum generation has been demonstrated in a Ti:sapphire
laser that was operated in the self-Q-switching regime [28]. However, in this
regime, the spectrum is unstable and therefore not useful for femtosecond
comb applications. As described in Chapter 3, intracavity continuum
generation has also been observed in high-repetition-rate lasers [29], but
there was insufficient power at the octave points to implement ν-to-2ν
stabilization. A 2ν-to-3ν scheme had to be used [30].
Finally, we would like to comment on the definition of “octave
spanning.” The span of a spectrum is often taken as the width at some power
below the peak (often full-width at half-maximum or perhaps even at the 10
dB points). However, for femtosecond comb applications, there is a good
operational definition of octave spanning, namely that it is possible to obtain
ν-to-2ν beats. A slightly stronger version of this is that the beats are
sufficiently strong to be used to stabilize the offset frequency of the laser.
This criterion can be met even when the intensity at the octave points is as
much as 40 dB below the peak.
3.1 Octave-spanning laser using prisms
The early designs for broadband Kerr-lens–mode-locked Ti:sapphire

lasers [31] used a two-prism sequence for dispersion compensation [32].
Some mechanism for generating anomalous dispersion is needed to correct
for the chirp acquired by the pulse as it passes through the normal dispersion
Ti:sapphire crystal (and air for extreme bandwidth). The development of
dispersion-compensating mirrors (also called “chirped mirrors”) [33], which
have anomalous dispersion, meant that broadband mode locking could be
achieved without prisms [34]. Dispersion-compensating mirrors not only
provide anomalous dispersion, but they also have a much larger reflection
bandwidth than standard dielectric mirrors. Incorporating both prisms and
dispersion-compensating mirrors added additional flexibility in the design
and enabled the production of sub-two-cycle pulses [10]. In this section, an
octave-spanning laser that incorporates both prisms and dispersion-
compensating mirrors is presented. To meet the operational definition of
“octave spanning,” we show that clean ν-to-2ν beats can be obtained and
used to lock the offset frequency of the laser comb.
3.1.1 Design
The prism-based–octave-spanning Ti:sapphire laser is an x-folded cavity

that uses commercially available negatively chirped mirrors and intracavity
CaF2 prisms for dispersion compensation, as shown in Figure 2-7. M1 and
M2 (10 cm radius of curvature) are both double-chirped mirrors (DCM)
designed to be used with CaF2 prisms [10]. M3 and EM are both negatively
chirped LayertecTM mirrors,* part numbers 100466 and 101515, respectively.
The output coupler (OC) is G034-007 from Spectra Physics* and the prism
separation between the two CaF2 prisms is ~82 cm. The laser crystal is 2.3
mm long and the pump focusing lens has a focal length of 10 cm. When the
laser is pumped with 5.5 W of 532 nm light, the spectrum, at its broadest,
extends from ~570 to ~1280 nm (-35 dB from the 800 nm portion of the
spectrum). The average ML power is > 400 mW (100 mW cw) at an 88 MHz
repetition rate. Mode locking is easily initiated by quickly changing the
insertion of the prism furthest from EM.
OC
EM
P2 P1
PUMP
M3
M1 M2
BBO
filter
(580 nm)
polarizer fo locking
PMT electronics
Figure 2-7. Schematic of laser and ν-to-2ν detection of f0.
The continuum generated by the laser relies on the interplay between an

enhancement of SPM and dispersion compensation, where the latter allows
for the existence of the former. The exact choice of chirped mirrors, the
prism separation, and the prism insertion appear not to be very important.
Different combinations of mirrors for M1 and M2 (Layertec 101568), M3
(DCM, Layertec 100466 and 101515) and EM (Layertec 100466 and
*
This information is given for technical purposes only and does not represent an endorsement
on the part of the National Institute of Standards and Technology.
66 Chapter 2
101515), and a protected silver mirror were tested and yielded octave-
spanning spectra. Also, two OC’s (Spectra Physics G034-007 and CVI Laser
PR2 — 2% transmission @ 800nm) resulted in an octave of bandwidth. We
have not been able to reproduce the octave spectrum using fused silica
prisms. Apparently, in this case, the continuum generation does not result
from meticulous dispersion compensation. This is not surprising since the
continuum is created on a single pass through the crystal. As a result, careful
dispersion compensation may only be necessary over the bandwidth
indicated by the OC reflectivity and not the entire spectrum.
The position of the curved mirrors is critical for producing an octave-
spanning spectrum. Typically, KLM is obtained by displacing one curved
mirror away from the optimum mirror position for cw operation. Here, the
mirror furthest from the pump is translated inward to the point that the laser
is on the edge of stability for cw operation. This arrangement increases the
discrimination between mode-locked and cw operation and requires the
formation of a strong Kerr-lens to correct for the misalignment with respect
to the pump. To optimize the bandwidth, both curved mirrors are then
translated in the same direction away from the pump, just before the point
where the laser begins Q-switching.
3.1.2 Carrier-envelope phase stabilization
ν-to-2ν beats
fr
30 dB
Frequency
Figure 2-8. Rf spectrum showing a repetition rate peak and ν-to-2ν heterodyne beats.
To measure and stabilize the carrier-envelope phase in this laser, we

implement the simplest self-referencing scheme, which compares the second
harmonic on the low frequency extreme of the laser spectrum with the
fundamental on the extreme high end of the spectrum (see Chapter 2). A ν-
to-2ν interferometer is used to measure f0. Specifically, f0 is obtained via
optical heterodyne between the fundamental light of the laser spectrum at
~580 nm and doubled light at ~1160 nm, see Figure 2-7. For the
measurements presented here, the signal-to-noise ratio of the offset-

frequency beat note, using the octave spectrum (see Figure 2-8), is greater
than 30 dB at 100 kHz resolution bandwidth. Once measured, f0 is then
stabilized by actively tilting the end mirror (EM) [35] with an in-loop-
accumulated phase error of 0.18 rad (1.65 mHz–102kHz), as shown in
Figure 2-8. Because this is an in-loop measurement of the phase noise, it
therefore may not be used alone to characterize the noise on the output of the
laser. However, using the octave spectrum from the laser directly, we have
eliminated the out-of-loop phase noise generated by microstructure fiber
[36]. Also, because the prism-based ν-to-2ν interferometer used in the
measurement of f0 minimizes non-common-mode mirrors, the dominant
source of out-of-loop phase noise on the laser results from the feedback
electronics.
Observation Time (s)
3 2 1 0 -1 -2 -3 -4 -5
10 10 10 10 10 10 10 10 10
Accumulated phase (rad)

unlocked
Phase noise (rad2/ Hz)
0.30
5
10 0.25
0.20
0
10
0.15
-5 locked 0.10
10
0.05
0.00
-3 -2 -1 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10
Frequency (Hz)
Figure 2-9. Power spectral density of phase noise (left axis) and accumulated rms phase
fluctuations (right axis).
3.1.3 Frequency-dependent spatial mode
Since the light in the wings of the spectrum is generated by self-phase

modulation and is outside the gain bandwidth, it is not truly lasing and does
not have to obey resonance conditions in the cavity. This means the spatial
mode in the wings can vary significantly. This variation is relevant to f0
detection as it can degrade the spatial-mode matching and thus limit the
signal-to-noise ratio of the f0 beat note. It may also pose problems to pulse
compression. This mismatch in mode profiles is depicted Figure 2-10. Figure
2-11 shows the result of a 90/10 knife-edge method that was used to perform
a second moment measurement of the beam profiles to yield the beam spot
sizes as a function of wavelength [37]. From this result, we see that the
spectral extremes exhibit a significant increase in asymmetry and deviation
68 Chapter 2
away from the resonant Gaussian TEM00 mode and its corresponding 1/λ
diffraction limit. G
587nm 612nm 664nm 690nm
587 nm 612 nm 664 nm 690 nm

779nm 850nm 1000nm
720 nm 779 nm 850 nm 1000 nm

1080nm 1100nm 1151nm
1080 nm 1100 nm 1120 nm 1051 nm
Figure 2-10. Beam profiles at various wavelengths.
Laser Spectrum (Arb. Units)

10
4 10
-1
Spot Size (mm)
sagittal plane
tangential plane -2
3 10
-3
10
2
-4
10
1 10
-5
600 800 1000 1200

Wavelength (nm)
Figure 2-11. Measured sagittal and tangential spot sizes as a function of wavelength (squares,
left axis). Line shows laser spectrum (right axis).
Measurement of the beam parameters at the output coupler, as shown in

Figure 2-12, explains the increase in spot size in the wings of the spectrum.
The beam waist shows a sudden decrease of the nonresonant as compared to
the resonant modes. This decrease should cause a greater divergence in the
wings of the laser spectrum, thereby leading to the larger spots sizes
observed in the far field (see Figure 2-10). Figure 2-13 compares the
measured beam divergence to that which would result from the propagation
of a Gaussian TEM00 mode given the respective measured beam waists in
Figure 2-12. The ratio of the former to the latter gives the M2 value as a
function of wavelength, indicating the strength of the non-TEM00-Gaussian
beam propagation (non-Gaussian modes and/or higher modes).
Waist Size (102 µm)

7
OC Transmission (%)
60
6
5 40
4
20
3
2
0
600 700 800 900 1000 1100
Wavelength (nm)
Figure 2-12. Waist size at output coupler as a function of wavelength for sagittal and
tangential planes (squares, left axis). Output coupler transmission is line (right axis).
Beam Divergence (mrad)
10
sagittal plane
tangential plane
8
600 700 800 900 1000 1100

Wavelength (nm)
Figure 2-13. Measurement of beam divergence (squares) and predicted divergence based on
waist size (lines).
The mode profiles and the sudden decrease in waist in the extremes of
the spectrum still need to be explained. For resonant modes, the cavity
geometry determines the waist size and divergence. The beam characteristics
of the nonresonant modes, however, can be directly influenced by the Kerr
lens in the crystal and by diffraction and aperturing of optics without the
self-correcting effects of the cavity. The Kerr lensing is considerable since
intracavity continuum generation, for this laser, requires significant
realignment away from ideal cw operation. This realignment results in a
highly asymmetric cw beam (tangential waist/sagittal waist = 1.56) and a
large discrimination in average output power between cw and mode-locked
operation (4:1). The asymmetry implies that the nonlinear correction for the
cw asymmetry may result in the formation of an asymmetric Kerr lens,
which may explain the observed asymmetry in the wings of the spectrum.
This mirror positioning also results in a different initial cw condition than for
70 Chapter 2
“normal” KLM because it gives rise to a much more tightly focused cw

beam than the pump beam (~ ratio of 4:1, calculated using ABCD cavity
propagation based on the observed cw asymmetry). This leads us to the
conclusion that the effect of the Kerr lens should be one that increases the
waist of the mode-locked beam. The increased waist size improves the
efficiency with which power is extracted from the pump beam and thus still
acts as an effective saturable absorber. This is consistent with the observed
beam waists (Figure 2-12), as the frequency at the very edges of the
spectrum should be produced by a weaker Kerr lens than those of the
resonant modes [37]. Complete modeling of the relationship between
nonlinear effects in the crystal and pulse formation, however, is beyond the
scope of this chapter but has been addressed elsewhere [16].
3.2 Prismless octave-spanning laser
We have also built octave-spanning lasers where dispersion

compensation is completely achieved by double-chirped mirror pairs [38]
and thin BaF2 wedges for fine adjustment of the dispersion [26, 39].
Prismless lasers do have the advantage of being scalable to GHz repetition
rates because of the compact dispersion compensation via chirped mirrors.
Higher repetition rates mean higher power per mode at the same average
output power, which is advantageous for frequency metrology. Also, the
compactness of the cavity makes it less sensitive to temperature variations
and environmental perturbations.
3.2.1 Design
The lasers consist of astigmatically compensated x- or z- folded cavities.

The x-folded version is shown in Figure 2-14(a). The lasers have a 2 mm
long Ti:sapphire crystal with an absorption of α =7 cm-1 at 532 nm. They are
pumped by a diode-pumped, frequency-doubled Nd:vanadate laser. The
radius of curvature (ROC) of the folding mirrors is 10 cm, and the pump lens
has a 60 mm focal length. All mirrors in the cavity, with the exception of the
end mirrors, are type I (gray) and type II (black) double-chirped mirrors
(DCM) that generate smooth group-delay dispersion when used together in
pairs [38]. Figure 2-9(b) shows the reflectivity and group delay of the DCM
pairs. The total group delay is smooth from 600 to 1200 nm. The average
reflectivity over the full octave can be as high as 99.9%. One cavity end
mirror is a silver mirror and the other is a broadband output coupler, made of
either ZnSe/MgF2 (80 MHz repetition-rate laser) or TiO2/SiO2 (150 MHz
repetition-rate laser) coatings with 1% transmission. The output power in cw
operation is typically 40 mW with 4.4 W pump power. In mode-locked
operation, the average power is 90 mW. The output power is limited by low
output coupling and scattering losses in the OC coating and the silver mirror.
Presumably, further optimization of output coupling and elimination of the
silver mirror would increase laser efficiency.
a) b)
PUMP 1.0 100
ϕ 0.8 80
Group Delay (fs)

Reflectivity
θ 0.6 60
1mm BaF 0.4 40
OC
0.2 20
Silver 0.0 0
BaF Wedges Mirror 600 800 1000 1200
Wavelength, nm
Figure 2-14. (a) Schematic diagram of octave-spanning prismless laser. Gray and black
mirrors are type I and II double-chirped mirrors (DCMs), respectively. The BaF2 wedges are
used for fine-tuning of the dispersion. (b) Reflectivity (left scale) of the type I DCMs with
pump window shown as thick solid line. The group-delay design is given by the thick, dash-
dotted line. The individual group delays (right scale) of type I and II DCMs are shown as thin
lines and the average as a dashed line (almost identical to the design goal over the wavelength
range of interest from 650–1200 nm). The measured group delay, using white-light
interferometry, is shown as the thick solid line from 600–1100 nm. Beyond 1100 nm, the
sensitivity of the Si detector limited the measurement.
In one round trip of the laser pulse through the cavity, the total number of
bounces on DCMs (12 bounces) generates the precise negative dispersion
required to compensate for positive second- and higher-order dispersion
caused by the laser crystal, the air path in the cavity, and the BaF2 wedge
pairs used to fine-tune the intracavity dispersion.
We used BaF2 for dispersion compensation because it has the lowest ratio
of third- to second-order dispersion in the wavelength range from 600–1200
nm, and the slope of the dispersion of BaF2 is nearly identical to that of air.
These features make it possible to scale the cavity length and repetition rate
without changing the overall intracavity dispersion. To achieve mode-locked
operation, it is necessary to reduce the amount of BaF2 inside the laser cavity
(by withdrawing one of the wedges).
The broadest spectrum can be achieved by optimizing the insertion of the
BaF2 wedge, so that the laser operates very close to the zero average-
dispersion point. In this case, the spectral width of the laser is critically
dependent on the dispersion balance. With the prismless lasers, adjusting the
dispersion does not significantly change the cavity alignment. Figures 2-
15(a) and (b) show the spectrum under broadband operation of an 80 MHz
and a 150 MHz laser, respectively. The octave is already reached at a
72 Chapter 2
spectral density about 25 dB below the average power level. The same plots
also show the corresponding OC transmission curves. The detailed shape of
this transmission curve is a determining factor in the width of the output
spectrum, since it significantly enhances the spectral wings. A roll-off of the
output coupler matched to the intracavity spectrum is necessary to generate
the octave-spanning spectrum.
a)
Power Spectrum (dB)
OC Transmission (%)
80
-10
-20 60
-30 40
-40 20
-50
600 800 1000 1200
Wavelength (nm)
Figure 2-15. (a) Measured output spectra for lasers with 80 MHz repetition rate and (b) 150
MHz repetition rate. In both cases, the octave is reached at approximately 25 dB below the
average power level. The average mode-locked power is 90 mW for both lasers. Also shown
(dashed lines) are the respective output coupler (OC) transmission curves for the ZnSe/MgF2
Bragg-stack (a) and a broadband SiO2/TiO2 stack (b). Both have about 1% transmission at
800 nm.
3.2.2 Carrier-envelope phase stabilization
We directly measure f0 by the ν-to-2ν technique described earlier, as

shown in Figure 2-16(a). Using a dichroic beamsplitter, we split the laser
output into long- and short-wavelength components and recombine them
after insertion of a proper time-delay stage. The delay stage is essential to
compensate for the difference in group delay between the 580 and 1160 nm
radiation in the optical components (splitters, OC and BBO crystal). The
recombined beam is then focused into a 1 mm BBO crystal cut for type I
second harmonic of 1160 nm. The resulting output is sent through a 10 nm
wide spectral filter centered at 580 nm. After projecting the doubled 1160
nm light and the fundamental into the same polarization to enable
interference, the beat signal is detected by a photomultiplier tube (PMT).
Figure 2-16(b) shows the detected f0 beat signals with 10 kHz and 100 kHz
measurement bandwidths. The signal-to-noise ratios are 30 dB and 40 dB,
respectively. The f0 beat is very stable and the laser stays mode locked over
many days. In the absence of active feedback and without any temperature
control, the f0 frequency stays within the bandwidth of the bandpass filter
following the PMT, which is 8MHz, for more than 10 hours. On the time
scale of seconds, the beat can jitter by about 100 kHz. The f0 lock only
breaks because of a slow drift of the beat note out of the filter bandwidth of
the 8 MHz filter, which is not yet controlled.
(a)
SM
AOM
OC
Loop
filter 580 nm
DM
580 nm Filter
Digital
Phase detector
PMT
DM
λ/2 plate BBO
1160 nm
LO
(b)
RF Spectrum [dB]
-40
100 kHz RBW

-60 10 kHz RBW
-80
6
0 20 40 60 80x10
Frequency [Hz]
Figure 2-16. (a) Setup for f0 detection and lock. (b) Measured carrier-envelope beat signal
from the 80 MHz laser.
To demonstrate control over the comb, we filter the component fr + f0 at

130 MHz from the PMT signal and phase lock it to a stable rf synthesizer
using a phase-locked loop. To increase the locking range, the 130 MHz
component of the signal is properly filtered and amplified, sent to a thirty-
two-fold frequency divider, and then compared with the synthesizer signal in
a digital phase detector. The properly filtered and amplified phase-error
signal controls the pump power via an acousto-optic modulator that directly
controls the carrier-envelope frequency of the mode-locked laser.
Figure 2-17 shows the measured phase-error spectral densities of the f0
beat with and without active stabilization. The integrated phase error for the
in-loop measurement is 0.2 rad (from 1 Hz to 10 MHz). The major
contribution to this phase-noise error comes from the yet unsuppressed pump
noise. We expect further improvement in the residual carrier-envelope phase
74 Chapter 2
fluctuations with an improved loop filter design for stronger noise

suppression.
Figure 2-17. Spectral densities of the phase-error signal at the output of the digital phase
detector for the f0 lock (solid line) and free running f0 (dashed line).
4. CARRIER-ENVELOPE PHASE DYNAMICS
The carrier-envelope phase evolution in a mode-locked laser depends on

the pulse energy. This is not surprising as there is a nonlinear contribution to
the phase shift of the intracavity pulse as it passes through the gain crystal.
At the simplest level, this nonlinear phase shift happens because of the Kerr
effect, i.e., the index of refraction depends on intensity and hence so does the
optical path length. This is clearly an over simplification. Better insight is
based on treating the pulse as a soliton. However, care is needed as classical
soliton theory does not include the fact that the group velocity also depends
on intensity (known as the “shock” term) or the fact that dispersion and
nonlinearity in the laser are not constant as a function of position in the
cavity. The intensity dependence of the carrier-envelope phase has both
positive and negative aspects. On the positive side, it provides a parameter
by which the carrier-envelope phase can be controlled. On the negative side,
it means the amplitude noise will be converted to phase noise.
One of the first experiments to measure carrier-envelope phase evolution
(using a cross-correlator) observed its intensity dependence [12]. However,
the slope, d∆φce/dI, had the opposite sign from what was expected based on
classical soliton theory. This result was explained as being caused by an
intensity-dependent shift in the spectrum.
Subsequent theoretical work included the effect of the shock term [40]. It
was found that the shock term causes the group velocity to change twice as
fast with intensity as does the phase velocity. This fact results in an opposite
sign of d∆φce/dI from what it would be if the shock term were ignored.
Further experiments using self-referencing to detect ∆φce showed that

d∆φce/dI could actually change signs [41]. The results were consistent with
the shifting of the spectrum with intensity also changing signs. These
measurements also showed that for shorter pulses, ∆φce was much less
sensitive to changes in the pulse intensity. This finding is consistent with
theoretical work [40] but may also be simply due to clamping of the
spectrum because of implicit spectral filtering in the laser.
Numerical simulations of nonlinear pulse propagation in sapphire also
show a change in the sign of d∆φce/dI that arises from warping of the
envelope at high intensities [42]. There was no corresponding change in the
direction of spectral shift, as was observed experimentally. However, this
work only modeled propagation in sapphire and did not treat the rest of the
laser cavity.
More recent theoretical treatment based on a multiple-scales treatment of
dispersion-managed solitons included both managed dispersion and
nonlinearity [43]. Both shock and third-order dispersion were considered as
well. The results confirmed the earlier conclusions [40], although the details
varied because of differences in the treatment of and inclusion of additional
effects.
At this point, a number of phenomena have been identified in theory and
experiment, although a clear connection between them remains elusive. A
further understanding is important as these effects, which are used to control
∆φce, can also corrupt its stability and hence that of the generated spectral
comb.
5. SUMMARY
The recent explosion in work on carrier-envelope phase stabilization of

ultrashort pulses and the resulting stable frequency comb has been built on a
foundation of mode-locked laser technology. The culmination of this has
been the development of lasers that directly generate an octave-spanning
spectrum. At the same time, the need for stabilization has pushed mode-
locked laser technology in new directions and forced the development of an
improved understanding of the lasers.
ACKNOWLEDGEMENTS
S.T.C. acknowledges the contributions of T. M. Fortier, D. J. Jones, P.

M. Goorgian, B. Ilan and M. Ablowitz to this work and support by NIST,
ONR and NSF. E. P. I. and F. X. K. acknowledge contributions from T.
76 Chapter 2
Schibli, L. Matos and O. Kuzucu. Work at MIT was supported by NSF-ECS-

0322740, AFOSR-FA9550-04-1-0011and ONR-N00014-02-1-0717.
REFERENCES
[2] D. K. Negus, L. Spinelli, N. Goldblatt, and G. Feugnet, in Advanced Solid-
State Lasers (OSA, 1991), Vol. 10; F. Salin, J. Squier, and M. Pichè, Opt.
Lett. 16, 1674-1676 (1991); M. Pichè, Opt. Commun. 86, 156-160 (1991).
[3] H. A. Haus, IEEE J. Quantum Electron. 11, 736-746 (1975).
[4] J. H. B. Nijhof, N. J. Doran, W. Forysiak, and F. M. Knox, Electron. Lett.
33, 1726-1727 (1997).
[5] K. Tamura, H. A. Haus, and E. P. Ippen, Electron. Lett. 28, 2226-2228
(1992).
[6] Y. Chen, F. X. Kärtner, U. Morgner, S. H. Cho, H. A. Haus, E. P. Ippen,
and J. G. Fujimoto, J. Opt. Soc. Am. B 16, 1999-2004 (1999).
[7] J. P. Zhou, G. Taft, C. P. Huang, M. M. Murnane, H. C. Kapteyn, and I. P.
Christov, Opt. Lett. 19, 1149-1151 (1994).
[8] A. Stingl, C. Spielmann, F. Krausz, and R. Szipocs, Opt. Lett. 19, 204-206
(1994).
[9] D. H. Sutter, G. Steinmeyer, L. Gallmann, N. Matuschek, F. Morier-
Genoud, U. Keller, V. Scheuer, G. Angelow, and T. Tschudi, Opt. Lett. 24,
631-633 (1999).
[10] U. Morgner, F. X. Kärtner, S. H. Cho, Y. Chen, H. A. Haus, J. G. Fujimoto,
E. P. Ippen, V. Scheuer, G. Angelow, and T. Tschudi, Opt. Lett. 24, 411-
413 (1999).
Opt. Lett. 26, 373-375 (2001).
Opt. Lett. 21, 2008-2010 (1996).
[13] A. Baltuška, T. Fuji, and T. Kobayashi, Phys. Rev. Lett. 88, art. no.-133901
(2002).
[14] L. Xu, C. Spielmann, F. Krausz, and R. Szipocs, Opt. Lett. 21, 1259-1261
(1996).
[15] I. P. Christov, H. C. Kapteyn, M. M. Murnane, C. P. Huang, and J. P. Zhou,
Opt. Lett. 20, 309-311 (1995).
[16] I. P. Christov, V. D. Stoev, M. M. Murnane, and H. C. Kapteyn, Opt. Lett.
21, 1493-1495 (1996).
[17] I. P. Christov and V. D. Stoev, J. Opt. Soc. Am. B 15, 1960-1966 (1998).
[18] C. Spielmann, P. F. Curley, T. Brabec, and F. Krausz, IEEE J. Quantum
Electron. 30, 1100-1114 (1994).
[19] K. Tamura, E. P. Ippen, H. A. Haus, and L. E. Nelson, Opt. Lett. 18, 1080-
1082 (1993).
[20] S. M. J. Kelly, Electron. Lett. 28, 806-807 (1992); J. P. Gordon, J. Opt.

Soc. Am. B 9, 91-97 (1992).
[21] K. Tamura, E. P. Ippen, and H. A. Haus, App. Phys. Lett. 67, 158-160
(1995).
[22] Y. Chen and H. A. Haus, J. Opt. Soc. Am. B 16, 24-30 (1999).
[23] R. Reinisch and G. Vitrant, Prog. Quantum Electron. 18, 1-38 (1994).
[24] E. J. Grace, A. Ritsataki, P. M. W. French, and G. H. C. New, Opt.
Commun. 183, 249-264 (2000).
[25] T. M. Fortier, D. J. Jones, and S. T. Cundiff, Opt. Lett. 28, 2198-2200
(2003).
[26] L. Matos, O. Kuzucu, T. R. Schibli, J. Kim, E. P. Ippen, D. Kleppner, and
F. X. Kärtner, Opt. Lett. 29 (2004).
[28] J. P. Likforman, A. Alexandrou, and M. Joffre, App. Phys. Lett. 73, 2257-
2259 (1998).
[29] A. Bartels and H. Kurz, Opt. Lett. 27, 1839-1841 (2002).
[30] T. M. Ramond, S. A. Diddams, L. Hollberg, and A. Bartels, Opt. Lett. 27,
1842-1844 (2002).
[33] R. Szipocs, K. Ferencz, C. Spielmann, and F. Krausz, Opt. Lett. 19, 201-
203 (1994).
[34] A. Stingl, M. Lenzner, C. Spielmann, F. Krausz, and R. Szipocs, Opt. Lett.
20, 602-604 (1995).
[35] S. T. Cundiff, J. Ye, and J. L. Hall, Rev. Sci. Instrum. 72, 3746-3771
(2001).
[36] T. M. Fortier, J. Ye, S. T. Cundiff, and R. S. Windeler, Opt. Lett. 27, 445-
447 (2002).
[37] S. T. Cundiff, W. H. Knox, E. P. Ippen, and H. A. Haus, Opt. Lett. 21, 662-
664 (1996).
[38] F. X. Kärtner, U. Morgner, R. Ell, T. Schibli, J. G. Fujimoto, E. P. Ippen,
V. Scheuer, G. Angelow, and T. Tschudi, J. Opt. Soc. Am. B 18, 882-885
(2001).
[39] T. R. Schibli, O. Kuzucu, J. W. Kim, E. P. Ippen, J. G. Fujimoto, F. X.
Kaertner, V. Scheuer, and G. Angelow, IEEE J. Sel. Top. Quantum
Electron. 9, 990-1001 (2003).
[41] K. W. Holman, R. J. Jones, A. Marian, S. T. Cundiff, and J. Ye, Opt. Lett.
28, 851-853 (2003); K. W. Holman, R. J. Jones, A. Marian, S. T. Cundiff,
and J. Ye, IEEE J. Sel. Top. Quantum Electron. 9, 1018-1024 (2003).
[42] P. M. Goorjian and S. T. Cundiff, Opt. Lett. 29, 1363 (2004).
[43] M. J. Ablowitz, B. Ilan, and S. T. Cundiff, Opt. Lett. 29, 1808-1810 (2004).
Chapter 3
GIGAHERTZ FEMTOSECOND LASERS
The tools for precise optical frequency metrology
Albrecht Bartels
Time and Frequency Division, National Institute of Standards and Technology
Abstract: Femtosecond lasers with a repetition rate of approximately 1 GHz are

commonly used frequency comb generators for precise optical frequency
metrology. They are conveniently compact, yield unambiguous frequency
readings with the help of a commercial wavemeter, and can yield greater
heterodyne beat signals against a cw laser than systems with lower repetition
rates. This chapter reviews the technology of oscillators based on Ti:sapphire
and Cr:forsterite that operate at repetition rates of up to 3.5 GHz. Aside from a
discussion of these “standard” lasers with a typical full-width-at-half-
maximum (FWHM) bandwidth of 30 nm, one section is dedicated to the
generation of a broadband continuum with a 1 GHz oscillator. This laser
allows frequency measurements without additional spectral broadening and
can be phase locked to a reference oscillator for uninterrupted periods
exceeding one day.
Keywords: mode-locked lasers, optical atomic clocks, optical frequency metrology
1. INTRODUCTION
The first report of a femtosecond laser based on Ti:sapphire as the gain

material was published in 1991 [1]. Since then a number of research groups
have attempted to increase the repetition rate from 100 MHz to even higher
values. In 1998, right before the dawn of optical frequency metrology with
femtosecond laser combs, the oscillators by Ramaswamy et al. [2] and Stingl
et al. [3] were conceptually the most intriguing for realizing higher repetition
rates. Both researchers employed a new, more compact method for
intracavity dispersion management, replacing the long prism sequences that
had been part of every femtosecond laser until then. The prior configuration
had prevented repetition rates higher than ≈300 MHz.
3. GIGAHERTZ FEMTOSECOND LASERS 79
Figure 3-1. Illustration of the advantages for frequency metrology of the frequency comb
emitted by a 1 GHz femtosecond laser (left) compared to that of a 100 MHz laser (right). The
length of the two sets of comb teeth (solid vertical lines) scales with the power per mode. The
lengths of the gray horizontal bars represent the necessary resolution for premeasurement of
the frequency to be measured (fT, indicated by the dashed line) to determine the index k of the
mode against which fT is beat.
In 1999 Udem et al. [4] demonstrated an optical frequency measurement

with a femtosecond laser-based frequency comb for the first time. The comb
was used to bridge the 18 THz wide gap between a calibrated reference
frequency and the measured frequency fT . Soon afterwards, Jones et al. [5]
showed that a femtosecond laser can be “self-referenced.” Self-referencing
implies that both degrees of freedom of the optical frequency comb, the
laser’s repetition rate fr and its carrier-envelope offset frequency f0, are
accessible and controllable. Because f0 and fr are electronically manageable
microwave frequencies, the entire optical frequency comb with a multitude
of components νn = f0 + nfr (where n is a large integer) in the hundreds of
terahertz range can be directly linked to a cesium clock and subsequently
used for absolute measurements of optical frequencies. Since then, many
groups have successfully used femtosecond lasers for optical frequency
measurements and to realize optical atomic clocks [6].
Even before these early experimental demonstrations of femtosecond
laser-based optical frequency metrology, Reichert et al. [7] pointed out that
repetition rates higher than the usually available 100 MHz would be greatly
advantageous because they would yield a wider-spaced frequency comb.
Typically in a frequency measurement, the result is given by fT = f0 + kfr + fb,
where fb is the frequency of the heterodyne beat between fT and the
neighboring frequency comb component with index k. While f0, fr, and fb can
easily be measured, an unambiguous result for fT requires determination of k
via a premeasurement of fT with a resolution of better than ± fr/2. In contrast
to a 100 MHz system with a 1 GHz laser, this task is easily solved using a
commercial wavemeter. (The best available wavemeter resolution within the
80 Chapter 3
wavelength range of Ti:sapphire-based frequency combs is ≈100 MHz.)

Also, with a cavity length of 30 cm, a 1 GHz oscillator is conveniently
compact (commercially available models come in a ≈32×20×10 cm box) as
compared to a 3 m long 100 MHz resonator. See Figure 3-1 for a comparison
of the two systems.
A femtosecond laser with a 1 GHz repetition rate yields ≈100 times more
usable optical power per comb component (mode) than does a 100 MHz
laser. This additional power increases the signal-to-noise ratio of a
heterodyne beat signal against a continuous-wave (cw) laser, which is used
as a local oscillator to realize fT, by the same amount. The factor of 100 is
caused by two features of the femtosecond laser: First, the average output
power of a well-designed Ti:sapphire laser is usually on the order of 25% of
the absorbed pump power, regardless of the repetition rate. At a repetition
rate 10 times higher, the average output power is distributed to one-tenth the
number of modes, increasing the power per mode by a factor of 10. Second,
the spectrally narrow output of a typical laser (FWHM ≈30–60 nm) needs to
be broadened to span at least one octave via self-phase modulation (SPM) in
a microstructure fiber [8] to allow self-referencing. However, nonlinear
amplification of shot noise present on the input light limits the pulse energy
that can be launched into a microstructure fiber and still attain usefully quiet
output light [9, 10]. Above a threshold of ≈300 pJ of pulse energy, the excess
noise on the fiber output grows exponentially. This behavior implies that a
maximum average power of 300 mW in a 1 GHz system can be launched
into a microstructure fiber. In comparison, only 30 mW can be sent into
microstructure fiber in a 100 MHz system.
Some problems remain, related mainly to the long-term reliability of
frequency metrology systems and their ease of use. A great improvement
was achieved in 2004 with a 1 GHz laser that readily spans a broadband
continuum and can be self-referenced without the use of microstructure fiber
for uninterrupted periods in excess of one day. This oscillator will be
addressed in the last section.
2. HIGH-REPETITION-RATE RING OSCILLATORS
2.1 Design criteria and basic resonator layout
The repetition rate fr of a passively mode-locked femtosecond laser is

determined by the physical cavity length L and the round-trip-averaged
group velocity v g :
vg
fr = . (1)
L
Thus, increasing the repetition rate means shrinking the cavity length. The
realization of gigahertz repetition rates for Ti:sapphire femtosecond lasers
has been straightforward. A few simple specific design criteria now exist.
They can be discussed in the context of a steady-state duration τ of a pulse
circulating in a cavity in which only SPM, provided by the nonlinear
refractive index n2 ≈ 3×10-16m2/W of the Ti:sapphire gain crystal, and
negative group-delay dispersion (GDD), designated as D2, shape the pulse
(see, for example, Reference [11]). This limit applies in good approximation
to all Kerr-lens–mode-locked femtosecond lasers as long as higher-order
dispersion is negligible:
(
τ = 4 ln 1 + 2 ) dn E
D2 λ w02
, (2)
2 P
where λ is the carrier wavelength, w0 is the waist radius inside the laser
crystal, d the length of the crystal, and EP the pulse energy.
When shrinking a 100 MHz cavity to yield a 1 GHz repetition rate, the
pulse energy E P is reduced by a factor of 10 (assuming constant average
power). This results in a steady-state pulse duration 10 times longer. The
combination of lower pulse energy and greater pulse duration reduces the
peak by a factor of 100. This alone does not present a fundamental problem
for achieving higher repetition rates. However, in practice Kerr-lens mode
locking (KLM) is required to maintain stable pulsed operation over cw
operation. A Kerr lens is induced by a high-intensity Gaussian beam profile
in the Ti:sapphire crystal via its nonlinear refractive index. To achieve KLM,
the cavity is arranged in a way that the Kerr lens modulates the net gain or
loss of the cavity for a pulsed beam. Either it reduces the net cavity losses by
increasing the transmission through a hard aperture (e.g., a slit) at an
appropriate position in the resonator (hard-aperture KLM), or it increases the
net gain by increasing the overlap with the finite gain volume inside the
crystal (soft-aperture KLM). KLM gets more efficient with higher peak
intensity. (High pulse energies and short pulse widths in the cavity are
essential.) Thus the reduction of peak power due to the increased repetition
rate must be compensated for. This is accomplished by reducing the amount
of output coupling with respect to a 100 MHz oscillator (from ≈10% to ≈1%)
and shrinking the waist radius w0 inside the gain crystal by tighter focusing.
The second major concern is the achievement of negative net-GDD in the
cavity. The standard method of employing a four-prism sequence (or a
82 Chapter 3
double-passed, two-prism sequence) fails for high repetition rates because

even with highly dispersive materials, an apex distance of ≈30 cm per prism
pair is required [12]. This requirement sets an upper limit of ≈300 MHz on
the repetition rate. An elegant way to replace a prism sequence is by using
negative dispersion mirrors that reflect light with longer wavelengths from a
deeper region in the coating than light with shorter wavelengths. A variety of
structures has been developed, of which only Gires-Tournois interferometer
(GTI) mirrors and chirped mirrors are used here [13].
Figure 3-2. (a) Four-mirror cavity with the gain crystal enclosed by focusing mirrors M1 and
M2. The flat mirror M3 and the output coupler (OC) form the collimated part of the cavity.
The pump light enters the crystal through lens L and M1. (b) Equivalent to (a) but with extra
folding using flat mirrors M4 and M5.
The ring oscillators shown in Figure 3-2 are the basis for all lasers
presented here. A ring configuration has been chosen for geometrical
reasons. With a given number of components, it allows a higher repetition
rate than a linear cavity. Also, ring cavities are less sensitive to back-
reflections. In the following sections, the cavities will be discussed in detail
along with actually realized lasers.
As with all Kerr-lens–mode-locked femtosecond lasers, these oscillators
need an initial perturbation, such as tapping a mirror mount, to generate an
intracavity power fluctuation that builds up to a stable circulating
femtosecond pulse. As opposed to cw operation, which is bidirectional,
mode-locked operation is unidirectional and starts in a random direction
[14].
2.2 Standard Ti:sapphire lasers for 0.3–3.5 GHz

repetition rate
The four-mirror cavity, as shown in Figure 3-2(a), is usually employed to

achieve the highest repetition rates because it can be made the most compact.
The cavity contains a Ti:sapphire crystal of length d=1.5 mm (absorption
coefficient α=5 cm-1) that is pumped with a frequency-doubled Nd:YVO4
laser at 532 nm through lens L with fL = 30 mm. The crystal is placed at
Brewster’s angle between mirrors M1 and M2. The pump beam radius inside
the crystal is ≈10 µ m. Mirrors M1 and M2 have a focal length of 15 mm,
leading to a calculated waist radius of the resonator mode inside the crystal
between 8 and 20 µm for repetition rates ranging from 0.5 to 3.5 GHz. The
angles at which the intracavity beam is reflected off M1 and M2 are set such
that the cavity is astigmatically compensated [15]. The crystal introduces a
GDD of ≈87 fs2 at 800 nm wavelength. Chirped mirrors with a GDD of ≈
-45 fs2 at 800 nm are used for M1–M3 to achieve a net GDD of ≈ -48 fs2 (the
GDD of the OC is negligible). The output coupler has a reflectance of 98%,
and the cavity length is set to yield a repetition rate of 2 GHz.
This laser is operated at pump powers between 1.8 and 5.5 W. Figure 3-3
shows output power as a function of pump power with a rather low slope
efficiency of 13.6%. The low efficiency is due to the short gain crystal,
whose length is limited by the constraint to achieve a negative net GDD.
However, if the absorbed pump power is considered, the slope efficiency
rises to 25.8%.
Figure 3-4(a) shows a measured autocorrelation trace of the laser output
when operating at 5.5 W pump power after compensation of extracavity
dispersion by multiple reflections off a pair of chirped mirrors. The envelope
of a 24 fs long pulse, assuming a sech2(t) intensity profile, is fitted to the
trace, showing good agreement. Figure 3-4(b) shows the corresponding
output spectrum with a FWHM of 32 nm. The time-bandwidth product of
0.35 deviates slightly from the ideal of 0.315. The inset of Figure 3-4(a)
shows the pulse duration of the same laser operated at different intracavity
pulse energies. (Variation of the pulse energy was attained by changing the
pump power.) The theoretical pulse durations as a function of pulse energy,
as in Equation (2), are also plotted in the inset and agree well with the
experiment. (w0 was calculated using ABCD matrices to generate the
theoretical curve.) The agreement between theory and experiment confirms
that SPM and GDD dominantly shape the pulses in this laser; thus, pulse
width and bandwidth can be easily and well controlled.
84 Chapter 3
4-mirror, 2 GHz, 2% output coupler, η=13.6%
output power (W)

6-mirror, 1 GHz, 2% output coupler, η=20.0%
2.0
1.5
1.0
0.5
0.0
0 2 4 6 8 10
pump power (W)
Figure 3-3. Laser output power as a function of incident pump power for the four-mirror
(triangles) and six-mirror (squares) lasers described in the text. The solid lines are linear fits
to the datasets.
Attempts to further shorten the pulse duration were undertaken by

reducing D2 . We found a limit at 14 fs (at a spectral bandwidth of 70 nm
FWHM) due to higher-order terms in the intracavity dispersion.
Figure 3-4. (a) Interferometric autocorrelation trace of the pulses from the four-mirror laser at
5.5 W pump power. Inset: Pulse duration as a function of intracavity pulse energy (squares)
and theoretical values according to Equation (2). (b) Corresponding output spectrum. Inset:
Series of output spectra of a continuously tunable version of the six-mirror cavity [16].
Often neglected in scientific papers, but important for real-world

applications, is the beam quality of a femtosecond laser. Figure 3-5(a) shows
a beam profile of the four-mirror cavity recorded at the focus after focusing
the laser output with a lens of 200 mm focal length. It shows no noticeable
deviation from circular symmetry. Figure 3-5(b) shows a cross section
through the beam profile with a fitted Gaussian, again with no noticeable
deviation from ideal. A measurement of the M2-factor has been performed,
yielding values of 1.1 for both tangential and saggital planes. (See Reference
[17] for a definition of M2.) The excellent beam quality is attributed to the
fact that spatially dispersive elements are absent in the cavity.
With the four-mirror cavity, repetition rates between 1 GHz and 3.5 GHz
were attained. The upper limit was set by geometrical constraints, i.e., it has
not been possible to move the components closer together. It is anticipated
that even higher repetition rates of up to 10 GHz are feasible by minimizing
mechanical components and further reducing the focal lengths of M1 and
M2. The disadvantage of this cavity is that the short-gain crystal results in a
relatively low output power.
Figure 3-5. (a) Mode profile of a four-mirror laser at the focus of a 200 mm lens. (b) Cross
section (squares) through the profile shown in (a) and fit with a Gaussian (solid line).
Folding with two additional negative dispersion mirrors in the cavity as

in the six-mirror configuration shown in Figure 3-2(b) allows one to use a
longer Ti:sapphire crystal, achieve more gain, and, at the same time, ensure a
negative net GDD in the cavity. This approach is favored when higher output
powers and moderate repetition rates are required. The oscillator depicted in
Figure 3-2(b) has a crystal of length d = 2.5 mm and an absorption α = 5 cm-
1
. It employs the same chirped mirrors that were used for the four-mirror
cavity in positions of M1–M5 and has a repetition rate of 1 GHz. The net
cavity GDD in this case is -80 fs2. With a 2% output coupler, the output
power as a function of incident pump power is shown in Figure 3-3. The
slope efficiency is 20%; with respect to the absorbed pump power, it is 28%,
comparable to that of the four-mirror laser. At the highest available pump
power of 10.5 W, a remarkably high output power of 2.1 W is attained.
With respect to output spectra, pulse duration, and beam quality, the six-
mirror cavity performs equivalently to the four-mirror cavity. It has been
realized at repetition rates between 300 MHz and 2 GHz.
86 Chapter 3
The two oscillators described in this section are widely used frequency
comb generators for optical frequency metrology. They are commonly
operated between 0.8 GHz and 1 GHz. This range has proven to be the best
compromise between high-repetition rate and sufficient pulse energy to
generate an octave-spanning spectrum in common microstructure fibers.
(See e.g., Reference [18] for a typical octave-spanning spectrum from a 1
GHz four-mirror laser and a microstructure fiber.)
The six-mirror cavity has also been realized in a wavelength-tunable
modification at a repetition rate of 1 GHz [16]. A half-Brewster prism was
used to introduce spatial dispersion into the resonator with the effect that
horizontally tilting a cavity mirror allowed continuous tunability from 733 to
850 nm with pulse durations of ≈40 fs. A series of output spectra is shown in
the inset of Figure 3-4(b).
2.3 Cr:forsterite oscillator at 433 MHz — extension to

telecommunication wavelengths
It is a logical step to extend the concept that has successfully been used
with Ti:sapphire to gain media that operate at longer wavelengths. One
reason is to obtain access to the telecommunication bands and thus be able to
calibrate and characterize system components, such as wavelength
references or wavelength-division multiplexing filters, in that region. A
second goal is to generate a mode-locked spectrum that is adjacent to and
that can be phase coherently combined with the output spectrum of a
Ti:sapphire laser (specifically the one described in the next section) to
generate an extraordinarily broad “superfrequency comb.” The gain material
that is best suited for both tasks is Cr:forsterite, which has a gain spectrum
extending from wavelengths of ≈1150 to 1350 nm.
A femtosecond laser with Cr:forsterite was realized at a repetition rate of
433 MHz in the six-mirror configuration [see Figure 3-2(b)] [19]. The 10
mm gain crystal with an absorption coefficient of α=1.1 cm-1 at 1075 nm is
significantly longer than in the Ti:sapphire lasers to allow for sufficient
pump light absorption. The crystal is pumped with 10 W from an Yb:glass
fiber laser emitting at 1075 nm through a lens with fL = 40 mm. The crystal is
cooled to 0° C to increase the gain and to reduce problems with thermal
lensing. (Cr:forsterite has a thermal conductivity ≈5 times lower than
Ti:sapphire.) The focal lengths of mirrors M1 and M2 are 25 mm; they are
longer than what has been used for the Ti:sapphire lasers to better match the
confocal length of the cavity mode to the length of the gain crystal. With
chirped mirrors at positions M1–M3, a GTI mirror at M4, a low-dispersive
high reflector at M5, and the GDD of the gain crystal of 185 fs2 [20], the net
GDD is ≈ -260 fs2. A 1.5% transmission output coupler at 1280 nm is used.
The mode-locked output power of this laser is 620 mW. A spectrum of

the output pulses is shown in Figure 3-6. The spectral bandwidth is 59 nm
centered around 1280 nm. The pulse duration of 30 fs was determined with
an intensity autocorrelation measurement.
Alignment of this laser for stable mode locking is more difficult than for
Ti:sapphire lasers. Ti:sapphire lasers consistently mode lock at the inner
edge of the cavity’s stability range, as expected from modeling KLM. This
systematic guideline for finding mode-locked operation does not work in
Cr:forsterite because of thermal lensing. However, once a working point has
been found, the laser’s operation is reliable and reproducible. Mode-locked
operation has been observed uninterrupted for up to seven days.
power per 433 MHz mode (µW)
1
10
0
10
-1
10
-2
10
-3
10
900 1200 1500 1800 2100
wavelength (nm)
Figure 3-6. Output spectrum of the Cr:forsterite laser (solid line) and spectrum after
broadening in a highly nonlinear Ge-doped silica fiber (HNLF) (dashed line).
The output from this laser was successfully broadened via SPM in highly
nonlinear Ge-doped silica fiber (HNLF) to form a frequency comb that spans
more than one octave [19, 21], as shown in Figure 3-6. This frequency comb
has been referenced to the National Institute of Standards and Technology’s
(NIST’s) Ca optical frequency standard and a cesium atomic clock. Its utility
for optical frequency measurements was demonstrated by Corwin et al. [22]
who used the broadened Cr:forsterite comb for the characterization of
telecommunication-band frequency standards.
The creation of a phase-coherent superfrequency comb, ranging from
wavelengths of 570 to 1450 nm, was demonstrated by actively linking the
output of the Cr:forsterite laser, described above, and a broadband
Ti:sapphire laser described in the next section [23].
88 Chapter 3
3. BROADBAND TI:SAPPHIRE OSCILLATOR
A new type of extremely broadband femtosecond laser has been realized

with a slight modification of the four-mirror cavity described earlier [see
Figure 3-2(a)]. The modified cavity has a convex mirror with a radius of
curvature of 1 m at the position of M3. M1, M2, and M3 are chirped mirrors.
The output coupler has a transmission of 2% at 800 nm. The net intracavity
dispersion of the resonator, including air and the Ti:sapphire crystal (d = 2.0
mm), and the mirror reflectivities are shown in Figure 3-7. The net GDD at
800 nm is ≈ -20 fs2. When the cavity is operated close to the inner edge of its
stability range, continuous translation of mirror M2 towards M1 results in
bidirectional mode locking and stages of stable and chaotic multipulsing in
two directions and in one direction until a state with a single stable pulse
circulating unidirectionally in the cavity is reached. At a pump power of 10.5
W, the mode-locked output power is ≈1 W. The cavity length is adjusted to
yield a 1 GHz repetition rate.
Figure 3-7. Left panel: Net cavity group-delay dispersion (GDD) of the broadband laser.
Right panel: Reflectance of the chirped mirrors (solid line) and the output coupler (dashed
line).
The output spectrum of the laser is shown in Figure 3-8. It spans from
570 to 1100 nm at a power level of 1 nW per 1 GHz frequency mode. (1 nW
per mode is the power typically required to measure a beat note against a
milliwatt-scale cw laser with sufficient signal-to-noise ratio for a frequency
measurement.) The spectrum has pronounced maxima at 670 nm, 845 nm,
900 nm, and 927 nm. The peak around 670 nm can be isolated by a double
reflection off a pair of 657 nm high reflectors, leading to the spectrum shown
in the inset of Figure 3-8(b). It contains 450 mW of average power.
The duration of the output pulses has only been roughly characterized by
using an intensity autocorrelation measurement. Assuming a temporal sech2
shape, it is 11 fs when the nonlinear crystal is angle-tuned to optimally phase
match the central part of the spectrum. The peak around 670 nm yields
bandwidth-limited pulses of 33 fs after isolation from the longer-wavelength
part of the spectrum.
Figure 3-8. (a) Output spectrum of the broadband laser on a logarithmic scale (solid line). At
a level of 1 nW per 1 GHz mode, it extends from 570 to 1100 nm. The dashed line shows a
simulated spectrum, offset by 20 dB for clarity. (b) Output spectrum of the broadband laser on
a linear scale. Inset: spectrum after reflection off a pair of 657 nm high reflectors.
3.1 How does it work?
The performance of this laser is extraordinary in a number of respects:

First, when mode locking is stopped (e.g., by briefly intersecting the pump
beam) and the laser falls back into cw operation, the output power drops by
more than an order of magnitude to about 60 mW. Assuming that a Kerr-
lens-based effective saturable absorber is responsible for this behavior, a
simple rate equation model leads to the conclusion that ≈30% saturable
absorption is required to explain this drop. This high saturable absorption
contrasts with what is usually observed for standard Ti:sapphire femtosecond
lasers, which have an effective saturable absorption of ≈1–2%. Second,
considering the fact that Ti:sapphire has no gain below 690 nm, the 450 mW
of power contained in the peak around 670 nm is very remarkable and
indicates that this strong component is generated during only a few round
trips through the cavity. Finally, assuming that SPM and GDD dominate
pulse shaping, mode locking of the laser at such extreme bandwidth should
not be possible. The strong GDD modulation and the positive values outside
a band ranging from 700 to 850 nm should quickly spread the pulse and
prevent mode locking (see Figure 3-7 for the net intracavity dispersion).
To get an idea of how this laser functions, it is helpful to review the two
practical routes towards shorter pulses and broader spectra in a femtosecond
laser. One approach is to engineer the dispersion inside the resonator in a
way that short pulses with extremely broad spectra do not spread in time
during a cavity round trip. This involves careful design of mirrors to have
negative GDD over the desired large bandwidth and compensate for the
positive-gain crystal dispersion with a low net-higher-order dispersion. This
90 Chapter 3
approach has been taken with great success by Ell et al. [24]. Their carefully
engineered intracavity dispersion is capable of generating spectra exceeding
one octave with 5 fs pulses at a repetition rate of 64 MHz. In our case,
however, because of the narrow bandwidth of the chirped mirrors, both with
respect to their dispersive and reflective properties, this limit is inapplicable.
Alternatively, one can allow higher-order dispersion in the resonator but, at
the same time, ensure that the leading and trailing edges of the temporally
spreading spectrum are sufficiently suppressed, such that only a short pulse
remains stable in the cavity. This effect can be attained by employing an
effective fast saturable absorber. In the case of our Ti:sapphire lasers, this
effective saturable absorber is provided by a soft aperture KLM (see Section
2.1). While this effect is actually a self-gain modulation, it can theoretically
be treated as an equivalent fast saturable absorber. These effects are more
generally referred to as self-amplitude modulation (SAM).
There are strong indications that the broadband laser described above
operates in the limit of an increased SAM. The strongest support for this idea
is the experimental observation that the mode-locked output power and the
continuous output power of the laser differ by more than an order of
magnitude. This difference indicates that the Kerr-lens-induced effective
saturable absorber has a saturable absorption of approximately 30%.
Theoretical calculations of the change in beam waist diameter inside the gain
medium (i.e., the soft-aperture Kerr-lens effect) show that replacing flat
mirror M3 with a slightly convex mirror (in our case with a radius of
curvature of 1 m) can increase the SAM of pulses circulating in the ring
cavity [25]. While the theory applied to calculate this effect is valid only at
power levels of about a factor of four below the power at which our
oscillator operates, it still has provided a clue to understanding this new type
of femtosecond oscillator.
A full theory explaining why the convex mirror leads to a stronger
effective saturable absorber does not yet exist. However, modeling the
resonator with a split-step Fourier simulation allows us to understand why
this laser functions. The model includes the Ti:sapphire gain spectrum and
dispersion, all reflective amplitude and phase properties of the cavity
mirrors, SPM, and a fast saturable absorber (of the same type as used by
Chen et al. [26]). The dashed curve in Figure 3-9 shows a simulation of a
cavity with the same mirrors as the broadband laser cavity but with a shorter
crystal (d = 1.5 mm) and a saturable absorption of qsat. = 2%. Its net GDD at
800 nm is -45 fs2. It represents a standard laser with a relatively narrow
spectrum of 45 nm bandwidth. An attempt to get shorter pulses from this
laser by increasing the net cavity GDD to ≈ -20 fs2 by using a longer crystal
(d = 2.0 mm) fails because the simulation does not yield a stable solution
anymore. Higher-order dispersion spreads the seed pulse and randomizes its
phase. However, when qsat. is increased to 30% for the d = 2 mm

configuration, the model again delivers a stable solution, shown as the solid
line in Figure 3-9. The solution corresponds to the new broadband laser
where, despite dominant higher-order dispersion, a stable mode-locked
solution is obtained because of the pulse-cleaning effect of the enhanced
SAM. This simulation agrees very well with the experimental spectrum in its
main features, which include the pronounced peaks at ≈670 nm and ≈900
nm, the asymmetric central contribution around 835 nm, and the spectrum’s
approximate width [see Figure 3-8(a)].
power per 1 GHz mode (µW)
1
10
0
10
-1
10
-2 d=1.5 mm, qsat.=0.02
10 d=2.0 mm, qsat.=0.3
-3
10
600 700 800 900 1000 1100
wavelength (nm)
Figure 3-9. Modeled output spectra of a standard laser (d = 1.5 mm, qsat.= 0.02) and a
broadband laser (d = 2.0 mm, qsat. = 0.3) using a split-step Fourier simulation.
3.2 Application in frequency metrology and optical

clocks
As previously explained, optical frequency metrology and optical-clock

designs require measurements of both degrees of freedom of a femtosecond
laser, its repetition rate fr and its carrier envelope offset frequency f0. This
section briefly discusses how the broadband 1 GHz laser described in the last
section is different from (and offers advantages over) the standard 1 GHz
laser because of the way that the new laser’s offset frequency is measured.
The standard ν-to-2ν-scheme for measuring f0 is to frequency double a
low-frequency portion of the laser spectrum (typically ≈1100 nm) and beat it
against a spectral matching, high-frequency part of the fundamental
spectrum (≈550 nm) [5]. This approach requires a spectrum covering at least
one octave. For standard lasers, the only way to attain such a broad spectrum
92 Chapter 3
has been SPM in microstructure fibers. However, coupling light into the core
of fibers with a diameter less than 2 µ m is a difficult task. As SPM is a third-
order nonlinearity, slight changes in coupling efficiency can result in large
changes of the attained spectrum and usually limit the time over which a
measurement of f0 can be maintained with sufficient signal-to-noise ratio.
Apart from that, microstructure fibers also limit the amount of useful
average power achievable for the continuum spectrum because the nonlinear
broadening process inherently amplifies both technical and shot noise
present on the input light. (References 10 and 27 provide an overview on
noise amplification during supercontinuum generation in microstructure
fibers.)
The broadband laser allows one to circumvent these problems with
microstructure fiber because its broad spectral coverage allows for a direct
measurement of its offset frequency. Although the output spectrum has
wavelength components that are one octave apart, the power contained in
them is too low to facilitate a measurement of f0 using the standard ν-to-2ν
method. Therefore, we employ a 2ν-to-3ν method: A portion around 930 nm
is frequency tripled and beat against the second harmonic of a portion
around 620 nm to yield a signal at f0. Details of this setup can be found in
Ramond et al. [28].
Although the 2ν-to-3ν method is more complicated than the standard
method, the broadband laser has an unprecedented long-term stability as a
frequency measurement tool. First, it routinely stays mode locked for periods
of many days (>1 week demonstrated) without noticeable changes to its
output power or spectrum. Second, the f0 signal usually has a signal-to-noise
ratio of 25–30 dB in the 300 kHz bandwidth and does not degrade over time
because of the absence of microstructure fiber in the measurement apparatus.
In a test of its long-term stability, we have operated the laser in an optical
clock configuration. Here, the repetition rate fr of the broadband laser is
phase locked to a cavity-stabilized single-frequency laser diode at frequency
fLD = 456 THz that represents the optical frequency standard that would be
used in an optical clock. Specifically, a beat between the laser diode and the
neighboring component of the frequency comb with frequency fb has been
phase locked to a synthesizer at 600 MHz by feedback to the femtosecond
laser’s cavity length via a mirror mounted onto a piezoelectric transducer.
(All synthesizers were referenced to the NIST primary standard.)
Additionally, f0 was phase locked to a synthesizer at 100 MHz by feedback
to the pump power via an acousto-optic modulator. It can be shown that fr
now derives its frequency and instability entirely from the laser diode or, in
the case of an optical clock, from the optical frequency standard and is
effectively the microwave output of the optical clock [23]. We then counted
f0, fb and fr simultaneously with frequency counters at 10 s gate time. The
offset of the counter readings from their preset values (f0 = 100 MHz, fb =
600 MHz, set by the synthesizers; fr = 998,092,449.54 Hz, defined by fLD and
the choices of f0 and fb) over a period of 21 hours are displayed in Figure 3-
10. The data set shows uninterrupted and hands-off operation of the f0 phase
lock for 21 hours with counter-resolution-limited deviations on the millihertz
scale until the system was turned off. The phase lock on fb operated similarly
for ≈14 h with only one detected feedback-loop error (cycle slip) at ≈4 h, i.e.,
≈10 h of continuously phase-locked data were obtained. The failure at ≈14 h
was likely caused by the failure of the stabilization of fLD to the Fabry-Perot
cavity that prevented the feedback loop from tracking the diode laser
frequency rather than failure of the femtosecond laser itself. As fr represents
a measurement of the laser diode frequency (fLD = f0 + nLD fr + fb, where nLD is
the mode number of the frequency comb component against which fLD is
beat), the time record of the offset of fr from 998,092,449.54 Hz can be
multiplied with nLD = 456,857 to give a record of the temporal drift of the
Fabry-Perot cavity on the right axis of the graph in Figure 3-10. The cycle
slip in the fb feedback loop does not appear in the fr record because the effect
of the 120 mHz excursion in fb results in an error of 260 nHz in fr, which is
below our measurement limit.
These results show that the broadband laser facilitates a phase-coherent,
cycle-slip-free link between an optical oscillator at 456 THz and the 1 GHz
laser repetition rate for 10 h. In other words, we have the ability to count
1.6×1019 optical cycles at 456 THz without ever losing track of the
oscillation. It has also been shown that fr and f0 of the broadband laser can be
continuously phase locked to a synthesizer for more than 48 hours.
4. CONCLUSION
Ti:sapphire ring oscillators at 1 GHz repetition rate are currently the best
available frequency comb generators for optical frequency metrology. They
are compact, have a conveniently large comb spacing, and yield 100 times
more power per mode than conventional 100 MHz lasers. Their simple ring-
laser architecture employing negative dispersion mirrors for GDD control
allows repetition rates between 300 MHz and 3.5 GHz with Ti:sapphire. The
concept is extendable to other gain media, as has been demonstrated with
Cr:forsterite at 433 MHz repetition rate.
94 Chapter 3
5
a)
f0 - 100 M Hz
(m Hz)
0
-5
b)
f b - 600 M Hz
100
(m Hz)
50
0
f R - 998,092,449.54 Hz
30 0
0.6 c)
f LD Drift (kHz)
0.4 20 0
(Hz)
0.2 10 0
0.0 0
-0.2 -100
0 2 4 6 8 10 12 14 16 18 20
Time (h)
Figure 3-10. (a) A record of consecutive counter readings at 10 s gate time for the phase-
locked f0. The preset value of 100 MHz has been subtracted. (b) The simultaneously measured
offset of fb from 600 MHz. In this plot, one cycle slip of the phase-locked loop controlling fb
has occurred. (c) A time record of the offset of the laser repetition rate from 998,092,449.54
Hz (left scale) that can be translated into a drift of fLD (right scale).
The most critical development has been the 1 GHz broadband Ti:sapphire
laser that eliminates troublesome microstructure fibers from optical-
frequency-measurement setups. This laser has enabled the construction of an
optical clockwork with an unprecedented continuous operation time in
excess of 48 hours. This laser might also become interesting in basic
research as a source of femtosecond pulses in the range from 620 to 690 nm.
ACKNOWLEDGEMENTS
T. Dekorsy and H. Kurz (Institute of Semiconductor Electronics, RWTH

Aachen) made invaluable contributions to the first stages of this work that
are highly appreciated. I also thank Christof Janke (GigaOptics GmbH),
Isabell Thomann, Tanya Ramond, Scott Diddams, Leo Hollberg, Brian
Washburn, Nathan Newbury and Kristan Corwin (NIST) for their important
contributions. Jeff Nicholson provided the fiber used for the data presented
in Figure 3-6.
REFERENCES
[2] M. Ramaswamypaye and J. G. Fujimoto, Opt. Lett. 19, 1756-1758 (1994).
[3] A. Stingl, C. Spielmann, R. Szipocs, and F. Krausz, in Conference on
Lasers and Electrooptics (Opt. Soc. Am., 1996), p. 66
3568-3571 (1999).
84, 5102-5105 (2000); M. Niering, R. Holzwarth, J. Reichert, P. Pokasov,
T. Udem, M. Weitz, T. W. Hänsch, P. Lemonde, G. Santarelli, M. Abgrall,
P. Laurent, C. Salomon, and A. Clairon, Phys. Rev. Lett. 84, 5496-5499
(2000); S. A. Diddams, T. Udem, J. C. Bergquist, E. A. Curtis, R. E.
Drullinger, L. Hollberg, W. M. Itano, W. D. Lee, C. W. Oates, K. R. Vogel,
and D. J. Wineland, Science 293, 825-828 (2001); J. Stenger, T. Binnewies,
G. Wilpers, F. Riehle, H. R. Telle, J. K. Ranka, R. S. Windeler, and A. J.
Stentz, Phys. Rev. A 63, 021802 (2001); T. Udem, S. A. Diddams, K. R.
Vogel, C. W. Oates, E. A. Curtis, W. D. Lee, W. M. Itano, R. E. Drullinger,
J. C. Bergquist, and L. Hollberg, Phys. Rev. Lett. 86, 4996-4999 (2001); G.
D. Rovera, F. Ducos, J. J. Zondy, O. Acef, J. P. Wallerand, J. C. Knight,
and P. S. Russell, Meas. Sci. Techn. 13, 918-922 (2002).
172, 59-68 (1999).
[8] J. K. Ranka, R. S. Windeler, and A. J. Stentz, Opt. Lett. 25, 25-27 (2000); J.
C. Knight, T. A. Birks, P. S. Russell, and D. M. Atkin, Optics Letters 21,
1547-1549 (1996).
[9] L. Hollberg, C. W. Oates, E. A. Curtis, E. N. Ivanov, S. A. Diddams, T.
Udem, H. G. Robinson, J. C. Bergquist, R. J. Rafac, W. M. Itano, R. E.
Drullinger, and D. J. Wineland, IEEE J. Quantum Electron. 37, 1502-1513
(2001).
[10] K. L. Corwin, N. R. Newbury, J. M. Dudley, S. Coen, S. A. Diddams, K.
Weber, and R. S. Windeler, Phys. Rev. Lett. 90, 113904 (2003).
[11] T. Brabec, C. Spielmann, and F. Krausz, Opt. Lett. 17, 748-750 (1992).
[13] F. Gires and C. R. Tournois, Science 258, 6112 (1964); R. Szipocs, K.
Ferencz, C. Spielmann, and F. Krausz, Opt. Lett. 19, 201-203 (1994).
[14] A. Bartels, T. Dekorsy, and H. Kurz, Opt. Lett. 24, 996-998 (1999).
[15] H. W. Kogelnik, C. V. Shank, A. Dienes, and E. P. Ippen, IEEE J.
Quantum Electron. 8, 373 (1972).
[16] A. Bartels, T. Dekorsy, and H. Kurz, in Conference on Lasers and Electro-
Optics (OSA Technical Digest, 2000), p. CMF3
96 Chapter 3
[17] A. E. Siegman, Lasers (University Science Books, Mill Valley, California,

1986).
[18] S. A. Diddams, T. Udem, K. R. Vogel, C. W. Oates, E. A. Curtis, R. S.
Windeler, A. Bartels, J. C. Bergquist, and L. Hollberg, in Laser Frequency
Stabilization: Standards, Measurement and Applications, edited by J. L.
Hall and J. Ye (SPIE, San Jose, 2001), Vol. SPIE 4269, p. 77-83.
[19] I. Thomann, A. Bartels, K. L. Corwin, N. R. Newbury, L. Hollberg, S. A.
Diddams, J. W. Nicholson, and M. F. Yan, Opt. Lett. 28, 1368-1370
(2003).
[20] I. Thomann, L. Hollberg, S. A. Diddams, and R. Equall, Appl. Optics 42,
1661-1666 (2003).
[21] J. W. Nicholson, M. F. Yan, P. Wisk, J. Fleming, F. DiMarcello, E.
Monberg, A. Yablon, C. Jorgensen, and T. Veng, Opt. Lett. 28, 643-645
(2003).
[22] K. L. Corwin, I. Thomann, T. Dennis, R. W. Fox, W. Swann, E. A. Curtis,
C. W. Oates, G. Wilpers, A. Bartels, S. L. Gilbert, L. Hollberg, N. R.
Newbury, S. A. Diddams, J. W. Nicholson, and M. F. Yan, Opt. Lett. 29,
397-399 (2004).
[23] A. Bartels, N. R. Newbury, I. Thomann, L. Hollberg, and S. A. Diddams,
Opt. Lett. 29, 403-405 (2004).
Opt. Lett. 26, 373-375 (2001).
[25] A. Bartels and H. Kurz, Opt. Lett. 27, 1839-1841 (2002).
[26] Y. Chen and H. A. Haus, J. Opt. Soc. Am. B 16, 24-30 (1999).
[27] N. R. Newbury, B. R. Washburn, K. L. Corwin, and R. S. Windeler, Opt.
Lett. 28, 944-946 (2003).
1842-1844 (2002).
Chapter 4
MICROSTRUCTURE FIBER AND WHITE-
LIGHT GENERATION
Alexander L. Gaeta1 and Robert S. Windeler2

1
School of Applied and Engineering Physics, Cornell University
2
OFS Laboratories
Abstract: Microstructure fibers have played a key role in the production of coherent
frequency combs that span more than an octave of bandwidth. In this chapter,
we review the fabrication process for such fibers and their linear and nonlinear
optical properties. We also discuss the underlying physical processes that give
rise to supercontinuum generation.
Key words: microstructure fiber, supercontinuum generation, ultrashort pulse propagation,

fiber dispersion
1. INTRODUCTION
The unlikely merger of ultrafast optics, microstructure fibers, and

nonlinear optics with frequency stabilization techniques has led to a
revolution in frequency metrology. Development over the past decade of
ultrafast solid-state laser systems, such as Ti:sapphire, has resulted in laser
oscillators that can produce sub-100 fs, nanojoule-energy mode-locked
pulses in the near infrared (IR). In parallel, novel single component fibers
with cross sections consisting of regularly spaced air holes were created [1]
that for large air holes allowed the creation of waveguides that strongly
confine the light field to a small area and shift the zero dispersion point of
the fiber close to the operating wavelength of the Ti:sapphire oscillators. The
combination of the high peak intensities due to strong light confinement and
the large effective interaction lengths in the fibers results in a highly intense,
nonlinear interaction that can generate a remarkably broad-bandwidth
spectrum spanning more than an octave of the central laser frequency [2].
98 Chapter 4
This coherent white-light spectrum is precisely what is needed to stabilize

the underlying frequency comb of the mode-locked laser oscillator. It has
also led to other applications such as optical coherence tomography [3].
In this chapter, we review the basic properties of microstructure fibers,
how they are fabricated, and the underlying physics of supercontinuum
generation using ultrashort laser pulses in these fibers.
2. MICROSTRUCTURE FIBER FABRICATION
Microstructure fibers have unique properties and can deliver

functionalities superior to many of the best transmission and specialty fibers.
Their unique properties are obtained from an intricate cross section of high-
and low-index regions that traverse the length of the fiber. The vast majority
of these fibers consist of silica for the high-index region and air for the low-
index region. These fibers are known by several different names including
microstructure fiber, holey fiber, and photonic crystal fiber.
The index profiles that make these fibers unique can also lead to
inherently high loss. Loss occurs at connections launching light in and out of
the fiber and along the length of the fiber. Light is launched into these fibers
by free-space optics, low-temperature fusion splicing, or butt coupling. For
microstructure fibers with small cores, coupling with free-space optics
produces the smallest insertion loss. Loss due to significant mode mismatch
occurs when splicing or butt coupling standard fibers to small-core
microstructure fibers. Attenuation along the fiber length can occur because
of impurities, hole surface roughness, or poor confinement but is presently
low enough that it is not important for most device applications.
Like traditional fibers, microstructure fibers are fabricated in a two-step
process. First, a large-scale template of the fiber, called a preform, is
fabricated. Second, the preform is drawn (stretched) into a fiber, typically
kilometers in length. However, new preform fabrication and draw techniques
had to be developed to incorporate air holes in the preform and have them
remain through draw.
2.1 Preform fabrication
Preforms are long cylinders, typically less than a few centimeters in

diameter, that closely match the structure of the desired fiber. Their large
scale, compared to fiber, makes them easy to handle when assembling the
desired microstructure pattern. They are typically composed of silica with air
holes running uniformly along their length. Other amorphous materials are
4. MICROSTRUCTURE FIBER AND WHITE-LIGHT GENERATION 99
also being explored such as polymers and highly nonlinear multicomponent

glasses.
The basic parameters of the fiber are hole size, hole position, pitch
(center-to-center hole spacing), core diameter, and number of layers. They
are usually determined by the geometry of the holes in the preform.
However, changes in the size and shape of the fiber’s holes can be made
purposely or accidentally during draw, causing deviations between the fiber
and preform profiles.
The most common method used to fabricate microstructure fibers is to
stack tubes, rods, and core rods (rods containing doped cores) in a closely
packed arrangement such as a triangular or hexagonal lattice (Figure 4-1).
This assembly is bound together by a large tube, called an overclad tube.
Fibers can be made with complicated or asymmetric index profiles by
strategically placing tubes with the same outer diameter (for good stacking)
but with different inner diameters to change the index profile in that region.
For example, a birefringent fiber can be produced by placing smaller holes
on opposite sides of the core.
Figure 4-1. Left: Drawing of a preform fashioned from a close-pack arrangement of tubes and
rods. Right: A scanning-electron-microscope (SEM) micrograph showing the fiber after the
preform is drawn.
The main advantages of stacking are that no special equipment is needed

for fabricating preforms and that doped cores can be easily added. However,
there are several disadvantages: First, the stacking method is limited to
simple geometries of the air holes because the tubes are stacked in a closely
packed arrangement. Second, unless hexagonal tubes are used, interstitial
areas are created between the tubes that may not be desired in the final fiber.
Third, the method is labor intensive and requires significant glass handling,
making it difficult to produce ultralow-loss fiber because of impurities.
Several alternative methods are being used to fabricate microstructure
performs, including casting, extrusion, and drilling. Like stacking, each
process has its advantages and disadvantages. The methods most likely to
100 Chapter 4
replace or coexist with stacking are extrusion and casting. These processes
have the advantage of not requiring precision tubing and are ideal for
polymer, sol-gel slurry, or low melting-point glass. The main advantage of
extrusion and casting is that complicated structures can be fabricated in
which the position, size, and shape of the air regions are independent of one
another. These methods may become more common as complex air
structures are needed to make advanced microstructure fibers.
Another method consists of drilling holes in a traditional preform or rod.
Drilling is well understood and is used for other specialty fibers. The
advantage of this method is that it is easy to put variously sized holes in any
position in a preform, including doped regions. The disadvantages are that
the holes cannot be drilled very deep compared to a standard preform length;
the distance between holes may be limited due to cracking; and the fiber may
experience high loss due to surface roughness of the holes and impurities
incorporated during drilling.
2.2 Fiber draw
Draw is the most difficult part of making a microstructure fiber. There

are typically many holes in the preform that all have a propensity to shrink to
decrease the surface tension of the preform. There are two techniques that
can be use to minimize or stop hole collapse. The first is to draw the preform
under very high tension by operating the draw furnace at a low temperature.
Minimal hole collapse will occur if the draw tension is significantly larger
than the surface tension. However, this method also results in a substantial
increase in the number of breaks during draw. This is not important if only
small lengths are needed, but it is not practical as a production method.
The second method is to pressurize the holes with an external source to
counteract the surface tension. The holes can be made larger or smaller
during draw by changing the pressure. However, the process can become
unstable because a large hole needs a lower pressure to maintain its size than
a small hole. If holes of different sizes exist, a larger hole will grow at the
expense of a smaller one regardless of the pressure used. To minimize the
instability, this method is typically performed in conjunction with relatively
high-tension drawing.
3. MICROSTRUCTURE FIBER TYPES
The different types of microstructure fiber are achieved by varying the

size, spacing, and pattern of the air holes within the fiber. Most
microstructure fiber types guide light by total internal reflection. Large air
holes in the fiber create substantial refractive-index differences that result in

large numerical apertures and optically isolated regions within the fiber.
Fibers with large holes surrounding a small core create ideal nonlinear fibers
at short wavelengths [Figure 4-2(a)]. As discussed in Section 5.2, these
fibers can be designed to have zero-group-velocity dispersion in the near-
visible regime, and their small core can result in very high intensities over
extended interaction lengths.
Figure 4-2. SEM photographs of several types of microstructure fiber. (a) High-nonlinearity
fiber. (b) Air-clad fiber. (c) Endlessly single-mode fiber [4]. (d) Band-gap fiber [6].
Air-clad fibers consist of large air holes closely packed around an inner
cladding containing a doped core [Figure 4-2(b)]. These fibers have inner
claddings with very high numerical apertures and are ideal for generating
efficient, high-power cladding pump amplifiers and lasers.
At the other extreme, fibers with small air holes have been designed [4]
such that the cladding index changes with wavelength to create fibers that
are endlessly single moded [Figure 4-2(c)]. These fibers can have very large
core diameters and therefore can propagate high powers at low intensities.
Fibers with precise, periodically spaced index regions in the cladding can
create band-gap guidance, allowing light to be guided in a core with an index
lower than the cladding [Figure 4-2(d)]. Guidance in an air core has been
demonstrated [5] and has the potential to have an attenuation over an order
of magnitude lower than the best traditional transmission fiber [6].
102 Chapter 4
4. LINEAR OPTICAL PROPERTIES OF

MICROSTRUCTURE FIBER
The success of using small-core microstructure fibers with large core-

cladding index contrast to produce frequency combs that extend for more
than an octave at optical frequencies can be attributed to two properties that
occur simultaneously when the core diameter is made small: (1) strong
waveguide confinement that enhances the effective nonlinearity of the fiber,
and (2) the shift of the zero-group-velocity dispersion point to a wavelength
that coincides with that of Ti:sapphire femtosecond laser oscillators.
Microstructure fibers can be modeled reasonably well as a step-index
fiber with a glass core and a surrounding cladding region with an effective
refractive index governed by the air-filling fraction of the glass lattice. As
such, the mechanism that effectively guides light in these fibers is total
internal reflection. However, unlike conventional fibers where the contrast of
the refractive indices of the core and cladding is small, the index contrast for
microstructure fibers can vary over a much larger range and can be
effectively as large as that for a glass rod in air. Many of the optical
properties can be well approximated by the effective index model in which
the effective index neff of refraction of the cladding region is given by
2
neff = f + (1− f ) nglass
2
, (1)
where f is the air-filling fraction of the cladding region and nglass is the index
of the glass that comprises the structure. The solutions for the modes and the
dispersion using this simple step-index model can be obtained analytically,
and the dispersion for the fundamental mode agrees very well with the
predictions of the full-vector model [7]. It is apparent by calculating the
effective V-number [8] that even for small cores and reasonably large air-
filling fractions (f > 0.5), the fiber is multimode (i.e., V > 2.405).
Nevertheless, since the difference in the propagation constant between the
fundamental mode and the higher-order modes is large, a strong perturbation
is required to couple energy from the fundamental mode to the higher-order
modes. Thus if the fundamental mode is initially excited, the energy does not
easily couple to the higher-order modes [9]. In addition, the large index
contrast also leads to a strong confinement of the fundamental mode with
little energy lying outside the core. This results in a large effective
nonlinearity that allows for strong nonlinear interactions with relatively
modest pulse energies such as those from a femtosecond Ti:sapphire
oscillator.
A key property for determining the propagation of ultrashort pulses in the

fiber and for generation of a broad-band frequency comb is the group-
velocity dispersion (GVD) β2 = d2k/dω2 of the propagating mode, where
k(ω) = neff(ω) ω/c is the wave number for the fundamental mode. The two
sources of dispersion in the fiber are material dispersion and waveguide
dispersion. For bulk fused-silica glass, the wavelength at which the GVD is
zero is λ = 1.27 µm. In conventional step-index fibers with small index
contrast, material dispersion dominates and is only slightly modified by the
waveguide contribution. However, in small-core microstructure fibers with a
large air-filling fraction, the waveguide dispersion dominates over most of
the visible and near-IR regimes. Figure 4-3(a) is a plot using the simple step-
index model of the predicted GVD parameter D = (−2πc/λ2)β2 for an air-
filling fraction of unity for various values of the core diameter. For D > 0
(<0) the GVD is termed anomalous (normal). For comparison, the dispersion
for bulk-fused-silica glass is also shown. As the core diameter is reduced, the
zero-GVD point shifts to shorter wavelengths and the maximal value of the
anomalous dispersion increases. For sufficiently small cores, the zero-GVD
point can be shifted completely into the visible regime, allowing for the
generation of solitons in the near IR [9] and potentially the visible regime.
For a 2 µm diameter core, the zero-GVD point is close to the central
operating wavelength of femtosecond Ti:sapphire laser oscillators. Figure 4-
3(b) shows the dispersion for a fiber with a 2 µ m core for various values of
the cladding’s air-filling fraction f. The primary effect of decreasing f is to
decrease the maximal anomalous dispersion and change the slope of the
GVD (i.e., the third-order dispersion) at the zero-GVD point. The shift in the
zero-GVD wavelength is relatively small. Thus, a “rule of thumb” in
designing these high air-filling fraction fibers for a particular application is
that the location of the zero-GVD point is controlled by the size of the core
and the third-order dispersion at this point is controlled by the air-filling
fraction. Such a picture has been verified by experiments that have
characterized the dispersion properties of the fibers with a range of core
sizes [10] and provides insight for the creation of microstructure fibers with
a relatively flat GVD over a large wavelength span.
104 Chapter 4
Figure 4-3. Plots of the group-velocity-dispersion parameter D for (a) various values of the
diameters d and an air-filling fraction f = 1 and (b) various values of f for a core diameter d =
1.7. The dashed line represents the dispersion curve for bulk-fused-silica glass.
5. SUPERCONTINUUM GENERATION
5.1 Nonlinear envelope equation
For subpicosecond optical pulses propagating in optical fibers, the

dominant optical nonlinearity is the intensity-dependent refractive index.
This nonlinear-index change leads to self-phase modulation that results in a
time-dependent phase that spectrally broadens the laser pulse. The addition
of new frequency components significantly outside the bandwidth of the
pulse can also occur via four-wave mixing. The efficiency of these nonlinear
processes can be greatly altered by the linear dispersion of the fiber since it
leads to a temporal broadening of the pulse and a corresponding decrease in
the peak intensity. As a result of these nonlinear interactions, the spectral
bandwidth of a pulse propagating within the fiber can increase substantially
and, most importantly for frequency comb applications, become greater than
the center input frequency ω0 of the incident pulse. The broad spectral
radiation that is produced under such conditions is termed supercontinuum
generation, or white-light generation. A theoretical description of
propagation under these conditions becomes difficult since the concepts of a

pulse-amplitude envelope and a carrier frequency break down. Nevertheless,
much of the underlying physics describing supercontinuum generation can
be described within the framework of the one-dimensional nonlinear
envelope equation (NEE) [11] with the inclusion of the effects of stimulated
Raman scattering.
In this model, the pulse is assumed to propagate along the z-axis with a
wave vector amplitude k0 = n0ω0/c where n0 is the linear refractive index of
the material at the central frequency ω0 of the pulse. For an input pulse with
a peak amplitude A0 and a pulse duration τp, the equation for the normalized
amplitude u(z, t) = A(z, t)/ A0 can be expressed as
∂u
∑ i ∂
Lds 
= −i sgn( β 2 ) + i  1+  p , (2)
∂ζ n=2 L
( n)
 ω 0τ p ∂τ  nl
ds
where Lds = τp2/|β2| is the dispersion length, ζ =z/Lds is the normalized

propagation distance, β2 is the group-velocity dispersion (GVD),
Lds(n) = τpn/βn is the nth-order dispersion length, τ = (t − z/vg)/τp is the
normalized retarded time for the pulse traveling at the group velocity vg, and
pnl is the normalized nonlinear polarization. Inclusion of both instantaneous
(i.e., electronic) and noninstantaneous (i.e., nuclear) nonlinear refractive
index changes in the nonlinear polarization yields
τ
Lds  
pnl = (1 − f )|u| + f
2
∫ dτ g (τ − τ ') | u |2  u , (3)
Lnl  −∞ 

where Lnl = (c/ω0 n2I0) is the nonlinear length, I0 = n0c|A0|2/2π is the peak
input intensity, f is the fractional contribution of Raman scattering to the
nonlinear refractive index, and g(τ) is the Raman-response function [12].
The presence of the operator 1+ i∂/ω0τp∂τ in the nonlinear polarization term
accounts for self-steepening effects and allows for the modeling of the
propagation of pulses with spectral widths comparable to the central
frequency ω0.
5.2 Spectral superbroadening
For a femtosecond pulse propagating in a fiber, the central wavelength of

the pulse relative to the zero-GVD point strongly determines the nature of
the spectral broadening and supercontinuum generation. For the case in
which the bandwidth of the pulse overlaps the zero-GVD point, the
106 Chapter 4
qualitative characteristics of the supercontinuum generation can be attributed

to the combined action of self-phase modulation and third-order dispersion
[13]. Figure 4-4 plots the supercontinuum spectrum and temporal profile at
two distances within the medium under conditions similar to that of
Reference [2] of an initially 100 fs, 10 kW pulse at 770 nm inside a fused-
silica microstructure fiber with a core size of 1.7 µ m in diameter. That is, Lds
= 500 cm, ω0τp = 140, Lds/ Lds(3) = 0.5, Lds/ Lds(4) = −0.005, and f = 0.15. After
only a 2 mm propagation distance [Figure 4-4(a)], the pulse spectrum has
broadened to an octave. The temporal profile [Figure 4-4(b)] is highly
complicated with no well-defined pulse structure. At longer distances
[Figure 4-4(c)], the spectral “envelope” exhibits little change; however, a
significant amount of substructure is seen. This substructure can be
understood as arising from the formation of well-defined pulses [Figure 4-
4(d)], known as soliton fission [14], that is accompanied by nonsolitonic
radiation at the short-wavelength side. Although stimulated Raman
scattering and self-steepening result in quantitative changes in the spectrum,
in the limit where the GVD is small, the basic shape of the supercontinuum
spectrum depends primarily on the amount of third-order dispersion and, to a
lesser extent, the amount of fourth-order dispersion.
Figure 4-4. Theoretically predicted supercontinuum spectra and temporal profiles for z = 0.02
cm (a,b) and z = 1.2 cm (c,d).
For the case in which picosecond and nanosecond pulses are used, four-
wave mixing and Raman scattering can produce a broadband
supercontinuum in long fiber lengths [15]. Tailoring the dispersion of the
fiber can result in control of the continuum such that certain spectral parts of
the continuum are suppressed and others are enhanced [16]. The generated
supercontinuum has also been found to be highly dependent on polarization
because of the birefringent nature of microstructure fibers [17].
5.3 Continuum instability and noise
Although at these powers, changes in the input pulse energy do not

significantly alter the spectral envelope of the supercontinuum. However,
numerical simulations show that the fine substructure is highly sensitive to
the initial pulse energy. Figures 4-5(a) and (b) show spectra under identical
conditions, except that the curve in (b) is for 0.1% higher input pulse energy,
which corresponds to the fluctuations in a typical Ti:sapphire femtosecond
oscillator. Although the gross spectral shape in both cases is nearly the same,
the fine substructure for the two cases is completely different. This fine
substructure and spectral instability were not observed in early
supercontinuum-generation experiments either because of insufficient
spectral resolution or as a result of the spectra consisting of an average over
many laser pulses. Later experiments [18], in which single-pulse spectra
were obtained, confirmed the existence of this fine substructure and its
sensitivity to input pulse energy, as shown in Figure 4-6. The same
wavelength subsection is shown for two different single-shot
supercontinuum spectra, and the fine substructures are observed to be
uncorrelated. Also shown is an average of four single-shot spectra. For
certain applications, this sensitivity of the supercontinuum spectra to input
fluctuations could impose limitations on its applicability.
Another important issue for frequency-locking applications is noise
added as a result of the nonlinear interaction, even for the case in which the
input pulse is perfectly stable. Stimulated Raman scattering and
modulational instability arising from four-wave mixing both result in
amplification of vacuum fluctuations and spontaneous Raman scattering.
These processes lead to a reduction in the degree of coherence of the
continuum spectrum [19] and to noise as much as 40 dB above the shot-
noise limit in the frequency regions between the modes of the frequency
comb [20]. The conclusion from these experiments and theoretical analysis
is that for frequency metrology applications, it is desirable to use laser pulses
with durations of less than 50 fs to minimize the effects of modulational
instability and stimulated Raman scattering.
108 Chapter 4
Figure 4-5. (a) Output spectrum for an input peak power P = 16 kW and the propagation
distance 1.5 cm. (b) Same as (a), but with 0.1% higher peak power. (c) High resolution
window of the spectra in (a) (solid line) and (b) (dotted line).
Figure 4-6. Experimental observation of spectral sensitivity to input pulse fluctuations, as

described in Reference 21. The dashed and dotted curves are single-shot spectra taken
seconds apart. The solid curve is the average of four spectra.
6. CONCLUSIONS
We expect that advances in supercontinuum generation in microstructure

fibers will continue to be made and that applications, such as carrier-
envelope phase control for attosecond pulse generation [21], will be
developed for this coherent white-light source. Future advances could
include designing both the transverse and longitudinal structure of the fiber
to tailor the shape of the continuum for specific applications. For example,
by using a combination of both anomalous and normal dispersion sections,
the fine substructure and overall sensitivity of the continuum to input
fluctuation could be minimized [22]. Other designs could be developed to
lower the threshold power, which could help avoid long-term damage to the
fiber endfaces that are crucial for frequency-stabilization techniques
requiring extended time periods. Further developments of femtosecond fiber
laser systems with nanojoule energies also promise to create highly compact,
stable, and coherent white-light systems that can be used under
environmentally challenging field conditions.
ACKNOWLEDGEMENTS
A.L.G. gratefully acknowledges discussions over the past few years on

these topics with F. Ahmad, S. Cundiff, J. Dudley, M. Foster, K. Koch, K.
Moll, D. Ouzounov, J. Ranka, R. Trebino, and J. Ye.
110 Chapter 4
REFERENCES
[1] J. C. Knight, T. A. Birks, P. S. Russell, and D. M. Atkin, Opt. Lett. 21,
1547-1549 (1997).
[3] I. Hartl, X. D. Li, C. Chudoba, R. K. Ghanta, T. H. Ko, J. G. Fujimoto, J.
K. Ranka, and R. S. Windeler, Opt. Lett. 26, 608-610 (2001).
[4] T. A. Birks, J. C. Knight, and P. S. Russell, Opt. Lett. 22, 961-963 (1997).
[5] J. C. Knight, J. Broeng, T. A. Birks, and P. S. J. Russel, Science 282, 1476-
1478 (1998).
[6] C. M. Smith, N. Venkataraman, M. T. Gallagher, D. Muller, J. A. West, N.
F. Borrelli, D. C. Allan, and K. W. Koch, Nature 424, 657-659 (2003).
[7] R. D. Meade, A. M. Rappe, K. D. Brommer, J. D. Joannopoulos, and O. L.
Alerhand, Phys. Rev. B 48, 8434-8437 (1993).
[8] G. P. Agrawal, Nonlinear Optics (Academic Press, San Diego, 2001).
[9] J. K. Ranka, R. S. Windeler, and A. J. Stentz, Opt. Lett. 25, 796-798
(2000).
[10] J. C. Knight, J. Arriaga, T. A. Birks, A. Ortigosa-Blanch, W. J. Wadsworth,
and P. S. Russell, IEEE Photonics Technol. Lett 12, 807-809 (2000); D.
Ouzounov, D. Homoelle, W. Zipfel, W. W. Webb, A. L. Gaeta, J. A. West,
J. C. Fajardo, and K. W. Koch, Opt. Commun. 192, 219-223 (2001).
[11] T. Brabec and F. Krausz, Phys. Rev. Lett. 78, 3282-3285 (1997).
[12] R. H. Stolen, J. P. Gordon, W. J. Tomlinson, and H. A. Haus, J. Opt. Soc.
Am. B 6, 1159-1166 (1989).
[13] A. L. Gaeta, Opt. Lett. 27, 924-926 (2002).
[14] A. V. Husakou and J. Herrmann, Phys. Rev. Lett. 8720, art. no.-203901
(2001); J. Herrmann, U. Griebner, N. Zhavoronkov, A. Husakou, D. Nickel,
J. C. Knight, W. J. Wadsworth, P. S. J. Russell, and G. Korn, Phys. Rev.
Lett. 88, art. no.-173901 (2002).
[15] J. M. Dudley, L. Provino, N. Grossard, H. Maillotte, R. S. Windeler, B. J.
Eggleton, and S. Coen, J. Opt. Soc. Am. B 19, 765-771 (2002); S. Coen, A.
H. L. Chau, R. Leonhardt, J. D. Harvey, J. C. Knight, W. J. Wadsworth,
and P. S. J. Russell, J. Opt. Soc. Am. B 19, 753-764 (2002).
[16] K. M. Hilligsoe, T. V. Andersen, H. N. Paulsen, C. K. Nielsen, K. Molmer,
S. Keiding, R. Kristiansen, K. P. Hansen, and J. J. Larsen, Opt. Expr. 12,
1045-1054 (2004).
[17] Z. M. Zhu and T. G. Brown, Opt. Expr. 12, 791-796 (2004).
[18] X. Gu, L. Xu, M. Kimmel, E. Zeek, P. O'Shea, A. P. Shreenath, R. Trebino,
and R. S. Windeler, Opt. Lett. 27, 1174-1176 (2002).
[19] J. M. Dudley and S. Coen, Opt. Lett. 27, 1180-1182 (2002); X. Gu, M.
Kimmel, A. P. Shreenath, R. Trebino, J. M. Dudley, S. Coen, and R. S.
Windeler, Opt. Expr. 11, 2697-2703 (2003).
[20] K. L. Corwin, N. R. Newbury, J. M. Dudley, S. Coen, S. A. Diddams, K.
Weber, and R. S. Windeler, Phys. Rev. Lett. 90, 113904 (2003); J. N.
Ames, S. Ghosh, R. S. Windeler, A. L. Gaeta, and S. T. Cundiff, Appl.

Phys. B 77, 279-284 (2003).
[21] A. Baltuska, T. Udem, M. Uiberacker, M. Hentschel, E. Goulielmakis, C.
Gohle, R. Holzwarth, V. S. Yakoviev, A. Scrinzi, T. W. Hänsch, and F.
Krausz, Nature 421, 611-615 (2003).
[22] T. Hori, J. Takayanagi, N. Nishizawa, and T. Goto, Opt. Expr. 12, 317-324
(2004).
Chapter 5
OPTICAL COMB DYNAMICS AND
STABILIZATION
Günter Steinmeyer 1 and Ursula Keller2

1
Max-Born-Institut für Nichtlineare Optik und Kurzzeitspektroskopie
2
Institut für Quantenelektronik, ETH Zürich
Abstract: The spectrum of a mode-locked laser consists of a comb of equidistantly

spaced frequencies. This comb has only two degrees of freedom, its offset
frequency at zero and the spacing of the teeth of the comb. While the spacing
of the frequencies is simply determined by the repetition rate of the laser and
can be relatively easily controlled, the offset frequency is governed by phase
differences between the carrier and the envelope of the pulses during one
round trip through the laser cavity. This carrier-envelope offset (CEO) phase is
measured via heterodyning different harmonics of the mode-locked laser
spectrum. In an unstabilized laser, this CEO phase exhibits very strong noise
and can fluctuate several thousand radians in only one second. We provide an
analysis of CEO phase-noise contributions and track their origin. The passive
stability with respect to CEO fluctuations can be greatly improved by suitable
cavity design, which greatly simplifies the stabilization of the CEO phase.
Recent efforts on carrier-envelope stabilization are reviewed and some
limitations are outlined.
Key words: dispersion, carrier-envelope phase, phase noise, phase stabilization
1. INTRODUCTION
Ultrashort pulse generation reached pulse durations of a few
femtoseconds around the turn of the century [1]. Some of the shortest pulses
have a width of only two optical cycles [2-4]. When pulse durations
approach this regime, the commonly used approach of the slowly varying
envelope approximation (SVEA) starts to fail. Nonlinear optical effects are
then expected to depend not only on the envelope structure of the pulses, but
also on the structure of the electric field itself, including its relative phase to
5. OPTICAL COMB DYNAMICS AND STABILIZATION 113
the envelope, which has been referred to as the absolute phase. In 2001 and
2003, the first experimental evidence for the failure of the SVEA was
reported [5].
In this chapter, we describe how to monitor the relative phase between
the envelope and the carrier of an optical pulse train from a mode-locked
optical oscillator. The chapter is organized as follows: First, we introduce the
definition of the carrier-envelope-offset (CEO) phase and frequency. We
show how both entities are connected with intracavity dispersion. Methods
to measure the CEO frequency are discussed, with an emphasis on the most
wide spread ν-to-2ν scheme. Careful characterizations of the CEO-phase-
noise spectra are used to isolate the physical mechanisms behind the
excessive fluctuations of the CEO phase of a free-running oscillator. This
insight allows building femtosecond lasers with an increased passive
stability of the CEO and forms the basis for subsequent stabilization of the
CEO frequency. Finally, we discuss how to optimize control of the CEO
phase and how to push residual phase jitter into the attosecond range.
2. COMB PARAMETERS AND THEIR

CONNECTION TO INTRACAVITY DISPERSION
2.1 Carrier-envelope-offset phase and frequency in the

time domain
Figure 5-1 shows the envelope and the carrier of two subsequent pulses
from a mode-locked laser. The envelope travels at group velocity
υ g = c / n g = c /(n + ω dn / dω ) and repeats itself after the cavity round-trip
time TR. The underlying carrier propagates at phase velocity υ p = c / n .
Generally, υ p ≠ υ g in any dispersive medium. This means that the electric-
field structure of the pulse will undergo a permanent change. The drift of the
relative phase between carrier and envelope can be tracked down to the
dn / dω term in the definition of the group velocity.
When propagating through a dispersive material with an index of
refraction n(z ) along the axis z, the pulse will accumulate a phase offset
between the carrier and envelope of
114 Chapter 5
φce
TR = 1/fr
Figure 5-1. The electric field E (t ) of two subsequent pulses from a mode-locked laser (solid
line). The envelope ± A(t ) is shown as dashed lines. The electric-field patterns of the pulses
experience a pulse-to-pulse phase shift ∆ φ ce according to Equation (1).
 2π L  ω 2 L dn( z ) 
∆φce =  ∫ n g ( z ) − n( z ) d z  mod 2π =  ∫ d ω d z  mod 2π .
(1)


λ 0  

c 0 
Here L is the length of the dispersive material. For the case of a linear cavity,
L takes the role of twice the cavity length, and the carrier-envelope offset
(CEO) phase ∆φce is the change of the phase φ ce per round trip:
∆φce (t ) = φ ce (t ) − φ ce (t − TR ) . (2)
The CEO phase ∆φ ce must not be confused with the phase φ ce , which is
typically defined such that a pulse with φ ce = 0 has the largest possible value
of the electric field [6]. Some authors have referred to φ ce as the absolute
phase. An example for such a pulse is shown as the left pulse in Figure 5-1.
∆φ ce , however, is defined as the difference of the absolute phase of two
subsequent pulses. It is useful to introduce the CEO frequency [7]
∆φce
f0 = fr , (3)
2π
where fr equals the inverse round-trip time 1/TR of the cavity and f0 is time
dependent unless the intracavity dispersion and the cavity length are
absolutely constant with time. Changes of f0 or ∆φce tend to be unnoticeably

small on a pulse-to-pulse time scale but can reach significant magnitudes on
millisecond time scales, rendering stabilization of these parameters a
nontrivial task. For a detailed derivation of the fundamental comb
parameters from periodic waveforms and their connection to intracavity
dispersion, we refer the reader to Reference [8].
2.2 The frequency comb and its dynamics
fr
f0
Figure 5-2. Equidistant frequency comb of a mode-locked laser. The comb lines are spaced
by the repetition rate fr and exhibit a nonvanishing offset frequency f0 at zero frequency unless
the electric-field pattern exactly reproduces from pulse to pulse (compare to the time domain
picture in Figure 5-1).
For the following considerations, it is useful to revisit the scenario of

Figure 5-1 in the Fourier domain. As the pulses follow each other at a
constant delay TR, their spectrum consists of a comb of equidistantly spaced
frequencies with a separation fr = 1/TR. This frequency comb must not be
confused with the modes of a linear cavity, which are only equidistant in the
absence of intracavity dispersion. In contrast, if the spacing between the
teeth in the mode-locked frequency comb were not constant, different
Fourier components of the pulse in Figure 5-1 would travel at different
repetition rates inside the cavity, and the pulse would slowly, but surely, drift
apart. The fact that the separation of the frequencies is constant over the
entire comb has been experimentally checked to better than 10-15 [9].
Differing phase and group velocity cause a translation of the entire
frequency comb by the carrier-envelope-offset frequency f0 . The frequencies
of the ith comb component can therefore be written in the form:
116 Chapter 5
ν i = f 0 + if r . (4)
This leaves only two degrees of freedom for the dynamics of the frequency
comb, translation via f0 and breathing via fr, as illustrated in Figure 5-3. Any
kind of perturbation of the cavity, e.g., by a thermal change of the refractive
index of the laser crystal, will typically affect both the repetition rate and the
CEO frequency.
breathing
translation
Figure 5-3. Comb dynamics [10]. The frequencies inside the comb structure are determined
by two parameters, f0 and fr . This gives rise to a translational degree of freedom and a
breathing mode. Noise contributions will induce a characteristic linear combination of both
degrees of freedom.
However, there is always a fixed frequency fx that remains unaffected by a

distortion X [11].
−1
∂f 0  ∂f r 
f x = f0 + fr   , (5)
∂X  ∂X 
where X could be any physical parameter of the cavity, e.g., its length or the
temperature of the laser crystal. Let us illustrate the concept of a fixed point
by choosing X as the cavity length. Cavity length fluctuations only affect the
repetition rate fr of the laser but leave the per-round-trip phase shift ∆φce
between envelope and carrier unchanged. Inserting Equation (3) into
Equation (5) yields f x = 2 f 0 , i.e., a value very close to zero frequency. A
complementary example would be an effect that causes only a change of the
cavity group delay but leaves the phase delay unchanged. This could be
achieved, e.g., by tilting a mirror in an intracavity prism sequence with the
pivot point adjusted to the center frequency of the mode-locked spectrum
[12]. Retarding the group by one cycle relative to the phase changes f0 by
one free spectral range (i.e., the repetition rate), whereas the repetition rate
itself only changes by a very small amount. In this case, one calculates that fx
equals the carrier frequency of the pulse. Most environmental contributions
to comb dynamics, such as thermal or nonlinear changes of the intracavity
refractive indices, have an f x located between zero frequency and the
carrier frequency [11]. This means that they neither add a pure contribution
to the group delay nor do they only affect the phase delay of the group.
Measuring the fixed frequency f x can help to pinpoint the source of
dominant frequency comb dynamics.
3. MEASUREMENT OF THE CEO FREQUENCY

An early approach for measurement of f0 employed an interferometric
method based on second-harmonic generation (SHG) cross-correlation
between two subsequent laser pulses [13]. For vanishing ∆φce , the cross-
correlation signal is identical to the interferometric autocorrelation, with a
symmetric fringe pattern. In all other cases, the fringe pattern appears shifted
with the fringe maximum located at ∆φce / ω c and the cross-correlation is
asymmetric. Even though this measurement in the time domain works in
principle [14], it is very susceptible to offset errors. Any offset between
group and phase delay in the long arm of the cross-correlator will induce a
measurement error in determining f0. Therefore, phase-coherent methods are
indispensable for precise control of the CEO phase as was suggested in
Reference [6].
3.1 Heterodyning different laser harmonics
Figure 5-4 provides the key to the measurement of f0 by heterodyning

harmonics from different parts of the mode-locked spectrum. The technique
was first proposed by Telle et al. [6]. Taking the Nth harmonic of a comb line
Nν m1 = Nf 0 + Nm1 f r and beating it with the Mth harmonic of another comb
line Mν m2 = Mf 0 + Mm2 f r yields
Nν m1 − Mν m2 = ( N − M ) f 0 , (6)
which requires that Nm1 = Mm2 . Equation (6) is the key to any measurement
of the carrier-envelope offset and was used in the first experimental
demonstrations by Jones et al. [14] and Apolonski et al. [15] for the case of
118 Chapter 5
N = 1. The beat note of Equation (6) delivers the carrier-envelope-phase-

slippage rate, either directly or as one of its harmonics. An example for a
measurement of the CEO frequency, which is based on this scheme, is
ν2 2ν1
f0
ν1 = f0 + n fr 2ν1 = 2f0 + 2n fr
ν2 = f0 + 2n fr
Figure 5-4. Scheme for measuring the CEO-frequency of a laser comb for the case of
heterodyning the fundamental and the second harmonic, i.e., N = 1 and M = 2 in Equation (6).
Graphically, this scheme mirrors the origin at f = ν1, transferring the f0 beat from dc into a
region with nonvanishing spectral content.
However, the scheme of Equation (6) requires a certain minimum

spectral width of the comb ∆ν ν = 2 (N − M ) (N + M ) ; e.g., beating of the
fundamental and second harmonic requires an optical octave of bandwidth
with ∆ν ν = 0.67 . The situation of second-harmonic generation is
illustrated in Figure 5-4. Often, an octave-spanning spectrum is not directly
available from the oscillator. Alternatively, one can meet the bandwidth
requirement by additional external broadening, as shown in the schematic
setup in Figure 5-6, which is actually the most commonly used scheme to
measure the CEO frequency of a laser.
0
-10
CEO-beats
intermode beat
Power density (dBc)
-20
spurious
-30
-40
-50
-60
0 20 40 60 80 100 120
Frequency (MHz)
Figure 5-5. Typical rf spectrum of the CEO beat note signal. This signal was measured at a
Ti:sapphire laser heterodyning the fundamental and the second-harmonic-generation (SHG)
signal from a continuum generated in a microstructure fiber [16]. The CEO beat is located at
35 MHz with a signal-to-noise ratio of >45 dB in a 100 kHz bandwidth. Its mirror frequency
is also visible at 65 MHz. The laser has a 100 MHz repetition rate. Some spurious
contributions have been generated by nonlinear electronic mixing processes in the detector
circuitry.
structure fiber 1060 nm

800nm
Laser
SHG
530 nm
530 nm CEO
Figure 5-6. Schematic drawing of a practical implementation of Equation (6) and Figure 5-4.
The laser is spectrally broadened to more than an optical octave using continuum generation
in a microstructure fiber. Two wavelengths forming one octave are separated. The long-
wavelength component is frequency doubled and heterodyned with the fundamental signal.
The beat note contains an rf component at the CEO frequency. Specific wavelengths shown
are meant as example values.
3.2 Transfer oscillators and interval bisection
Let us briefly mention other ways to circumvent the bandwidth

bottleneck [6]. These alternative methods rely on transfer oscillators and
interval bisection. A transfer oscillator is an additional single-frequency cw
laser with frequency νtrans. This frequency or one of its harmonics has to be
locked to one of the comb frequencies. At the same time, the fundamental
120 Chapter 5
transfer oscillator is used for sum-frequency generation from the low-

frequency part of the comb to the high-frequency part. The beat notes of
both then allow extraction of the CEO frequency. In particular, when the
second or third harmonic of the transfer oscillator can be generated at
sufficient power, this scheme can be operated with smaller bandwidth
requirements. If the second harmonic of the transfer oscillator is locked to
the comb, only 0.6 of an optical octave is required in the comb. For third-
harmonic generation, only 0.4 optical-octave bandwidth is required.
Frequency-interval bisection [17] can be employed to divide the required
comb width. The simplest scheme of this type uses a one-octave interval-
divider stage that generates the frequency νa at the midpoint of νb and 2νb.
This can be done by phase locking the second harmonic of νa to the sum
frequency of νb and 2νb to yield
ν a 3ν b 2 3
= = . (7)
2ν b 2ν b 4
This way, one now has available two phase-locked oscillators at only half-
an-octave spectral separation. Locking one to the comb and phase
comparison of the other allow extraction of the CEO frequency of the comb.
4. CEO PHASE NOISE
From the discussion so far, it should be clear that the carrier-envelope

phase of a femtosecond oscillator is extremely sensitive to any kind of
environmental influence and changes of the laser parameters such as pulse
duration or power fluctuations. In this section, we discuss measurements of
the carrier-envelope-phase noise of a laser oscillator. From these
measurements, one can judge the severity of the noise problem. It is also
helpful to reach an understanding of the mechanisms behind the carrier-
envelope-phase noise before attempting to stabilize the CEO phase. For this
reason, we discuss measurements of the CEO phase noise for different laser
configurations. We restrict ourselves to measurements of Ti:sapphire lasers,
as the CEO-phase noise in these lasers has been analyzed with great scrutiny.
The general carrier-envelope-phase noise shows striking similarities with
timing-jitter noise in mode-locked lasers [18, 19]. We therefore use a similar
formalism to describe and analyze the dynamics of the carrier-envelope-
offset frequency and phase.
4.1 Noise densities and rms phase jitter
The easiest way to characterize the CEO-noise properties consists of

frequency-to-voltage conversion of the CEO beat noise and subsequent
spectral analysis with a Fourier analyzer or a similar spectrum analyzer.
Multiplying the measured voltage noise by the conversion factor of the
frequency-to-voltage converter yields the single-sideband frequency noise
density σ f0 ( f ) in units Hz Hz vs Fourier frequency f [16]. Other ways
to measure the frequency noise density have been described in [20].
However, the latter approach relied on constant amplitude-to-phase coupling
over the entire measurement range. The frequency-noise density can be
easily converted into a phase-noise density using the identity
σ f0
σ φce = . (8)
f
For an interpretation of the noise data, it is sometimes more useful to

integrate the noise densities according to
2
 σ f0
fr 2 fr 2

δf 0(f low ) = 2 ∫ σ 2f 0 d f , δφ ce(f low ) = 2 ∫ 
 f


df . (9)
flow flow  
The integration spans from a lower frequency flow given by the inverse
measurement time to an upper bound that is ideally half the repetition rate
itself. Typically, one can only carry the integration to a few tens or hundreds
of kilohertz, which is normally considered sufficient as the noise rolls off
very rapidly at high frequencies. The integrated noise densities δφ ce ( f low )
and δf 0 ( f low ) can be interpreted as rms widths of the fluctuation range of
phase or frequency, respectively.
4.2 CEO-phase noise of mode-locked oscillators

Examples of phase-noise measurements are shown in Figures 5-7 and 5-8
[16, 21]. From the data in Figure 5-7, one can see that the noise generally
rolls off above 1 kHz. At the very end of the measurement range shown in
Figure 5-7, there seems to be an increase of noise that may be explainable by
relaxation oscillations, similar to observations of timing-jitter noise [19]. The
major noise contributions, however, are located in the range of a few 100 Hz
to 1 kHz. For an interpretation of the severity of the noise, it is useful to
122 Chapter 5
inspect the integrated phase noise starting at high frequencies and locating
the point where the δφce approaches unity. In Figure 5-8, one can clearly see
that the integrated phase noise already reaches one radian at several kHz
Fourier frequency. Around 1 kHz, the noise grows dramatically, reaching
values of several hundred to thousands of radians at 100 Hz offset frequency.
In summary, this means that severe phase-noise contributions accumulate
within approximately 1 ms of measurement time. For a successful phase lock
of the CEO frequency, however, it is mandatory to keep residual jitters
below δφce ≈ 0.3 rad [22]. This requirement is only set by the cycle-slip-free
functioning of the phase-locked loop, whereas applications may demand an
even tighter locking to smaller jitter values. These considerations make it
clear that for any meaningful stabilization of the CEO frequency, servo
bandwidths of 10 kHz or more are required, which makes the use of acousto-
optic or electro-optic controls preferable to a mechanical adjustment of
intracavity dispersion.
4
10 without prisms
with prisms
Freq. noise density (Hz/÷Hz)
3
10
2
10
phase locked
1
10
0
10
0 1 2 3 4 5
10 10 10 10 10 10
Offset frequency (Hz)
Figure 5-7. Frequency noise density of different Ti:sapphire lasers [16, 21]. The top trace
shows a measurement for a laser with intracavity prisms; the middle trace is measured for a
prismless variant of the same laser. The bottom trace shows how the measured frequency
noise density drops farther when a phase lock to a reference oscillator is activated. This
measurement has to be interpreted as a noise floor, as it is limited by the stability of the local
oscillator in the measurement.
Moreover, Figures 5-7 and 5-8 contain measurements for both oscillators
with and without prisms for intracavity dispersion compensation. In these
measurements, an identical pump laser was used for the prismless and the
prism setup. It becomes evident that prism-based femtosecond oscillators are
about ten times as noisy as prismless lasers. This can only be explained by
an additional physical mechanism present because of the intracavity prisms.
Figure 5-8. Integrated phase-noise density of different Ti:sapphire lasers [16, 21]. The top and
middle traces have been computed from the data displayed in Figure 5-7 using Equations (8)
and (9). The bottom trace is based on a direct phase comparison of a stabilized oscillator and
an rf reference using an rf lock-in amplifier.
4.3 Physical mechanisms behind CEO fluctuations

The key to understanding the mechanisms forming the CEO noise is
Equation (1). Any change of temperature, air pressure, or laser power may
also affect ∆φce. For simplicity, let us assume that we have a laser cavity of
length L filled with a material of index n. We can then rewrite the
dependence of Equation (1) on any laser or environmental parameter X as
[13, 21, 23]:
∂ ∂ω ∂n ∂n ∂L ∂2n
∆φ ce = 2ω c c L + ω c2 + ω c2 L. (10)
∂X ∂X ∂ω ∂ω ∂X ∂ω∂X
These three partial derivatives account for changes of the center

frequency ωc [13], changes of the cavity geometry [21], and changes of the
first-order dispersion of the cavity [16, 23], respectively. Let us first consider
environmental effects, which equal X for temperature or air pressure.
Temperature changes of the laser crystal or intracavity prisms immediately
translate into changes of the CEO frequency. Nevertheless, such variations
are relatively slow and cannot fully explain the noise contributions at the
high Fourier frequencies well above 1 kHz in Figures 5-7 and 5-8. A similar
argument holds for pressure variations, which couple to the CEO frequency
by refraction changes caused by the air in the cavity. Another coupling
124 Chapter 5
mechanism is displacement of the cavity mirrors. The latter effect is much

less of a concern than for stabilization of single-frequency lasers, because
the fixed frequency of the comb dynamics lies close to zero frequency. Thus
there is only a negligible effect on f0. In summary, environmental
contributions are an important input to the low-frequency part of the CEO
noise spectrum. Nevertheless, they can be relatively easily reduced by
enclosing the laser in a box, thereby avoiding air turbulence. Typically,
environmental contributions can be suitably reduced by these passive
measures. Therefore they rarely represent an obstacle to successful
stabilization of the CEO frequency.
4.4 Amplitude-to-phase conversion effects
Amplitude-to-phase conversion (APC) is a special case of Equation (10).

The case where X is the intensity deserves special attention, as the resulting
fluctuations of ∆φce can be arbitrarily fast when electronic nonlinearities such
as the all-optical Kerr effect are mediating between amplitude fluctuations
and CEO-phase noise. APC effects are mainly taking place in the laser
crystal, as this is the position of highest intracavity intensities.
Spectral shifting of the laser spectrum has been proposed as the first
mechanism giving rise to APC effects [13]. The carrier frequency ωc shifts
with pump power or intracavity intensity, an effect that strongly depends on
the operating conditions of the laser. Generally, both effects seem to be
weaker when the laser bandwidth is wider. In a recent publication, spectral
shifting was observed for a 750 MHz repetition-rate laser below 50 nm
mode-locked bandwidth, whereas it did not appear to play a role in a 100
MHz repetition-rate laser with its stronger mode locking and higher pulse
energy [23]. APC coefficients ∂f 0 / ∂I on the order of 10-7 HzW/m2 were
observed when spectral shifting dominates the APC, resulting in the
prevalent contribution to Equation (9). In contrast, the APC coefficient drops
to a few 10-9 HzW/m2 in the absence of spectral shifting [16, 21, 23].
A second contribution to CEO-phase noise arises from geometrical
changes of the laser cavity affecting the total cavity length L. This
contribution is typically negligible in prismless cavities but can play a role in
cavities that use intracavity prism sequences for dispersion compensation [3,
21]. One potential mechanism behind such laser dynamics is beam-pointing
variations inside the laser cavity together with the directional sensitivity of
the dispersion of a prism compressor [21, 24]. If the beam direction inside
the prism sequence changes, this will also affect the net first-order dispersion
of the cavity via the second term in Equation (10). Beam-pointing variations
can be induced by changes of the refractive index of the laser crystal. If the
index of refraction of the laser crystal changes, Snell's law demands a change
of angles inside and outside the laser crystal [21]. Beam-pointing effects are
held responsible for an approximately tenfold increase of CEO-phase noise
of prism laser cavities as compared to prismless variants.
The third term in Equation (10) contains contributions to CEO-phase
noise via intensity-induced changes of the refractive index [25]. Nonlinear
refraction is well known as the all-optical Kerr effect [26], but according to
Equation (10), only the dispersion of the Kerr effect affects changes of the
CEO phase. The issue of dispersion of the Kerr effect has been addressed by
[27, 28]. According to Sheik-Bahae et al. [27], the main contribution to the
first-order dispersion of a dielectric medium well below half the band edge
stems from a Kramers-Kronig term induced by two-photon absorption. As
per their example, for sapphire at 800 nm, one calculates
∂ 2 n / ∂ω ∂I ≈ 10 −36 s m2/W rad. Inserting values for typical Ti:sapphire laser
cavities [4], one computes a theoretical estimate of ∂f 0 / ∂I =5×10-9 HzW/m2,
which agrees well with the lowest experimentally observed values of
∂f 0 / ∂I . Again, these low APC coefficients can only be reached in the
absence of geometrical effects and spectral shifting.
8
Coupling coeff. (10 Hz m /W)
measurement
2
6
-8
theoretical estimate
0
1 2 3 4
10 10 10 10
Modulation frequency (Hz)
Figure 5-9. Transfer function of laser intensity noise into fluctuations of the CEO frequency
[16]. These measurements were done in a prismless laser in the absence of spectral shifting.
The measurements reflect Kerr contributions to the APC and thermally induced amplitude-to-
phase conversion (APC) effects at low frequencies.
A measurement of the APC coefficient is shown in Figure 5-9. The

theoretical estimate from Reference [27] is shown as a dashed line. The
measurement was performed at different modulation frequencies. For
frequencies below 1 kHz, an APC coefficient of a few times 10-8 HzW/m2 is
observed. At higher modulation frequencies, the APC levels off to about 10-8
126 Chapter 5
HzW/m2. Spectral shifting does not seem to play a role in these experiments,
and geometrical effects are also not a concern because of a prismless cavity.
At lower frequencies, additional contributions from thermally induced
changes of the refractive index increase the APC effect. For modulation
frequencies of about 10 kHz or more, the coupling dynamics appear to be
restricted to a purely electronic-refractive nonlinearity.
From the experimental observations, some guidelines can be given on
how to keep APC effects to a minimum. The first recommendation is to use
a prismless cavity, which is also strongly supported by the data in Figures 5-
7 and 5-8. In prismless cavities, beam pointing does not translate into CEO-
phase noise [21]. Spectral shifting is the other APC effect that can be
avoided by suitable design of the laser. For a stable position of the laser
spectrum, a broad mode-locked bandwidth of more than 50 nanometers and
a high pulse energy appear to be favorable conditions [23]. If geometric
effects and spectral shifting can be avoided, the APC effects are restricted to
nonlinear refractive mechanisms, both Kerr-type and an additional thermally
induced mechanism at low Fourier frequencies. Values on the order of
∂f 0 / ∂I =10-8 HzW/m2 or less are indicative of a dominance of nonlinear
refraction in the APC dynamics.
5. STABILIZATION OF THE CEO FREQUENCY
In previous sections, we have introduced methods to measure the CEO

frequency. We have also analyzed sources of CEO-phase noise and given
guidelines on how to increase the passive stability of the CEO phase. Even if
such measures can be further improved, the residual noise of the free-
running laser is still prohibitive for many applications in extreme nonlinear
optics, which demand a stable CEO phase for seconds or minutes of
integration times.
5.1 Controlling the CEO frequency of a laser oscillator
The only missing link to stabilization of the CEO frequency is now a

mechanism for external control of the CEO frequency. Such a mechanism
allows closing the servo loop, forcing the CEO frequency into a lock with an
rf-reference oscillator. Ideally, a control mechanism should only act on the
CEO frequency and leave other cavity parameters unchanged
(orthogonality). If we leave this concern aside, all mechanisms causing APC
are suited, in principle, for control of the CEO frequency. As was discussed
previously, a servo bandwidth of more than 10 kHz is needed, which rules
out many slow mechanisms. Choice of the control mechanism is therefore a

trade-off between orthogonality and bandwidth.
In lasers with intracavity prism sequences, an elegant way of controlling
the CEO frequency of a laser without affecting other laser parameters is
offered. Tilting the end mirror after the prism sequence affects only the
difference between the group and phase delay in the cavity but leaves other
laser parameters widely unchanged [12, 29]. The tilt of the end mirror has to
be restricted to small excursions compared to the angular aperture of the
beam at the end mirror. Only then can one be sure that the intracavity power
is not also affected by the mirror tilt. Mirror excursion in the microradian
range is sufficient to control the CEO frequency within one spectral range.
This makes mirror tilting the method of choice for cavities with prisms.
However, it is typically very difficult to reach a servo bandwidth of more
than 1 kHz with mirror tilting because of mirror inertia. Reaching sufficient
bandwidth requires an optimized setup of the tilt actuator. Bandwidths up to
25 kHz have been demonstrated using a mirror of low mass directly mounted
on a split piezoelectric transducer (PZT) actuator [30].
common
w
mode
(w )
GD
differential
mode
Figure 5-10. Control of intracavity first-order dispersion by tilting of an end mirror after a
dispersive delay line. Translation of the mirror parallel to the optical axis acts on both group
and phase delay; tilting changes the difference between them. Provided a choice of the correct
pivot point has been made, tilting only affects the CEO-frequency.
Mirror tilting is not an option when a prismless setup is used. Then the
method of choice is modulation of the pump power either with an acousto-
optic modulator [16] or with an electro-optic device [20]. As the required
pump-power modulation is on the order of 10-3, it is typically very easy to
reach bandwidths of several tens to hundreds of kHz. Pump-power
modulation relies on the APC mechanisms discussed in the previous section
and is currently the most widespread mechanism for CEO-frequency control.
5.2 Performance of CEO phase locks
Several detailed investigations on stabilization of the CEO frequency and

the resultant residual CEO phase noise have been published [16, 20, 21, 30].
All these authors used microstructure fibers for additional external
128 Chapter 5
broadening of the laser spectra. A first attempt to directly stabilize the CEO
frequency of an octave-spanning laser was reported by Morgner et al. [3].
However, because this laser only spanned the octave at about -40 dBc, the
authors achieved a CEO beat note that was considered minimum for a robust
stabilization. The authors thus proposed a 2ν-to-3ν scheme. This scheme has
been carried out recently with resulting small residual timing jitters and
excellent long-term stability [31]. Still, a direct stabilization is more
challenging than stabilization based on additional spectral broadening.
Another important issue is the setup of the locking electronics. For any
meaningful application, a phase lock to an rf-reference source is required. A
phase lock can be as simple as that depicted in Figure 5-11, which consists
of a double-balanced mixer and some means to adjust the servo loop gain.
The gain has to be optimized for a sufficient phase margin of the loop to
prevent self-oscillation of the servo circuit.
APD Bandpass-filter
Ref. oscillator
AM
AOM Driver
Pump Beam
Ti:Sa
Figure 5-11. Simple phase-lock circuit used for stabilization of the CEO frequency [16]. The
avalanche photo detector (APD) measures the beat note signal [Equation (6), Figure 5-5].
Suitable bandpass filtering isolates the beat note and suppresses mirror frequencies and
spurious contributions. After mixing the signal with the reference oscillator, the mixing
product is directly fed back via an acousto-optic modulator. The servo loop gain has to be
adjusted for sufficient phase margin.
The simple circuit of Figure 5-11 has only a very limited capture range
and may not be able to avoid cycle slips in the presence of strong CEO-
phase noise. An alternative is usage of a phase detector with enhanced
capture range [32]. Such a circuit is based on an electronic counter and can
boost the phase capture range to tens or hundreds of π. This strategy comes
at the price of decreased sensitivity that will ultimately limit the overall
performance of the lock. The general recommendation is to reduce noise
mechanisms as far as possible by enhancing the passive stability of the laser.
Capture range enhancement should only be used as a last resort and then
moderately; otherwise, extra noise of the stabilized laser will result.
Some of the best results in terms of residual phase noise were achieved
with the simple double-balanced mixer (Reference [16], Figure 5-11). The
measured data is also shown in Figures 5-7 and 5-8. From the data in Figure
5-8, one can conclude that the residual phase jitter in these measurements
was only about 20 mrad in a 10 kHz to 0.01 Hz interval. This corresponds to
residual timing jitters of only 10 as associated with the CEO phase.
5.3 Limitations of CEO control

Using additional external broadening in a piece of microstructure fiber is
currently the most widespread method for measuring and stabilizing the
CEO frequency. As APC effects play a strong role in intracavity CEO
dynamics, it would be surprising if the strong Kerr nonlinearities involved in
continuum generation would not also give rise to APC effects extracavity.
There is an important difference that immediately explains why extracavity
APC effects are a much lesser concern than the effects discussed in Section
4.4. If the same element with identical amplitude-to-phase coupling is
moved from intracavity to extracavity, it generates CEO-phase noise via
1 (intracavity)
σ φ(extracavity)
(f) = σ (f). (11)
ce
f r f0
The highest measured noise densities of the prismless oscillator of

σ (intracavi
f0
ty)
≈ 1 kHz/ Hz would therefore translate into extracavity-phase
noise densities of only σ φ(ceextracavity ) ≈ 10 −5 rad Hz .
Several approaches to experimentally quantify the strength of extracavity
APC have been reported. Fortier et al. measured a coupling coefficient of 3.8
rad/pJ for a 100 MHz laser [33]. These measurements relied on an out-of-
loop characterization of the stabilized oscillator, employing a second totally
independent microstructure fiber and a second ν-to-2ν interferometer. A
differential CEO-phase measurement using two independent CEO-
measurement schemes was reported in [8]. In this measurement, two
independent CEO measurements on the same source were compared. A
differential phase-noise spectrum is displayed in Figure 5-12. One can
immediately see that the residual uncertainty of the CEO-phase measurement
is several orders of magnitude smaller than the phase noise of the free-
running oscillator. The phase noise shown in Figure 5-12 adds up to an rms
value of about π in a 5 s integration time [compare Equation (9) and Figure
5-8]. The differential phase noise is mainly due to slow drifts that can
corrupt measurements at long integration times.
130 Chapter 5
Figure 5-12. Differential phase-noise spectrum of two independent measurements of the CEO
phase.
Several ways have been suggested for overcoming these residual effects.
Rather than using the laser pulses directly, one can use the white-light
continuum pulses for any kind of CEO-sensitive application. Using the
spectrum directly from an octave-spanning oscillator would also rule out
APC effects in the microstructure fiber. Still, both these solutions suffer
from interferometer drift. One way to strongly reduce drift effects is a
common-path interferometer, as suggested by Kakehata et al. [34] for single-
shot CEO-phase measurements.
6. SUMMARY
We have discussed fluctuations of the CEO phase in oscillators and
explained their origin. The CEO phase turns out to be a very sensitive
parameter that is easily influenced by nearly all laser and environmental
parameters. Many of these contributions either give rise to slow drift effects
or can be easily shielded by enclosing the laser. From the perspective of
stabilizing the CEO frequency of the laser, however, amplitude-to-phase
noise conversion turns out to be a much more significant problem that
cannot totally be avoided. Several mechanisms contribute to the conversion
of laser amplitude noise into CEO frequency fluctuations. Again, some of
these contributions, such as spectral shifting or beam pointing, can be
avoided, or at least reduced, by construction and choice of favorable
operating conditions. With all these measures in place, it is possible to
provide a tight lock to an external rf reference, with resulting residual timing
jitters between carrier and envelope of the laser of only a few attoseconds.
Even with such superior performance, one has to be careful to avoid intrinsic
sources of CEO phase noise in the measurement set-up itself. Amplitude-to-
phase conversion also takes place in external continuum generation, which is
often used to broaden the laser spectrum to an optical octave. The ν-to-2ν
interferometer is another weak point and can give rise to a small residual
drift. These residual effects are relatively weak and do not appear to corrupt
most applications of frequency combs demonstrated to date. Nevertheless,
they can be avoided. An improved control of frequency comb parameters
offers even higher precision in metrology applications and opens up novel
applications in extreme nonlinear optics. Understanding the dynamics of the
comb is the key to further progress in these areas.
REFERENCES
[1] G. Steinmeyer, D. H. Sutter, L. Gallmann, N. Matuschek, and U. Keller,
Science 286, 1507-1512 (1999); U. Keller, Nature 424, 831-838 (2003).
[2] M. Nisoli, S. Desilvestri, O. Svelto, R. Szipocs, K. Ferencz, C. Spielmann,
S. Sartania, and F. Krausz, Opt. Lett. 22, 522-524 (1997); A. Baltuška, M.
S. Pshenichnikov, and D. A. Wiersma, Opt. Lett. 23, 1474-1476 (1998); A.
Baltuška, T. Fuji, and T. Kobayashi, Opt. Lett. 27, 1241-1243 (2002); R.
Ell, U. Morgner, F. X. Kärtner, J. G. Fujimoto, E. P. Ippen, V. Scheuer, G.
Angelow, T. Tschudi, M. J. Lederer, A. Boiko, and B. Luther-Davies, Opt.
Lett. 26, 373-375 (2001).
[3] U. Morgner, R. Ell, G. Metzler, T. R. Schibli, F. X. Kärtner, J. G. Fujimoto,
H. A. Haus, and E. P. Ippen, Phys. Rev. Lett. 86, 5462-5465 (2001).
[4] D. H. Sutter, G. Steinmeyer, L. Gallmann, N. Matuschek, F. Morier-
Genoud, U. Keller, V. Scheuer, G. Angelow, and T. Tschudi, Opt. Lett. 24,
631-633 (1999).
Priori, and S. De Silvestri, Nature 414, 182-184 (2001); G. G. Paulus, F.
Lindner, H. Walther, A. Baltuška, E. Goulielmakis, M. Lezius, and F.
Krausz, Phys. Rev. Lett. 91, 253004 (2003).
[6] H. R. Telle, G. Steinmeyer, A. E. Dunlop, J. Stenger, D. H. Sutter, and U.
Keller, Appl. Phys. B 69, 327-332 (1999).
[8] F. W. Helbing, G. Steinmeyer, and U. Keller, IEEE J. Sel. Top. Quantum
Electron. 9, 1030-1040 (2003).
883 (1999).
[10] T. F. Albrecht, K. Bott, T. Meier, A. Schulze, M. Koch, S. T. Cundiff, J.
Feldmann, W. Stolz, P. Thomas, S. W. Koch, and E. O. Göbel, Phys. Rev.
B 54, 4436-4439 (1996).
[11] N. Haverkamp, B. Lipphardt, J. Stenger, H. R. Telle, C. Fallnich, and H.
Hundertmark, In Conference On Ultrafast Phenomena, Vancouver, BC,
2002), P. Me31-31.
[12] K. F. Kwong, D. Yankelevich, K. C. Chu, J. P. Heritage, and A. Dienes,
Opt. Lett. 18, 558-560 (1993).
Opt. Lett. 21, 2008-2010 (1996).
132 Chapter 5

(2000).
[16] F. W. Helbing, G. Steinmeyer, U. Keller, R. S. Windeler, J. Stenger, and H.
R. Telle, Opt. Lett. 27, 194-196 (2002).
[17] H. R. Telle, D. Meschede, and T. W. Hänsch, Opt. Lett. 15, 532-534
(1990).
[18] U. Keller, K. D. Li, M. Rodwell, and D. M. Bloom, IEEE J. Quantum
Electron. 25, 280-288 (1989).
[19] R. P. Scott, C. Langrock, and B. H. Kolner, IEEE J. Sel. Top. Quantum
Electron. 7, 641-655 (2001).
Hansch, and F. Krausz, Appl. Phys. B 72, 373-376 (2001).
[21] F. W. Helbing, G. Steinmeyer, J. Stenger, H. R. Telle, and U. Keller, Appl.
Phys. B 74, S35-S42 (2002).
[22] H. R. Telle, In Frequency Control Of Semiconductor Lasers, Edited By M.
Ohtsu (John Wiley & Sons, New York, 1996), P. 137-167.
28, 851-853 (2003).
[24] R. L. Fork, O. E. Martinez, and J. P. Gordon, Opt. Lett. 9, 150-152 (1984);
R. E. Sherriff, J. Opt. Soc. Am. B 15, 1224-1230 (1998).
[25] J. Stenger and H. R. Telle, Opt. Lett. 25, 1553-1555 (2000); H. A. Haus and
E. P. Ippen, Opt. Lett. 26, 1654-1656 (2001).
[27] M. Sheikbahae, D. C. Hutchings, D. J. Hagan, and E. W. Van Stryland,
IEEE J. Quantum Electron. 27, 1296-1309 (1991).
[28] R. Desalvo, A. A. Said, D. J. Hagan, E. W. Van Stryland, and M. Sheik-
Bahae, IEEE J. Quantum Electron. 32, 1324-1333 (1996).
172, 59-68 (1999).
Lett. 27, 1436-1438 (2002).
[31] A. Bartels, S. A. Diddams, T. M. Ramond, and L. Hollberg, Opt. Lett. 28,
663-665 (2003).
[32] U. Tietze, C. Schenk, and E. Schmid, Electronic Circuits: Design And
Applications (Springer, New York, 1991).
447 (2002).
[34] M. Kakehata, Y. Fujihira, H. Takada, Y. Kobayashi, K. Torizuka, T.
Homma, and H. Takahashi, Appl. Phys. B 74, S43-S50 (2002).
Chapter 6
FEMTOSECOND NONCOLLINEAR
PARAMETRIC AMPLIFICATION AND
CARRIER-ENVELOPE PHASE CONTROL
Takayoshi Kobayashi
Department of Physics, Faculty of Science, University of Tokyo
Abstract: This chapter presents the basic principles for three parametric interactions that
enhance bandwidth to obtain short pulses while maintaining phase matching.
To extend the bandwidth, we introduced a noncollinear configuration between
the pump and signal. The idea is used in three different parametric processes:
optical parametric generation (OPG), optical parametric amplification (OPA),
and optical parametric oscillation (OPO). Using noncollinear phase matching,
we developed a noncollinear-optical-parametric amplifier (NOPA) that
delivers 4 fs visible-near-infrared pulses. We designed geometrical and
temporal configurations of the NOPA that broaden the gain bandwidth in
excess of 250 THz. The main requirements for bandwidth enhancement
include (1) phase matching, (2) group-velocity matching, (3) pulse-front
matching, and (4) optimization of the angular dispersion of the pump. To
achieve the extended-gain bandwidth, full phase adjustment is performed by
several compensators, including a prism pair, a grating-mirror system
equivalent to a grating pair, chirped mirrors, and a deformable mirror. By
adding these devices to the NOPA system, we obtained pulse widths of 3.9 fs
in the visible and NIR spectral range.
Key words: carrier-envelope phase, optical parametric amplification
1. INTRODUCTION
Femtosecond laser systems are experiencing explosive growth in both the

number of users and variety of applications. In the process, pulse duration is
continually being reduced. Optical pulses with a duration of only a few
optical cycles have been achieved with the proliferation of diverse types of
Ti:sapphire lasers. Sub-5 fs pulses have become available with 800 nm
134 Chapter 6
Ti:sapphire lasers that employ self-phase modulation (SPM) and spectral

broadening in both quartz [1-3] and hollow waveguides [4-6]. The
broadband pulses are compressed to a nearly Fourier-transformed pulse with
novel pulse compression techniques such as chirped mirrors [7].
Even though sub-5 fs pulse durations have been achieved [8], methods
utilizing SPM for spectral broadening have two major drawbacks. First, they
do not offer tunable pulses. Only limited tunability can be achieved (at the
expense of reduced bandwidth) by filtering out various components of an
SPM-broadened spectrum. Second, is the spectral modulation inherent in the
SPM process. This limitation is perhaps the most fundamental. On the one
hand, the interplay between material dispersion and SPM can substantially
improve the smoothness of the broadened spectra [9]. On the other hand,
pulse stretching due to dispersion lowers the peak intensity and makes it
impossible to sustain strong SPM action over a long propagation distance.
The result is limited spectral broadening.
However, the invention of microstructure fiber [10] led to the generation
of extremely broadband white-light continua with spectra that could support
2-fs pulses. This remarkable feat is a consequence of the dramatic lowering
of fiber dispersion. The same feature responsible for such unprecedented
spectral broadening also yields substantial spectral modulation. The latter
makes it nearly impossible to use these pulses in many demanding
applications of nonlinear spectroscopy, which are sensitive to both the time
and frequency distributions of the electric field. This feature also makes it
difficult to use microstructure fiber in pump-probe experiments.
Measurements of time-resolved spectra are subject to large errors induced by
fluctuations (hugely enhanced by enormous nonlinearity) of the sharp
(spiky) spectral structure.
Unlike these schemes, which involve cumbersome fiber-chirping stages,
remarkable progress in the generation of record-breaking short pulses has
been made directly with Ti:sapphire lasers [11]. Few-cycle pulses at the few
nJ level have been reported by several research groups. Even though short-
pulse generation directly from cavities is simpler than the elaborate
extracavity compression [1, 6], the two methods share the same spectral
range and the lack of wavelength tunability, except when the latter is
implemented at the expense of shortness of pulse duration. In 2001-2002, a
different approach to generating even shorter pulses directly from laser
cavities was developed. Researchers demonstrated that it was possible to
phase lock two independent Ti:sapphire oscillators [12] or two oscillators
based on different gain media [13]. This approach could provide a
straightforward way to generate an ultimately short, single-cycle optical
pulse by coherently combining the output from several lasers employing
different gain media [8]. Both the intracavity and extracavity methods of
6. FEMTOSECOND NONCOLLINEAR PARAMETRIC 135
AMPLIFICATION AND CARRIER-ENVELOPE PHASE CONTROL
pulse shortening should make it possible to generate broad laser spectra at

the laser’s fundamental frequency or around one of its harmonics; however,
this makes continuous wavelength tuning impossible in the shortest pulse
regime. This difficulty, however, can be readily overcome in parametric
pulse generation and amplification.
In the 1990s, optical parametric oscillation (OPO) and optical parametric
amplification (OPA) had become popular methods for tunable femtosecond
pulse generation, especially since the appearance of femtosecond Ti:sapphire
lasers [14-16]. The high power and excellent stability of a Ti:sapphire
system make optical parametric conversion the most promising method for
obtaining tunable ultrashort pulses. In addition, important progress was made
during the 1980s in the growth of nonlinear optical crystals with large
nonlinear coefficients and high damage thresholds. These advances opened
the door to new and versatile applications of these crystals in the new
century [17].
Beginning about 1994, with an OPA pumped by a Ti:sapphire amplifier,
it became possible to create ultrashort pulses tunable outside the 800 nm
region with a few-µJ pulse energy. An OPA pumped with a Ti:sapphire laser
with peak wavelength around 800 nm generates near-infrared (NIR) pulses
between 1 and 4 µm, whereas an OPA pumped at about 400 nm (the second
harmonic of Ti:sapphire) can cover the shorter, visible–NIR optical
frequency region between 450 nm and 3.5 µm [15, 16, 18-22]. Following the
difference frequency generation (DFG) of the signal and idler, they can be
used to cover the range from 200 nm to 12 µm, or almost the entire optical
frequency region [15, 16]. However, the shortening of the output pulse from
an OPA faces a problem with phase-matching bandwidth.
2. ADVANCES OF NONCOLLINEAR-PHASE-
MATCHED OPTICAL PARAMETRIC
CONVERSION
The idea of a noncollinear phase matching is not new for broadband

interaction. Takeuchi et al. [23] reported broadband generation of idler
pulses by focusing a white-light continuum with a large convergence angle
into a crystal to be phase-matched noncollinearly among a wide range of the
continuum. The bandwidth was increased to as broad as 1300 cm-1 in the
NIR. They used an optical parametric generator in a noncollinear
configuration to generate signal and idler beams just around the pump axis.
In the present work, noncollinear phase matching is investigated where the
136 Chapter 6
signal and idler are arranged with larger noncollinear angles, as shown in
Figure 6-1.
ks signal
vs
α kp
β
vp pump
ki
idler
vi
Figure 6-1. Geometric configuration of the wave vectors in an optical-parametric-

amplification (OPA) nonlinear crystal. The wave vectors of the pump (kp), signal [ks (λs )],
and idler [ki(λi)] are shown in the gain crystal. The angles α and β are internal noncollinear
angles between pump and signal beams and pump and idler beams, respectively. The group
velocities of the pump (νp), signal (νs), and idler (νi) are also shown by dashed lines.
Noncollinear phase matching has advantages over collinear phase

matching. In an OPO, noncollinear phase matching is often adopted to
compensate for spatial walk-off of the extraordinary ray in the crystal and to
increase the gain [14, 24-26]. Temporal walk-off caused by the group-
velocity mismatch can also be improved by noncollinear phase matching
[27-29]. Trapani et al. demonstrated noncollinear phase matching in an OPA
for satisfying the group-velocity matching between the pump pulse and
signal or idler pulses in a β-BaB2O4 (BBO) crystal to obtain higher
conversion efficiency with longer interaction length than for a collinear OPA
[27-29].
The first use of noncollinear phase matching to eliminate temporal walk-
off was reported by Gale et al. in 1995 [26]. They constructed a
synchronously pumped OPO using a BBO crystal. The noncollinear
configuration between the pump and oscillating signal beams resulted in
broadband phase-matching and generated pulses as short as 13 fs in the
visible. The generation of a broad spectrum of over 100 nm in a nanosecond
OPO based on the same configuration was also reported [30]. We
demonstrated a noncollinearly phase-matched OPA (NOPA) using a BBO
and achieved tunable 14 fs pulse generation in the visible with 3–8 µJ pulse
energies in 1997 [31, 32]. Although there are restrictions for synchronizing
the pump and signal pulses and bandwidth limitations due to cavity mirrors
in a synchronously pumped OPO [33], a femtosecond NOPA will likely
generate shorter pulses with multi microjoule pulse energies and broad
tunability. Extensive subsequent investigations have been performed [34-
38], and the NOPA continues to attract attention owing to its remarkable
advantages.
We will now describe the development of a noncollinear parametric
amplifier. First, we introduced a pulse-front-matching geometry [36] to
achieve a sub-5 fs light source using methods other than the conventional
continuum-compression scheme. In the nonlinear configuration, the group
velocity of a faster-traveling wave (the idler) is projected at an angle onto a
slower wave (the signal), thus improving the temporal overlap of the two and
achieving broadband phase matching. The importance of parametric
conversion in the visible and near infrared has been dramatically increased
by the discovery of the unique phase-matching conditions in a type-I BBO
crystal pumped by the second harmonic of the Ti:sapphire laser [36, 38]. As
a result, widely tunable sub-20 fs NOPAs have become more or less
routinely used [34, 39]. Tunable operation in a 10 fs regime in both the
visible and infrared has also been demonstrated [32, 36, 37, 40]. The
development of sophisticated pulse-compression schemes has made phase
correction over the entire parametric-amplification bandwidth possible,
resulting in the realization of sub-5 fs pulse generation in the visible–NIR
region [38, 41-43].
The great potential of sub-10 fs NOPAs was proven by several
spectacular applications of nonlinear spectroscopy such as time-domain
studies of ultrafast molecular dynamics in the condensed phase [42, 44].
Although other methods of ultrashort pulse generation may, in some cases,
be better than NOPAs for both pulse length and spectral width [45],
parametric amplifiers produce noticeably smoother spectra with high
stability. The smooth, stable spectra play a vital role in time- and frequency-
resolved spectroscopy with high time resolution and with relatively high
spectral resolution when the spectrometer is placed after the sample. The
demand for continuing improvement of the NOPA pulse quality is evident.
The second step in obtaining short optical pulses is to control the phase
of the field to be constant over the whole gain bandwidth. A constant field
phase leads not only to improved pulse compression, but also to easier
dispersion manipulation in phase-sensitive applications of nonlinear
spectroscopy. Several dispersion-control techniques have been employed to
obtain the shortest pulses from NOPAs. For instance, sub-5 fs pulses have
been achieved both by using a combination of a 45°-angled-prism
compressor and a set of ultrabroad-chirped mirrors (UBCM) [38] and by
using custom-designed UBCMs alone [43]. While it is feasible to fabricate a
set of “ideal” UBCMs, the fixed dispersion of such multilayer dielectric
structures makes it impossible to introduce fine wavelength-selective control
of group delay, which is required in the daily optimization of a pulse
138 Chapter 6
compressor. Similarly, typical elements of tunable negative dispersion, such

as the prism and grating pairs [46-48], are rendered virtually nonadjustable
because of the large spectral bandwidth of a sub-5 fs pulse.
Thus, additional improvement of pulse quality cannot be achieved
without implementing a flexible phase-correction method. The usefulness of
a flexible phase method has been demonstrated in a computer-controlled
shaping of tunable 16 fs pulses from a NOPA using a liquid crystal mask
[49]. Micromachined flexible mirrors with electrostatic actuators, invented
in the late 1990s, offer the possibility of adaptive pulse shaping. An
inexpensive, commercially available membrane of this type was used to
obtain nearly ideal spectrum-limited pulses of 15 fs duration from a
Ti:sapphire laser [50] and subsequently employed for shaped-pulse
optimization of the coherent emission of soft x-rays produced by high-
harmonic generation [51]. By implementing this compression method for a
NOPA signal wave, a visible 7 fs pulse was obtained [52]. A successful
optimization of the compression in a similar 10 fs tunable NOPA has been
reported by Baum et al. [53]. Previously, we have briefly described the use
of this technique to compress the NOPA pulses [54] and discuss it here in
detail. We demonstrate how computer-controlled optimization combined
with second-harmonic frequency-resolved optica-gating [55] allows us to
reduce the output pulse duration down to about 4 fs at the central wavelength
near 600 nm.
3. PRINCIPLE OF PARAMETRIC AMPLIFICATION
We describe the principle of optical parametric interaction, in particular

noncollinear interaction, in this section. The requirements for efficient
spectral conversion in optical parametric interactions are given by the two
conditions among wave vectors and frequencies,
κ p = κ s + κi , (1)
ω p = ω s + ωi . (2)
Here κj and ωj are the wave vector and angular frequency with the suffixes j
= p, s, and i corresponding to the pump, signal, and idler, respectively. The
first is a phase-matching condition corresponding to momentum
conservation among the relevant three photons; the second is the
requirement of energy conservation. Hereafter, we assume a plane-wave
interaction among them. This assumption is well satisfied in the case when
the beam diameter (including the beam waist at the focal point) is much
larger than the wavelength. By neglecting the pump depletion and spectral
width, the well-known parametric gain G is [56, 57]
G = (Γl c sinh (gl c ) gl c ) + 1

2
(
g = Γ 2 − (∆k 2 ) )
2 1/ 2
, Γ = 2d eff
ω sω iΦ
ε 0 n s ni n p c 3
, (3)
where Γ is the coupling constant, deff is the effective second-order nonlinear

coefficient of the nonlinear optical crystals, c is the vacuum light velocity, ε0
is the vacuum dielectric constant, Φ is the pump intensity, lc is the crystal
length, and nj is the refractive index of the corresponding wave The phase 㧚
mismatch ∆κ along the pump direction is
∆κ = κ p − κ s cos α − κ i cos β , (4)
where α and β are the noncollinear angles between the wave vectors of the
pump and signal and those between pump and idler, respectively. Figure 6-1
illustrates the arrangement of the general noncollinear configuration.
In the conventional collinear geometry (α = β = 0), the expression is
simplified. By the Taylor expansion in powers of the angular frequencies
around the central frequencies ω s 0 − ω i 0 , ∆κ is given by
 1 1  ∂ 2κ s δ 2κ i 
(∆ω s )2 + ...
1
∆κ =  − ∆ω s
 −  + (5)
ν s νi  2  δω 2 δω i2 
where ∆ω ≡ ω s − ω s 0 and ∆ω ≡ ω i 0 and we use ∆ω i = − ∆ω s . Within the

−1
 δκ j 
monochromatic pump approximation, as assumed above, ν j =  is
 δω 
 j 
the group velocity of a jth component at the central frequency. The first-order
derivative includes the parametric bandwidth [57],
∆ν parametric = 0.53 Γ lc (c(1 ν s − 1 ν i )) , (6)

140 Chapter 6
which is defined as the full width at half maximum (FWHM) of the gain
spectrum G( ω s ). This represents the group-velocity mismatch (GVM)
between the signal and idler in the nonlinear optical crystal that determines
the bandwidth of an OPA. The GVM is an intrinsic effect for the ultrashort
pulse interaction and determines the parametric bandwidth around 200–500
cm-1 in standard OPAs [57]. Previously reported femtosecond OPAs with
collinear geometry suffered from this problem, and the shortest pulse
durations were limited to 40 fs in the visible [20] and to 30 fs in the NlR [16,
21]. Broadening of the parametric bandwidth is essential for obtaining a
tunable sub-10 fs light source. Sosnowski et al. [58] demonstrated a spectral
broadening of the signal to about 800 cm-1 in the visible region by multistage
amplification in which different spectral regions were amplified at each
stage with slightly different crystal angles. However, this method has
shortcomings, such as being rather complicated. In the late 1990s, novel NIR
OPAs were reported [59, 60]. For example, Fournier et al. used the
effectively cascading third-order process of second-harmonic generation of a
signal and obtained self-compression to 20 fs signal pulses [59]. Nisoli et al.
[60] demonstrated 14.5 fs signal pulse generation around 1.5 µm using an
ultrashort 18 fs pump source from a Ti:sapphire amplifier followed by a
hollow-fiber compressor. Both methods seem to be difficult and too sensitive
to the pump energy. Third-order effects play an essential role in the
characteristic pulse propagation in their schemes. Simpler and more robust
methods are strongly desired for shorter pulse generation.
3.1 Noncollinear-optical-parametric amplification

(NOPA)
In a noncollinear geometry, phase mismatch is expressed with more

parameters than in the collinear geometry. Angles α and β are wavelength
dependent and are expanded around the central frequency as
 δα  1  δ 2α 
 (∆ω s )2 + ...
cos α = cos α 0 − sin α 0  ∆ω s − cos α 0 
  2 
(7a)
 δω s  2  δω s 0
 δβ  1 δ 2β 
 (∆ω s )2 + ...,
cos β = cos β 0 − sin β 0  ∆ω s − cos β 0 

(7b)
 δω s  2  δω 2 
 s 0
where the subscript 0 denotes the central value. From the phase-matching
condition of Equation (1) we obtain
κ p = κ s 0 cos α 0 + κ i 0 cos β 0 , (8a)
κ s 0 sin α 0 = κ i 0 sin β 0 . (8b)
By substituting Equations (7) and (8) into (4) we can expand the phase
mismatch as
∆κ = − 
 cos α
νs
0
−
cos β 0
νi

− κ s 0 sin α 0 
∂α 



+ κ i 0 sin β 0 
∂β  
  ∆ω s

+ K

  ∂ω s 0  ∂ω i  0 
(9)
Thus, the first-order coefficient can include the dispersion of the

noncollinear angles. It means that the signal-idler–group-velocity mismatch
in (6) can be modified by the wavelength-dependent noncollinear angles. If
the condition
cos α 0 cos β 0  ∂α   ∂β 
− − κ s 0 sin α 0   + κ i 0 sin β 0 
  ∂ω


=0 (10a)
νs νi  ∂ω s  0  i 0
is satisfied, then the first-order phase mismatch is eliminated and extremely

broadband phase matching is expected. The corresponding condition for the
component perpendicular to the pump can be derived in the same way as
sin α 0 sin β 0  ∂α   ∂β 
+ + κ s 0 cos α 0   + κ i 0 cos β 0 
  ∂ω


= 0. (10b)
νs νi  ∂ω s  0  i 0
We can define the group-velocity matching in the noncollinear geometrical

configuration by Equations (10a) and (10b). For a conventional collinear
geometry ( α = β = 0 ), this condition is satisfied only when
ν s =νi (11)
is fulfilled. This condition is only satisfied for the case of the degeneracy

(ω s = ωi = ω p 2) in a type-I interaction (e o + o or o e + e). In a type-

II interaction (e o + e or o o + e) the GVM is usually larger than that in
a type-I interaction because of the large birefringence in the crystals [22, 57]
142 Chapter 6
used in type II interactions. In a noncollinear geometry, the angular
dispersion can eliminate the GVM and a broadband phase matching can be
attained. The schematic of this configuration is shown in Figure 6-2.
ks(λs)
ki(λp
α β (λi) )
y
(λs) kp
θ
Figure 6-2. Signal-idler–group-velocity matching in an OPA nonlinear crystal with a general

geometry. The wave vectors of the pump (kp), signal (ks(λ)), and idler (ki(λi)) are shown in the
gain crystal. The angles α(λs) and β(λi) are internal noncollinear angles between pump and
signal beams and between pump and idler beams, respectively. Coordinates x, y, and z and
angle θ denote the crystal axes and angle between the optical axis of the crystal and a pump
beam, respectively.
The angular dispersions satisfying this condition are calculated from (10) as
 ∂α  ν s−1 cos (α 0 + β 0 ) − ν i−1

  = (12a)
 ∂ω
 s

0 κ i 0 sin (α 0 + β 0 )
 ∂β  ν i−1 cos(α 0 + β 0 ) − ν s−1

  = . (12b)
 ∂ω
 i

0 κ i 0 sin (α 0 + β 0 )
This kind of phase matching is well known as achromatic phase matching

[58, 59] and has been used to obtain broadband wavelength conversion of
ultrashort pulses [23, 60-62].
For practical applications of the amplified signal or generated idler, it is
better to remove the angular dispersion. Thus we consider the case where the
signal has no angular dispersion ( α = α 0 ). The schematic is shown in
(Figure 6-2). For α = α 0 , (12) yields the simple group-velocity matching
condition
ν s = ν i cos(α + β ) , (13)
with
∂β 1  sin α sin β  tan(α + β )

=−  + = , (14)
∂ω i κ i cos β  ν s ν i  κ iν i
where the subscript 0 is dropped for simplicity. The physical meaning of this
condition can be stated as the projection of the idler group velocity on the
signal direction is equal to the signal group velocity. The phase mismatch in
Equation (9) can then be rewritten
cos(α + β )  1
∆κ = −  −
1 
∆ω s +
cos β  ν s ν i cos(α + β ) 

K. (15)
Group-velocity matching is satisfied at the expense of the collimation of the

idler beam. This means that the angular dispersion of the idler given by
Equation (12) absorbs the group-velocity mismatch between the signal and
idler. We use the term "signal-idler–group-velocity matching" in this special
case. For simplicity, the generalized group-velocity matching is called
"achromatic phase matching" hereafter.
Under the condition of a small noncollinear angle α, which is satisfied in
cos(α + β )
many experimental situations, then ≈ 1 can be used. It is
cos β
convenient to use the general group-velocity mismatch defined as:
1 1
GVM s −i = − . (16)
νs ν i cos (α + β )
The GVM between the pump and signal, which limits the interaction length
in the crystal, is [27]
1 1
GVM p − s = − . (17)
νp ν s cos α
As clearly seen from Equations (16) and (17), both GVMs are dependent on
the noncollinear angle α. Each of the GVMs can be eliminated by selecting a
geometrical arrangement satisfying Equation (16) or (17). Group-velocity
matching between the pump and signal or between the signal and idler using
noncollinear geometry in the case of ν s > ν i has been reported; an order-of-
144 Chapter 6
magnitude longer interaction length than in the collinear geometry was also
obtained [27-29]. Our current interest is the signal-idler–group-velocity
matching, which is possible if ν s > ν i . The bandwidth is broadened by more
than tenfold and only limited by the group-velocity dispersion (GVD, as
indicated in Equation (5) [57]; ultrashort pulse generation is thus expected.
4. SIGNAL-WAVELENGTH-INSENSITIVE PHASE
MATCHING
In this section, we discuss signal-wavelength-insensitive phase matching

or, in other words, broadband phase-matching. Recently, a simple expression
for the phase-matching angle for different collinear and noncollinear type-I
nonlinear-optical interactions in a negative uniaxial crystal was presented
[63]. The phase matching and the energy-conservation conditions among the
interacting beams are
k1e = k 20 + k 30 , (18)
ω1 = ω 2 + ω 3 . (19)
Here, ki and ωι (λi) are the wave vector and the frequency (wavelength),
respectively, of i-th beam, and i = 1, 2, and 3. Solving Equations (18) and
(19), we obtain the following expression for the phase-matching angle θ that
is defined as the internal angle made by the extraordinary (e) polarized pump
beam with the optic axis of the crystal:
θ = cos −1 (A Y ) e 2
−1
. (20)
(A e
A )
O 2
−1
Here,
Y= 

(k ) + (k )
o 2
2
o 2
3 + 2k 2o k 3o cosψ  ,


(21)
ψ =α + β , (22)
β = sin-1[(k2ok3o)sinα], (23)
and Ao = 2π(n1o/λ1), Ae = 2π(n1e/ λ1), k2o = k2o =2π(n2o/λ2), and k3o =  k3o
=2π(n3o/λ3); n1, n2, and n3 are refractive indices of the three interacting
radiations with wavelengths λ1, λ2, and λ3. The superscripts o and e
correspond to the ordinary and extraordinary polarizations, respectively, and
α (β) is the noncollinear angle between the pump and the signal (idler)
beams. Here k2o and k3o are the wave numbers corresponding to the signal
and idler beams, respectively. Y is defined as the ‘average’ wave number of
the signal and idler beams parallel to the pump beam. Considering the phase
matching along the direction perpendicular to k1, Equation (23) is obtained.
In the case of monochromatic pump, from Equation (20) we observe that the
phase-matching angle θ will be independent (to first order) of the signal
wavelength or signal frequency (ω2) if ∂θ/∂ω2 = 0, for any value of ω2. In
Equation (20), all the parameters except Y are independent of ω2. Therefore,
the condition may be expressed as follows:
∂Y/∂ω2 = 0. (24)
It can be shown easily that
Y = k 2o cos α + k 3o cos β . (25)
If the incident angle of the tunable seed beam has no frequency or,
equivalently, wavelength dependency, the noncollinear angle α between the
pump and tunable seed pulses will remain unchanged even with the change
of the signal frequency, i.e., ∂α/∂ω2 =0. From Equations (24) and (25)
(cosα/vg2 − cosβ/vg3) − k3osinβ(∂β/∂ω2) = 0, (26)
where, vgi = ∂ωi/∂ki, is the group velocity of the i-th beam (i = 2 and 3) with
frequencies of ωi. To obtain Equation (26), we have used ∂ω2 = −∂ω3, since
ω1 = constant because the pump is monochromatic. From Equation (23)
k 3o sin β = k 2o sin α . (27)
Differentiating Equation (27) with respect to ω2 and considering ∂α/∂ω2 = 0

and ∂ω2 = −∂ω3, it can be shown that ∂β/∂ω2 = (sinα /vg2 + sinβ /vg3)
/(k3ocosβ). Substituting ∂β/∂ω2 in Equation (26) and after some algebraic
simplification, we get (cosψ/vg2 − 1/vg3)/cosβ = 0, which is equivalent to
146 Chapter 6
vg2 = vg3 cosψ. (28)
Equation (28) can also be derived [31] by equating the first-order derivative
of the phase mismatch with respect to ω2 to zero, assuming a monochromatic
pump and setting ∂α/∂ω2 = 0. Using this method, we found that for a NOPA,
the effective inverse group-velocity mismatch between the signal and the
idler pulses [31]
GVM2–3 = [1/vg2 – 1/(vg3 cosψ)]. (29)
We observe from Equation (29) that GVM2–3 can be eliminated if Equation

(28) is satisfied. Thus the condition given by Equation (28) is considered to
be the requirement for group-velocity matching in a NOPA [31]. Therefore,
with earlier methods [31], it can be shown only implicitly that Equation (28)
is the condition for the insensitivity of the phase-matching angle with respect
to the variation of signal frequency or wavelength. However, the derivation
above is straightforward and simple starting from the expression of the
phase-matching angle θ. We show explicitly that to achieve the insensitivity
of the phase-matching angle with respect to the variation of signal frequency
or wavelength of the NOPA, Equation (28) must be satisfied, i.e., the
components of the group velocities of the idler beams with a broad spectrum
along the signal direction must be equal to that of the corresponding signal
beam. We studied the optical parametric properties of a relatively recently
developed NLO crystal LB4 and found that Equation (28) is also satisfied if
this crystal is employed as a NLO amplifier crystal in a 395 nm pumped
type-I NOPA with a particular value of the noncollinear angle α [63].
5. GROUP-VELOCITY MATCHING IN β-BAB2O4
This section describes achromatic phase matching in a BBO crystal,

which is widely used as the gain crystal in an OPA. A BBO crystal has the
advantages of broad tunability, a high damage threshold, and a large
nonlinear coefficient [57]. A type-I (e → ο + ο ) OPA is more suited to
ultrashort pulse generation than a type-II (e → ο + e ) OPA because of the
smaller GVM and larger effective nonlinear coefficient [22, 57, 60] in the
former case. From Equation (8), the angle θ between the pump beam and the
z axis (optic axis) and the idler noncollinear angle β satisfying the type-I
phase matching are respectively

ϑ (λs ,α ) = cos −1 
 1
×
−2
( ) −2
 n0 λ p − ne λ p ( )
1 1
−
(λ )
2
(30a)
2
2
 nc2
λ 0  n0 (λs )cos α / λs + (n0 (λi ) / λi ) − (n0 (λs )sin β / λ )
2

p
 
β (λ s ,α ) = sin −1 
 n0 (λ0 ) λi sin α . (30b)
 n0 (λi ) λs


Here the three wavelengths ( λ j ’s) satisfy the relation, λi = λ−p1 − λ−s 1 ( )
−1
, and
no and ne are the refractive indices of the ordinary and extraordinary rays,
respectively. The wavelength dependency of the refractive indices is
obtained by the Sellmeier equations.
3.0
2.5
Wavelength ( µm)
2.0 α=2° α=4°
1.5 α=0°
α=6°
1.0
0.5
opacity region
0.0
22 26 30 34 38
θ (deg)
Figure 6-3. Theoretical phase-matching curves. Phase-matched probe wavelength dependence

on the angle θ between the crystal axis Z and the pump beam in a type-I BBO OPA pumped at
395 nm with several different values of the signal noncollinear angle α. The signal branches
(solid curves) and idler branches (dashed curves) are shown. The area in gray indicates the
region where the achromatic phase matching is possible. The opacity region indicates that the
OPA gain cannot be obtained because of the corresponding idler absorption.
148 Chapter 6
The curves of θ(λs, α) for various α with pump wavelength at 395 nm are
shown in Figures 6-3 and 6-4. There are two branches corresponding to the
signal and idler in the phase-matching curve for each non-zero α. These
figures show that the most characteristic feature of the phase-matching curve
is the broad spectral range. The region filled in gray in Figure 6-3 can satisfy
the phase-matching condition by a noncollinear interaction. For a given θ,
the signal and idler waves can be simultaneously emitted in a broad range of
wavelength as shown in Figure 6-5. For a smaller angle between the pump
beam and the crystal axis θ < 29.4° , the range of the spectrum is partially
limited, whereas for θ > 29.4° , optical modes extending from 450 nm to 3
µm are excited. This limit is due to the absorption of the idler in the crystal.
The broad bandwidth can be explained in terms of noncollinear
configuration with an extended group-velocity matching condition.
900
Signal Wavelength (nm)
800
GVM s-i =0
700 6°
600 5°
α =0°
4°
500 2°
3°
24 26 28 30 32 34 36 38
θ (deg)
Figure 6-4. Signal branches of the theoretical phase-matching curves magnified in a

wavelength region between 450 and 900 nm. The gray curve indicates the signal-idler–group-
velocity matching points.
By utilizing a noncollinear geometry, group-velocity matching can take

place for λ s < 790 nm, as indicated by a dashed curve in Figure 6-4. The
wavelength dependence of α that yields GMVs −i = 0 is calculated. It is
found that over a spectral range broader than 100 nm, the noncollinear angle
of α needed for phase matching varies from 3.7° only by less than 0.1° . A
broad plateau at θ = 31.5 ± 0.1° between 520 to 710 nm appears in the
phase-matching curve for α = 3.7° , as presented in Figure 6-6.
Figure 6-7 shows parametric gain curve given by Equation (3) calculated
for the configuration of θ = 31.5° with effective angular dependent
nonlinear coefficient d eff = 1.6 cos θ pm V [64] and with the pump intensity
level of Φ = 50 GW/cm2. The gain extends the extremely broadband in the
visible and NIR range from 520 to 750 nm with a small intensity variation
and structure. The bandwidth reaches a breadth of 160 THz (5300 cm-1), and
from the Fourier transform of the gain spectrum shortest possible pulse
width is estimated to be 4.4 fs. Note again that the idler waves are also
generated with the same broad bandwidth in the NIR with the angular
dispersion to be phase-matched with the broad spectrum of the signal waves
(see Figure 6-6).
ks(λs)
ki(λp)
α = fixed β (λi)
y
θ kp
Figure 6-5. Signal-idler–group-velocity matching in an OPA with a special geometry of

constant pump-signal angle (α = fixed). The wave vectors of the pump (kp), signal [ks(λs)],
and idler [ki(λi)] are shown in the gain crystal. The angles α and β(λi) are internal
noncollinear angles between pump and signal beams and pump and idler beams, respectively.
Coordinates x, y, and z denote the crystal axes and q is the angle between the optical axis of
the crystal and the pump beam.
37 20
Idler Noncollenear Angle
Crystal Angleθ (degree)
36
35 15
G
34 G
T
I
G
33 10 F

32
β
31 5
30
29 0
450 500 550 600 650 700 750 800
Wavelength (nm)
Figure 6-6. Noncollinear phase-matching curves. Wavelength dependence of pump incident

angle θ and idler noncollinear angle β in the case of the pump-signal noncollinear angle of α
= 3.7°.
These features indicate that a NOPA is a promising light source for

generating ultrashort pulses over a wide visible and NIR range in a sub-10 fs
regime. In the present work this characteristic signal-idler group-velocity
matched interaction is focused on and the applicability to the ultrashort pulse
generation is extensively investigated and explored.
150 Chapter 6
Figure 6-7. Wavelength dependence of the parametric gain [Equation (3)] in the type-I
NCPM for the pump-signal noncollinear angle of α = 3.7°. Here the experimental parameters
Lc = 1 mm and Φ = 50 GW/cm2 are used.
6. FEMTOSECOND NOPA BASED ON β -BAB2O4
6.1 Broadband amplification of a single-filament

continuum
Figure 6-8 shows the experimental setup for the noncollinear-phase-

matched OPA (NOPA) developed in our group. The center wavelength,
pulse energy, and duration of the regenerative amplifier pulse Ti:sapphire
laser are about 790nm, 300 µJ, and 120 fs, respectively. The second
harmonic at 395 nm is generated with a 1 mm thick BBO crystal by type-I
phase matching with a cut angle of Φ = 29.4°. The second harmonic
generated with a 100 µJ pulse energy is used as a pump. The pulse width of
the second harmonic is estimated to be 150 fs by the cross-correlation trace
with the fundamental pulse. After passing through a delay line for
synchronization, the pump beam is telescoped to obtain a peak intensity of
~300 GW cm-2 and pumps a 1 mm thick BBO crystal (type I, Φ = 30°). The
small beam diameter of 0.5 mm is used to suppress the effect of the pulse-
front tilting [29]. The crystal is placed with the xz-plane lying in the plane of
the page, and the polarization of the pump beam is in the plane as an e-ray.
The thickness of the BBO crystal is determined so that the temporal walk-off
due to GVM p − s (~100 fs/mm) is well below the duration of the pump pulse.
A small fraction of the pulse energy (1–2 µJ at 790 nm) is focused onto a
2 mm thick sapphire plate and converted to a single-filament continuum to
be used as the signal beam [19-21, 65]. The continuum is obtained by
carefully adjusting the incident pulse energy to be only slightly higher than
the threshold of continuum generation. The stabilities of the intensity and
spectrum and the spatial coherence along the cross section are both excellent
and suited to be used as the seed of an OPA. The spectrum of the single-
filament continuum covers a broad spectral range from 450 nm to longer
than 1000 nm. There is a strong spike around 790 nm because of the
fundamental. Outside this region, the spectrum is smooth and flat, i.e., well-
suited for obtaining a smooth, short pulse without temporal structure because
of the Fourier-transformation relation.
BS 1kHz, 790nm
790nm SHG
HS 300µJ, 120fs
1µJ
395nm sapphire
100µJ plate white
light prism-pair
x compressor
NF y
pump z 550-690nm output
inverted 2-3µJ
D telescope signal αext PS
BBO
idler
Figure 6-8. Experimental setup for the noncollinear-phase-matched OPA (NOPA). SHG:
nonlinear crystal for second-harmonic generation; BS: beam splitter; HS: fundamental and
second-harmonic separator; NF: notch filter centered at 800 nm; D: variable optical delay
line; PS: periscope for rotating the polarization of the signal. The conelike parametric
fluorescence with the minimized dispersion (see text) is illustrated with the external cone
angle αext. Also illustrated are the crystal axes x, y, and z.
The sum-frequency pulse of the continuum at different wavelengths and

120 fs gate pulses at 790 nm are generated in a 0.5 mm thick BBO crystal
(type I, Φ = 35°) and probed to measure the group delay. The wavelength-
dependent group delay is determined as the center-of-mass delay position in
the corresponding cross-correlation trace [66]. The effect of dispersion in air
is corrected using refractive index data in the literature [67]. Because of
unsaturated self-phase modulation, the single-filament continuum does not
posses a linear chirp across the whole range of the spectrum as is well
known in optical fibers [1, 68]. The continuum beam is collimated and
passes through a notch filter with a peak reflectance of -90% at 800 nm to
reduce the fundamental pulse energy. It is then injected to propagate along
the dispersion-minimized cone surface. The direction is in the xz-plane to be
matched with the Poynting vector of the pump for the walk-off
compensation [24]. A broad spectral range of the continuum is
152 Chapter 6
noncollinearly amplified to a 2–3 µJ pulse energy with a short-term peak-to-

peak fluctuation of about 10%.
6.2 Output properties
The center wavelength of the amplified signal depends sensitively on the

delay of the pump pulse owing to the chirp of the continuum. By sweeping
the delay line length by 50 µm, the signal is stably and continuously tuned
from 550 to 690 nm, as shown in Figure 6-9. Each delay yields a broad
spectrum with the bandwidth up to 2000 cm-1. The small peaked structures
around 530 nm and 760 µm are caused by the imperfect flatness of the
phase-matching curve and the strong seed intensity near the fundamental,
respectively. The notch filter suppresses the unexpected dual-wavelength
amplification by decreasing the fundamental intensity and by separating
temporally the two spectral components in the single beam.
Figure 6-9. Spectra of the amplified signal. The center wavelength can be tuned only by
scanning the delay line of the pump by 50 µm.
In this group-velocity-matched geometry, the idler is broadly generated

and fan shaped with a spanning angle of about 7° in the xz-plane, which can
be observed by the second harmonic of the idler generated from the same
BBO crystal. Figure 6-10 shows the wavelength dependence of the external
noncollinear angle with respect to the pump direction.
Figure 6-10. Angular-dispersion property of the idler. Measured external noncollinear angles
βext with respect to the pump are shown (full circles at λs = 680 nm and open squares at λs =
600 nm). Also shown is the calculated phase-matching curve (solid curve).
Figure 6-11 shows the background-free intensity autocorrelation trace at

625 nm after the pulse compression. A 100 µm thick BBO (Type I, θ = 45° )
is used for the second-harmonic generation. The pulse compressor is a pair
of BK7 Brewster prisms. It is optimized to a 37 cm slant length with a
minimum insertion. By assuming that the envelope function is a sech2 pulse,
a least-squares best fit yields a pulse width of 14 fs. The spectral width of a
66 nm at FWHM corresponds to a temporal pulse width as short as 8 fs,
indicating the pulse is far from transform-limited.
2 2
Intensity (arb. units)
1 0
500 600 700
Wavelength (nm)
0
-100 -50 0 50 100
Delay (fs)
Figure 6-11. Intensity autocorrelation trace of the amplified signal after pulse compression
(full circles). Autocorrelation measured with a second-harmonic generation crystal. The
sech2-fit (solid curve) pulse width is 14 fs (FWHM). The spectrum is shown in the inset.
154 Chapter 6
The measured pulse width and time-bandwidth product over the pump-
delay-tuning range are shown in Figure 6-12. The bandwidth is 700 cm-1 at a
central wavelength of 550 nm. The bandwidth reaches up to 2000 cm-1 when
the center wavelength is tuned to 690 nm. For both tunings, the pulse
duration is sub-20-fs. The time-bandwidth product varies from 0.6 to 1.1,
except at 550 nm where it is 0.4. In the long wavelength region, the spectrum
spreads to beyond 800 nm with a non-negligible spike that degrades the
time-bandwidth products to be larger than 1. The large time-bandwidth
product is partly due to the pulse-width measuring apparatus that may
overestimate the pulse width. The devices responsible could include the
autocorrelator composed of a 100 µm thick BBO with a group-velocity
mismatch of 40 fs at 600 nm and dispersive media, such as the lens and
beam splitter. Using the 3.3 fs/step delay stage, the time resolution of the
pulse stage used for the delay line also hinders accurate measurement of the
pulse width. The wings on both sides observed in the autocorrelation traces
also indicate higher-order dispersion. These problems have been eliminated
as discussed below.

U
-1

H
O

E

J
V
F
J
V K
F 9

K
Y
sub-20fs G
U
F
P N
C W
2
$

%GPVGT9CXGNGPIVJ
PO
Figure 6-12. Wavelength dependence of the pulse width (full circles, sech2-fit) and time-
bandwidth product (full squares). The dotted line indicates the product of transform-limited
pulse (0.315) in the case of a sech2-pulse envelope.
The low conversion efficiency (-5%) can be improved by additional two-

stage amplification, where both pump and signal beams are reflected back to
the crystal twice by plane mirrors with small, vertical tilt angles. Even
though the gain of the first stage is as high as 104, the gain in the succeeding
stage is smaller than 3. The reason for this is not clear but may be caused by
the degradation of the beam quality from higher-order nonlinear effects. The
resultant pulse energy is up to about 8 µJ, resulting from a longer pulse
duration (~18 fs) and an increased time-bandwidth product of larger than 1.
7. LIMITATION OF PULSE WIDTH BECAUSE OF

PULSE-FRONT MISMATCH
The NOPA based on the signal-idler–group-velocity matching can

generate sub-20 fs pulses with a wide tuning range in the visible regime.
However, from the broad spectra with bandwidths exceeding 2000 cm-1,
much shorter pulses of less than 10 fs duration are expected. The
unexpectedly large time-bandwidth products are partly due to the non-
optimized autocorrelator and pulse compressor. However, it seems difficult
to explain a broadening factor of 2–3 from the transform-limited duration.
The most probable reason for the broadening is pulse-front tilting, an
intrinsic effect of the noncollinear interaction of ultrashort pulses [29]. In a
noncollinear geometry, the pulse fronts of the pump and signal cannot
overlap fully with each other. Noncollinear amplification causes a tilted gain
volume in the signal beam (with a noncollinear angle α), resulting in the
generation of a signal tilted by the same angle.
The external tilt angle γ is simply given from the relation
νs
tan α = tan γ . (31)
c
The group velocity ν s describes the refraction. The pulse-front tilting

induces a temporal delay across the beam diameter d by
d tan γ
δttilt = , (32)
c
which gives the maximum pulse broadening via time-space coupling. In a

femtosecond NOPA, the tilting is a more serious problem than in a
noncollinear OPO because an order-of-magnitude larger spot size is used in
OPAs [29]. In the present NOPA for α = 3.7° , the tilt angle of the exit
signal is calculated to be 6.3° at 600 nm. The 0.5 mm diameter of the pump
can cause broadening up to 180 fs. This value is too large to explain the
present experiments. An interaction within a smaller cross section is taking
place in the crystal (see Figure 6-13). The divergence of the amplified signal
and the relatively low conversion efficiency support this explanation.
The most evident feature is the spatial chirp of the amplified signal. The
pulse-front tilting is accompanied with a dispersion of the exit angle ε [47]
156 Chapter 6
dε tan γ
=− . (33)
dλ λ
Figure 6-13. Schematic of the noncollinear interaction between the pump pulse and signal
pulse. The crystal is on the left side. The volumes occupied by the pump pulse are shown by
rectangular shapes elongated perpendicular to the propagation direction. Those for probe
pulse are shown by closed gray rectangles. The gain volume introduced by the pump pulse
causes the tilting of the wave front of the signal pulse by α in the crystal, resulting in the
special and angular dispersions of the exiting signal pulse.
This feature originates from the inhibition of confined beam propagation

with pulse-front tilting from the wave front, which is analogous to
propagation through a prism [46] or grating [69]. The spatial chirp of the
amplified signal is measured by scanning a 100 µm slit on the xz-plane over
the beam cross section after collimating. The spectrum passing through the
slit exhibits a position-dependent feature (Figure 6-14). The spatial chirp is
estimated by a weighted average of the spectra. The result shows a
characteristic quasi-linear dependence on the exit angle (Figure 6-15). The
slope expected from the pulse-front tilting with a tilt angle of γ = 6.3° shows
an extremely good agreement with the measured spatial chirp.
2
total
slit position

1
0
500 550 600 650 700
Wavelength (nm)
Figure 6-14. Spatial chirp of the signal spectra after passing through the vertical slit. Slits are
located at three different positions (thin curves) after the nonlinear crystal of the NOPA. The
case of the fully open slit is shown in bold.
620
Wavelength (nm)
610
600
γ =6.3°
590 ( α=3.7°)
580
-5 -2.5 0 2.5 5
Relative Exit Angle (mrad)
Figure 6-15. Dependence of the center-of-mass wavelength of signal spectra as a function of

the relative exit angle. The dashed line indicates the expected spatial chirp of the signal tilted
by 6.3°.
7.1 Tilted pump geometry for pulse front matching
The pulse-front matching geometry is schematically shown in Figure 6-

16. The pump beam passes through a prism with incident and exit angles c1
and c2, respectively, causing the pulse-front to tilt. The tilt angle γprism just
after the prism is [47]
dϕ 2 sin α apex dn
tan γ prism = −λ =− λ ', (34)
dλ cos φ1 cos φ 2 dλ '
'
158 Chapter 6
where n is the refractive index of the prism, α apex is the prism apex angle,
and φ1 ' is the internal incident angle. The tilt angle is then decreased and the
pulse width broadened during propagation in free space by the spectral
lateral walk-off over the beam cross section [69]. The following telescope
recollimates the spectral lateral walk-off and images the tilted fronts on the
focal plane with a longitudinal magnification factor, M = f1 / f 2 , which
gives the tilt angle γ ext at the crystal position
f1
tan γ ext = tan γ prism . (35)
f2
The internal tilt angle γ int is reduced by refraction with the relation
νg
tan γ int = tan γ ext . The pulse-front matching condition γ int = α is then
c
v g f1
tan α = tan γ prism . (36)
c f2
Because the spectral lateral walk-off spectrum (lateral dispersion) in the

horizontal direction is insignificant in the present experimental conditions,
the beam is nearly optimally recollimated in both horizontal and vertical
directions. Since the spot size is reduced by the lateral magnification factor
M-1 = f2 / fl', the design of the telescope is essentially determined by the
requirement of the efficient amplification and dimensional restriction.
white light continuum
signal
output
confocal position
fused silica
BBO f=70 f=200 45 prism
49 pump
idler
130µJ, 150fs
100mm 88mm 270mm 50mm
@395nm
magnification
γext=6.4 γprism=2.3
Figure 6-16. Geometry of pulse-front matching optical system. L: lens; M: mirror; CYM:
cylindrical mirror; SM: spherical mirror; γint: internal tilt angle of pulse front of probe pulse in
the nonlinear crystal of NOPA; γext: external tilt angle of pulse front of probe pulse out of the
nonlinear crystal of NOPA; BBO: nonlinear crystal (β-barium borate).
8. CHIRP PROPERTY OF SIGNAL
Because the seeded continuum has a chirp, the OPA is described as

chirped-pulse amplification [70], and the amplified signal initially possesses
the same chirp. In addition, the signal experiences phase distortion caused by
the nonlinear effect in the BBO crystal and by the propagation through
dispersive media such as the crystal, air, and the beam splitters in the
frequency-resolved auto-correlation (FRAC). To attain a transform limited
sub-10 fs pulse, accurate and precise phase correction is required over the
whole spectral range.
The dependence of the phase φ(ω) ( ω: angular frequency) of the pulse on
the wavelength is expanded around a central angular frequency ω0
dφ  1  d 2φ 
φ (ω ) = φ (ω 0 ) +  ( )  (ω − ω 0 )

 ω − ω + 
 2 
2
 dω  ω 0 2  dω ω
0
0
(37)
1d φ  1d φ
K
3 4
+   (ω − ω 0 ) +   (ω − ω 0 ) +
3 4
3 
3!  dω ω 4!  dω 4 ω
0 0
The central wavelength may not be uniquely defined in the case of an

extremely short pulse, which may have an irregular spectral shape with
several spikes and a complicated structure.
The distortion of the pulse shape is not caused by the dispersion of the
phase velocity but by that of the group velocity. Thus the group delay (GD)
T (ω ) =
dφ  d 2 φ
=
dω  dω 2



(ω − ω 0 ) +
1  d 3φ 
 ( ω − ω0
2  dω 3 
)
2
 ω0
(38)
1  d 4φ 
+  (ω − ω 0 ) +
6  dω 4 
3
L
is used for pulse-propagation characterization. The constant term induces
only a temporal shift during the pulse propagation without any shape change
L
and is thus neglected in Equation (38). The coefficients of the expansion
φ '' (ω ), φ ''' (ω 0 ), φ '''' (ω 0 ), are called GD dispersion (GDD), third-order
dispersion (TOD), fourth-order dispersion (FOD), and so on, respectively.
In the propagation through a medium with the path length l and wave
vector κ = (ω ) = n(ω )ω / c , the phase shift φ (ω ) = κ (ω )l yields
160 Chapter 6
T (ω ) =
l
, (39)
ν g (ω )
which gives the intuitive representation of the GD. However, the concept of
the GD can be represented in terms of wavelength-dependent optical paths
P(ω ) such as a grating pair [46] or prism pair [48].
Because the higher terms, such as a TOD and FOD, become significant in
a sub-10 fs regime with a broad spectral range, a more precise phase
correction is needed for the compression to the transform-limit.
8.1 Compression to the sub-5 fs regime
The GD compensation of the signal pulses using a pair of

ultrabroadband-chirped mirrors (UBCMs) is performed taking into account
the higher-order dispersion of the whole system including a 45° fused-silica
prism pair, air along the optical path, and a beam splitter. By global
optimization across the whole spectral range, the best compression is
attained in the case of a 1 m slant length and a 6.0 mm internal path length of
the prism pair at 650 nm and eight reflections in the four round trips between
the two UBCMs in the pair, as shown in Figure 6-17.
Frequency (THz)
550 500 450 400
100
0
-100
-200 prism pair
-300 prism separation=1 M
material amount=6.7mm
200
100
-100
UBCM
8 reflections
150
air
100 4.5 M
50
50
40 beam splitter
30 0.5 mm
20
10
0
-10
100
50
0
-50
-100
-150
-200 Total
-250 -(measured GD)
-300
500 550 600 650 700 750 800
Wavelength (nm)
Figure 6-17. Group-delay property of the compressor. The group delays of the prism pair, the
UBCM pair with four round trips (thin, solid curve), the prism pair (dashed curve), air with a
4.5 m path length through the air, the beam splitter in the FRAC, and the whole compressor
system are shown from top to bottom. Also shown is the measured group delay of the signal
(full circles) in the bottom box with the sign reversed.
To obtain the full-bandwidth amplification, the compressor is divided

into two portions before and after the BBO crystal. The experimental setup
for the sub-5 fs NOPA system is shown in Figure 6-18. The pre-compressor
is composed of one UBCM pair in a single round-trip configuration, and the
main compressor has another pair in a three round-trip configuration with
other elements. The insertion of the UBCM pair before the crystal reduces
the seed chirp to about 10 fs GD difference between 530 and 770 nm, and
the inequality ∆T + ∆TBBO < τ p / 2 + (GMPp − s ) MAX is fulfilled.
162 Chapter 6
LBO HS 790nm
P2 395nm
TP
L1 <5fs, 5µJ
CM
He-Ne laser
VND WSM
BS pump
~1µJ 100µJ P1 CM
S L2 FRAC
PS
CM TBS
CF signal PD
CM BBO
idler BBO CCM
D BPF
2ω FD
120fs,400µJ PMT CCM
1kHz,790nm SM SM
Figure 6-18. Schematic of the visible sub-5-fs pulse generator. L: lens; M: mirror; CYM:
cylindrical mirror; SM: spherical mirror; BS: beam splitter; HS: harmonic separator; TP:
prism for pulse-front tilting; L1, L2: lenses for the telescope; SMs: spherical mirrors (γ = 100
mm); VND: variable neutral-density filter; WSM: spherical mirrors (γ = 120 mm); CF: cut-off
filter; D: optical delay line; PS: periscope; P1,P2: 45° fused silica prisms; and CCMs: corner-
cube mirrors.
By tailoring the experimental conditions and parameters as described

above, full-bandwidth amplification by the double-path amplifier has been
achieved. As evident in the chirp properties of the signal and idler inspected,
effects such as a SPM or a cascaded second-order process [71] seem to be
weak under our experimental conditions. Therefore the reduction of the
initial chirp effect does not significantly change the net pulse parameters,
including the overall chirp property.
The pulse energy varies from about 6 to 7 µJ just after the BBO crystal;
after the main compressor with about 70% throughput, it becomes about 5
µJ. The highest throughput of about 80% is obtained by using properly
selected UBCMs. Then only the reflections on the surfaces of the prisms
cause loss.
The spectrum of the signal is shown in Figure 6-19. The spectral shape
depends on the positions of the delay lines, which are fixed to maximize the
bandwidth. The FWHM is as broad as 240 nm, corresponding to 150 THz,
and the transform-limited pulse width is calculated to be 4.4 fs. The
bandwidth is slightly narrower than that reported in Reference [37]. The cut-
off filter of the seed effectively suppresses the amplification around 790 nm
where the chirp exhibits a large nonlinearity. The collimated seed-pump
beam interaction also avoids the spectral broadening of the signal caused by
beam divergence [66] that is accompanied by an undesirable spatial chirp.
Careful attention to both of these details is essential to obtain a signal
compressible to the transform-limit.
Frequency (THz)
600 500 400
2
240 nm (150 THz)

0
500 550 600 650 700 750 800
Wavelength (nm)
Figure 6-19. Spectrum of the signal pulses under a full-bandwidth operation.
The pulse shape is measured by FRAC. The pulse energy is appropriately

reduced by adjusting the radius of open aperture of an iris. Figure 6-20
shows the FRAC trace of the signal after optimization of the parameters and
alignment. The sech2-fit pulse width is as short as 3.5 fs, which is shorter
than the transform-limited width. Because of the significant modulation of
the spectral intensity, such an assumption cannot give an accurate width [1].
In the present work the pulse width is estimated by Fourier-transformation of
the spectrum including the wavelength dependence of the phase in two ways.
First, the phase is calculated from the properties of the measured group
delay of the signal T pulse (ω ) and the group delay of the compressor
Tcompresssor (ω ) (see Figure 6-17) yielding
φ (ω ) = ∫
ω
[T pulse (ω ) + Tcompressor (ω )]dω . (40)
The scattered values of the measured group delay are caused partly by the
poor time resolution and by intensity fluctuation. The measured group delay
is smoothed by fitting to a cubic-polynomial function. Using a calculated
phase, the FRAC trace of the compressor output is then constructed and
depicted by open circles in Figure 6-20.
164 Chapter 6
8
DIFT
7 (4.7fs)
Normalized Intensity
6 sech2-fit
5 (3.5fs)
4
3
2
1
0
-15 -10 -5 0 5 10 15
Delay Time (fs)
8
1
0
-40 -30 -20 -10 0 10 20 30 40
Delay Time (fs)
Figure 6-20. Frequency-resolved auto-correlation (FRAC) trace of the compressor output.

Traces are (a) for │τ│<15 fs and (b) for │τ│<40 fs (τ: delay time). Measured trace (solid
curve), calculated trace with measured group-delay (GD) of signal and of compressor (open
circles), calculated trace from the FT of measured spectrum (full circles), and calculated trace
using sech2 with 3.5 fs width (dashed curve) are shown. The calculated pulse width is 4.7 fs,
while that obtained by Fourier transformation of measured spectrum is 4.7 ±0.1 fs.
The second approach is to fit the measured FRAC trace with parameters
describing the phase. The complicated wavelength dependence of the group
delay of the UBCMs is included. The residual phase can be reasonably
assumed to have a smooth property and is determined by fitting the trace to a
cubic-polynomial function of the frequency [5, 6]. Because an
autocorrelation is used, the direction of the time axis is not determined by
this process alone. By comparison with the calculated phase mentioned
above, the direction can be determined. The fitted FRAC trace is shown in
Figure 6-20 with full circles.
Both phase profiles show similar behavior. The deviation is within π/2
radian over the whole spectral range from 510 to 790 nm. The oscillations
are due to the group delay of the UBCMs. The large deviations in the regions
of < 500 nm and > 800 nm are due to the limited bandwidth of the UBCMs.
However, UBCMs are less effective for pulse-width broadening because the
spectral intensity in these regions is weak. The difference between both
phases is mainly caused by the uncertainty of the measured group delay
because of a lack of information in the whole spectral range. It is also

difficult to determine the precise values of the compressor parameters such
as the internal path length.
The calculated and fitted FRAC traces have nearly the same shapes and
agree very well with the experimental trace even in the side wing regions.
The estimated intensity profiles are shown in Figure 6-21. The calculated
profile gives a 4.7 fs pulse duration, whereas the fitted one gives a 4.7 ± 0.1
fs pulse duration. The result indicates that the phase correction with the
highest precision is attained using the prism-chirped mirror compressor;
almost transform-limited pulses are obtained. It is the first demonstration
that the UBCMS can be applied to the visible range over more than 200
THz. And, the electric field oscillates only for 2.2 cycles in the FWHM of
the intensity profile.
2
Intensity (a. u)
1 4.7 fs
(TL 4.4 fs)
0
-20 -10 0 10 20
Time (fs)
Figure 6-21. Intensity profile of the compressed pulse. The transform-limited pulse width
determined by the spectrum is 4.4 fs (thin solid curve), whereas the calculated (thick solid
curve) and Fourier-transform-fit to FRAC with phase parametric (dashed curve) pulse widths
are 4.7 fs and 4.7 ± 0.1 fs, respectively.
The difference in the intensity profiles clearly indicates the limitation in

extracting the pulse shape from a FRAC trace. The phase information is
mainly included in the fringes of the side wings, which cannot be measured
with high enough signal-to-noise ratios in many cases. Uncertainty results
particularly in the shape of the trailing pulses. However, the pulse width,
which is defined as the FWHM, is reasonably estimated by this method with
a 0.1 fs accuracy. The pulse shape and phase can be precisely determined by
other methods such as frequency resolved optical gating (FROG) [3].
166 Chapter 6
9. SECOND-GENERATION NONCOLLINEAR
PARAMETRIC AMPLIFIER
The NOPA we developed consists of three parts: The first element is a

parametric amplifier with a gain bandwidth sufficient to support sub-4 fs
operation. The second component is a grating-chirped-mirror compressor,
used for rough group-delay compensation; it has a flexible mirror that allows
for fine spectral phase adjustment. The third part comprises a pulse
diagnostic set-up based on second-harmonic generation (SHG) FROG with
feedback to a personal computer that also controls the actuators of the
flexible mirror. An overview of the setup is presented in Figure 6-22. The
system is pumped by a 120 fs, 1 kHz repetition rate regenerative amplifier
[CPA1000 (Clark MXR)] that is seeded by a fiber laser oscillator (IMRA,
Clark). We now address separately the design and functions of each stage of
the setup.
Pulse diagnostic & Pulse Double-pass non-collinear Vis-NIR

adaptive feedback compressor optical parametric amplifier
Electrostatic Razor blade Variable delay

actuators Flexible P2
membrane
Quartz block Periscope
Spherical L=100 mm White-light pulse
P1
mirror pre-compressor
Input: 120fs
SH pulse: 0.15 mJ @790 nm
39-channel 0.05 mJ Spatial filter SHG crystal
HV driver @395 nm
White-light SP
Grating CM2
generator
SHG FROG Signal wave:
Periscope λ/2
CM1 5 µJ, 480-800nm, 300 fs
P3
BS2 0.5 µJ, <4 fs 1-mm BBO
Ultrabroadband Prism for SH
BS1 chirped mirrors angular dispersion
Periscope
To spectrometer
Figure 6-22. Schematic of experimental setup of sub-4 fs NOPA. λ/2: 800 nm wave-plate; SP:
2 mm sapphire plate; P1, P2: 45°quartz prisms; P3: 69° quartz prism (the distance from P3 to
the NOPA crystal is 80 cm); CM1,2: ultrabroadband chirped mirrors; GR: 300 lines/mm ruled
diffraction grating (Jobin Yvon); SM: spherical mirror, R = -400 mm; BS1,2: chromium-
coated d = 0.5 mm quartz beam splitters; SHG crystal: 0.4 mm θ = 29° BBO (EKSMA);
NOPA crystal: 1 mm θ = 31.5°BBO (Casix); and SHG FROG crystal: θ = 29° BBO wedge
plate d = 5÷20 µm (EKSMA). Spherical mirrors around the NOPA crystal are R = -200 mm.
Thick arrows on the left indicate the data flow from the pulse diagnostic setup (SHG FROG)
and the feedback to the flexible mirror.
The discovery of “magic” phase-matching conditions in a Type I-BBO

crystal pumped by a 400 nm light [26, 72] opened the door to producing
amplified visible pulses with a bandwidth of nearly 200 THz [36, 37, 41].
The uniqueness of this pumping arrangement comes from the fact that the
pump inclination to the direction of the seed (~ 3.7°) matches almost
perfectly the angle of birefringent walk-off between the o and e waves inside
the crystal. As a result, BBO as long as 1–2 mm can be employed even in 5
fs NOPAs. The implications of noncollinear phase matching are well
understood and discussed in numerous papers [26, 29, 31, 32, 34, 36-40, 66,
72].
We paid further attention to the subtleties of parametric amplification in a
noncollinear configuration [36, 38], by considering the effect of pulse-front
tilting of the signal wave on the ability to compress the signal into a sub-5 fs
pulse. To prevent the tilting of the signal pulse in space, an effect that also
results in angular dispersion of the amplified pulse, several researchers
proposed to use a pump beam with a tilted wave front. This configuration,
named pulse-front-matching, was implemented by sending the pump beam
through a prism and adjusting the pulse tilt with a telescope consisting of
two convex lenses [38, 41, 73].
We took care to find a balanced Mach-Zehnder interferometric
autocorrelator [74] to characterize the NOPA pulses [38, 41, 42]. This
particular autocorrelator does not have spatial resolution and, therefore, is
insensitive to wave-front tilt. This does not necessarily mean the real pulse
width does not suffer from broadening of the pulse width due to pulse-front
tilting, however. The required spatial sensitivity can also be achieved by
using an asymmetric number of reflections in the interferometer.
Together with the effect of pulse-front matching, the angular dispersion
of the pump beam in the sub-5 fs NOPA is an important factor in terms of
enhancing the phase-matching bandwidth. Even with relatively thick second-
harmonic crystals (1–2 mm LBO or BBO) and comparatively long (120–150
fs) pulses from standard regenerative amplifiers employed to pump the
NOPAs [37, 38, 41, 42], the resulting second-harmonic radiation has a
bandwidth of several nanometers. Wide-bandwidth parametric amplification
can subsequently be achieved by pointing the seed beam in a specific
direction for each spectral component of the pump, as schematically shown
in Figures 6-23 and 6-24. The adjustment of the pump beam dispersion,
required for phase-matching optimization, is obtained by selecting the apex
angle of a prism in the second-harmonic pathway [Figure 6-23(b)] and the
distance from the prism to the focusing optic. The required second-harmonic
(SH) dispersion
168 Chapter 6
 l
δα (λ ) = atan tan(γ (λ0 ) − γ (λ ) ) , , (41)
 f
where λ is the pump (SH) wavelength, λ0 denotes the central component of

the SH spectrum, l and f are the distance from the prism to the focusing optic
and the focal distance, respectively. The exit angle of the SH beam behind
the prism is
   sin γ 0 
γ (λ ) = asinn(λ ) sin θ apex − asin  . (42)
   n(λ ) 
In Equation (42), γ 0 is the angle between the SH beam and the normal to the
input face of the prism (incidence angle onto the prism), θ apex stands for the
apex angle of the prism, and n(λ ) is the refractive index of glass.
(a)
(b)
Figure 6-23. Extension of parametric bandwidth by adjusting the incidence angle of

individual pump wavelengths onto a NOPA crystal. (a) Schematic representation of
noncollinear pump geometry. (b) Second-harmonic dispersion adjustment based on the use of
a prism and a focusing optic.
In our second-generation NOPA system, the distance between the

Brewster-angled prisms for second-harmonic dispersion and a focusing
mirror (Figure 6-22) is 80 cm. The radius of curvature for the concave mirror
is R = -200 mm. Standard formalism for type I phase-matching in crystals
[17] was used in this calculation to obtain the results depicted in Figure 6-24.
We assumed that the efficiency of the parametric frequency conversion is
not high. Thus we could disregard any pulse-reshaping effect that can lead to
an effective reduction of the group-velocity mismatch between the pump and

signal waves because of a kind of nonlinear pulling effect in the reshaping
process. As a result, we believe the data shown in Figure 6-24 represent a
low estimate of the signal bandwidth that can be simultaneously phase-
matched in the NOPA. The pairs of curves with dark shaded areas in
between composing dark stripes in Figure 6-24 represent cases of individual
pump wavelengths. The cases of wavelengths of 389, 392, and 395 nm are
shown. The thickness of each dark stripe gives the FWHM of the angular
phase matching [17] and is determined by the nonlinear interaction length,
assumed to be equal to the crystal thickness of 1 mm. The whole shaded area
in Figure 6-24 corresponds to the combined bandwidth produced by the
entire second-harmonic spectrum, shown as a shaded contour in the inset.
Frequency [THz]
650 600 550 500 450 400
λp1=389 nm
λp2=392 nm
λ p3=395 nm
32
Phase-matching angle [deg]
31 1
Intensity
6 nm
Direction of
seed light
30
0
375 380 385 390 395 400 405 410
SH wavelength [nm]
500 550 600 650 700 750 800
Wavelength [nm]
Figure 6-24. Phase matching in a 1 mm Type I BBO crystal, θ = 31.5°. Dark-shaded contours
indicate FWHM of the angular phase matching for individual monochromatic pump
wavelengths after optimization of their incidence angles. Dashed line denotes the direction of
the seed beam. Inset depicts experimentally measured second-harmonic spectrum used to
pump the NOPA (shaded contour) and calculated conversion efficiency of a 0.4 mm Type I
BBO second-harmonic-generation crystal.
To maximize the effect by employing the widest region of pump

frequencies, we used a thinner, 0.4-mm, BBO for the frequency doubling of
the fundamental 120 fs pulses. This crystal provides an ~30% conversion
efficiency in contrast to the higher value of about 50% in the 1 mm BBO
used in our first NOPA; however, it ensures a larger spectral width of the
second harmonic. The corresponding frequency conversion efficiency is
given in the inset to Figure 6-24 by a solid curve and indicates that this
170 Chapter 6
crystal causes no significant reduction of the resulting second-harmonic

bandwidth.
Analogous ideas about bandwidth extension have been applied in other
parametric schemes. Angular dispersion of the frequencies in the
fundamental beam has been used to obtain achromatic phase matching for
second-harmonic generation [62], whereas a multipass OPA using slightly
different beam directions in every pass [58] has been shown to dramatically
broaden the resultant bandwidth of the amplified pulse. Recently, a single-
stage OPA pumped by two noncollinear beams has also been demonstrated
as an efficient means of boosting the spectral width [75].
Another method for obtaining extremely broad bandwidth with short
pulses is to use a two-stage compressor. This device incorporates a flexible
mirror for adaptive pulse shaping plus a simple and trustworthy feedback
loop based on a one-dimensional spectral measurement. Our rapid numerical
algorithm for adaptive control of the flexible mirror is superior to more
complex search routines, which are less resistant to laser intensity
fluctuations. The automated optimization results in the generation of two-
cycle pulses with a carrier wavelength near 600 nm. The absence of deep
modulation on the amplified spectrum in combination with adaptive phase
correction lead to a high-quality temporal profile and allow the concentration
of more than 90% of the pulse energy within a 7.5 fs time window.
10. CONCLUSIONS AND OUTLOOK
In this work, we report the design of a sub-5 fs visible–near-IR

parametric amplifier with adaptive pulse tailoring that permits a 20-fold
compression of the pump pulses from a standard Ti:sapphire chirped-pulse-
amplification system. Whereas smooth parametric spectra were generated
that correspond to 3.5 fs pulses (if ideally compressed), we managed to
obtain 3.9 fs pulses after passing the signal beam through a hybrid
compressor based on a diffraction-grating telescopic dispersion line,
specially designed multilayer-dielectric-chirped mirrors, and a computer-
controlled flexible mirror.
As has been shown in this chapter, efficient use of all pump frequencies
leads to the broadening of the amplification band. One of the fundamental
issues for further progress in the development of a NOPA with even shorter
pulse duration is the difficulty with the pulse diagnostics. The type I phase-
matching SHG FROG, employed in this work, is stretched to the limit
because of the finite phase-matching bandwidth of the SHG crystal.
Implementation of a heterodyne third-order FROG technique, based on
automatically phase-matched, frequency-resolved pump-probe in a

transparent medium [76] may provide an excellent alternative.
We would like to mention another fascinating use of the NOPA
technique. Remarkably, parametric amplifiers provide an alternative
approach to the generation of few-cycle optical pulses by using multiple
source spectra rather than a single one [77]. This becomes possible as a
result of unique phase relations among signal, pump, and idler waves in a
white-light-seeded NOPA [78]. Methods for obtaining monocycle pulses
include merging the signal and idler spectra [38, 41, 42] or merging the
spectra of the signal and the fundamental broadened around 800 nm [79] into
a single, coherent bandwidth since the carrier-envelope phase [80] of those
waves exhibit identical drift. Moreover, octave-spanning carrier-envelope–
phase-locked infrared idler pulses [78] and carrier-envelope-phase–self-
referenced visible signal pulses [81] from NOPAs have already been
demonstrated. These developments are opening up the possibility of using
NOPAs in field-sensitive applications of nonlinear optics [8, 81].
ACKNOWLEDGEMENTS
The author is grateful to Drs. A. Shirakawa, A. Baltuška, and P.

Kumbakhar for their essential contributions to the work presented in this
chapter. The work presented here was partly sponsored by the “Research for
the Future” program of the Japan Society for the Promotion of Science
(Grant No. JSPS-RFTF-97P-00101).
REFERENCES
[1] A. Baltuška, Z. Wei, M. S. Pshenichnikov, D. A. Wiersma, and R. Szipocs,
Appl. Phys. B 65, 175-188 (1997).
[2] A. Baltuška, Z. Y. Wei, M. S. Pshenichnikov, and D. A. Wiersma, Opt.
Lett. 22, 102-104 (1997).
[3] A. Baltuška, M. S. Pshenichnikov, and D. A. Wiersma, Opt. Lett. 23, 1474-
1476 (1998).
[4] M. Nisoli, S. DeSilvestri, and O. Svelto, App. Phys. Lett. 68, 2793-2795
(1996).
[5] M. Nisoli, S. DeSilvestri, O. Svelto, R. Szipocs, K. Ferencz, C. Spielmann,
S. Sartania, and F. Krausz, Opt. Lett. 22, 522-524 (1997).
[6] M. Nisoli, S. Stagira, S. DeSilvestri, O. Svelto, S. Sartania, Z. Cheng, M.
Lenzner, C. Spielmann, and F. Krausz, Appl. Phys. B 65, 189-196 (1997).
172 Chapter 6
[7] R. Szipocs, K. Ferencz, C. Spielmann, and F. Krausz, Opt. Lett. 19, 201-
203 (1994); R. Szipocs and A. Köházi-Kis, Appl. Phys. B 65, 115-135
(1997); R. Szipocs and F. Krausz, Dispersive dielectric mirror, USA Patent
5734503 (1998).
[10] J. K. Ranka, R. S. Windeler, and A. J. Stentz, Opt. Lett. 25, 25-27 (2000); J.
K. Ranka, R. S. Windeler, and A. J. Stentz, Opt. Lett. 25, 796-798 (2000).
[11] I. D. Jung, F. X. Kärtner, N. Matuschek, D. H. Sutter, F. Morier-Genoud,
G. Zhang, U. Keller, V. Scheuer, M. Tilsch, and T. Tschudi, Opt. Lett. 22,
1009-1011 (1997); Y. Chen, F. X. Kärtner, U. Morgner, S. H. Cho, H. A.
Haus, E. P. Ippen, and J. G. Fujimoto, J. Opt. Soc. Am. B 16, 1999-2004
(1999); U. Morgner, F. X. Kärtner, S. H. Cho, Y. Chen, H. A. Haus, J. G.
Fujimoto, E. P. Ippen, V. Scheuer, G. Angelow, and T. Tschudi, Opt. Lett.
24, 411-413 (1999); G. Steinmeyer, D. H. Sutter, L. Gallmann, N.
Matuschek, and U. Keller, Science 286, 1507-1512 (1999).
Ye, Science 293, 1286-1289 (2001); R. K. Shelton, S. M. Foreman, L. S.
Ma, J. L. Hall, H. C. Kapteyn, M. M. Murnane, M. Notcutt, and J. Ye, Opt.
Lett. 27, 312-314 (2002).
[13] Z. Y. Wei, Y. Kobayashi, Z. G. Zhang, and K. Torizuka, Opt. Lett. 26,
1806-1808 (2001).
[14] P. E. Powers, R. J. Ellingson, W. S. Pelouch, and C. L. Tang, J. Opt. Soc.
Am. B 10, 2162-2167 (1993).
[15] V. Petrov, F. Seifert, O. Kittelmann, J. Ringling, and F. Noack, J. Appl.
Phys. 76, 7704-7712 (1994); M. K. Reed and M. K. Steiner-Shepard, IEEE
J. Quantum Electron. 32, 1273-1277 (1996).
[16] K. R. Wilson and V. V. Yakovlev, J. Opt. Soc. Am. B 14, 444-448 (1997).
[17] V. G. Dmitriev, G. G. Gurzadyan, and D. N. Nikogosyan, Handbook of
Nonlinear Optical Crystals (Springer-Verlag, Berlin, 1997).
[18] F. Seifert, V. Petrov, and F. Noack, Opt. Lett. 19, 837-839 (1994); M.
Nisoli, S. De Silvestri, V. Magni, O. Svelto, R. Danielius, A. Piskarskas, G.
Valiulis, and A. Varanavicius, Opt. Lett. 19, 1973-1975 (1994).
[19] M. K. Reed, M. K. Steiner-Shepard, and D. K. Negus, Opt. Lett. 19, 1855-
1857 (1994).
[20] M. K. Reed, M. S. Armas, M. K. Steiner-Shepard, and D. K. Negus, Opt.
Lett. 20, 605-607 (1995).
[21] V. V. Yakovlev, B. Kohler, and K. R. Wilson, Opt. Lett. 19, 2000-2002
(1994).
[22] S. R. Greenfield and M. R. Wasielewski, Opt. Lett. 20, 1394-1396 (1995).
[23] S. Takeuchi and T. Kobayashi, J. Appl. Phys. 75, 2757-2760 (1994).
[24] E. S. Wachman, W. S. Pelouch, and C. L. Tang, J. Appl. Phys. 70, 1893-
1895 (1991).
[25] Q. Fu, G. Mak, and H. M. van Driel, Opt. Lett. 17, 1006-1008 (1992); A.
Shirakawa, H. W. Mao, and T. Kobayashi, Opt. Commun. 123, 121-128
(1996); D. T. Reid, C. McGowan, M. Ebrahimzadeh, and W. Sibbett, IEEE

J. Quantum Electron. 33, 1-9 (1997).
[26] G. M. Gale, M. Cavallari, T. J. Driscoll, and F. Hache, Opt. Lett. 20, 1562-
1564 (1995).
[27] P. Di Trapani, A. Andreoni, G. P. Banfi, C. Solcia, R. Danielius, A.
Piskarskas, P. Foggi, M. Monguzzi, and C. Sozzi, Phys. Rev. A 51, 3164-
3168 (1995).
[28] P. Di Trapani, A. Andreoni, P. Foggi, C. Solcia, R. Danielius, and A.
Piskarskas, Opt. Commun. 119, 327-332 (1995).
[29] P. Di Trapani, A. Andreoni, C. Solcia, P. Foggi, R. Danielius, A. Dubietis,
and A. Piskarskas, J. Opt. Soc. Am. B 12, 2237-2244 (1995).
[30] J. Wang, M. H. Dunn, and C. F. Rae, Opt. Lett. 22, 763-765 (1997).
[31] A. Shirakawa and T. Kobayashi, IEICE Trans. Electron. E81C, 246-253
(1998).
[32] A. Shirakawa and T. Kobayashi, App. Phys. Lett. 72, 147-149 (1998).
[33] F. Hache, M. Cavallari, and G. M. Gale, in Ultrafast Phenomena X, edited
by P. F. Barbara, J. G. Fujimoto, W. H. Knox and W. Zinth (Springer-
Verlag, Berlin, 1996).
[34] T. Wilhelm, J. Piel, and E. Riedle, Opt. Lett. 22, 1494-1496 (1997).
[35] G. Cerullo, M. Nisoli, and S. De Silvestri, App. Phys. Lett. 71, 3616-3618
(1997).
[36] A. Shirakawa, I. Sakane, and T. Kobayashi, Opt. Lett. 23, 1292-1294
(1998).
[37] G. Cerullo, M. Nisoli, S. Stagira, and S. De Silvestri, Opt. Lett. 23, 1283-
1285 (1998).
[38] A. Shirakawa, I. Sakane, M. Takasaka, and T. Kobayashi, App. Phys. Lett.
74, 2268-2270 (1999).
[39] E. Riedle, M. Beutter, S. Lochbrunner, J. Piel, S. Schenkl, S. Sporlein, and
W. Zinth, Appl. Phys. B 71, 457-465 (2000).
[40] G. Cerullo, M. Nisoli, S. Stagira, S. De Silvestri, G. Tempea, F. Krausz,
and K. Ferencz, Opt. Lett. 24, 1529-1531 (1999); G. Cerullo, M. Nisoli, S.
Stagira, S. De Silvestri, G. Tempea, F. Krausz, and K. Ferencz, Appl. Phys.
B 70, S253-S259 (2000).
[41] T. Kobayashi and A. Shirakawa, Appl. Phys. B 70, S239-S246 (2000).
[42] T. Kobayashi, A. Shirakawa, and T. Fuji, IEEE J. Sel. Top. Quantum
Electron. 7, 525-538 (2001).
[43] M. Zavelani-Rossi, G. Cerullo, S. De Silvestri, L. Gallmann, N. Matuschek,
G. Steinmeyer, U. Keller, G. Angelow, V. Scheuer, and T. Tschudi, Opt.
Lett. 26, 1155-1157 (2001).
[44] G. Cerullo, G. Lanzani, M. Muccini, C. Taliani, and S. De Silvestri, Phys.
Rev. Lett. 83, 231-234 (1999); A. Sugita, T. Saito, H. Kano, M. Yamashita,
and T. Kobayashi, Phys. Rev. Lett. 86, 2158-2161 (2001); T. Kobayashi, A.
Shirakawa, H. Matsuzawa, and H. Nakanishi, Chem. Phys. Lett. 321, 385-
393 (2000); T. Fuji, T. Saito, and T. Kobayashi, Chem. Phys. Lett. 332,
174 Chapter 6
324-330 (2000); T. Kobayashi, T. Saito, and H. Ohtani, Nature 414, 531-

534 (2001); H. Kano, T. Saito, A. Ueki, and T. Kobayashi, Int. J. Mod.
Phys. B 15, 3817-3820 (2001); H. Kano and T. Kobayashi, J. Chem. Phys.
116, 184-195 (2002); A. Sugita, T. Furuhi, M. Yamashita, and T.
Kobayashi, J. Phys. Chem. A 106, 581-588 (2002); T. Saito and T.
Kobayashi, Opt. Mater. 21, 301-305 (2003); S. Adachi, V. M. Kobryanskii,
and T. Kobayashi, Phys. Rev. Lett. 89, - (2002); T. Kobayashi, T. Saito,
and S. Adachi, J. Lumin. 102, 722-726 (2003).
[45] M. Wittmann, A. Nazarkin, and G. Korn, Opt. Lett. 26, 298-300 (2001).
[46] E. Treacy, IEEE J. Quant. Electr. 5, 454 (1969).
[47] Z. Bor and B. Racz, Opt. Commun. 54, 165-170 (1985).
[49] D. Zeidler, T. Hornung, D. Proch, and M. Motzkus, Appl. Phys. B 70,
S125-S131 (2000).
[50] E. Zeek, K. Maginnis, S. Backus, U. Russek, M. Murnane, G. Mourou, H.
Kapteyn, and G. Vdovin, Opt. Lett. 24, 493-495 (1999); E. Zeek, R.
Bartels, M. M. Murnane, H. C. Kapteyn, S. Backus, and G. Vdovin, Opt.
Lett. 25, 587-589 (2000).
[51] R. Bartels, S. Backus, E. Zeek, L. Misoguti, G. Vdovin, I. P. Christov, M.
M. Murnane, and H. C. Kapteyn, Nature 406, 164-166 (2000).
[52] M. R. Armstrong, P. Plachta, E. A. Ponomarev, and R. J. D. Miller, Opt.
Lett. 26, 1152-1154 (2001).
[53] P. Baum, S. Lochbrunner, L. Gallmann, G. Steinmeyer, U. Keller, and E.
Riedle, Appl. Phys. B 74, S219-S224 (2002).
[54] A. Baltuška, T. Fuji, and T. Kobayashi, Opt. Lett. 27, 306-308 (2002).
[55] R. Trebino and D. J. Kane, J. Opt. Soc. Am. A 10, 1101-1111 (1993); R.
Trebino, K. W. DeLong, D. N. Fittinghoff, J. N. Sweetser, M. A.
Krumbugel, B. A. Richman, and D. J. Kane, Rev. Sci. Instrum. 68, 3277-
3295 (1997).
[56] R. Byer and R. L. Herbst, in Nonlinear Infrared Generation, edited by Y.
R. Shen (Springer-Verlag, Berlin, 1997).
[57] R. Danielius, A. Piskarskas, A. Stabinis, G. P. Banfi, P. Di Trapani, and R.
Righini, J. Opt. Soc. Am. B 10, 2222-2232 (1993).
[58] T. S. Sosnowski, P. B. Stephens, and T. B. Norris, Opt. Lett. 21, 140-142
(1996).
[59] S. Fournier, R. Lopez-Martens, C. Le Blanc, E. Baubeau, and F. Salin, Opt.
Lett. 23, 627-629 (1998).
[60] M. Nisoli, S. Stagira, S. De Silvestri, O. Svelto, G. Valiulis, and A.
Varanavicius, Opt. Lett. 23, 630-632 (1998).
[61] G. Szabo and Z. Bor, Appl. Phys. B 50, 51-54 (1990); R. A. Cheville, M. T.
Reiten, and N. J. Halas, Opt. Lett. 17, 1343-1345 (1992); T. Hofmann, K.
Mossavi, F. K. Tittel, and G. Szabo, Opt. Lett. 17, 1691-1693 (1992); S. H.
Ashworth, M. Joschko, M. Woerner, E. Riedle, and T. Elsaesser, Opt. Lett.
20, 2120-2122 (1995).
[62] B. A. Richman, S. E. Bisson, R. Trebino, M. G. Mitchell, E. Sidick, and A.

Jacobson, Opt. Lett. 22, 1223-1225 (1997); B. A. Richman, S. E. Bisson, R.
Trebino, E. Sidick, and A. Jacobson, Opt. Lett. 23, 497-499 (1998).
[63] P. Kumbhakar and T. Kobayashi, J. Appl. Phys. 94, 1329-1338 (2003).
[64] D. Eimerl, L. Davis, S. Velsko, E. K. Graham, and A. Zalkin, J. Appl.
Phys. 62, 1968-1983 (1987).
[65] M. K. Reed, M. K. Steiner-Shepard, M. S. Armas, and D. K. Negus, J. Opt.
Soc. Am. B 12, 2229-2236 (1995).
[66] G. M. Gale, F. Hache, and M. Cavallari, IEEE J. Sel. Top. Quantum Electr.
4, 224-229 (1998).
[67] Handbook of Chemistry and Physics (CRC Press, Cleveland, 1988-1989).
[68] W. J. Tomlinson, R. H. Stolen, and C. V. Shank, J. Opt. Soc. Am. B 1, 139-
149 (1984); R. L. Fork, C. H. B. Cruz, P. C. Becker, and C. V. Shank, Opt.
Lett. 12, 483-485 (1987).
[69] O. E. Martinez, Opt. Commun. 59, 229-232 (1986).
[70] A. Dubietis, G. Jonusauskas, and A. Piskarskas, Opt. Commun. 88, 437-
440 (1992).
[71] F. Hache, A. Zeboulon, G. Gallot, and G. M. Gale, Opt. Lett. 20, 1556-
1558 (1995).
[72] T. J. Driscoll, G. M. Gale, and F. Hache, Opt. Commun. 110, 638-644
(1994).
[73] R. Danielius, A. Piskarskas, P. Di Trapani, A. Andreoni, C. Solcia, and P.
Foggi, Opt. Lett. 21, 973-975 (1996).
[74] C. Spielmann, L. Xu, and F. Krausz, Appl. Optics 36, 2523-2525 (1997).
[75] E. Zeromskis, A. Dubietis, G. Tamosauskas, and A. Piskarskas, Opt.
Commun. 203, 435-440 (2002).
[76] S. Yeremenko, A. Baltuska, F. de Haan, M. S. Pshenichnikov, and D. A.
Wiersma, Opt. Lett. 27, 1171-1173 (2002).
[77] O. Albert and G. Mourou, Appl. Phys. B 69, 207-209 (1999); Y. Kobayashi
and K. Torizuka, Opt. Lett. 26, 1295-1297 (2001).
[78] A. Baltuška, T. Fuji, and T. Kobayashi, Phys. Rev. Lett. 88, art. no.-
133901 (2002).
[79] M. S. Pshenichnikov, personal communication 2001.
Chapter 7
OPTICAL FREQUENCY MEASUREMENT
Thomas Udem, Marcus Zimmermann, Ronald Holzwarth, Marc Fischer,

Nikolai Kolachevsky, Theodor W. Hänsch
Abstract: The new alliance between the time and frequency domains in laser
spectroscopy has made it possible to develop ultrafast counting schemes that
can keep track of single optical oscillations. With this achievement, counting
optical oscillations of more than 1015 cycles in one second has become a
simple task. High-resolution spectroscopy for basic research and metrology
greatly benefit from this technology as it has allowed the highest possible
precision. This development has also led to the construction of all-optical
atomic clocks that are expected to eventually outperform current state-of-the-
art cesium clocks. The possibility to measure almost any frequency ratio with
very high precision can be used to search for possible variations of natural
constants.
Key words: hydrogen spectroscopy, optical atomic clocks, optical frequency metrology,
optical frequency synthesis
1. FREQUENCY COMBS
Frequency can be measured with by far the highest precision of all

physical quantities. In the radio frequency (rf) domain (say up to 100 GHz),
frequency counters have existed for a long time. Almost any of the most
precise measurements in physics have been performed with such a counter
that uses an atomic clock as a time base. To extend this accurate technique to
higher frequencies, so called harmonic frequency chains have been
constructed since the late 1960s [1]. In such a chain, nonlinear elements
produce frequency multiples (harmonics) of a given oscillator to which a
subsequent oscillator is phase locked. The latter is necessary because
7. OPTICAL FREQUENCY MEASUREMENT 177
nonlinear devices usually produce weak signals, at least when they are
driven with a continuous wave (cw). Electronic phase-locked loops can be
used to stabilize any kind of oscillator, even lasers, provided their intrinsic
stability is sufficient so that there is no need for very rapid frequency
corrections. Repeating the multiplication and phase-lock procedure many
times makes it possible to convert a reference radio frequency, say from an
atomic clock, to much higher frequencies. Because of the large number of
steps necessary to build a long harmonic frequency chain, it was not before
1995 when visible laser light was first referenced phase coherently to a
cesium atomic clock using this method [2].
The disadvantage of these harmonic frequency chains was not only that
they could easily fill several large laser laboratories at one time, but that they
could be used to measure only a single optical frequency. Even though
mode-locked lasers for optical frequency measurements were used in
rudimentary form in the late 1970s [3], this method did not become practical
until the advent of femtosecond (fs) mode-locked lasers. Such a laser
necessarily emits a very broad spectrum, comparable in width to the optical
carrier frequency. Currently the field’s work horse is the Ti:sapphire Kerr-
lens mode-locked laser, but fiber-based lasers are expected to take over for
frequency metrology applications.
In the frequency domain, a train of short pulses from a femtosecond
mode-locked laser is the result of a phase-coherent superposition of many cw
longitudinal cavity modes. These modes at ωn1 form a series of frequency
spikes called a frequency comb. The individual modes can be selected by
phase locking other cw lasers to them. As has been shown, the modes are
remarkably uniform, i.e., the separation between adjacent modes is constant
across the frequency comb [4, 5]. This strictly regular arrangement is the
most important feature used for optical frequency measurement and may be
expressed as:
ω n = nω r + ω CE . (1)
Here the mode number n of some 105 may be enumerated such that the
frequency offset ωCE lies in between 0 and ωr = 2π/T. The mode spacing is
thereby identified with pulse repetition rate, i.e., the inverse pulse repetition
1
The notation used in this chapter corresponds to that of other chapters through the following
relationships:
ωn = 2πνn
ωr = 2πfr
ωCE = 2πf0
∆φ = ∆φce
178 Chapter 7
time T. With the help of Equation (1), two radio frequencies ωr and ωCE are
linked to the optical frequencies ωn of the laser. For this reason, mode-
locked lasers are capable of replacing the harmonic frequency chains of the
past.
To derive these frequency comb properties [6, 7], as detailed by Equation
(1), it is useful to consider the electric field E(t) of the emitted pulse train. If
the pulses were exact time-shifted copies, E(t) = E(t - T), a simple Fourier
transformation would yield a strictly periodic spectrum with a mode
separation of ωr and a zero comb offset ωCE = 0. However, this is not what
occurs in a real laser. Because of intracavity dispersion, the group and phase
velocities do not match for the pulse that is stored in the cavity. This causes
the carrier wave to continuously shift with respect to the pulse envelope (see
Figure 7-1). The pulses that emerge at the output-coupling mirror after each
round trip show a discrete pulse-to-pulse carrier-envelope phase shift of ∆φ.
E (t)
∆7 2∆7
Fourier
transformation
E (1)
1CE
1r 1c 1
Figure 7-1. Three consecutive pulses of the pulse train emitted by a mode-locked laser and
the corresponding spectrum. The pulse-to-pulse phase shift ∆φ results in an offset frequency
ωCE = ∆φ/T because the optical carrier wave at ωc moves with the phase velocity while the
envelope moves with the group velocity.
The difference between phase and group velocity is determined by the

dispersion properties of the cavity as a whole. The remarkable property of
Kerr-lens mode-locked lasers, such as the Ti:sapphire laser, is that this
difference is the only effect of intracavity dispersion on the pulse train.
Higher-order dispersive terms that would continuously deform the pulse are
strongly suppressed. For this reason, the properties of the pulse train are
most conveniently derived by separating the optical carrier wave that
propagates with its phase velocity at ωc from the pulse-envelope function:
E (t ) = A(t )e −iω ct . The pulse repetition time T = 2π / ω r is determined from
the group velocity by demanding that A(t) = A(t - T). Since it is strictly
periodic it may be written in terms of a Fourier series
E (t ) = A(t )e − iωct + c.c. = ∑A e

q
q
− i (ω c + qω r ) t
+ c.c. , (2)
where Aq are Fourier components of A(t). This equation represents a comb of

laser frequencies with spacing ωr that is shifted as a whole from the
harmonics of ωr, just as in Equation (1). The only difference is the
numbering of modes, which clearly is a matter of convention.
2. THE CESIUM D1 LINE AND THE FINE

STRUCTURE CONSTANT
The first application of a fs mode-locked laser in optical frequency

metrology occurred in 1998 with the determination of the transition
frequency of the cesium D1 line at 335 THz (895 nm) [8]. In that experiment,
we compared the frequency of the cesium D1 resonance line with the 4th
harmonic of a CH4-stabilized 3.39 µm He-Ne laser. The frequency of this
reference laser was known to within 2.6 parts in 1013. To bridge the large
frequency gap of 18.39 THz between the 4th harmonic near 848 nm and the
895 nm Cs D1 line, we used the frequency comb generated by a Mira 900
system (Coherent Inc.) delivering 70 fs pulses with a repetition frequency of
75 MHz. Even though this was already the largest frequency gap bridged
coherently at that time, it was clear that the method could do much better
with shorter pulses.
The cesium D1 line was observed in a 7.5 cm long cell at room
temperature. To probe the cesium D1 transition, we used a saturation
spectrometer with two linearly polarized counter-propagating laser beams
with equal intensities (10 µW/cm2). With this Doppler-free method, we
observed 4 hyperfine components of the single stable isotope 133Cs for the
transitions from the ground states Fg = 3 and Fg = 4 to the excited states Fe =
3 and Fe = 4. The observed line width was about 6 MHz (FWHM),
somewhat larger than the natural line width. Stray magnetic fields were
reduced by a double-cylindrical µ-metal shielding to values below 2 µT
along the laser beam axis. We found the line center of the resonances by
180 Chapter 7
fitting a Lorentzian with a linear-frequency-dependent background to it as

saturation signal [arb. units]
-30 -20 -10 0 10 20 30

detuning from 335 122 537.894 MHz [MHz]
Figure 7-2. The Fg = 4 →

Fe = 4 component of the cesium D1 transition obtained from a
saturation spectrometer and a Lorentzian fit.
The cesium D1 frequency can be used to derive a value for the fine
structure constant α by taking advantage of the extremely well-known
Rydberg constant, R∞ = α 2 cme / 2h . Steven Chu and collaborators at
Stanford University have measured the photon recoil shift of the D1 line
f rec = f D21 h / 2mCs c 2 with an atom interferometer aiming for an accuracy
near the parts-per-billion (ppb) level. Together with the very precise values
for the proton–electron mass ratio mp /me [9] and the cesium–proton mass
ratio mCs /mp [10], the fine structure constant is derived according to
2 R∞ h 2 f c mCs m p
α2 = = 2 R∞ rec . (3)
c me f D21 m p me
The atomic mass ratios are measured with high precision in terms of the
ratios of their cyclotron frequencies in Penning traps. Please note that all
necessary ingredients to Equation (3), including the Rydberg constant that is
determined by spectroscopy on atomic hydrogen, are based on intrinsically
accurate frequency measurements. Because α scales all electromagnetic
interactions, it can be determined by a variety of independent physical
methods. Unfortunately, different values measured with comparable
accuracy disagree with each other by up to 3.5 standard deviations [11], and
the derivation of the currently most accurate value of α from the electron g -
2 experiment relies on extensive QED calculations [12]. The evaluation via

Equation (3) provided the first independent confirmation of the g - 2 value at
the ppb level of accuracy [11, 13].
3. OPTICAL SYNTHESIZERS
The availability of octave-spanning frequency combs allows the

construction of a simple optical synthesizer. This synthesizer is capable of
converting a given radio frequency phase coherently into the optical domain
by virtue of Equation (1). To do so, one has to measure and/or phase-lock
both ωr and ωCE to a radio frequency reference. While ωr may be measured
with a photo detector anywhere in the output beam of the fs laser, the carrier-
envelope offset frequency ωCE is not determined that easily unless one has an
octave-spanning frequency comb.
I(1)
n1r+1CE 2n1r+1CE 1
x2
2(n1r+1CE)
beat frequency = 1CE
Figure 7-3. The offset frequency ωCE that displaces the modes of an octave-spanning
frequency comb from being exact harmonics of the repetition rate ωr is measured by
frequency-doubling some modes at the “red” side of the comb and beating them with modes
at the “blue” side.
In that case, a group of modes from the low-frequency end of the comb
around nLω r + ω CE is frequency doubled in a nonlinear crystal and beat
notes with another group of modes around nH at the high-frequency end of
the comb are created. As sketched in Figure 7-3, this technique creates
signals at the difference frequency of two optical waves on a photodetector:
2(n Lω r + ω CE ) − ( n H ω r + ω CE ) = ( 2nL − nH )ω r + ω CE . These beat
frequencies are created for all possible combinations of nL and nH. However,
the detected beat notes cannot exceed the bandwidth of the photodetector.
One can further restrict the number of signals to precisely one by using an rf-
182 Chapter 7
low-pass filter. If this filter has a cut-off frequency of ωr/2, only the beat
note at ωCE, belonging to all combinations with 2n L − nH = 0 , remains. The
octave-spanning comb ensures that these combinations exist, i.e., that the
modes nL and nH with nH = 2nL are simultaneously active modes. Knowing
both ωCE and ωr means that frequencies of all modes according to Equation
(1) are known and may be used for optical frequency measurements. This is
accomplished by creating another beat note ωb between an unknown laser at
ωl and a nearby mode of the comb. The frequency of that laser can then be
determined by ω l = nω r + ω CE + ω b . The mode number n may be
determined by a coarse measurement of ωl with a wave meter or by
repeating the measurement with slight variation in ωr.
This technique is called self-referencing [5, 14] and requires a frequency
comb spanning a full optical octave in its simplest form. Alternatively, any
combination of two different harmonics from the same comb will serve that
purpose. Self-referencing has been performed with the beating of the 2nd
and 3rd harmonic of comb components [15] and by comparing modes with
frequency ratio of 7:8 [16]. The frequency comb, even without the
knowledge of ωCE, can be thought of as a ruler in frequency space that
allows the precise measurement of large optical frequency differences. The
initial utilization of frequency combs was for that purpose [8].
To stabilize the frequency comb to a precise rf reference, it is not
sufficient to measure the two comb parameters ωr and ωCE; they also have to
be controlled separately. The repetition frequency is most conveniently
controlled via the cavity length through a piezoelectric transducer. For a
typical laser, the intrinsic stability of the repetition rate is high enough that
the attainable bandwidth of the transducer is sufficient to keep it in phase
with a good rf reference. The other comb parameter ωCE is controlled by
changing the intensity of the stored pulse. It was shown that for a soliton-like
laser such as the Ti:sapphire Kerr-lens mode-locked laser, the group and
phase velocity depend in a different way on the peak intensity of the stored
pulse [17]. Therefore one can adjust the pulse-to-pulse slippage of the
carrier-envelope phase, and thereby ωCE, by controlling the average laser
power. For a Ti:sapphire Kerr-lens mode-locked laser, this can be done by
controlling the power of the pump laser. In this way the comb offset, just
like the repetition frequency, can be adjusted to stay in phase with the rf
reference. The two controls are not completely independent, but they affect
the round-trip group delay and the round-trip phase delay differently.
Keeping the servo bandwidths for ωr as low as possible and for ωCE as high
as possible usually decouples the two servo systems. Having two two-phase
servo systems operational that lock ωr and for ωCE to a radio frequency
reference such as an atomic clock ensures that, by virtue of Equation (1) and
the large integer n, the radio frequency reference is up-converted phase

coherently to the optical region.
4. OCTAVE-WIDE FREQUENCY COMBS
Only now octave-spanning fs Ti:sapphire lasers are becoming possible.

Therefore, external spectral broadening has been key to the simple ν-to-2ν
self-referencing scheme described in the previous section. That spectral
broadening employs self-phase modulation, whose origin is a slight intensity
dependence of the refractive index of most materials. After propagating a
length l, the intensity-dependent refractive index n(t ) = n 0 + n 2 I (t ) leads to
a self-induced phase shift of the field in Equation (2) of
Φ NL (t ) = − n2 I (t )ω c l / c , with I (t ) = A(t ) .
2
(4)
This time-dependent phase shift leads to a frequency modulation that is

proportional to the time derivative of the self-induced phase shift ∂ tΦ NL (t ).
For fused silica with its positive Kerr coefficient n2 = 3.2 × 10 −16 cm 2 /W ,
the leading edges of the pulses create extra frequencies shifted to the red
(∂ tΦ NL (t ) < 0) , while the trailing edges cause blue-shifted frequencies to
emerge. Self-phase modulation modifies the envelope function according to
A(t ) → A(t )e iΦ NL (t ) . (5)
Because Φ NL (t ) has the same periodicity as A(t), the comb structure of the
spectrum, as previously derived, is not affected.
In an optical fiber, this process can be quite efficient as compared to bulk
material, even though the nonlinear coefficient in fused silica is
comparatively small. This efficiency arises because of the fiber core carrying
a high-intensity pulse over an extended length. The details of fiber actions on
the pulse train are, in fact, more complicated because other nonlinear effects
such as Raman scattering may contribute. Whatever the fiber does to the
pulses, however, a periodic input should produce a periodic output of some
kind if the fiber action is the same for all the pulses. Therefore the comb
structure, as derived from the periodic envelope function, is maintained.
Effective self-phase modulation in a regular single-mode fiber takes
place only over a limited fiber length. Dispersion causes the pulse to broaden
temporally. This broadening continuously lowers the peak intensity as the
pulse travels along the fiber until the nonlinear interaction comes to a halt.
184 Chapter 7
Very efficient spectral broadening can be observed in photonic crystal fibers

(PCFs) that can suppress this effect to some extent [18]. A PCF uses an array
of submicron-sized air holes that surround the fiber core and run the length
of a silica fiber to change its effective dispersion. This can be used to
maintain high peak power over an extended propagation length and to
significantly increase the spectral broadening. With these fibers, it has
become possible to broaden low-power, high-repetition-rate lasers to beyond
one optical octave.
Effective spectral broadening of femtosecond light pulses was performed
in very much the same way a long time before the invention of PCFs.
Formerly, spectral broadening was accomplished with very powerful laser
pulses obtained with laser post-amplifiers. However, so far this technique
only works by reducing the pulse repetition rate, thereby concentrating the
available power into fewer pulses. This reduction is disastrous for frequency
metrology because it also reduces the power that is left per mode. It would
also be very difficult to separate single modes out of such a dense comb. The
PCFs allow spectral broadening of fs pulses to more than an optical octave at
repetition rates of 1 GHz.
Meanwhile, some lasers are reaching an octave-spanning spectrum even
without using external self-phase modulation. Some designs use intracavity
self-phase modulation [19] while others employ specially designed cavities
[20] or mirrors [21]. These lasers avoid mechanical stability problems
associated with the tiny core of PCFs. However, their power is generally
weak at the spectral edges that are used for self-referencing. So far only one
laser system had been phase stabilized in this way [20].
5. APPLICATION TO HYDROGEN
In our lab, high-resolution spectroscopy on atomic hydrogen was the

incentive to develop wide-span optical frequency combs. Initially, the task
was to bridge a gap of about 2 THz in our previous harmonic frequency
chain to determine the sharp hydrogen 1S–2S transition frequency relative to
a very precise methane-stabilized He-Ne laser. This laser operates at 3.39
µm, which is close to the 28th subharmonic of the 1S–2S transition at 121
nm. The accuracy-limiting factor at that time was the frequency mismatch
between the 28th harmonic of the He-Ne laser standard and the hydrogen
transition. For the first measurement of this kind, 6,372 longitudinal modes
from a precisely calibrated cavity were used to determine the frequency gap
[22]. This method introduced an uncertainty of 45 kHz.
In the following years, phase-coherent methods were developed to bridge
optical frequency gaps in the THz range. The first method applied in our
laboratory was the use of optical-frequency interval dividers [23, 24]. The
concept was to phase lock the second harmonic of a laser 2f to the sum
frequency generated by two additional lasers at f1 + f2. In doing so, the laser
frequency f is set at the precise midpoint between f1 and f2: f = (f1 + f2)/2. In
this way, an arbitrary large frequency gap f2 - f1 can be divided in two. These
dividers can be cascaded such that n stages would divide an optical
frequency gap by 2n. With a 5-stage interval-divider chain, we have divided
the aforementioned gap to become accessible to radio frequency counting
techniques. At the time, the accuracy of the experiment was limited by the
reproducibility of our He-Ne laser standard to about 840 Hz [6].
Already then it was clear that, given a way to measure frequency gaps
that are comparable in size to the optical frequency itself, absolute optical
frequency measurements could be performed using the ν-to-2ν method
described above [23]. The first electro-optic frequency comb was brought to
our lab by Motonobu Kourogi from the Tokyo Institute of Technology in
1997. This device was capable of measuring frequency gaps of several THz
using a single laser. It replaced the much-more-complicated interval divider
stages but was itself soon replaced by a femtosecond frequency comb. With
the frequency comb, it was possible for the first time to measure frequency
gaps comparable to the optical carrier frequency and thereby convert the 1S–
2S transition frequency phase coherently to the radio frequency domain [16].
Given a precise radio frequency reference, the accuracy was limited by the
hydrogen spectrometer.
6. THE FIRST OPTICAL SYNTHESIZER
In 1999 [25], the (1S , F = 1, m F = ±1) → (2 S , F ′ = 1, m′F = ±1) two-

photon transition frequency in atomic hydrogen was compared phase
coherently to the frequency of the ground-state hyperfine splitting in 133Cs.
For this purpose, one of the most precise Cs atomic fountain clocks (FOM)
[26], built in the group of A. Clairon at the Bureau National de Métrologie –
Systèmes de Référence Temps Espace (BNM-SYRTE) in Paris, was brought
to our laboratory. A photonic crystal fiber that would have allowed us to
produce an octave-spanning frequency comb was not at our disposal at that
time. Therefore, we set up a self-referencing scheme that compared modes
with a frequency ratio of 7:8, as shown in Figure 7-4. The frequency comb
was produced by a Mira 900 (Coherent Inc. with about 70 fs pulse duration)
with some spectral broadening in a regular single-mode fiber. The total span
of the frequency comb was in excess of 1/14 of the dye laser’s frequency, or
44 THz.
186 Chapter 7
In 2003 [27] we repeated this measurement with a much simpler ν-to-2ν

self-referencing scheme as described above (see Figure 7-3). This
modification did not improve the accuracy of the frequency measurement of
the dye laser, which was in both cases determined by the Cs fountain clock,
but vastly simplified the experimental setup.
hydrogen 1S-2S hydrogen

interval: 28f-8∆f spectrometer
14f-4∆f
x2
3.39 µm
NeNe laser
486 nm
f 7f-2∆f dye laser
4f-∆f
x2
x4
848 nm 851 nm
973 nm
counter #1 counter #2 laser diode
I(λ)
0.5f 44.2THz
∆f 1THz
4f 4f-∆f single mode 3.5f-∆f

fiber λ
Cs
atomic mode locked
clock laser
Figure 7-4. Sketch of the first (still somewhat complicated) self-referenced frequency comb.
While the second harmonic of the dye laser at 486 nm is used to excite the 1S–2S transition,
its subharmonic 3.5f-∆f and the 4th harmonic of the He-Ne laser are compared to the
frequency comb. The ratio of these frequencies is 7:8 instead of the 1:2 in a much simpler ν-
to-2ν self-referencing scheme. Because the 3.39 µm He-Ne laser could not be tuned precisely
to the 28th subharmonic of the dye laser, a second, much smaller frequency gap ∆f was present
in this setup and was measured simultaneously with the same frequency comb. The oval
symbol represents an optical interval divider that fixes the center frequency 4f – ∆f relative to
its inputs 4f and 7f - 2∆f. All optical connections represent phase-locked lasers except where
the rf counters are located. These are used to measure beat notes with the nearest comb
modes. Once the two frequency gaps 0.5f and ∆f have been precisely measured with a Cs
clock-referenced frequency comb, the dye laser at 14f + 8∆f is readily calculated. In addition,
we can calculate the frequency of any other mode in the frequency comb and fix the carrier-
envelope phase by locking the comb with ωCE = 0.
7. THE HYDROGEN SPECTROMETER
The cw dye laser near 486 nm that is used to excite the hydrogen atoms
can be locked to a high-finesse reference cavity to narrow its line width and
fix its frequency. A new cavity made from Ultra-Low-Expansion (ULE)
glass was used for the 2003 measurement. Its drift was measured to be less
than 1 Hz s-1 for the entire time of the measurement. Due to better thermal
and acoustic isolation and improvements in the laser-locking electronics, the
laser line width was narrower than it had been in 1999. An upper limit for
the laser line width was deduced from an investigation of the beat signal
between two laser fields locked separately to independent Zerodur and ULE
cavities to about 120 Hz at a laser wavelength of 486 nm.
dye laser
486 nm
BBO
stabilization servo frequency doubler
AOM
253 nm
vacuum chamber
reference resonator
nozzle
frequency chopper
measurement Lα-detector rf dis-
charge
fork H
chopper
H2
Figure 7-5. Hydrogen spectrometer. The 486 nm light from the dye laser is frequency doubled
in a Barium β-Borate (BBO) crystal. The resulting radiation is used to drive the 1S–2S two-
photon transition that is free of the first-order Doppler effect if the two photons are counter-
propagating. The latter is ensured by a linear enhancement cavity in the vacuum chamber that
is held at a pressure of about 10-8–10-7 mbar. The acousto-optic modulator (AOM) allows the
dye laser to be tuned away from the resonance of the reference resonator to scan the hydrogen
line.
As shown in Figure 7-5, molecular hydrogen is dissociated in a gas

discharge and forms a cold atomic beam after contact with a copper nozzle
that is kept at 5–7 K. The region of the atomic beam is shielded from stray
electric fields by a Faraday cage. Some of the atoms are excited from the
ground state to the metastable 2S state by the absorption of two counter-
propagating photons from the laser field in the enhancement cavity. In this
188 Chapter 7
arrangement, the first-order Doppler effect of the two photons exactly

cancels each other. Those atoms that are flying close to the cavity axis can
enter the detection region where the 2S atoms are quenched in a small
electric field and forced to emit a Lα-photon, which is detected with a solar-
blind photo multiplier. We have periodically blocked the laser beam and the
atomic beam. This procedure eliminates background counts from the
exciting laser. The delay ∆t between blocking the 243 nm radiation and the
start of photon counting sets an upper limit on the velocity of the atoms that
contribute to the signal. Therefore, velocity-dependent systematic effects,
such as the second-order Doppler shift and the time-of-flight broadening, are
smaller for spectra recorded at larger ∆t. The hydrogen beam is blocked by a
fork chopper in less than 200 µs after the blocking of the excitation light to
prevent slow atoms from being blown away by fast atoms that emerge
subsequently from the nozzle. With the help of a multi-channel scaler, we
count all photons and sort them into 12 equidistant delay-time bins. From
each scan of the laser frequency over the hydrogen 1S–2S resonance, we get
12 distinct spectra, as shown for a sample line in the left panel of Figure 7-6.
To correct for the second-order Doppler shift, we use an elaborated theory
model [28] to fit all the delayed spectra simultaneously with one set of 7 fit
parameters. Besides the second-order Doppler effect, the other dominating
systematic effect is the dynamic AC Stark shift that scales linearly with the
excitation light intensity. We have compensated for that effect by varying
the laser intensity and extrapolating the transition frequency to zero
intensity, as shown in the right portion of Figure 7-6. The result of the fitting
procedure is the 1S–2S transition frequency for the hydrogen atom at rest.
16
2S signal [103 counts per second]
0.3 delay 0 µs 5.6

νH - 2 466 061 102 470 [kHz]
14 delay 1000 µs
0.2 delay 1200 µs

12
0.1
delay 1400 µs
delay 100 µs 5.4
delay 1800 µs
10 delay 1600 µs
0
8 -10 0 10 5.2
delay 200 µs
6
5.0
4 delay 1000 µs delay 400 µs
2 4.8
0
-40 -30 -20 -10 0 10 20 0 1 2 3 4 5 6 7 8
detuning @ 121nm [kHz] power @ 243 [a.u.]
Figure 7-6. Left: simultaneous fit of a 1S–2S transition spectrum recorded at different
detection delays. The nozzle temperature was 7 K. Right: AC Stark shift extrapolation to zero
power for all the lines recorded during one day of data taking.
To check for possible cycle slipping, all phase-locked frequencies were

counted to verify consistency with the phase-lock systems. Data points were
accepted only where all counter readings were consistent. Additionally, it

was verified that the dye laser was continuously locked to the reference
resonator during the measurement. To exploit the superior short-term
stability of that resonator, we measured the frequency of the standing light
field in the resonator first. To compensate for the slow drifts of the resonator,
we fit a second-order polynomial to that drift and then, knowing the AOM
detuning, we used the polynomial in the subsequent determination of the dye
laser frequency. The left part of Figure 7-7 shows the absolute frequency of
the standing light field inside our new ULE reference resonator. The right
hand side of the figure shows the normalized Allan variance of the dye laser
locked to the reference resonator, measured relative to the Cs fountain clock
FOM. For longer averaging times, the plot of the Allan variance is generated
by juxtaposing 1 s counter readouts. Whereas it is known that such a
procedure can alter the functional dependence of the Allan variance [29],
white frequency noise, as produced by the Cs fountain, is immune to this
form of bias. The observed τ -1/2 dependence coincides with the
independently measured fountain clock instability for averaging times
shorter than ≈ 10 s. The short-term stability of the laser system is better than
the stability of the fountain clock. However, the long-term stability is limited
by the drift of the ULE reference resonator. For the given stability of the Cs
fountain clock and the cavity, the optimum record length is around 500 s.
For longer averaging times, the Cs fountain is more stable than the drift-
corrected ULE cavity.
600
fcavity - 616 516 065 769 800 Hz [Hz]
normalized Allan variance
400
10-13
200
∼τ−1/2
0
-200
10-14
-400
-600
0 1000 2000 3000 4000 5000 1 10 100 1000
time [sec] gate time [sec]
Figure 7-7. Left: Absolute frequency of the standing light field inside the reference resonator
(AOM shifted dye laser, see Figure 7-5) as a function of time. The solid line is a parabolic fit
to the data. The linear contribution to the drift is less than 0.1 Hz per s in this example and
below 1 Hz per s at all times. Right: Normalized Allan variance vs averaging time computed
from a time series of 1 s counter readings with considerable dead time. The straight line
indicates the τ -1/2 dependence, which is the signature of the Cs fountain clock. The raw data
analysis (solid squares) shows that the stability for averaging times longer than 20 s is limited
190 Chapter 7
by the drift of the ULE reference cavity. Open squares represent data corrected for the
parabolic cavity drift.
8. THE 1S–2S TRANSITION FREQUENCY
We have measured the 1S–2S transition in atomic hydrogen during 10

days in 1999 and during 12 days in 2003. Both data sets have been analyzed
using the same theoretical line-shape model and are therefore comparable. In
Figure 7-8, the results of the extrapolation to zero-excitation light intensity
and the respective statistical error bars for each day are presented. Since
1999, the statistical uncertainty for each day of measurement has been
significantly reduced because of the narrower laser line width and better
signal-to-noise ratio; however, the scatter of the daily averages did not
reduce accordingly. We have tested several possible reasons for this
additional scatter, including an intrabeam pressure shift, a background gas-
pressure shift, Stark effects due to the rf gas discharge, and the DC Stark
shift. We have been able to exclude all of these effects, at least on a
conservative level of 10–20 Hz. A possible origin of the observed scatter can
be due to a residual first-order Doppler effect arising from a violation of the
axial symmetry of the enhancement cavity mode and the hydrogen atomic
beam. The scattering of the excitation light on intracavity diaphragms could
also cause slight changes of the field distribution and the corresponding first-
order Doppler effect. However, such changes should average to zero over
multiple adjustments of the hydrogen spectrometer because the shifts can
have both signs. As the scatter is the same for both the measurement sets, we
believe them to be equivalent. The main statistical and systematic
uncertainties of these measurements are collected in Table 7-1. The
averaging of the 1999 and 2003 daily data points was performed without
weighting them. 2 For both measurements, the dominating uncertainty arises
from the day-to-day scatter while the pure statistical uncertainty for each day
is significantly smaller. In fact, weighting of the daily data only slightly
influences the results (on the level of σ/2).
Comparing both measurements, we deduce a difference of
ν H , 2003 − ν H ,1999 equal to (-29 ± 57) Hz within 44 months. This difference
corresponds to a relative drift of νH against the 133Cs ground-state hyperfine
splitting of ∂ t (ln(ν H /ν Cs )) = ( −3.2 ± 6.3) × 10 −15 per year.
2
The result of 2 466 061 102 474 870 Hz was inadvertently described in [25] as “the
weighted mean value” but was calculated without consideration of the daily statistical
uncertainties.
Table 7-1. Results of the (1S, F = 1, mF = ±1 2S, F' = 1, m'F = ±1) transition frequency→
measurement (νH,1999, νH,2003) and uncertainty budgets (σH,1999, σH,2003) for the 1999 and 2003
measurements, respectively.
Contribution νH,1999 σH,1999 νH,2003 σH,2003

[Hz] [Hz] [Hz] [Hz]
Extrapolated value – 2 466 061 102 474 kHz 870 36 851 25
Background gas-pressure shift 10 10 0 2
Intrabeam pressure shift 0 10 0 10
Line-shape model 0 20 0 20
DC Stark shift 0 5 0 5
Blackbody radiation 0 1 0 1
Standing wave effects 0 10 0 1
Intensity zero uncertainty 0 1 0 0
Fountain clock uncertainty 0 5 0 5
Total – 2 466 061 102 474 kHz 880 45 851 34
5200 44 months
f(1S2S) - 2 466 061 102 470 000 Hz [Hz]
5100
5000
4870(36) Hz
4851(25) Hz
average
4900
average
4800
4700
4600
June 1999 - July 1999 February 2003
4500
→
Figure 7-8. Experimental results and averages for the 1999 and 2003 measurements of the
(1S, F = 1, mF = ±1 2S, F' = 1, m'F = ± 1) transition frequency in atomic hydrogen. Each
data point represents one day of averaging. The two results are unweighted mean values of the
daily averages.
192 Chapter 7
9. CHECKING FOR SLOW DRIFTS OF A NATURAL

CONSTANT
In the era of rapid development of precision experimental methods, the

stability of fundamental constants becomes a question of basic interest. Any
drift of nongravitational constants is forbidden in all metric theories of
gravity, including general relativity. The basis of these theories is Einstein's
Equivalence Principle (EEP), which states that weight is proportional to
mass and that in any local freely falling reference frame, the result of any
nongravitational experiment must be independent of time and space. This
hypothesis can be proven only experimentally because no theory exists that
predicts the values of fundamental constants. In contrast to metric theories,
string theory models, which aim to unify quantum mechanics and
gravitation, allow for, or even predict, violations of EEP. Limits on the
variation of fundamental constants might therefore provide important
constraints on these new theoretical models.
A recent analysis of quasar absorption spectra with red-shifted UV
transition lines indicates a variation of the fine structure constant on the level
of ∆α / α = ( −0.54 ± 0.12) × 10 −5 for a red-shift range (0.2 < z < 3.7) [30]. On
geological timescales, a limit for the drift of α has been deduced from
isotope abundance ratios in the natural fission reactor of Oklo, Gabon, which
operated about 2 Gyr ago. Modeling the processes that have changed the
isotope ratios of heavy elements gives a limit of
−8
∆α / α = ( −0.36 ± 1.44) × 10 [31]. In these measurements, the high
sensitivity to the time variation of α is achieved through very long
observation times at moderate resolution for ∆α. Therefore, they are
vulnerable to systematic effects. In fact, more recent observation of quasar
absorption spectra, performed by different groups, seem to rule out a
variation of α on the level observed by Webb et al. [32].
In contrast to astronomical or geological observations, laboratory
experiments can reach an accuracy of 10-15 with better controlled
systematics. For this reason, laboratory measurements can compete, even
though the temporal separation of observations is necessarily smaller by
many orders of magnitude. In general, frequency measurements must be
performed with any form of unit and therefore always yield the ratio of two
frequencies. If one compares with the ground-state hyperfine splitting of
133
Cs, for example, the measurement would be in units of Hz. However, for
the purpose of testing the stability of natural constants, any other frequency
ratio may be used. In fact, as will be shown below, measuring the ratio of
two optical frequencies could be advantageous. The frequency comb
provides a convenient way to measure optical frequency ratios. This
measurement is accomplished by locking one mode of the frequency comb

to one optical transition, referencing the carrier envelope offset frequency to
the repetition rate and observing a beat note between the comb and the other
optical frequency [33].
To map any possible variation of the measured frequency ratio on the
variation of natural constants, the following expression for electronic
(optical) transitions is commonly used:
ν opt = A R y Frel (α ) , (6)
where Ry is the Rydberg frequency expressed in hertz and Frel(α) takes into
account relativistic and many-body effects. The constant A is a pure
numerical factor and does not depend on any natural constant. For hydrogen,
one would use the Schrödinger energies with A = 1/n2 and Frel(α) = 1 to
describe the first-order scaling in α, which is sufficient here. The
corresponding expression for hyperfine transitions reads:
µn
ν hf = A′ R y α 2 ′ (α ) ,
Frel (7)
µB
with a different constant A′ and a different relativistic correction F’rel(α).

From these two equations, it is clear that the Rydberg constant cancels in any
frequency measurement. The last expression depends on the Bohr magneton
µB and the nuclear magnetic moment µn. The size of the latter depends
mainly on the strong coupling constant that may also be subject to a slow
drift. This is the reason why optical-to-optical comparisons are desirable.
However, having more than one optical-to-hyperfine comparison at hand, the
drift of the different constants may be separated [27].
Even though the frequency measurements in use for such an evaluation
are rather accurate, the derived drift is not. Therefore only low-order
numerical values for the derivatives of the relativistic corrections are used in
the evaluation. For the cesium ground state splitting, one finds [34]
α∂α ln FCs (α ) ≈ +0.8 . (8)
Whereas for the hydrogen optical transitions, the Schrödinger approximation

with
α∂α ln FH (α ) ≈ 0 (9)
194 Chapter 7
is sufficient. The relative drift of the hydrogen/cesium frequency ratio is

therefore approximated by
∂ t ln(ν H /ν Cs ) = −∂ t ln( µ Cs / µ B ) − 2.8 ∂ t ln(α ) . (10)
For clarity, we set x ≡ ∂ t ln(α ), and y ≡ ∂ t ln( µ Cs / µ B ) and get for an

experimental constraint with the result of the previous section:
y + 2.8 x = (3.2 ± 6.3) × 10 −15 yr −1 for H. (11)
To separate the relative drift rates of x and y, other transition frequency

ratios have to be observed over time. In fact, in similar experiments
performed at the National Institute of Standards and Technology (NIST) in
Boulder, Colorado (USA) and the Physikalisch-Technische Bundesanstalt
(PTB) in Braunschweig, Germany, other optical transition frequencies in a
single trapped Hg+ ion [35] and a single trapped Yb+ ion [36] have been
compared to the cesium ground-state splitting using a self-referenced
frequency comb. These measurements lead to the following experimental
constraints:
y + 6.0 x = (0.2 ± 7.0) × 10 −15 yr −1 for Hg+ , (12)
y + 1.9 x = (1.2 ± 4.4) × 10 −15 yr −1 for Yb+ . (13)
Since Equations (11) through (13) represent an over-determined set for

the two variables x and y, a least-square adjustment procedure yields the
most likely values and their uncertainties [36]:
α&
x = ∂ t ln(α ) = = ( −0.3 ± 2.0) × 10 −15 yr −1 , and (14)
α
y = ∂ t ln( µ Cs / µ B ) = (+2.4 ± 6.8) × 10 −15 yr −1 . (15)
Figure 7-9 shows the current limit for the drift rates from laboratory
measurements. The uncertainties of these limits reach within a factor of ten
to the uncertainties of the relative drift rates derived from astronomical
observations [30, 32]. In the near future, one can expect a significant
improvement of the laboratory measurements, even without an increase of

experimental accuracy, just because the available time interval increases.
One can also expect that additional optical transition frequencies may be
measured that will further restrict the extent of the ellipse in Figure 7-9. It
should also be pointed out that, since there is no accepted physical model
that would describe the drift of natural constants,3 there is no reason to
assume it to be linear in time. In this sense, astronomical and laboratory
measurements are somewhat complementary because they observe on
different time scales.
ln(µCs/µB) [10-14 yr-1]
4
Hg+
2 H
-2 Yb+
y=
-4
-6 -4 -2 0 2 4 6
x= ln(α) [10-15 yr-1]
Figure 7-9. Combining the drift rates of optical transitions in hydrogen, Hg+ and Yb+. The
experimentally allowed region for the drift rates of α and µCs/µB is shown in the form of an
ellipse.
It has also been argued that a possible drift of the coupling constants
must be correlated [37], at least if one believes in the existence of a Grand
Unified Theory. In such a theory, the fundamental interactions are described
within the same framework. The drift rates must therefore be synchronized
in some sense. According to Calmet et al. [37], the fractional time variation
of hadron masses and along the same line the nuclear magnetic moments,
should change about 38 times faster than the fractional time variation of the
fine structure constant α. The analysis presented here is independent of such
3
If there was such a model that describes the drift of natural constants in terms of other
constants, it would simply mean that our current constants are not fundamental.
196 Chapter 7
assumptions as it separates the drift rates of different constants. In that sense,

the conclusions are model independent.
REFERENCES
[1] L. O. Hocker, A. Javan, D. R. Rao, L.Frenkel, and T. Sullivan, Appl. Phys.
Lett. 10, 147-149 (1967); K. M. Evenson, J. S. Wells, F. R. Petersen, B. L.
Danielson, and G. W. Day, Appl. Phys. Lett. 22, 192 (1973).
Lett. 76, 18-21 (1996).
850 (1978).
883 (1999); S. A. Diddams, L. Hollberg, L. S. Ma, and L. Robertsson, Opt.
Lett. 27, 58-60 (2002); L. S. Ma, Z. Y. Bi, A. Bartels, L. Robertsson, M.
Zucco, R. S. Windeler, G. Wilpers, C. Oates, L. Hollberg, and S. A.
Diddams, Science 303, 1843-1845 (2004).
[5] R. Holzwarth, T. Udem, T. W. Hänsch, J. C. Knight, W. J. Wadsworth, and
P. S. J. Russell, Phys. Rev. Lett. 85, 2264-2267 (2000).
W. Hänsch, Phys. Rev. Lett. 79, 2646-2649 (1997).
172, 59-68 (1999).
3568-3571 (1999).
[9] P. J. Mohr, Rev. Modern Phys. 72, 351-495 (2000).
[10] M. P. Bradley, J. V. Porto, S. Rainville, J. K. Thompson, and D. E.
Pritchard, Phys. Rev. Lett. 83, 4510-4513 (1999).
[11] A. Wicht, J. M. Hensley, E. Sarajlic, and S. Chu, Physica Scripta T102, 82-
88 (2002).
[12] T. Kinoshita, Reports on Progress in Physics 59, 1459-1492 (1996); T.
Kinoshita, IEEE Trans. Instrum. Meas. 50, 568-571 (2001).
[13] J. M. Hensley, Ph.D Thesis, Stanford University (2001).
84, 5102-5105 (2000).
1842-1844 (2002).
[16] J. Reichert, M. Niering, R. Holzwarth, M. Weitz, T. Udem, and T. W.
Hänsch, Phys. Rev. Lett. 84, 3232-3235 (2000).
[18] P. Russell, Science 299, 358-362 (2003).

Opt. Lett. 26, 373-375 (2001).
(2003).
[21] T. Fuji, A. Unterhuber, V. S. Yakovlev, G. Tempea, A. Stingl, F. Krausz,
and W. Drexler, Appl. Phys. B 77, 125-128 (2003).
[22] T. Andreae, W. Konig, R. Wynands, D. Leibfried, F. Schmidtkaler, C.
Zimmermann, D. Meschede, and T. W. Hänsch, Phys. Rev. Lett. 69, 1923-
1926 (1992).
[23] T. W. Hansch, in The Hydrogen Atom, edited by M. Inguscio (Springer
Verlag, Berlin, 1989), p. 93.
(1990).
and A. Clairon, Phys. Rev. Lett. 84, 5496-5499 (2000).
[26] M. Abgrall, Ph.D Thesis, L'universite Paris (2003).
[27] M. Fischer, N. Kolachevsky, M. Zimmermann, R. Holzwarth, T. Udem, T.
W. Hänsch, M. Abgrall, J. Grunert, I. Maksimovic, S. Bize, H. Marion, F.
P. Dos Santos, P. Lemonde, G. Santarelli, P. Laurent, A. Clairon, C.
Salomon, M. Haas, U. D. Jentschura, and C. H. Keitel, Phys. Rev. Lett. 92,
230802 (2004).
[28] A. Huber, B. Gross, M. Weitz, and T. W. Hänsch, Phys. Rev. A 59, 1844-
1851 (1999).
[29] P. Lesage, IEEE Trans. Instrum. Meas. 32, 204-207 (1983).
[30] J. K. Webb, M. T. Murphy, V. V. Flambaum, V. A. Dzuba, J. D. Barrow,
C. W. Churchill, J. X. Prochaska, and A. M. Wolfe, Phys. Rev. Lett. 87,
091301 (2001).
[31] Y. Fujii, A. Iwamoto, T. Fukahori, T. Ohnuki, M. Nakagawa, H. Hidaka, Y.
Oura, and P. Moller, Nucl. Phys. B 573, 377-401 (2000).
[32] H. Chand, Astronomy and Astrophysics 417, 853-871 (2004); R. Quast,
Astron. and Astrophys. 415, L7-L11 (2004).
Wineland, Science 293, 825-828 (2001).
[34] V. A. Dzuba, V. V. Flambaum, M. T. Murphy, and J. K. Webb, Phys. Rev.
A 63, art. no.-042509 (2001).
[35] S. Bize, S. A. Diddams, U. Tanaka, C. E. Tanner, W. H. Oskay, R. E.
Drullinger, T. E. Parker, T. P. Heavner, S. R. Jefferts, L. Hollberg, W. M.
Itano, and J. C. Bergquist, Phys. Rev. Lett. 90, 150802 (2003).
[36] E. Peik, Phys. Rev. Lett., submitted (2004).
[37] X. Calmet and H. Fritzsch, Euro. Phys. J. C 24, 639-642 (2002).
Chapter 8
OPTICAL FREQUENCY MEASUREMENT
USING FREQUENCY MULTIPLICATION AND
FREQUENCY COMBS
Harald Schnatz, Jörn Stenger, Burghard Lipphardt, Nils Haverkamp, and

Carl-Otto Weiss
Physikalisch-Technische Bundesanstalt
Abstract: We review the development of phase-coherent optical frequency measurement

from its start with the harmonic frequency chain to the present use of optical
comb generators. We demonstrate that the frequency fluctuations of a
femtosecond comb laser—used in optical frequency measurement—can be
eliminated completely from the measurement process, allowing ultimate
frequency measurement performance. This principle was applied to the
frequency measurements of the Ca cloud standard and the Yb+ single-ion
standard. The latter measurements confirm the suitability of these optical
standards as future atomic clocks.
Key words: frequency comb, frequency chains, optical atomic clocks, optical frequency
metrology, optical frequency references
1. INTRODUCTION
In the past, the measurement of optical frequencies with reference to

standard rf frequencies was done analogously to microwave techniques.
Essentially, higher frequencies were measured by harmonic mixing with
lower frequencies, progressing from the microwave range through the
infrared (IR) to the visible spectral range. The oscillators used were solid
state and electron tube microwave sources up to 500 GHz and lasers for
higher frequencies. For nonlinear elements, diodes with progressively
smaller RC-time constants for increasing frequency were used, directly
scaled from the microwave range.
8. OPTICAL FREQUENCY MEASUREMENT USING FREQUENCY 199
MULTIPLICATION AND FREQUENCY COMBS
Commonly used devices included micron-size semiconductor Schottky

diodes (up to about 5 THz) [1] and metal-insulator-metal point contacts that
work up to about 200 THz [2]. Above 100 THz, second-harmonic generation
in bulk crystals was used. In this way it was possible to phase-coherently
relate the 657 nm cold Ca atom optical frequency standard with a standard
frequency generated by a cesium clock [3].
It may be worth remembering at this point in time, optical frequency
measurements were only a rudiment in a much more ambitious program. The
idea of scaling microwave techniques to the optical range arose from the
development of the laser out of the maser. Consequently, several groups
started to work on a program to develop “optical electronics” in the 1960s
(prominent among these groups was Javan’s group at MIT). This group
envisioned scaling electronic elements in such a way that they would be
capable of handling frequencies as high as visible optical radiation
frequencies; thus they envisioned the generation of coherent optical radiation
with transistor oscillators rather than with atom amplifiers using population
inversion. This project would have entailed, of course, guiding light by wires
or metal cables, which is possible over distances of small multiples of the
wavelength. Techniques of this kind resurfaced in the 1980s and 1990s in
near-field microscopy research.
The first steps in this program consisted of scaling diodes to high
frequencies because a transistor was basically composed of two diodes. This
research led to the successful implementation of optical frequency
measurements where diodes served as the necessary, fast harmonic mixers.
Somehow, probably also under the impression of the successes of laser
physics and technology, this ambitious program of “optical electronics” was
forgotten. No serious efforts were ever made to realize diodes of the
necessary small response time and negative resistance (not just a negative
differential resistance as is sufficient for harmonic mixing), as necessary for
amplification. Only half-hearted efforts were made (using metals with a
higher plasma frequency) to increase the frequency response of point contact
diodes beyond 200 THz, a limit imposed by the plasma frequency of
commonly used metals [4].
Nevertheless, the desirability of a technique of “optical electronics” is
still the same as it was in the 1960s; indeed, one might expect this program
to be taken up again in the future, reinvigorated by an urgent need for ever-
increasing speeds in electronics.
What was accomplished during this epoch, however, was the
demonstration that electronic techniques can – if not amplify optical
radiation – multiply and mix frequencies of radiation up to the visible
spectral range.
200 Chapter 8
2. FREQUENCY MEASUREMENTS BY REPEATED

HARMONIC MIXING (FREQUENCY CHAINS)
Most lasers in the IR that can be used for a “frequency multiplication

chain” are gas lasers. They are tunable only within approximately a Doppler-
gain linewidth, i.e., 10-6 of their carrier frequency. For all practical purposes,
this means a fixed frequency. Therefore the frequency multiplication chains
were essentially designed (and only useful) for measuring one particular
optical standard.
A certain flexibility was provided by the fact that, at around 30 THz, CO2
lasers were used that had more than 100 different gain lines with gaps
between 20 and 100 GHz. The chains aimed first at one particular CO2-laser
line. The continuation to higher frequencies was made possible by other
CO2-laser lines (which provided a certain frequency flexibility). The
frequency gap between the two CO2-laser lines used was bridged by using
harmonics of millimeter wave sources.
In the Physicalisch-Technische Bundesanstalt (PTB), the frequency
multiplication chain (Figure 8-1) was designed for the 657 nm Ca and the
3.39 µm CH4 transitions. The microwave/far-IR part aimed at the 12CO2
R(32) line to reach the CH4 transition and used the 12CO2 P(14) line for the
continuation to Ca frequency standard. The gap of ~1 THz between these
lines was bridged by the 8th harmonic of a 126 GHz oscillator. For details of
this optical frequency measurement chain see [5].
Figure 8-2 shows the stability of frequency measurements for the CH4
line at 3.39 µm and the Ca line at 657 nm. Both measurements are limited by
the available short-term stability of the low-frequency reference (an H
maser), which is also given.
The conclusion drawn from these results is that the combined instability
of the optical references and the frequency measurement is smaller than that
of the low-frequency reference. This effect occurred in spite of the fact that
cycle slips in the phase locks, particularly at the 400 GHz step, could not be
completely avoided. This result shows that optical atomic clocks are
feasible. For the results of repeated measurements of the Ca line, see Section
5.1.
Optical frequency chain measurements as described need essentially one
special set-up for an optical frequency range of only a few tens of GHz. The
arrangements are thus specialized, complex, occupy a large amount of space,
and consist of costly and difficult-to-operate lasers.
Ca Frequency Standard
PLL @ 60 MHz
455986.240494 GHz
2x
Diode Laser
227993.090247 GHz
PLL @ 100 MHz

2x
Color Center Laser
113996.495123 GHz
12C18O
2x 2 -Laser P*(20)
Counter
28534.284 GHz
2x
CO2-Laser P(14) 3x DRO
28464.684 GHz 23.171 GHz
PLL @ 10 MHz PLL @ 70 MHz
CO2-Laser P(14) OsO4 - stabilized

28464.674 GHz CO2 Laser
8x
PLL @ 10 MHz Gunn 126.5 GHz
phase control
CO2-Laser R(32)
Gunn
29477.165 GHz 3x 55.4 GHz
4x
PLL @ 100 MHz Klystron 73.35 GHz
13C16O -Laser P*(28) CH4 - stabilized

2
29770.665 GHz HeNe Laser
7x
Counter Gunn 8.99 GHz
Methanol Laser
4251.676 GHz
Counter frequency control
11x
Backward Wave Oscillator
386.5 GHz
PLL @ 80 MHz Cs Frequency Standard
17x
Gunn Oscillator 9.192 631 77 GHz
22.7 GHz
PLL @ 10 MHz
Quartz Oscillator
100 MHz Hydrogen Maser
Figure 8-1. Harmonic frequency chain to the Ca frequency standard. (PLL: phase-locked
loop; ⊗: harmonic mixers.)
202 Chapter 8
-12
10
-13
10
σy( 2, τ )
-14
10
1 10 100 1000 s
integration time
Figure 8-2. Allan standard deviation of Ca (■) and CH4 (□) frequency standards. For
comparison, the typical stability of an H maser is represented by the line without
measurement points.
3. FREQUENCY INTERVAL DIVISION APPROACH
Several approaches to overcome the limitations of optical frequency

measurement chains have been proposed. Among them are frequency
interval bisection [6], optical-parametric oscillators (OPO) [7], optical comb
generators [8, 9], sum- and difference-frequency generation in the near IR
[10], frequency-division-by-three [11], and four-wave mixing in laser diodes
[12]. All of these techniques rely on the principle of difference-frequency
synthesis, in contrast to the harmonic-generation method used in the
frequency chains.
Two techniques were primarily used: (a) frequency-interval division and
(b) passive optical-comb generation. In method (a), the concept is that any
optical frequency interval can be measured by successive frequency interval
division. An interval is halved in the first step, divided by 4 in the second
step, and then by 8, 16, etc. until the interval is small enough to be counted
electronically. If the initial interval is the one between ν and 2ν (i.e., created
by second-harmonic generation), then this process leads to the determination
of the optical frequency ν. This “interval bisection” was first demonstrated
by Telle et al. [6]. For the experiment, one uses a third laser (ν3) at the
arithmetic mean of the frequencies ν1 and ν2 to determine the interval
(Figure 8-3). The frequency ν3 bisects the interval between ν1 and ν2 if 2ν3 =
ν1 + ν2. Complete frequency chains have been planned using this approach
[6]. However, difficulties were encountered in obtaining the lasers in certain
ranges of the visible spectrum.
To measure the frequency of the 1S–2S transition in atomic hydrogen,
Hänsch’s group in Garching built a frequency chain starting from a CH4-
stabilized reference laser at 3.39 µm and reaching the UV at 243 nm. The
technique combined conventional harmonic mixing with optical interval
bisection [13, 14]. The CH4-reference frequency itself had to be calibrated
using the harmonic-mixing frequency chain at PTB and a transportable CH4-
frequency standard.
Laser 1 ν1
Laser 2 ν2
SUM
ν1+ν2 - 2ν3
Detector
Servo
SHG
Laser 3 ν3 = (ν1+ν2) / 2
Figure 8-3. Diagram of an optical difference-frequency divider for frequency-interval

bisection (adapted from Reference [6]).
An alternative method for frequency-interval division is to inject the two

frequencies forming the interval into a laser that emits roughly at the half-
interval frequency. Frequency components are generated via four-wave
mixing symmetrically around the mid-interval frequency and act as “seeds”
for the laser emitting near the mid-interval frequency. The four-wave-mixing
products are all of equal frequency when the laser emits the half-interval
frequency, and consequently a phase synchronization of the laser emission
occurs. Thus the interval is divided by the laser phase-coherently. Due to the
degeneracy of the two four-wave mixing processes occurring, there is phase
bistability for the laser emission. The phase bistability is the basis for the
formation of spatial solitons currently under consideration for optical
information processing [15].
204 Chapter 8
The second method (b) was conceived early [8]. Using an electro-optic
modulator, driven at a microwave frequency (typically 5–10 GHz) in an
optical resonator, a wide spectrum of modulation sidebands is produced. If
the round-trip time of the light in the optical resonator is synchronized with
the microwave period, the phase modulation of the light adds up over a
number of round trips. This means that the phase excursion of the phase
modulation is increased by a factor equal to the optical resonator finesse,
thereby increasing the number and strength of the modulation sidebands. In
this way, Kourogi [16] was able to generate frequency combs of 30 THz
widths. Such a frequency comb generator was used by Huber et al. [17] to
bridge a gap of ~2 THz in the 800 nm range. The width of the comb was
further increased by passing the optical radiation through a nonlinear fiber
that effects self-phase modulation.
During the last decade, a number of measurements of the properties of
hydrogen, such as the 1S–2S transition frequency [18], the isotope shift in
deuterium [19], the 1S Lamb shift [20], and the Rydberg constant [13, 21]
were performed via optical frequency measurements. Incorporating novel
techniques like Kurogi’s optical comb generator [16], the frequency chain in
Garching could be modified to measure the frequency of the In+ clock
transition at 236.5 nm [22]. More details of the various frequency
measurements performed with different concepts at different laboratories can
be found in recent review articles [23]. Although frequency-interval division
had proved its usefulness, no complete chain was ever built using this
principle.
4. OPTICAL FREQUENCY MEASUREMENT USING

FEMTOSECOND LASERS
Apart from sequential-interval bisection, a frequency comb with

precisely spaced comb lines may be used to measure the interval ν to 2ν.
The requirement for this is that the comb width is > (2ν – ν), i.e., larger than
an octave. Initially efforts were directed at expanding the comb width of
passive comb generators [16], but with the demonstration [24] that mode-
locked Ti:sapphire lasers emit excellent spectral combs, the efforts
concentrated on “active” comb generators, i.e., on mode-locked lasers. The
concept foresaw using the comb spectrum emitted directly by the mode-
locked Ti:sapphire laser, which is far less than an octave wide, and to reduce
the octave interval by dividing the interval by 2 or 4, as described above.
During the course of these efforts, however, evidence appeared of a more-
than-an-octave-wide “continua” generated by mode-locked pulses in fibers
with narrow cores [25, 26]. The suggestion was that these continua might
actually be coherent frequency combs if driven by mode-locked lasers,

which by themselves emit coherent combs. In other words, the suggestion
was that the generated octave-wide continua maintain the coherence of the
driving field. Experiments quickly showed that these continua can, in fact,
be coherent combs, and the first optical frequency measurements using
octave-spanning combs were demonstrated [27]. Tests showed pulse-to-
pulse coherence across the entire optical spectrum, and by comparing the
interval division method with a comb-frequency measurement, the coherent
nature of the comb was further ascertained [28].
In most frequency measurement schemes, errors can occur due to “cycle
slipping.” Considering that mode-locking a laser also involves a “phase-
locking” process, 2π phase slips are conceivable that would adversely affect
frequency measurement. A test, however, showed no indication of such
cycle slips. By measuring an octave interval [29], it was shown that the
frequency-measurement errors must be below ∆ν/ν = 10-18. In fact, the
observed 10-18 limit was clearly caused by slow phase changes due to slow
changes of optical path lengths because of temperature changes and can thus
be improved (see Section 6.2).
The mechanism(s) that generate octave-wide spectra in nonlinear fibers
are not entirely understood yet. In [30], it is argued that self-phase
modulation (SPM) that is commonly assumed to account for the spectral
broadening is by far too weak to generate octave-wide spectra. Model
calculations show that wide spectra occur only under a condition where the
pump pulse disintegrates into fundamental fiber solitons whose wavelength
red-shifts in time (to cover the wide spectrum towards the IR) while the
“dispersive waves” (i.e., the radiation “shedded” from the initial pulses in
the process of soliton formation) blue-shift in time covering the short
wavelength range. In a remotely related paper [31], it was shown that the
dispersive waves themselves can subsequently transform to solitons,
shedding additional light covering additional spectral ranges. Thus at the end
a wide spectrum is generated.
These model calculations seem realistic since they predict several
features to be critical that are in agreement with experimental results: (1) a
particularly short pulse is not important (indicating that SPM is not the main
mechanism for spectral broadening) and (2) wide spectra are only obtained
in the case where the center of the input spectrums is close to the zero-
dispersion point of the fiber. Therefore, it appears that octave-wide spectra
are produced by not just one mechanism. Rather, several effects contribute,
including Raman amplification, four-wave mixing, soliton formation, and
SPM.
206 Chapter 8
The question arises on how such incoherent processes can produce

coherent comb spectra. All observations point to the following explanation:
initially, in the interaction of the light with the fiber, SPM (which produces a
coherent spectrum) creates a field that, at all comb frequencies (i.e., even at
the edges of the width of an octave), exceeds the incoherent noise strongly.
Thus incoherent amplification mechanisms must start from a coherent
“seed” (instead of incoherent noise), resulting in a coherent octave comb.
Because of the complete phase coherence of the entire comb, the phase
fluctuations of one comb line are very small. The entire laser field acts as
one coherent field; thus the phase fluctuations induced by spontaneously
emitted photons (Shawlow-Townes phase diffusion) are exceedingly small
given the total laser power of several 100 mW. The comb is extremely
“rigid.”
We mention that this enormous rigidity, or phase coherence, is not, in
general, a property of mode-locked lasers. Mode-locked dye lasers, for
example, lock predominantly by slow cascaded four-wave-mixing processes,
where in each mixing process, (vacuum-) noise is added, so that in such
systems the coherence range in the comb is much smaller than the comb
width. It is this particular mode-locking mechanism of the Ti:sapphire laser
that utilizes the essentially instantaneous Kerr effect that produces an
instantaneous and completely synchronous refractive-index modulation for
all modes throughout the spectrum.
Generally, optical oscillators possess extreme spectral purity limited only
by spontaneous emission. One wonders if such high spectral purity could be
transferred to other spectral regions where applications require it, e.g., to the
rf/microwave range. To transfer the spectral purity to lower frequencies, a
“noiseless-optical-frequency divider” would evidently be needed; however,
such a device does not exist. Fortunately, as discussed above, the optical
frequency can be measured in terms of the pulse repetition rate (the comb
line spacing), such that the optical frequency is a (in general noninteger)
multiple of the pulse repetition frequency. In this sense, the mode-locked
laser contains a built-in divider. The latter is very low noise because of the
rigidity of the comb spectrum. The laser radiation contains rf frequencies in
the difference-frequency domain. However, these frequencies have to be
converted to electrical signals. Although this optical-to-electrical conversion
presents a limitation (photodiodes can only handle limited amounts of
optical power), it can be estimated that microwave signals thus generated
will be superior in phase noise to the fields of the best (cryogenic)
microwave oscillators.
We now briefly describe the basic properties of a Kerr-lens, mode-locked
laser that emits a periodic train of short pulses generating an optical
frequency comb of distinct lines with well-defined, equidistant spacing. The
spectral span of this comb reflects the duration of an individual pulse, while
the spacing between the lines is determined by the pulse repetition
frequency.
We start with a time domain description of the pulses emitted by a mode-
locked laser, as shown in the upper part of Figure 8-4. Every time the pulse
circulating inside the cavity hits the output coupler, the laser emits a pulse.
This process results in a train of pulses separated by time τ = lc / vg, where lc
is the round-trip length of the cavity and vg is the group velocity. Because of
unavoidable dispersion in the cavity, the group and phase velocities vp are, in
general, not equal. This inequality results in a shift of the carrier phase with
respect to the peak of the envelope for each round trip.
∆Φ
∆φce
Time domain:
time
τ == 1/f
∆t 1/frep
r
Frequency domain:
νν(m)
(m) ν(2n)
ν(2n)
νfceo
0
fr
frep
ν(m)
ν(m) = ν=ceof0++mmfrep
fr
SH frequency
ν=0
νfceo
0 == 2 ν(n)- -ν(2n)
2ν(n) ν(2n) f0
νceo νν(n)
(n) 2ν(n)
2ν(n)
Figure 8-4. Representation of the output field of a mode-locked laser in the time domain or
frequency domain, including the definition and measurement of the carrier-envelope-offset
frequency f0.
This shift between successive pulses is described by ∆φce, which is

proportional to the difference of the inverse of the group- and phase-
velocities. Taking the Fourier transform of a train of pulses, together with
the condition that the field reproduces itself after each round trip coherently,
yields a comb spectrum with angular frequencies
m ⋅ 2π ∆φce
ωm ≡ + . (1)
τ τ
Dividing by 2π yields the frequencies of the mth comb mode,

208 Chapter 8
∆φ
ν ( m) = mf r + f r = mf r + f 0 . (2)
2π
Therefore, the frequency of any line of the spectral comb emitted by such
a laser, as shown in the lower part of Figure 8-4, is given by an integer m,
the pulse repetition frequency fr of the laser, and the carrier-envelope offset-
frequency, f0, which accounts for an offset of the entire comb with respect to
the frequency zero. The frequency f0 represents the time derivative of the
optical (carrier) phase ∆φce as measured with respect to the pulse envelope.
The determination of the frequency of the mth comb mode ν(m) therefore
requires precision measurements of fr, f0, and m. The determination of m can
be performed by a wavelength measurement, for example. The measurement
of the pulse repetition rate fr is straightforward to carry out. However, since
fr enters the optical frequency measurement process with the large factor m,
the noise properties of the measurement are demanding.
The measurement of f0 is also a demanding task. Several more or less
complex schemes, depending on the comb span available, have been
proposed for the measurement of f0 [32]. The simplest concept requires an
octave span of the frequency comb that is not directly available from the
laser used. The available span has to be expanded, e.g., by external spectral
broadening. Microstructure air-silica fibers, which allow the tailoring of the
group-velocity dispersion (GVD) properties [26], are highly suited for SPM
of moderate peak power pulses. They provide both lateral and temporal
confinement of the pulses over long interaction lengths. If the octave span is
achieved, then the f0 measurement can be accomplished by second-harmonic
generation (SHG) of the spectral lines in the low-frequency wing of the
comb, as shown in Figure 8-4. This method has been developed at JILA [33]
and MPQ [34].
Whereas the frequencies of the comb lines are shifted by f0 with respect
to the origin, their second harmonics are shifted by 2f0 (Figure 8-4). Thus,
the beat notes between these harmonics and the corresponding high-
frequency-wing lines of the comb shows the desired component at f0 (n=2m)
δν = 2 f 0 + 2mf r − f 0 − nf r = f 0 + (2m − n ) f r . (3)
When all three quantities fr, f0, and m are known, any unknown external
optical frequency νext of a laser emitting within the spectral range of the
comb can, in turn, be determined by measuring the beat-note frequency ∆ν
with a suitable comb line. The absolute frequency νext is then expressed by
νext = ν(m) + ∆ν. Furthermore, it is possible to directly reference the comb
spacing (fr) and position (f0) to the microwave cesium time standard, thereby
determining the absolute optical frequencies of all of the comb lines.
This recent advance in optical frequency measurement technology has
facilitated the realization of the ultimate goal of a practical optical frequency
synthesizer: it forms a phase-coherent grid linking the entire optical
spectrum to a microwave standard, or vice versa.
5. OPTICAL FREQUENCY MEASUREMENTS
We have used such “self-referenced” combs to measure the frequencies

of the Ca cold atom cloud standard at 657 nm [35] and the single Yb+-ion
standard at 435 nm [36]. The arrangement in Figure 8-5 was used. The comb
generator used a 10 fs Kerr-lens mode-locked Ti:sapphire laser [32] and a
microstructure fiber for broadening the comb spectrum [26]. By coupling 30
mW into a 10 cm long fiber, we covered the spectrum from 500 to 1200 nm.
The pulses behind the fiber are frequency doubled and coherently
superposed with the fundamental comb radiation. Variable delay adjusts the
pulse arrival times so that the pulses of fundamental and frequency-doubled
radiation overlap in time. A grating filters out the high frequency end of the
spectrum that contains the f0 beat, which is detected by a photo detector. The
f0 beat frequency is filtered by a tracking oscillator as developed for classical
frequency multiplication work. It is an optimum filter strategy, for lasers and
oscillators exhibiting only frequency noise with Fourier components in the
kHz range [5].
The critical frequency to be measured is fr since the optical interval ν to
2ν is measured as a very high multiple of this quantity. Instead of fr itself, a
high harmonic of it in the 10 GHz range (~100th harmonic) is detected after
spectral filtering of the laser light by a solid etalon of 10 GHz free spectral
range. The 10 GHz beat is down-converted with a low noise 10 GHz signal
synthesized from the rf-frequency standard (a hydrogen maser referenced to
a cesium fountain clock). The difference frequency is again multiplied by 96
and by 3, reducing frequency-counter digitization errors to below the level of
the noise of the H maser and thereby, the frequency that is effectively
measured is 3 THz.
For the measurement of frequencies of optical standards, the light of
these standards is heterodyned with a narrow range of the comb spectrum
filtered out by interference filters around the standards’ frequencies. The
usual tracking oscillator is used for optimal filtering of the heterodyne
signal.
210 Chapter 8
Frequency Ti:Sa f/U ∆ x -Servo

Comb & Meas.
Generator
PLL Counter
PD
SESAM
PZT2 Filter
PZT1 Grating
LBO
MS Fiber
OC
Self-Phase-
10 GHz PM
fr -Servo
f REP Modulation
Etalon
& Meas.
PD
Counter PLL
PLL
f/U
10 GHz ν CEO
f0 -Measurement
× 96 ×3 Counter
DRO
100 MHz from H-maser From Optical Standard
Figure 8-5. Femtosecond laser as frequency comb generator and generated comb spectrum.
LBO: Lithium β-Borate crystal and SESAM: Semiconductor Saturable Absorber Mirror.
The optical standards that are used for absolute optical frequency
measurement at PTB are located in two different buildings, and the comb
generator in a third building. The standards’ radiation is brought to the comb
generator by ~250 m long single-mode fibers and the rf standard signal by
air-dielectric cables. Measurements have shown that the changes in the cable
and fiber lengths due to temperature changes, which produce temporal phase
changes (corresponding to frequency shifts), remain below ∆ν/ν = 10-15. At
the present level of precision for absolute frequency measurements, these
changes can be neglected.
5.1 Ca frequency measurement
An improved Ca spectrometer uses an ensemble of cold Ca atoms

prepared in a magneto-optical trap (MOT) [37]. The contributions to the
uncertainty of the improved standard were investigated earlier [38], resulting
in an overall uncertainty of 8 Hz (δν/ν = 2 x 10-14). A summary of the
historical record of frequency measurements of laser-cooled Ca frequency
standards is shown in Figure 8-6. Earlier frequency measurements based on
the harmonic frequency chain (see 1995–1997 in Figure 8-6) led to a
fractional uncertainty of 2.5 x 10-13. The first measurements of the Ca
standard with a femtosecond laser comb agreed within the combined
uncertainties (September 2000) [39]. In the meantime further measurements
were performed. These measurements were limited mainly by the

uncertainties resulting from the corrections due to magnetic quadrupole
fields of the MOT and AC-Stark shifts (June 2001). A frequency
measurement with reduced uncertainty was then performed [35] where the
spurious phase shifts resulting from misalignment and curvature of wave
fronts were reduced and the systematic shifts due to the quadrupole MOT
field and the AC-Stark shift were eliminated.
For a weighted mean of all frequency measurements shown in Figure 8-
6, the measurements of June 2001 were averaged, and the uncertainty of the
single measurements of 23 Hz kept as they were taken on a single day with
no changes in the settings, so that the systematic shifts were not independent.
The same was true for the two days of measurements in October 2001 that
were averaged and their uncertainty determined to 8 Hz. With a new comb
system, additional frequency measurements were done in October 2003, and
individual data points were treated in the way described above, resulting in a
total uncertainty of these measurements of 6 Hz. The weighted mean of all
the frequency measurements performed at PTB with the harmonic chain and
the femtosecond frequency comb is thus,
νCa = ( 455 986 240 494 149 ± 6) Hz
with a relative standard uncertainty 1.3 x 10-14.
300
νCa - 455 986 240 494 000 Hz
Hz
200
100
1995 1997 Sep '00 Jun '01 Oct '01 Oct '03
Figure 8-6. Frequency measurements of laser-cooled Ca optical frequency standards over

the past eight years. The full-square data points represent data from PTB [38] while the single
open-dot data point represents the frequency value reported by the NIST group [40]. Prior to
1999: a harmonic frequency chain; since 2000: femtosecond optical combs.
212 Chapter 8
The Ca frequency measurement is in good agreement with recent

measurements made at the National Institute of Standards and Technology
(NIST), Boulder, also using a femtosecond comb generator [40]. The
consistency of the Ca results between the two laboratories, over time and
with different apparatus, demonstrates the good reproducibility of these
standards with respect to the cesium primary frequency standards.
5.2 Yb+ frequency measurement
Several measurements of the absolute frequency of the Yb+ ion trap

frequency standard were performed in 2001 and 2003. The results are shown
in Figure 8-7. By averaging the data over the measuring period in 2001, a
value for the 6s 2S1/2 (F = 0) – 5d 2D3/2 (F = 2) electric-quadrupole clock
transition of the 171Yb+ ion of
νYb+ = 688 358 979 309 312 Hz ± 6 Hz
was derived [41]. An additional measurement of its frequency was

performed in 2003 with a new frequency comb generator. The results of this
measurement deviated from the previous one by (–2.3 ± 6.4) Hz.
40
Hz
30
ν - 688 358 979 309 312 Hz
20
10
0
-10
-20
-30
-40
12 / 00 01 / 01 11 / 03
Figure 8-7. Frequency measurements of laser-cooled Yb+ optical frequency standards over
the past three years. The first three data points represent published data from Reference [41].
The combined 1σ-uncertainty of 6.4 Hz of the latest measurement is

given by statistical and systematic contributions. Sources of systematic
uncertainties are the realization uncertainty of the cesium fountain frequency
standard [42] and the 171Yb+ frequency standard. The dominant contribution
to the total uncertainty of the frequency value is the statistical uncertainty of
the frequency measurement of 5 Hz because of the limited time of
observation. The most significant systematic contribution to the uncertainty
of 3 Hz is due to a quadrupole shift caused by unknown electric field
gradients in the trap. This quadrupole Stark effect has recently been
measured [43]. The reproducibility of frequency comparisons of two traps
over an extended time, which included reloading traps, seems to indicate that
this quadrupole Stark shift would, with the traps used presently, be below 1
Hz.
The weighted mean of the frequency of electric-quadrupole clock
transition of the 171Yb+ ion is
νYb+ = 688 358 979 309 311 ± 6 Hz.
The measurement of the frequency of the electric-quadrupole clock

transition of the 171Yb+ ion with a relative uncertainty of 1 x 10-14
demonstrates the potential of the 171Yb+ standard as a precise optical
frequency reference.
In accordance with the measurements using the traditional frequency
chain on the Ca standard and on the CH4 IR standards [3, 44], the absolute
frequency measurements are dominated by the noise of the low-frequency
reference (the hydrogen maser) for short integration times.
6. TEST ON THE PRECISION OF FREQUENCY

MEASUREMENT WITH FREQUENCY COMBS
6.1 Transfer technique
Ideally in a frequency measurement, the measurement instrument itself,

i.e., the frequency multiplication chain or frequency comb, should not limit
the measurement in any way. Obviously any laser setup exhibits
frequency/phase-fluctuations. Even if one has the necessary “transducers”
for phase locking all desired frequencies, the residual phase excursions dφ/dt
still allow changes of the frequencies.
The faster the transducers (i.e., typically piezo transducers limited in
speed by the mirror masses), the smaller the phase excursions can be held.
(Evidently, with the increase of control speed, an increased signal-to-noise
ratio of the control signals is required.)
214 Chapter 8
In general, to become independent of transducer speed in frequency

synthesis, the “transfer technique” is used. As an example, one may consider
a frequency multiplication step (Figure 8-8). The frequency f2 of a laser is
measured against a reference near its frequency. (In a frequency
multiplication chain this reference would be a harmonic of a lower
frequency f1 that is already known.) The beat signal b1,2 between the
reference and the laser contains all the frequency fluctuations of the transfer
laser.
f3 stabilized
b2,3 = mf2 - f3 m
*m 1
f2 transfer osc.
f2 - f3/m
b1,2 = nf1 - f2
*n
f1 stabilized fc = nf1 - f3/m
Figure 8-8. Using an intermediate oscillator in the transfer mode. The signal fc is independent
from f2.
The laser’s frequency is then multiplied m-fold and its harmonic

compared with the frequency of another stabilized laser f3 of approximately
m-fold higher frequency. The latter beat signal b2,3 contains the frequency
fluctuations multiplied m times. Thus the frequency fluctuations of the
transfer laser can be eliminated completely from the measurement by
dividing the beat signal frequency b2,3 electronically by m and subtracting it
(e.g., by a mixer that produces a difference frequency) from the beat signal
frequency b1,2. The resulting difference frequency then measures directly the
difference of the nth harmonic of the reference and the laser frequency
divided by m. All frequency fluctuations of the transfer laser itself are
eliminated purely electronically, and no fast transducers are needed. The
transfer laser itself does not enter into the measurement. It just “transfers”
information from the reference frequency level by a (n x m)-multiple to the
laser frequency level. Thus the name: “transfer technique.” We have used
the scheme of phase-coherent transfer in our conventional frequency
multiplication chain [5].
For any phase-coherent frequency measurement, the maintenance of the

coherence or “phase-tracking-ability” of the harmonic mixing signals is the
essential requirement for a frequency measurement. Simply stated: if a
frequency is multiplied by a factor N, its period duration is divided by N. If
the low starting frequency shows a certain level of phase noise (i.e., the zero
crossings of the field jitter by an amount ∆t), then there is a critical factor
Ncrit above which ∆t exceeds the period duration of the multiplied frequency.
This means that, on average, the period to which a zero crossing belongs is
no longer defined. (This phenomenon is called “coherence collapse.”) If an
oscillator “transfers” information between two other oscillators (relating
their frequency in some way), the transfer technique can be generalized. In
the way, the femtosecond comb generator can act as a transfer oscillator, and
its frequency fluctuations thus do not enter the measurement process.
For the realization of absolute frequency measurement using the transfer
technique, the scheme in Figure 8-9 was used. This scheme realizes
physically the relation of the frequency ν of an optical frequency standard
with the three measured frequencies fr, f0, and ∆ν: ν = m fr + f0 + ∆ν. To keep
the requirements on the phase detector stabilities reasonable, the pulse
repetition frequency is measured as a high harmonic (m1 ~ 100). The detector
output frequency around 11 GHz is then down-converted to a range of a few
MHz using a suitable harmonic of the absolute reference (hydrogen maser
referenced to a cesium clock). The resolution of the frequency measurement
is further increased by a factor m3 using an additional tracking generator
with a harmonic phase-locked loop. In this kind of harmonic generation, the
harmonic number must be limited because of the decrease of the signal-to-
noise ratio and of the necessary increase of the tracking bandwidth with the
harmonic number that would both ultimately cause cycle slips.
On the other hand, the optical frequencies can be divided by the factor m
directly. However, this reduces the resolution and the phase noise of the
dividers will provide a serious limitation. Therefore, the two “optical” beats
are summed and divided by m2. This procedure leads to the optimized
combination m = m1 ⋅m2 ⋅m3. For practical reasons, we choose m1 and m3 in
the range of 100 and 294, respectively.
In this way, the frequency νc represents the beat between the up-
converted low-frequency reference (hydrogen maser referenced to cesium
clock) and the down-converted optical-frequency standard. All information
is contained in the frequency νc that is independent of the fluctuations of the
comb. Since the input signals ν and fLO are stable, it is possible to filter the
signal νc by a filter of a few Hz(!) bandwidth, increasing the SNR ratio
enormously.
216 Chapter 8
This technique differs completely from the conventional approach in its

handling of the technical frequency noise contributions. One radio frequency
signal containing all necessary information about the external optical signals
is synthesized and stabilization of the comb is not necessary. Technically,
the complete cancellation of all femtosecond comb fluctuations requires a
processing of the optical signals with a sufficiently high bandwidth.
Considering the typical Fourier frequency range of the noise of f0 and the
beat signal ∆ν of a few MHz, this requirement can be easily fulfilled.
m1 xxffr
m ffLO
1 rep LO
Local Oscilator
Local Oscillator
Harmonic
Harmonic PLL
PLL m33
xxm
A ν = (m f – f
1 r LO ) m3
νA= (m 1frep- fLO)m3
ννCC==νAνA++ννBB
νfCEO
0
νB = (νBCEO
ν = (f + ∆ν ) / m2
f0 +0 ∆ν) /m2
÷m
÷ m22
ν
∆∆ν
νν == m22ννCC ++ m2m3ffLO
LO
= m1 ·m2 ·m3 fr + f0 + ∆ν
Figure 8-9. Linking microwave and optical frequencies: signal-processing scheme.
6.2 Frequency ratio
Any absolute frequency measurement requires that the frequency ν is

measured and expressed in the SI-unit Hertz. The primary unit Hertz is
technically realized by a cesium-clock controlled H-maser generating a
standard frequency of 100 MHz, for example. Therefore, any absolute
frequency measurement is principally limited by the frequency instability of
the hydrogen maser. However, optical frequencies are measured (in many
cases) with respect to an optical reference frequency by measurement of
their frequency difference [45]. Then, only a fraction of the H-maser noise
enters the measurement process and is given by the ratio between the
frequency difference and the absolute frequency.
Measuring the ratio of two frequencies does not require a frequency
reference. Because the ratio is a unitless number, it is not referred to an
absolute frequency and its stability is that of the optical frequency standard
with the larger instability. Thus, frequency ratios for oscillators with better
stability can be determined with smaller uncertainty if a technique is
available to realize the frequency ratio without introducing additional noise.
Such a technique is the transfer oscillator concept, described above, which
has been generalized for signals with rational frequency ratios [29, 46].
∆1
ννCC==νν –ν
A A- νB B
νν
A A== ff00 ++∆∆11
νCEO ==νν – ν x m1/m
1 1 ν2 ×2 m1 /m
- 2 3
ν2ν2
f0 νν11
∆2 ∆
∆
∆1 ∆2
2
νB = (ννCEO
B0 = ( f02 + ∆2 )1 x m
f + ∆ ) × m /m2 /m 1
1 2
Figure 8-10. Modified transfer concept for the measurement of optical frequency ratios. νC
represents the beat note between arbitrary optical frequencies ν1 and ν2.
If, as with the optical comb generator, two frequencies are related by the
comb through a “rational” harmonic ν2/ν1 = m2/m1, then the two beats have to
be treated accordingly. One of the beat frequencies has to be multiplied by
m2/m1, and the result has to be subtracted from the other beat frequency.
Figure 8-10 shows the scheme for the measurement of the ratio of the
frequencies of two optical standards whose frequencies are given by
ν1 = m1 ⋅fr + f0 + ∆1 and ν2 = m2 ⋅fr + f0 + ∆2 .
Three signals are detected: f0 and two beat signals of two external
frequency standards with the modes of the femtosecond frequency comb, ∆1
and ∆2. The sum frequency of ∆1 and f0 is generated by a mixer. The sum
frequency of ∆2 and f0 is additionally processed with a direct-digital
synthesizer (DDS). Such a device generates an output signal from an input
signal with a frequency ratio given by a long digital tuning word. It
numerically approximates the ratio of the two integers, m1 and m2, by j/2n,
where j is an integer and n the bit length of the tuning word. The resulting
possible error is negligible for commercially available n = 48 bit devices.
Generating the difference frequency of the signals νA and νB with the
help of a mixer, results in the ratio
218 Chapter 8
ν 2 m2 1   m ν
+  ∆2 + f 0 − 2 (∆1 + f 0 ) = 2 + C .
m
=
ν 1 m1 ν 1  m1  m1 ν1
The frequency νC is a measure of the (small) deviation of the optical

signals frequency ratio ν2/ν1 from m2/m1. Since νC<<ν1, ν2, the requirements
on the radio frequency reference are, in general, not demanding. The
frequency νC carries the full information of the fluctuations of the ratio ν2/ν1
and can be considered as the frequency of the beat note between ν1 and ν2 as
projected to the vicinity of ν1. However, the uncertainty of ν1 enters and
limits the precision of the ratio if the term in the brackets is not zero. By
proper choice of fr and f0, this term can be made equal to zero and the ratio of
optical frequency standards can be measured without knowledge of the
comb-related frequencies (except for the definite determination of the mode
numbers m1 and m2). By measuring νC, this scheme allows one to realize a
relative frequency measurement of an unknown frequency ν2 with respect to
an arbitrary frequency reference ν1.
As well as measuring the ratio of two optical frequencies, the method
allows the measurement of the ratio of an optical and a radio frequency. This
rf reference can be related to a primary standard, thereby realizing the
absolute frequency measurement described in the previous paragraph.
The measurement of a frequency ratio has been used with a subharmonic
of the Yb+ ion-frequency standards νYb+ at 871 nm [47], and an iodine-
stabilized Nd:YAG laser ν0 at 1064 nm [48]. Referring to Figure 8-10, the
beat note of νYb+ with a comb mode was ∆2 and that of ν0 was ∆1. The output
of the last mixer in Figure 8-10 is equivalent to the beat signal between the
iodine and Yb+ signals at a virtual frequency of νC = 43 THz.
Figure 8-11 shows the resulting Allan standard deviation of this ratio
measurement (open squares) and of an absolute optical frequency
measurement of the Yb+ optical frequency standard (full dots). The accuracy
of the absolute frequency measurement is limited by the noise of the
hydrogen maser (solid line). In contrast, the frequency noise of the beat note
between the Yb+ and iodine signals is expected to be substantially smaller
than that of the H-maser, at least at Fourier frequencies f > 0.1 Hz. The direct
frequency comparison of two independent I2-stabilized standards shows
frequency instabilities similar to that found for the Yb+/I2-ratio measurement,
indicating that this ratio measurement was limited by the noise of the I2-
stabilized standard. The dashed line assumes a white noise level of ≈3 x 10-14
at 1 s.
This experiment demonstrated the capability of phase-coherently linking
optical frequencies from very different spectral regions without introducing
additional noise from the measurement process. Thus frequency-ratio
measurements between optical frequencies with short-term instabilities

superior to that of a microwave reference can be carried out.
Using the setup we just described, it was not possible to investigate any
possible noise contribution from the femtosecond comb; the stability of even
the high-quality reference-signal pair was not good enough. Therefore, in a
second experiment, we used the signal ν0 and its second harmonic νSH that
was generated in a nonlinear crystal. The frequency-doubling process
enabled the generation of a pair of reference frequencies with almost
completely correlated fluctuations, leading to a fixed frequency ratio of 2:1
even for the shortest averaging times.
Referring to Figure 8-10, the beat note of νSH with a comb mode was ∆2
and that of ν0 was ∆1. The DDS was replaced by a frequency divider of 2.
Both comb lines generating the beat notes at the fundamental and the second
harmonic were themselves generated in the microstructure fiber.
-12
10
-13
10
-14
10
-15
10
σy( 2, τ )
-16
10
-17
10
-18
10
0,01 0,1 1 10 100 s 1000
integration time
Figure 8-11. Allan standard deviation of the Yb+ frequency standard (full dots) with respect
to an H-maser. The frequency instability of a typical hydrogen maser reference (solid line) is
given for comparison. The open squares represent the stability of an I2-stabilized–frequency-
doubled Nd:YAG laser as derived from the frequency ratio with respect to Yb+ standard. The
triangles show the Allan standard deviation for the frequency ratio between the second
harmonic and the fundamental of the I2-stabilized Nd:YAG laser.
The triangles in Figure 8-11 show the Allan standard deviation for this
measurement. The instability is more than two orders of magnitude smaller
than for the νYb+/ν0 measurement, for which high-quality oscillators were
already used. The optical frequency ratio can be measured at averaging times
of 100 s to a level of ~10-18 corresponding to mHz fluctuations. The residual
fluctuations at the large averaging times do not indicate a limitation of the
220 Chapter 8
frequency measurement but are the result of the slow optical phase
variations caused by the changes in the optical path lengths between the
optical elements on the table where the system was set up, in temperature, in
the motion of the air, and so forth. These variations could evidently be
reduced further so that 10-18 does not represent a limit in measurement
precision.
In any case, the lower curve shows unambiguously that the transfer
method works well and that a frequency comb used in the transfer mode is
adequate for forming a part of future atomic clocks based on optical
transitions; it would not limit the transfer of optical stability to the rf range,
even at the projected precision and stability of 10-18.
Figure 8-12 shows a Fourier transform of the measurements represented
by the lowest curve of Figure 8-11. The 9 mHz spectral line representing the
frequency measurement uncertainty corresponds to the “virtual measurement
frequency” of 43 THz. Converted to the measured optical frequency of 282
THz it would show a spectral line of 60 mHz width (still remarkably narrow)
at 282 THz as characterizing the measurement, or the coherence of the
measurement process.
Figure 8-12. Power spectrum of the beat note between sub-harmonics of the outputs of an I2
and an Yb+-frequency standard taken at a virtual frequency of 43 THz as calculated from the
recorded phase excursions by fast Fourier transform.
7. SUMMARY
Frequency chains have been developed and operated for many years in
several laboratories. However, during the last few years, great progress has
been achieved in the development of optical frequency standards and optical
frequency synthesis. These advances are based on precision spectroscopy

with cold absorbers and improvements of phase-coherent optical frequency
measurements. In the past, each laboratory focused on different target
frequencies and hence different frequency standards. With the development
of optical frequency measurements based on femtosecond lasers, a new era
in optical frequency metrology was started. In fact, optical frequency
measurements based on femtosecond lasers will soon be available in many
laboratories. This concept is well suited for measuring the absolute
frequency of essentially any optical frequency standard with reproducibility
at the level of the present primary atomic frequency standards.
The control of the comb frequencies, especially the carrier offset
frequency f0, which arises from the pulse to pulse phase shift of the carrier
envelope phase, allows one to investigate extreme nonlinear processes that
are directly sensitive to the electric field of each pulse such as x-ray
generation, and it can be used to tightly synchronize independent mode-
locked lasers operating in different spectral regions [49].
With the femtosecond comb, it is possible to compare different optical
frequency standards by beat frequency measurements, and the
reproducibility of stabilized lasers can now be determined over long periods
of time by comparisons with the cesium atomic clock. Furthermore, the use
of active optical-frequency comb generators allows one to develop accurate
clocks operating in the optical frequency range.
It can be envisaged that the uncertainty of the best optical frequency
standards will eventually become lower than that of the present realization of
the second. This new situation may lead to a replacement of the (cesium-
based) definition of the second by an optical reference frequency. With the
continuing fast progress in the development of optical frequency standards,
the selection of the “best reference” will be a challenging and exciting task.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge the contributions of C. Degenhardt,

J. Helmcke, E. Peik, F. Riehle, U. Sterr, H. R. Telle, and C. Tamm.
REFERENCES
[1] C. O. Weiss and A. Godone, IEEE J. Quantum Electron. 20, 97-99 (1984).
[2] D. J. E. Knight, K. I. Pharaoh, and M. Zucco, in 5th Symposium of
Frequency Standards and Metrology, Woods Hole, MA, 1995), p. 465-467.
222 Chapter 8

Lett. 76, 18-21 (1996).
[4] H. U. Daniel, M. Steiner, and H. Walther, App. Phys. 25, 7-12 (1981).
[5] C. O. Weiss, G. Kramer, B. Lipphardt, and H. Schnatz, in Frequency
Measurement and Control Advanced Techniques and Future Trends,
2000), Vol. 79, p. 215-247.
(1990).
[7] N. C. Wong, Opt. Lett. 15, 1129-1131 (1990).
2693-2701 (1993); M. Kourogi, T. Enami, and M. Ohtsu, IEEE Photonics
Technol. Lett. 6, 214-217 (1994).
[9] L. R. Brothers, D. Lee, and N. C. Wong, Opt. Lett. 19, 245-247 (1994).
[10] D. A. van Baak and L. Hollberg, Opt. Lett. 19, 1586-1588 (1994).
[11] O. Pfister, M. Murtz, J. S. Wells, L. Hollberg, and J. T. Murray, Opt. Lett.
21, 1387-1389 (1996); P. T. Nee and N. C. Wong, Opt. Lett. 23, 46-48
(1998).
[12] C. Koch and H. R. Telle, J. Opt. Soc. Am. B 13, 1666-1678 (1996).
[13] T. Andreae, W. Konig, R. Wynands, D. Leibfried, F. Schmidtkaler, C.
Zimmermann, D. Meschede, and T. W. Hänsch, Phys. Rev. Lett. 69, 1923-
1926 (1992).
W. Hänsch, Phys. Rev. Lett. 79, 2646-2649 (1997).
[15] V. B. Taranenko, K. Staliunas, and C. O. Weiss, Phys. Rev. Lett. 81, 2236-
2239 (1998).
[16] K. Imai, M. Kourogi, and M. Ohtsu, IEEE J. Quantum Electron. 34, 54-60
(1998).
[17] A. Huber, T. Udem, B. Gross, J. Reichert, M. Kourogi, K. Pachucki, M.
Weitz, and T. W. Hänsch, Phys. Rev. Lett. 80, 468-471 (1998).
[18] F. Schmidt-Kaler, D. Leibfried, S. Seel, C. Zimmermann, W. Konig, M.
Weitz, and T. W. Hänsch, Phys. Rev. A 51, 2789-2800 (1995).
[19] F. Schmidt-Kaler, D. Leibfried, M. Weitz, and T. W. Hänsch, Phys. Rev.
Lett. 70, 2261-2264 (1993).
[20] M. Weitz, A. Huber, F. Schmidtkaler, D. Leibfried, and T. W. Hänsch,
Phys. Rev. Lett. 72, 328-331 (1994).
[21] F. Nez, M. D. Plimmer, S. Bourzeix, L. Julien, F. Biraben, R. Felder, O.
Acef, J. J. Zondy, P. Laurent, A. Clairon, M. Abed, Y. Millerioux, and P.
Juncar, Phys. Rev. Lett. 69, 2326-2329 (1992).
[22] J. von Zanthier, J. Abel, T. Becker, M. Fries, E. Peik, H. Walther, R.
Holzwarth, J. Reichert, T. Udem, T. W. Hänsch, A. Y. Nevsky, M. N.
Skvortsov, and S. N. Bagayev, Opt. Commun. 166, 57-63 (1999).
(2001); J. Ye, H. Schnatz, and L. W. Hollberg, IEEE J. Sel. Top. Quantum
Electron. 9, 1041-1058 (2003).
[24] T. Udem, J. Reichert, R. Holzwarth, and T. W. Hänsch, Phys. Rev. Lett.
82, 3568-3571 (1999); T. Udem, J. Reichert, R. Holzwarth, and T. W.
Hänsch, Opt. Lett. 24, 881-883 (1999); J. Stenger and H. R. Telle, in Laser
Frequency Stabilization: Standards, Measurement and Applications, edited
by J. L. Hall and J. Ye (SPIE, San Jose, CA, 2001), Vol. 4269, p. 72.

1547-1549 (1997); N. G. R. Broderick, T. M. Monro, P. J. Bennett, and D.
J. Richardson, Opt. Lett. 24, 1395-1397 (1999).
[27] S. A. Diddams, D. J. Jones, L. S. Ma, S. T. Cundiff, and J. L. Hall, Opt.
Lett. 25, 186-188 (2000); S. A. Diddams, D. J. Jones, J. Ye, S. T. Cundiff,
J. L. Hall, J. K. Ranka, R. S. Windeler, R. Holzwarth, T. Udem, and T. W.
Hänsch, Phys. Rev. Lett. 84, 5102-5105 (2000).
[28] R. Holzwarth, M. Zimmermann, T. Udem, and T. W. Hänsch, IEEE J.
Quantum Electron. 37, 1493-1501 (2001).
073601 (2002).
[30] A. V. Husakou and J. Herrmann, Phys. Rev. Lett. 8720, art. no.-203901
(2001).
[31] W. S. Man, H. Y. Tam, M. S. Demokan, and D. Y. Tang, Opt. Quant.
Electron. 33, 1139-1147 (2001).
Keller, Appl. Phys. B 69, 327-332 (1999).
[35] J. Helmcke, G. Wilpers, T. Binnewies, C. Degenhardt, U. Sterr, H. Schnatz,
and F. Riehle, IEEE Trans. Instrum. Meas. 52, 250-254 (2003).
[36] C. Tamm, T. Schneider, and E. Peik, in Frequency Standards and
Metrology, Proceedings of the 6th Symposium, edited by P. Gill (World
Scientific Publishing, 2001), p. 369-384.
[37] F. Riehle, G. Wilpers, U. Sterr, T. Binnewies, and J. Helmcke, in
Frequency Standards and Metrology, Proceedings of the 6th Symposium,
edited by P. Gill (World Scientific Publishing, 2001), p. 339-357.
[38] G. Wilpers, T. Binnewies, C. Degenhardt, U. Sterr, J. Helmcke, and F.
Riehle, Phys. Rev. Lett. 89 (2002).
[39] J. Stenger, T. Binnewies, G. Wilpers, F. Riehle, H. R. Telle, J. K. Ranka, R.
S. Windeler, and A. J. Stentz, Phys. Rev. A 63, 021802 (2001).
[40] T. Udem, S. A. Diddams, K. R. Vogel, C. W. Oates, E. A. Curtis, W. D.
Lee, W. M. Itano, R. E. Drullinger, J. C. Bergquist, and L. Hollberg, Phys.
Rev. Lett. 86, 4996-4999 (2001).
[41] J. Stenger, C. Tamm, N. Haverkamp, S. Weyers, and H. R. Telle, Opt. Lett.
26, 1589-1591 (2001).
[42] S. Weyers, U. Hubner, R. Schroder, C. Tamm, and A. Bauch, Metrologia
38, 343-352 (2001); T. Heindorff, A. Bauch, P. Hetzel, G. Petit, and S.
Weyers, Metrologia 38, 497-501 (2001).
[43] C. Tamm, personal communication 2004.
[44] F. Riehle, H. Schnatz, B. Lipphardt, G. Zinner, T. Trebst, and J. Helmcke,
IEEE Trans. Instrum. Meas. 48, 613-617 (1999).
[45] G. Kramer, B. Lipphardt, and U. Hubner, in Joint Meeting of the EFTF and
the IEEE International Frequency Control Symposium (IEEE, Besancon,
1999), Vol. 99CH36313.
224 Chapter 8
[46] H. R. Telle, B. Lipphardt, and J. Stenger, Appl. Phys. B 74, 1-6 (2002).
[47] C. Tamm, D. Engelke, and V. Bühner, Phys. Rev. A 61, 053405 (2000).
[48] P. Cordiale, G. Galzerano, and H. Schnatz, Metrologia 37, 177-182 (2000).
[49] R. K. Shelton, S. M. Foreman, L. S. Ma, J. L. Hall, H. C. Kapteyn, M. M.
Murnane, M. Notcutt, and J. Ye, Opt. Lett. 27, 312-314 (2002); R. K.
Shelton, L. S. Ma, H. C. Kapteyn, M. M. Murnane, J. L. Hall, and J. Ye,
Science 293, 1286-1289 (2001).
Chapter 9
FEMTOSECOND LASERS FOR OPTICAL
CLOCKS AND LOW NOISE FREQUENCY
SYNTHESIS
Scott A. Diddams1, Jun Ye2 and Leo Hollberg1

1
Time and Frequency Division, National Institute of Standards and Technology
2
Abstract: In the late 1990’s mode-locked femtosecond lasers were introduced as an

important new tool for the synthesis and measurement of optical frequencies.
The simplicity, robustness and improved precision of femtosecond lasers have
now led to their prominence in the field of optical frequency metrology. In
addition, their use is developing significant new time-domain applications
based on the precise control of the carrier-envelope phase. It is anticipated
that narrow linewidth lasers referenced to optical transitions in atoms and ions
will soon be the best electromagnetic frequency references of any kind, with
projected fractional frequency instability below 1µ10-15τ-1/2 and uncertainties
approaching 1µ10-18. When used in conjunction with such ultraprecise
frequency standards, the femtosecond laser serves as a broadband synthesizer
that phase coherently converts the input optical frequency to an array of
optical frequencies spanning hundreds of terahertz and to countable
microwave frequencies. The excess fractional frequency noise introduced in
the synthesis process can approach the level of 1µ10-19.
Key words: optical atomic clocks, optical frequency metrology, mode-locked lasers,
frequency comb, optical frequency synthesis
1. INTRODUCTION
Recent developments in time and frequency metrology have brought us

to the point where we can begin to address the possibility of using atomic
clocks “ticking” at optical frequencies (~500 THz) to measure time intervals
226 Chapter 9
with uncertainties approaching 1 part in 1018. This represents a dramatic

shift from the past fifty years of precision time keeping, where atomic clocks
Nonlinear Frequency
Mode-locked Conversion and
Solid State Lasers Spectral Broadening
Optical Frequency Metrology

And Optical Clocks
Hz-linewidth Laser Cooled &

CW lasers High Resolution Trapped Ions
Spectroscopy and Atoms
Figure 9-1. The development of optical clocks and optical frequency metrology has arisen out
of many different sub-fields of atomic, molecular, and optical physics. While optical clocks
have their roots in the precision optical-frequency spectroscopy of cooled and trapped ions
and atoms, the emergence of robust mode-locked solid-state femtosecond lasers and nonlinear
spectral broadening/conversion techniques has provided new opportunities and invigorated
the field as a whole.
have exclusively operated at microwave frequencies, with the 9,192,631,770

Hz ground-state hyperfine transition in cesium-133 being the internationally
recognized definition for the SI second. The jump from microwave to
optical frequencies has been jointly fueled by advances in several branches
of atomic and optical physics, including laser cooling and trapping, precision
spectroscopy, frequency stabilized lasers, nonlinear fiber optics, spectral
broadening and frequency conversion, and femtosecond mode-locked lasers
(see Figure 9-1). Over the past 30–40 years, developments in precision
optical frequency spectroscopy with highly stabilized lasers have formed a
solid basis for the surge of activity in optical frequency metrology since
1999. Pioneers of this field [1-8] recognized that the spectroscopy of cooled
and trapped atoms with narrow linewidth lasers would be valuable for
precision measurements in general, and even 20–30 years ago, prototype
“optical clocks” were being developed [6, 9]. While it was clear that clocks
operating at optical frequencies had much potential for improved stability
and accuracy, a longstanding problem in the development of such clocks has
been a reliable and straightforward means of counting the extremely rapid
9. FEMTOSECOND LASERS FOR OPTICAL CLOCKS AND LOW 227
NOISE FREQUENCY SYNTHESIS
optical oscillations. Solutions to this counting problem involved harmonic

frequency chains [10, 11] at several national facilities or an alternative
approach based on interval bisection [12]. However, both of these
techniques required significant resources and were rather complicated,
meaning it was unlikely they could ever become widespread and reliable. In
1998-99, this scenario was rapidly changed after the group of T. Hänsch
demonstrated that mode-locked femtosecond lasers could play an important
role in optical frequency metrology [13, 14]. With the subsequent
introduction of highly nonlinear optical fibers that could broaden the
femtosecond laser spectrum to bandwidths greater than one octave, all of the
pieces rapidly came together for the implementation of a simple and elegant
single-step phase-coherent means of counting optical frequencies [15, 16].
As illustrated by Figure 9-1 and this brief introduction, the science and
technology that underlie optical clocks are rich and varied. Several excellent
reviews and introductions to this topic already exist [17-22] and others are
found in this volume itself. Instead of attempting to provide a
comprehensive review, we will focus on the role that femtosecond lasers
play in emerging optical clocks and in low noise frequency synthesis. We
will begin by presenting the main components that make up an optical clock
and highlight some of the most relevant detail of the atomic references. The
bulk of this chapter will then describe in detail the role of the femtosecond
laser in optical clocks and low-noise optical and microwave frequency
synthesis. This description will include various methods by which the
femtosecond laser can be controlled as well as some tests that have searched
for inaccuracies in such laser systems when they are used in precision
frequency measurements. An overview of the array of different femtosecond
laser synthesizers that have been developed since 2000 will be presented.
And in conclusion, we will discuss recent experiments focused on the
distribution of optical clock signals through optical fiber networks using
mode-locked lasers.
1.1 Basic components of optical clocks
Most common clocks consist of two major components: an oscillator

that produces periodic events or “clock ticks,” and a counter for
accumulating and displaying each tick [23]. For example, the swing of a
pendulum provides the periodic events that are counted, accumulated, and
displayed via a set of gears driving a pair of clock hands. Similarly, in a
quartz watch, the mechanical vibrations of a small quartz crystal are
electronically detected, accumulated, and displayed to generate time.
However, because of the influence of environmental fluctuations, the nature
228 Chapter 9
of an artifact-based standard makes it very difficult for a quartz oscillator to

remain reproducible at a level of 1 µs per day. Atomic clocks add a third
component that solves this problem: the resonance of a well-isolated atomic
transition that is used to control the oscillator frequency. This is where
precision spectroscopy becomes vital to ensure that the evolution of a clock
signal is ultimately governed by fundamental quantum physics. If the
frequency of the oscillator is controlled to match the transition frequency
between two unperturbed atomic states (i.e., the oscillator is locked to the
atomic transition frequency), then the time generated can have the desired
long-term stability and accuracy. For an atomic clock based on a microwave
transition, high-speed electronics count and accumulate a defined number of
cycles of the reference oscillator to mark a second of time. As sketched in
Figure 9-2, the basic concepts are the same for an atomic clock based on an
optical transition at a much higher frequency. In this case, the oscillator is a
stabilized laser with its frequency locked to an optical transition. However,
no electronic device exists that can directly count the extremely rapid optical
oscillations. Until very recently, devices that could divide an optical
frequency down to a countable microwave (called “frequency chains”) were
complicated, large-scale devices requiring significant resources for operation
[11, 24, 25]. However, as will be described in detail in Section 3, the advent
of femtosecond-laser-based clockwork greatly simplifies this problem of
directly counting the optical frequency.
A tom s νo
Detector
& Laser ∆ν ν
Control
Atom ic Resonance
Frequency
Laser
Counter
O ptical Cavity
Figure 9-2. Schematic diagram of an optical clock. A laser is first stabilized to a Fabry-Perot
optical cavity that provides a means to narrow the laser linewidth leading to good short-term
stability. The center of a narrow resonance in an appropriate atomic sample then provides a
stable reference to which the frequency of the laser can be steered. Once its frequency is
locked to the center of the atomic resonance, a predetermined number of optical cycles are
counted to mark a second of time.
1.2 Uses of optical atomic clocks
While timekeeping is an important aspect of our everyday lives, most of

us do not really care or need to know the exact time to much better than ~60
seconds at any point in our daily routines. Computer networks that record
financial transactions and reservations may require timekeeping that is ~100
times better than this, but still at the modest level of millisecond accuracy.
However, many technological applications, such as communication and
navigation systems (e.g., the Global Positioning System, GPS [26]), are
much more demanding, requiring clocks that are stable and accurate at the
level of 10–100 nanoseconds per day. The most accurate clocks are
ultimately referenced to natural resonances in atoms, and they provide a
unique tool for the scientific exploration of basic atomic structure and the
physical interactions between atoms and their environment. It is necessary
for timekeeping purposes to have clocks that are accurate at the level of 100
picoseconds per day (1 part in 1015), but it is also arguable that clocks with
such performance are equally valuable as basic science research tools. The
next generation optical clocks that are currently being developed will move
beyond this already astounding level of performance, with projected
uncertainties approaching 1 part in 1018. In other words, if such a clock
could run indefinitely, it would neither gain nor lose one second in the
lifetime of our universe.
It is valid and important to ask why scientists, or society for that matter,
might want still better clocks. A brief look back in the history of
timekeeping provides some idea of how we might expect emerging optical
clocks to benefit society in the coming decades. In the 1950s when
microwave atomic clocks based on cesium (and other atoms and molecules)
were first developed, the situation was not so different from where we find
ourselves today. The first atomic clocks were quickly recognized as being
significantly better than the existing mechanical and quartz-based clocks, yet
at the same time, they were mainly a tool of scientific interest. At that point,
few people would imagine that just 30 years into the future, a constellation
of satellites containing cesium and rubidium atomic clocks would
circumnavigate the globe providing accurate time and position to all people
below. The GPS and its constituent atomic clocks are now an integral part
of our lives. While it is difficult to extrapolate to the next 50 years, it is
fully expected that optical clocks will find numerous applications in
advanced communications and navigation systems. For example, the very
stable optical clock ticks may be especially useful for tracking and
communication between satellites and spacecraft in the much greater
expanses beyond our planet. Indeed, stable and precise lasers are envisioned
230 Chapter 9
to provide both laser distance ranging (length metrology) and time keeping
simultaneously for satellites [27].
For the near future, it is already clear that optical clocks and optical
frequency metrology will provide interesting and new scientific avenues to
study our universe—pushing the limits on tests of the most fundamental
physical laws to new levels. Of particular interest is the continued
application of optical frequency standards in spectroscopy and the improved
determination of the fine structure constant α and the Rydberg constant R∞
[14, 28]. As measurement accuracy improves, metrologists may find
themselves in the unique position of being able to observe physical
“constants” evolve in time [29]. Indeed, laboratory tests on the possible
divergence of clocks based on different atomic transitions already provide
some of the most stringent constraints of the variation of α [30, 31].
Experiments of fundamental importance for which precision
clocks/oscillators are of value include searches for variations in the isotropy
of space, a preferred reference frame, and Lorentz and Charge-Parity-Time
(CPT) symmetry violation [32]. Following the recent trapping of cold
antihydrogen at CERN [33], in the coming years it may be possible to
compare optical clocks based on the 1S–2S transitions in both hydrogen and
anti-hydrogen [34]. Such measurements would provide precisions tests of
the fundamental symmetry between matter and antimatter [35].
1.3 A brief history of the development of optical clocks
In the decade following the invention of the laser, it became clear to

many researchers that a stable optical oscillator had great potential for
precision measurements including time keeping. The reader is referred to
the references [24] and the contributions of Chapter 8 in this volume for a
more detailed history of the development of optical frequency standards and
optical clocks. Here, we present only a brief overview of this topic touching
on a few of the most important advances.
Not long after the invention of the laser, efforts were in place to stabilize
its frequency to the center of its Doppler-broadened gain profile. From that
point forward, an increasingly sophisticated set of tools was used to improve
the resolution and fidelity with which a laser’s frequency could probe, and
be stabilized to, a narrow spectroscopic feature in an atom, ion, or molecule.
The proposal to use the divided-down output of a laser-based oscillator
suitably locked to a quantum reference as a time and/or frequency standard
was put forth by pioneers such as Javan [36] and Dehmelt [1]; however, the
difficult task of counting optical cycles was a formidable challenge that
could be addressed by only a few research groups and national laboratories
with sufficient resources. This effort resulted in the so-called harmonic
frequency chains that employed successive nonlinear steps to multiply the

frequency of a cesium-based primary standard up to the IR and then,
ultimately, the optical domain. At NIST (then National Bureau of
Standards), a cesium-referenced frequency measurement of the 88 THz
methane-stabilized helium-neon laser was combined with a krypton-
referenced wavelength measurement of the same stabilized laser to provide a
hundredfold improvement to the measurement of the speed of light. This
was a conclusive demonstration of the value of laser-based frequency
standards, and soon thereafter, the speed of light was defined as a fixed value
such that an optical frequency standard could also serve as an absolute
length standard (e.g., λ = c/f ). The result of this advance cannot be
overstated. The ubiquitous 633 nm helium-neon laser, with its frequency
stabilized to a known transition in iodine, has served for the past 20 years as
the length standard for wide variety of industrial and scientific applications.
The widespread use of atomically stabilized lasers for time standards has
been more difficult to realize. Beyond the serious problem of counting
optical cycles, this is due, at least in part, to the continued improvements and
excellent performance accessibility of cesium-based microwave frequency
standards (the latest developments of cesium standards around the world can
be found in Reference [37]). Nonetheless, in the 1980s at least two groups
developed operational “optical clocks” based on the 88 THz methane-
stabilized helium-neon laser and conventional frequency chains [6, 9]. The
fractional short-term stability for these devices was already in the 10-12
regime where the limitation arose from the microwave standard against
which the optical clock was compared. In the 1990s laser cooling and
trapping techniques were introduced and refined frequency chains were
developed to connect optical transitions to the microwave domain.
Noteworthy in this respect were the efforts surrounding Ca (456 THz) [10]
and Sr+ (445 THz) [11] at the Physikalisch-Technische Bundesanstalt of
Germany and the National Research Council of Canada, respectively.
2. THE ATOMIC REFERENCE
Every optical atomic clock needs a quantum reference, which is most

commonly an electronic transition in an atom, molecule, or ion. As might be
expected, many factors go into the choice of a specific reference, and atoms,
molecules, and ions all have advantages and disadvantages relative to one
another. In this section, we will first describe the benefit one gains by
choosing an optical reference as opposed to a microwave reference. Then
we will spend some time describing references based on single trapped ions,
232 Chapter 9
an ensemble of laser-cooled and trapped atoms, and molecular gases.

Specific examples from laboratory systems will also be provided.
When the center-of-mass motion of the quantum reference for the clock
is well controlled, the coherent interaction time in both optical and
microwave domains can be the determining factor of the spectral resolution.
However, the optical part of the electromagnetic spectrum provides
~100,000 times higher operating frequencies. Therefore, the resonance
quality factor, Q = ν/ν0, of an optical clock transition is expected to be
significantly higher than that available in the microwave domain. (Here, ν0
is the atomic transition frequency and ∆ν is the transition linewidth.) A
superior Q factor provides a more stable frequency standard, and is essential
for making a more accurate standard as well. This is seen in the Allan
deviation σy(τ) which provides a convenient measure of the fractional
frequency instability of a clock as a function of the averaging time τ [38].
For an oscillator locked to an atomic transition, the Allan deviation is
∆ν rms 1 T
σ y (τ ) ≈ ≈ , (1)
ν0 τ
π ⋅ Q τN
where ∆νrms is the measured frequency fluctuation, N is the number of atoms,

and T is the cycle time (i.e., the time required to make a determination of the
line center) with τ >T. This expression assumes that technical noise is
reduced to a sufficiently small level that the quantum-mechanical atomic
projection noise is the dominant stability limit [39]. In this limit, σy(τ)
decreases as the square root of the averaging time for all clocks, so a tenfold
decrease in the short-term instability leads to a hundredfold reduction in
averaging time τ to reach a given stability and uncertainty. This point is
particularly important if one ultimately hopes to reach a fractional frequency
uncertainty of 10-18, which is the anticipated level for ion-based optical
clocks. In this case, an extremely small short-term instability, i.e., σy(τ) ≤
1×10-15 τ -1/2 , is clearly desirable to avoid inordinately long averaging times
(see Figure 9-3).
Since σy(τ) scales as 1/Q, all else being equal, the shift from microwave to
optical frequencies should improve the short-term stability by a factor of 105.
Thus, with a linewidth of about 1 Hz and 106 atoms detected every 0.5
seconds, theoretically, these systems could support an instability σy(τ) º 5 µ
10-19τ -1/2. This simplistic estimate ignores significant complications that will
degrade the performance. Nonetheless, it promises that in the years ahead
there will be plenty of room for improvement using optical frequency
standards. The current status and near-term prospects of the Allan deviation
for optical and microwave frequency standards is shown in Figure 9-3
-13
10
Allan Deviation — Instability — σy (τ)
Cs Fountain Clock
Iodine
-14 (predict) Iodine
10
-15
10
Optical Calcium
-16
10 Cavity
Ca, Sr Hg+ (predict)
-17
(predict)
10 1 day 1 month
-2 0 2 4 6
10 10 10 10 10
Averaging Time (s)
Figure 9-3. Fractional frequency instability as characterized by the Allan deviation of a

microwave frequency standard (cesium fountain clock) and several promising optical
frequency standards. The plotted curves are based on measured data, except where predicted
curves are given as dashed lines.
2.1 Single ion references
Trapped ions, particularly single laser-cooled ions, have numerous

advantages as optical frequency standards and clocks [1, 40]. Perhaps one of
their most important attributes is the nearly ideal independence between the
trapping potential for the center-of-mass motion and the internal atomic level
structure that provides the clock reference. Importantly, ions can be confined
in an rf trap and laser-cooled so that the amplitude of the residual motion is
much less than the optical wavelength of the probe radiation (the Lamb-
Dicke limit). This nearly eliminates the velocity-dependent Doppler
broadening and shifts associated with motion of the ion relative to the
probing radiation. In a cryogenic environment, the ion is nearly unperturbed
by atomic collisions, and the effects of blackbody radiation are also very
small. The storage time of a single ion in a trap can be months; hence, the
probe interaction time is usually constrained by other technical issues such
as the laser coherence time or, fundamentally, by the natural lifetime of the
transition under study. This permits extremely high-Q resonances to be
234 Chapter 9
observed. All of these factors are critically important for the achievement of
the highest accuracy.
The technical challenges of making an optical frequency standard based
on a single ion are formidable but single-ion standards have now been
achieved in a handful of laboratories around the world. At NIST, an optical-
frequency standard based on a single trapped 199Hg+ ion is being developed.
The performance of this standard is immediately competitive with the
performance of the best microwave standards and has the potential to surpass
those standards in terms of stability, frequency reproducibility, and accuracy.
For this standard, a single 199Hg+ ion is trapped in a small rf Paul trap (≈1
mm internal dimensions) and laser cooled to a few milliKelvins using 194
nm radiation. A highly stabilized dye laser at 563 nm with a linewidth of
less than 1 Hz is frequency doubled to 282 nm (1,064 THz) to probe the
clock transition [41]. Measured linewidths as narrow as 6.7 Hz on the 282
nm transition have been reported [42]. For an averaging time τ in seconds,
the projected instability of an optical frequency standard using a single Hg+
ion is 1 x 10-15τ−1/2, and fractional frequency uncertainties approaching
1x10-18 seem feasible.
One apparent difficulty of modern optical standards is the requirement
that suitable transitions for laser cooling, fluorescence detection, and the
clock itself be present in the same atom or ion. While some atoms have
good clock transitions, their cooling and detection transitions might be less
than ideal. Wineland et al. have proposed an efficient solution to this
problem in the case of ion-based clocks that involves simultaneously
trapping a “clock” ion and a “logic” ion [43]. Using techniques developed
for quantum computation applications [44], the logic ion would provide the
functions of sympathetic cooling and detection, leaving more flexibility in
choosing the best clock ion with a narrow, unperturbed and accessible
transition. This approach is currently be implemented at NIST using 27Al+
as the clock ion and 9Be+ as the logic ion.
2.2 Neutral atom references
Some neutral atoms also have narrow optical transitions that are
relatively insensitive to external perturbations and are thus attractive as
optical frequency standards. Neutral atoms have some advantages and
disadvantages relative to ions. Using the well-established techniques of laser
cooling and trapping, they are fairly easy to confine and cool to low
temperatures. However, in contrast to ions, the trapping methods for neutrals
perturb the atomic energy levels, which is unacceptable for use in a
frequency standard. To avoid the broadening and shifts associated with the
trap, neutral atoms are released from the trap before the clock transition is
probed. The atoms fall from the trap under the influence of gravity and
expand with low thermal velocities (typically a few cm/s). The resulting
atomic motion brings with it limitations in accuracy (and even stability) that
are associated with velocity-dependent frequency shifts. Two of the more
troublesome effects are the limited observation time and the incomplete
cancellation of the first-order Doppler shift associated with wave-front
curvature and wave-vector mismatch. Reduced observation times limit the
line Q, the stability, and the accuracy. However, neutral atoms do have at
least one significant advantage: large numbers of atoms can be used,
producing a large signal-to-noise ratio (SNR) in a short time and the potential
for exceptional short-term stability.
The atomic Ca optical frequency standard [45, 46] is one of the most
promising and extensively studied cases. It has a 400 Hz wide clock
transition at 657 nm (1S0 – 3P1) that is reasonably immune from external
perturbations. It is readily laser cooled and trapped, and it is experimentally
convenient because the relevant transitions are accessible with tunable diode
lasers. Cooling and trapping of about 107 Ca atoms to mK temperatures can
be accomplished in a magneto-optic trap (MOT) with frequency doubled
diode laser tuned to the 423 nm 1S0 – 1P1 transition. With the cooling
radiation turned off, an injection-locked and stabilized diode laser at 657 nm
(456 THz) then probes the clock transition with the separated excitation
method of optical Ramsey-Bordé spectroscopy. Optical fringes with high
signal-to-noise ratio are observed using shelving detection on the cooling
transition. With this technique, the present Ca standard can provide short-
term fractional frequency instability of about 4 x 10-15 in 1 s of averaging.
Second stage cooling on the narrow 1S0 – 3P1 to temperatures ~10 µK has
been achieved with the aid of 552 nm light that quenches (i.e., depopulates)
the long-lived 3P1 state [47]. The 10-µK temperatures reduce velocity-
related systematic shifts, and it appears that uncertainties at or below 10-15
will be attainable with such a system.
Ultimate neutral atom-based systems with high accuracy will demand a
stringent separation between the external degrees of freedom (controlled by
the trapping potential) and internal level structure (clock transitions), similar
to that obtained with single trapped ions. A crucial step towards high
reproducibility and accuracy is to confine neutral atoms in the Lamb-Dicke
regime, while at the same time limiting the effect of the confining potential
to only the external degrees of freedom [48]. The fermionic isotope of 87Sr
has a nonzero nuclear magnetic moment (I = 9/2) that gives rise to magnetic
substructure in both the ground and excited states [49]. Moreover, 87Sr
possesses a doubly forbidden J = 0 to J = 0 clock transition, 1S0 (F = 9/2) –
3
P0 (F' = 9/2), which has a ~1mHz linewidth (corresponding to an intrinsic
236 Chapter 9
line Q of 4.3x1017) and is expected to be highly insensitive to external

electromagnetic fields and collisional shifts [50]. The Sr atoms are first
cooled in two stages to temperatures <1 µK [51]. These pre-cooled atoms
can then be efficiently transferred into a far-off-resonance optical lattice trap
where sideband cooling can bring the atoms to the motional ground state.
Using a magic wavelength (~800 nm) for the latter trap ensures that both the
ground and excited (3P0) state experience the same energy shift anywhere
inside the trap. The sensitivity to the light polarization is reduced because of
the fact that J = 0 for both clock states. The 87Sr 1S0 (F = 9/2) – 3P0 (F' = 9/2)
transition can then be interrogated free from broadening and shifts due to
Doppler, recoil, and trapping potential related shifts.
2.3 Molecular references
The abundance and relatively convenient gas-cell spectroscopy of

molecular resonances make them attractive references for optical clocks as
well. Indeed, IR transitions in molecules such as CH4, CO2, and OsO4 were
some of the first references explored [25, 52] and continue to be used in
several research programs around the world [53]. Their natural linewidths
range from a few megahertz to below a kilohertz, limited by molecular
fluorescent decay. Useable linewidths are usually ≥10 kHz due to the transit
of molecules through the light beam. Transitions to higher levels of these
fundamental rovibrational states, usually termed overtone bands, extend
these rovibrational spectra well into the visible with similar ~ kHz potential
linewidths. Until recently, the rich spectra of the molecular overtone bands
have not been adopted as suitable frequency references in the visible due to
their small transition strengths. However, with one of the most sensitive
absorption techniques, which combines frequency modulation with cavity
enhancement, an excellent SNR for these weak but narrow overtone lines can
be achieved, enabling the use of molecular overtones as standards in the
visible [54].
Perhaps the most widespread example of a molecular reference is the
iodine molecule. The narrow-linewidth I2 transitions in the visible
wavelength region have provided excellent cell-based optical frequency
references for laser stabilization. Frequency-doubled Nd:YAG/127I2 at 532
nm has proved to be one of the best portable optical frequency standards
with compact size, reliability, and high stability (<5 x 10-14 at 1 s) [55]. To
reach a better frequency stability, it is useful to explore I2 transitions towards
the dissociation limit at wavelengths below 532 nm where the natural
linewidths decrease at a faster rate than the line strengths. The systematic
variation of the I2 transition linewidths within the range of 498–532 nm has
been measured, with the linewidth decreasing by nearly sixfold when the
wavelength is decreased from 532 nm to near the dissociation limit [56]. The
high SNR results indicate that I2 transitions in the wavelength range 501–
532 nm hold great promise for future development of optical frequency
standards, especially with the advent of all-solid-state Yb:YAG lasers. One
exciting candidate is the 514.67 nm standard that has a projected instability
<10-14 at 1 s [57].
Table 9-1. Some of the promising optical frequency references for emerging and future
optical clocks. Presently achieved values of instability and uncertainty are listed, while the
projected values are given in parentheses. The research institute is given in the last column.
Frequency 1-s Fractional Fractional
Institute
(THz) Instability Uncertainty
Ions
Hg+ 1064 3µ10-15 1µ10-14 NIST
(1µ10-15) (<1µ10-17)
In+ 1267 2µ10-13 MPQ
Sr+ 445 5µ10-15 NRC, NPL
Al+ 1124 NIST
Yb+ 688 9µ10-15 PTB
(<1µ10-17)
Yb+ 642 2µ10-12 NPL
Neutral
Atoms
H 2466 2µ10-14 MPQ
Ca 456 4µ10 -15
2µ10-14 PTB, NIST
(<2µ10-16) (<1µ10-15)
Sr U. Tokyo,
JILA, SYRTE,
LENS
Yb KRISS, NIST
Molecules
OsO4 29 3µ10-13 Paris Nord
CH4 88 Lebedev, NPL,
PTB,
Novosibirsk
I2 563 4µ10-14 9µ10-12 JILA, PTB,
NMIJ, BIPM
238 Chapter 9
2.4 Local oscillator requirements
While it is necessary to start with a suitable atomic transition, whether it

be in an ion, atom, or molecule, as diagramed in Figure 9-2, another
important element of an optical clock is the local oscillator. For an optical
clock, the local oscillator is comprised of a continuous wave (cw) laser that
has its emission spectrum narrowed and stabilized to an isolated high-Q
Fabry-Perot optical cavity. The optical cavity must have sufficiently good
short-term stability to permit interrogation of a narrow spectroscopic feature
in the quantum reference (i.e., ion, atom, or molecule) with a good SNR. For
example, in the case of the single Hg+ ion, the optical transition linewidth is
just a few hertz wide, which requires a local oscillator with a sub-hertz
linewidth. Significant effort has been invested for the past few decades in
the reduction of the laser linewidth down to this sub-hertz level [41, 58]. A
key aspect in this development has been improved techniques for the
mechanical and thermal isolation of the Fabry-Perot optical cavity from the
surrounding laboratory environment [41]. A more recent advance in this
evolution focuses on reducing the acceleration sensitivity of the cavity
through geometrical design and by careful choice of how the cavity is
supported.
3. FEMTOSECOND LASER-BASED OPTICAL

FREQUENCY SYNTHESIZERS
The possibility of using a mode-locked laser as a tool for optical

frequency metrology was first demonstrated with picosecond lasers by
Hänsch and co-workers in the late 1970s [59]. The essence of this original
idea was to use the comb of frequencies emitted from a mode-locked laser as
a precise “optical frequency ruler.” The spacing of the tick marks of such an
optical frequency ruler is given by the repetition rate fr at which pulses are
emitted from the mode-locked laser, while the differential phase shift
∆φce between the pulse carrier and the pulse envelope each round trip
determines the overall offset of the comb elements f0 = fr (∆φce/2π). The
relationship between these two parameters and the nth element of the optical
frequency comb is given by the simple expression
νn = nfr + f0 . (2)
This technique lay largely dormant until experiments in 1998

demonstrated that the frequency comb associated with a femtosecond mode-
locked laser could be readily controlled and provided a more versatile and
precise method for counting optical frequencies than existing technologies

[13]. In fact, in just a matter of years, the new femtosecond comb
technology has fully replaced laboratory efforts that existed for decades; it is
now widely accepted that mode-locked femtosecond lasers will play a
critical role in the next generation of atomic clocks based on optical
frequencies [46, 60, 61]. Moreover, due to the simplicity and relatively low
cost, the techniques and tools described here can be implemented in
university and industrial research laboratories, which has resulted in a
variety of interesting and exciting new avenues of research [62-64].
The absence of robust solid-state femtosecond lasers likely contributed to
the twenty-year delay between the first conceptual experiments and practical
implementation of the frequency comb concept. Although the femtosecond
dye laser was developed throughout the 1970s and 1980s, it was not until the
1990s that robust and high-power femtosecond lasers based on titanium-
doped sapphire (Ti:sapphire) were perfected [65, 66]. (Chapter 2 of this
volume presents a detailed account of development of solid-state mode-
locked laser technology.) For comb generation, it was natural that
femtosecond Ti:sapphire would become the source of choice. When
combined with nonlinear microstructure and tapered fibers [67], octave-
spanning spectra could be generated. A few representative Ti:sapphire laser
configurations and resulting spectra are shown in Figure 9-4. It is important
to realize that the array of frequency modes given by Equation (2) reside
beneath the broad spectral envelopes shown in this figure (which were
recorded with a low resolution grating spectrometer). As will be described
in further detail below, such a simple means for generating very broad
spectra allows for the straightforward measurement and control of f0 in
addition to providing a network of useful comb lines spanning the visible
and near-IR.
In this section, we will not dwell on the actual generation of the
frequency comb from a femtosecond laser. The reader is referred to several
excellent review articles on this topic [17, 18, 21], in addition to the other
chapters in this book. Instead, we will explain the details of how a
femtosecond laser is actually used in an optical clock and for low noise
frequency synthesis. We first spell out some of the specific metrological
considerations that arise in the use and choice of a femtosecond laser for
optical clocks and synthesis applications. Next, we discuss the “mechanical”
details of actually using and controlling a femtosecond laser. We will then
present what is currently known about the achievable noise properties of the
laser itself. Finally, while Ti:sapphire femtosecond lasers continue to be the
most widely used options for optical clocks and frequency metrology, fiber-
based femtosecond laser sources and alternative solid-state lasers that can be
240 Chapter 9
efficiently pumped offer significant advantages in terms of size and cost.

We will spend the last part of this section looking at these sources and the
benefits they might have for future optical clocks.
3.1 Considerations in designing a femtosecond comb for

use in an optical clock
The first experiments [13, 15, 16] verifying the valuable contribution of
femtosecond lasers to optical frequency metrology all employed
femtosecond lasers that had been developed over the previous decade for
experiments in ultrafast science [66]. While such lasers worked very well
for the initial proof-of-principle experiments, and continue to be used in
many cases, they lacked other desirable qualities for frequency metrology.
The ultimate femtosecond laser for optical frequency metrology and optical
clocks is likely to be a continuously evolving device, but nonetheless, in this
section we attempt to lay out some of the desirable characteristics of these
devices as they relate to their use in frequency metrology and optical clocks.
1
10 (c)
-1
(a) 10
-3
P rism P air 10
Power per mode (microWatt)
Pum p Ti:S 10
-5
1
10 (d )
-1
10
-3
10
-5
(b) 10
Pump 1
600 800 10 00 120 0
Ti:S 10 (e )
-1
10
-3
10
-5
10
600 800 10 00 120 0
W avelength (nm )
Figure 9-4. Common Ti:sapphire femtosecond lasers used for frequency metrology and their
output spectra. (a) Conventional linear cavity design employing a combination of chirped
mirrors and prisms produces the broadband spectrum shown in (c). The ring laser shown in
(b) uses only chirped mirrors and can produce the spectrum show in (d). The spectrum shown
in (e) is produced using either a laser (a) or (b) in combination with nonlinear microstructure
optical fiber.
Repetition rate: There are a few factors that drive the consideration of
the best repetition rate. The most obvious feature is that for a constant
average output power from the laser, the mode comb of a high repetition rate
laser will have a proportionately higher power per mode. This is a clear
advantage for frequency domain applications where we rely on a good SNR
in the heterodyne beats between a cw laser and an individual tooth of the
comb itself. However, one cannot take this point to the extreme. For
example, if we again assume a constant output power, as the repetition rate
increases, the power per pulse decreases. This is an important consideration
if one is to rely on nonlinear processes to broaden the optical spectrum to an
octave. As a rule of thumb, with the typical microstructure fibers used with
Ti:sapphire lasers around 800 nm, one needs roughly a few hundred
picojoules of energy in a ~50 fs pulse to obtain an octave of spectral
broadening. This can be achieved with a 1 GHz repetition rate laser, but
given the same average power, it is not likely to be possible if the laser could
operate with a 10 GHz repetition rate.
Photodetectors & electronics: Since many uses of the femtosecond
laser comb ultimately involve a connection to the rf/microwave domain, the
photodetectors that provide this connection as well as the subsequent
electronics that are used must also be a consideration. While photodetectors
and microwave electronics with bandwidths up to ~50 GHz are
commercially available, devices that operate much above 10 GHz tend to be
more costly, more difficult to use, and less sensitive. This last factor can
actually result in a loss of SNR that one had hoped to achieve precisely by
choosing a higher repetition rate in the first place. This can be particularly
true with the photodetector that is used to detect the optical clock output at fr,
where extremely high SNR is required to generate stable electrical signals
[68]. Another important electronics-related consideration surrounding the
choice of repetition rate is the frequency one might ultimately choose for
connecting the fr output of the femtosecond laser to existing rf/microwave
sources and standards. The best choice of low-noise electronic synthesizers
that would divide or multiply fr up or down to common frequencies of 5, 100
or 9,192.631770 MHz will certainly be a consideration for the most
demanding applications.
Size, simplicity & robustness: While size, simplicity, and robustness
are not necessarily synonymous, it is often true that a smaller device of
simple construction can be more robust. For example, small size lends itself
to improved temperature stability, which directly impacts the operation of
any femtosecond laser and the required dynamic range for the various servo
controls. This is a point in favor of higher repetition rate systems, or at least
systems with smaller footprints. When it comes down to the actual design of
the laser, this factor would also tend to favor the use of chirped mirrors or
fully integrated optics (such as a fiber laser) over the use of prisms for
242 Chapter 9
dispersion control. The reliable generation of octave-spanning spectra

directly from the laser is also a significant simplification as it allows one to
eliminate sensitive microstructured nonlinear fibers from the apparatus.
Spectral coverage: A study of the different atoms, ions, and molecules
used in emerging optical-frequency standards (see Table 9-1) reveals that
many of the atomic transitions of interest fall in the 500–1100 nm range (or
harmonics thereof). Thus, Ti:sapphire-based femtosecond laser combs
centered at 800 nm are a good starting place and direct continuum generation
or subsequent broadening in nonlinear microstructure fibers provides overlap
with most of the wavelengths that are presently of interest. While nonlinear
microstructure fibers have performed amazingly well in all experiments to
date, they present challenging technical limitations on the ability to control
the femtosecond laser over long periods. Coupling of the light into their
~1.7 µm diameter cores is difficult to maintain for long times at a level
where an octave of spectrum is attained. An attractive alternative is recently
developed femtosecond lasers that directly emit octave-spanning spectra
suitable for self-referencing [69-71]. These lasers have slightly less
bandwidth than is typically achieved from nonlinear microstructure fibers;
however, initial experiments indicate the broadband lasers can be much more
robust [72, 73]. A full understanding and optimization of these extremely
broadband lasers is yet to emerge, but we anticipate they will play a very
important role in optical clocks, perhaps reducing the need for nonlinear
microstructure fibers.
Noise: Perhaps the most important consideration for any synthesizer is
its residual or excess noise. More specific details on the presently achieved
noise levels will be presented below, but from a general design point of
view, there are important considerations that should be pointed out with
regards to systems that employ nonlinear microstructure fibers. First, it is
important to construct a femtosecond laser with robust and stable mechanical
configurations and environmental shielding. Understanding the laser
dynamics and noise sources is also fundamental to design and
implementation of stabilization feedback loops [74, 75]. Because of its
highly nonlinear nature, microstructure fiber can significantly amplify both
technical and fundamental shot noise on the input light [76, 77]. With
careful attention, the most significant aspects of the technical noise can be
reduced; however, the shot noise on the laser output will always be present
and potentially represents a more formidable problem. Of particular
significance is the finding that longer pulses and chirped pulses generate
more noise when launched into nonlinear microstructure fibers [77, 78].
For example, 75 fs pulses containing a few nanojoule of energy can make a
stunning continuum stretching from 450 to 1400 nm, but excess broadband
noise makes it nearly useless for precision frequency metrology.
3.2 Frequency synthesis with a femtosecond laser
The basic function of the femtosecond laser in an optical clock is to

provide a phase-coherent link between the uncountable optical reference
frequency and the more accessible microwave domain. Using a mechanical
analogy, we can imagine the femtosecond laser to be a one-step reduction, or
multiplication, gear. Alternatively, one can discuss the femtosecond laser
as being an extremely broadband optical frequency synthesizer, with general
properties not so different from commonly used rf and microwave
synthesizers. While the mechanical analogy is straightforward conceptually,
the synthesizer picture is the more useful one as it allows us to employ
rigorous characterization techniques developed in the rf and microwave
domains.
optical
µ-wave out
reference Optical Synthesizer
1 3
I(f)
f0
2 fr 4
µ-wave
0 νn = nfr + f0
f optical out
reference
Figure 9-5. The femtosecond laser-based optical frequency synthesizer.
As shown in Figure 9-5, the femtosecond laser-based synthesizer can

employ either a microwave (µ-wave) or optical input reference that is phase
coherently linked to an optical or microwave frequency at the output. These
operations can all be carried out and understood by considering the
underlying frequency comb structure associated with the femtosecond laser.
The first uses of the femtosecond laser-based synthesizer involved supplying
a microwave reference at port 2 and obtaining coherently related optical
frequencies in the form of the optical comb of Equation (2) at port 4 [15, 16,
79]. This path relies on the femtosecond laser to effectively multiply fr by
approximately a factor of 106 to reach the optical domain. In principle, the
multiplicative factor could be a much smaller value, on the order of 102 or
103, to instead yield microwave frequencies in the hundreds of gigahertz (or
even terahertz range). This potentially interesting use of the synthesizer
Ø
(ports 2 3) has so far received less attention. When used to down-convert
Ø
an optical frequency input to a microwave output (ports 1 3), the frequency
244 Chapter 9
of the output is given by the repetition rate fr or one of its harmonics [46, 60,
61]. The synthesizer can also function to effectively synthesize new optical
frequencies that are displaced from the input optical reference by a freely
Ø
chosen offset plus-or-minus harmonics of fr (ports 1 4) [80].
3.2.1 Methods of control
Control of f0 : To accomplish all of these operations, it is required that at

least two of the three variables given in Equation (2), in addition to n, be
measured and controlled. For example, one might control fr and f0 and use νn
as an output. Alternatively, Equation (2) can be inverted such that one tooth
(n = N) of the comb νN and f0 are controlled and fr is then the output.
Whereas it is relatively straightforward to measure the repetition rate fr by
monitoring the pulse train with a sufficiently fast photodetector, it is more
challenging to measure the offset frequency f0 of the comb. In principle,
once fr is measured, it is sufficient to use a cw laser with well-known
frequency to measure and stabilize f0. While this might be practical in some
situations, it doubles the number of required frequency standards (one for f0
and a second for fr) required for using the synthesizer and potentially limits
the performance.
A more elegant means of measuring and controlling f0 involves using
nonlinear frequency generation to compare different regions of the frequency
comb [81]. As shown in Figure 9-6(a), if the laser spectrum covers more
than one octave, then the comb elements at the low-frequency end of the
spectrum can be doubled in a nonlinear crystal and subsequently
heterodyned against the high-frequency components of the comb to yield
2(nfr + f0) – (mfr + f0) = f0 when n = m/2. Spectral broadening of low power
Ti:sapphire lasers to more than an octave is accomplished through self-phase
modulation (SPM) in microstructure fibers [67]. Such an approach was first
demonstrated by Jones et al. [16] and continues to be the most common
manner in which to measure f0. This so-called “ν-to-2ν” self-referencing
scheme has also been demonstrated with low-power Cr:LiSAF [82] and Er-
doped fiber lasers [83]. It has also been shown that “ν-to-2ν” self-
referencing can be accomplished with Ti:sapphire lasers that directly emit an
octave-spanning spectrum [71, 84]. As summarized by Telle et al. [81], it is
not necessary to have an octave-spanning spectrum from the femtosecond
laser to measure f0. The price to be paid, however, is that extra steps of
nonlinear conversion must be employed. For example, with a spectrum
spanning 2/3 of an octave, f0 can be obtained through the comparison of the
second harmonic and third harmonic of separated portions of the optical
spectrum, e.g., 3(nfr + f0) – 2(mfr + f0) = f0 when n = 2m/3 [72, 84, 85].
Pump Power or Phase

Tilting Mirror
Detector
f0
νn=f0+nfr x2 ν2n=f0+2nfr
fr
(a) f
Pump Power or Phase

Tilting Mirror Detector
f0 = fb2 – 2fb1
fb1 fb2
(b) f
νs 2νs
Phase Ref.
PZT
Detector fr or n fr
(c) f
Phase Ref.
PZT
Detector
fb = ν N - ν s
(d) f
νs
Figure 9-6. Techniques for measuring and controlling f0 (a, b) and fr (c,d). See text for
details.
Once f0 has been measured, it is typically phase-locked to a stable rf

source or to a rational fraction of fr itself. Such a phase-lock requires an
actuator inside the laser cavity that provides a differential change to the
group and phase velocities of the circulating field. Two common modes of
246 Chapter 9
control have so far been used. The first employs modulation of intracavity
laser power [79, 86], and the second (which is applicable in linear cavities
employing prism pairs) involves the tipping of one of the cavity mirrors [18,
87]. While the merits and physical mechanisms of these differing techniques
remain a topic of discussion, both have been successfully implemented to
reduce the residual fluctuations of f0 to the millihertz level relative to a stable
reference. When f0 is phase-locked to D.C. or a harmonic of fr, precise
stabilization has the consequence of maintaining the relative carrier-
envelope phase at a fixed value for times approaching one hour [88]. This
has enabled the subsequent amplification and use of such lasers for
experiments in phase-sensitive extreme nonlinear optics experiments (see,
for example, Reference [64] and Chapter 10 of this volume).
Control of fr: As already mentioned, fr can be detected with a sufficiently
fast photodetector. The rf spectrum of the resulting photocurrent consists of
a series of harmonics of fr, one of which can be filtered and phase-locked to
a low-noise rf source (Figure 9-6(c)). In this case, it is common to use a
piezo-mounted cavity mirror as an actuator for changing fr. A more careful
consideration of Equation (2) highlights some of the difficulties and
disadvantages of controlling fr in the rf/microwave domain as just described.
Because the optical frequency νn scales as the multiplicative mode number n,
the spectral density of phase noise on the rf reference source to which fr is
phase-locked is multiplied by a factor of n2. For a mode in the optical
domain with n º500,000 (fr = 1 GHz), this implies that the phase noise of the
rf reference is increased by 115 dB when it is effectively multiplied up to the
optical domain using the femtosecond laser comb. A high-quality quartz-
based rf reference at 1 GHz might have a typical noise floor of –110 dBc/Hz,
but when it serves as the reference for fr, there will be nothing remaining of
the phase-coherent carrier in the optical mode νn. This is similar to the well-
known problem of “carrier collapse,” which places extreme demands on the
rf reference (as well as all intermediate electronic components) for fr.
There are, however, a few ways to minimize this phase-noise
multiplication problem. First of all, one can rely on the relatively good
short-term stability of the femtosecond laser itself. On time scales less than
~1ms, the noise of a typical Ti:sapphire laser has been measured to fall
below that of most high quality microwave sources [20, 68]. This means
that if one controls fr relative to a microwave reference, a control bandwidth
of §1 kHz is all that is required to remove the low-frequency thermal and
acoustic noise of the mode-locked laser. Use of higher bandwidth will
simply transfer the noise of the microwave reference to fr, which will
subsequently be multiplied up to the optical comb elements. Although the
observed optical linewidth may still be on the order of 0.1–1 MHz, one
generally finds that the fractional stability of the optical comb elements can
be equal to the fractional stability of the microwave reference. Of course, if

a microwave reference for fr with lower phase noise is available (such as a
cryogenic sapphire microwave oscillator), then one can potentially narrow
the optical linewidth further.
However, in the near future it is likely that the lowest-noise
electromagnetic oscillators of any kind will be based on cw lasers referenced
to stable Fabry-Perot optical cavities and/or atomic or molecular resonances.
In fact, cw lasers with sub-hertz linewidths have recently been demonstrated
[41], allowing the use of such a source to control fr in the optical domain.
-80
High Quality Quartz
-100
L(f) [dBc/Hz]
-120 Commercial Synthesizer
-140 Ca Optical
Sapphire Resonator
-160
-180
Hg+ Optical
Cavity
-200
0 1 2 3 4 5
10 10 10 10 10 10
Frequency [Hz]
Figure 9-7. Phase noise of various oscillators and synthesizers at 1 GHz. The quartz curve
corresponds to a high-quality 5 MHz oscillator and assumes noise-free multiplication to 1
GHz. The commercial synthesizer is residual synthesizer noise that would be added to the
reference oscillator (typically quartz). The sapphire resonator curve represents a state-of-the-
art commercial product with its frequency divided down from 10 GHz to 1 GHz. The dotted
Ca curve is a projection of what should be achievable, and the Hg+ optical cavity is estimated
from experimental data [41].
As seen in Figure 9-6(d), this is typically done by making an optical-

offset phase-lock between one tooth (mode N) of the femtosecond comb (νN)
and the stable cw reference laser νS. The resulting expression for fr is
fr = (νS - f0 + fb)/N , (3)

248 Chapter 9
where fb is the arbitrary (yet phase-locked) beat frequency between νS and

νN. This expression illustrates that any frequency fluctuations in νS are
divided by N before appearing in fr (spectral density of phase fluctuations are
divided by N2). Since both f0 and fb are phase-locked microwave frequencies
<1 GHz having fluctuations well below 1 Hz, the fluctuations of νS will
likely dominate the noise of fr. Assuming a perfect frequency divider, the
anticipated limits to the phase noise spectral density of a sub-hertz laser (Hg+
optical cavity) and the Ca optical standard are compared to other low-phase-
noise sources in Figure 9-7. Clearly, the phase noise of the optical standards
offers significant potential in this regard. At present, the noise from the
femtosecond laser and the optical detection of fr (1 GHz) has been measured
to be -120 dBc/Hz at 1 Hz offset, decreasing to ~ -155 dBc/Hz at 1 kHz [89].
It has been shown that the phase noise on the optical comb of the
femtosecond laser itself can be significantly less than this level [90], leading
to the conclusion that the dominant noise sources for optical-to-microwave
conversion with the femtosecond laser synthesizer are introduced in
photodetection of fr and subsequent electronic processing and measurement.
Alternative control methods: Instead of frequency-doubling a portion
of the octave spanning comb spectrum as described above, one can use a cw
laser and its second harmonic to access and control f0 and fr. [91]. This is
diagramed in Figure 9-6(b). Measurement of the heterodyne beat between
the cw laser frequency, νs, and the comb line n gives f b1 = ν s − (nf r + f 0 )
and between the second harmonic of the cw laser and comb line 2n gives
f b2 = 2ν s − (2nf r + f 0 ) . Mixing the beats with appropriate weighting factors
gives f b2 − 2 f b1 = 2ν s − (2nf r + f 0 ) − (2ν s − 2(nf r + f 0 )) = f 0 , which
represents another detection scheme of f0. On the other hand, it is now also
possible to establish a direct phase coherent link between optical and
microwave frequencies without the need of stabilizing f0. The following
expression, f b2 − f b1 = 2ν s − (2nf r + f 0 ) − (ν s − (nf r + f 0 )) = ν s − nf r ,
illustrates this principle. After appropriate processing, this error signal is
used to stabilize the phase of fr coherently to νs , thereby producing an output
clock signal in the rf domain derived from νs [61].
There are several other methods for removing the dependence of the
frequency comb on f0. For example, by using difference-frequency
generation (DFG) between different spectral portions of the same frequency
comb, a DFG comb independent of f0 in the IR spectral region with excellent
accuracy and stability can be generated [92, 93]. Furthermore, a DFG comb
(again, independent of f0) tuned to 800 nm might also pave the way to an all-
optical carrier-envelope phase-stabilization scheme by re-injecting this DFG
comb into the femtosecond laser cavity. Equivalent to the DFG approach,
sum-frequency generation (SFG) between an IR optical frequency standard
(e.g., a methane-stabilized He-Ne laser at 3.39 µm) with the long-

wavelength portion of a frequency comb yields an SFG comb that can be
tuned to spectrally overlap with the short-wavelength portion of the original
frequency comb. The resultant heterodyne beat between the two combs of
the same origin leads again to a carrier-envelope frequency-independent
optical clockwork. We have recently implemented such an optical clockwork
based on SFG between a cw optical parametric oscillator (OPO) at 3.39 µm
idler wavelength and a custom-tailored mode-locked Ti:sapphire laser with
two dominant spectral peaks at 834 and 670 nm. No additional single-
frequency lasers phase-locked to the Ti:sapphire comb are necessary.
Finally, it is important to note the clever “transfer oscillator” [94, 95]
concept in which a judicious choice of frequency-ratio mixing effectively
eliminates the noise properties of the unstabilized femtosecond laser when it
is used to determine the ratio of widely separated optical frequencies or even
optical and microwave frequencies. An advantage of this technique is that it
allows one to replace the more difficult and relatively slow (<100 kHz)
control of the femtosecond laser with more straightforward and faster (~1
MHz bandwidth) phase-tracking oscillators on the various heterodyne beats.
3.3 Testing the synthesizer
As might be expected, the introduction of mode-locked femtosecond

lasers into the field of optical frequency metrology was met with some initial
questions about how well such a new technology could perform relative to
what existed. Such healthy skepticism has led to some interesting and
valuable tests of Equation (2). Table 9-2 provides a summary of some of the
relevant experimental tests of the femtosecond laser synthesizer. It is not
always straightforward to understand precisely what is being tested in some
of these experiments, but in general they can all be viewed as placing
simultaneous constraints on fr and f0.
The most obvious question one can ask about Equation 2 is whether or
not the comb of optical frequencies is indeed uniform or evenly spaced. The
time domain perspective of how the mode-locked laser functions leads one
to the conclusion that this must be the case. An uneven spacing of the
modes would imply that different portions of the spectrum experience
different roundtrip delays in the cavity. Were this the case, the pulse would
rapidly spread and break apart, which is not consistent with the solitonlike
operation of mode-locked lasers. The first frequency-domain verification of
this operation was offered by the group of T. Hänsch, by comparing a 44
THz wide comb from a femtosecond laser with an optical interval divider,
thus confirming the uniformity at the level of a few parts in 1018 [13, 96].
250 Chapter 9
While these impressive experiments can also be viewed as a test of the

uniformity of f0, one apparent weakness is that they employed only a single
femtosecond laser and did not verify that the actual frequency position of the
modes of the comb could be controlled or reproduced at this level.
Table 9-2. Summary of various tests of the femtosecond laser synthesizers

What was tested Fractional Systems Used Reference
Uncertainty
Uniformity of fr 3µ10-18 fs laser broadened to 44 [13]
THz in standard fiber [96]
Microwave to 5.1 µ 10-16 ν-to-2ν fs synthesizer [79]
optical frequency with 3.5ν-4ν frequency
synthesis chain
Microwave to 1.6 µ 10-12 ν-to-2ν fs synthesizer and [97]
optical frequency harmonic frequency chain
synthesis
Optical-to-optical 4 µ 10-17 Two ν-to-2ν fs [98]
frequency synthesis synthesizers
Optical-to-optical 7 µ 10-19 Single fs laser and cw [94]
ratio and SHG laser +SHG
Microwave to 1 µ 10-15 Two ν-to-2ν fs [99]
optical frequency synthesizers
synthesis
Uncertainty of DFG 6.6 µ 10-21 Single fs laser and DFG [92]
relative to and SFG
fundamental comb
Optical-to-optical 1.4 µ 10-19 Two ν-to-2ν and two [100]
frequency synthesis 2ν-to-3ν fs synthesizers
To minimize the possibility of unknown systematic effects, a better test

of the mode-locked laser frequency comb is the comparison of several
independent devices. A few tests have been performed using microwave
standards to reference both a femtosecond laser-based synthesizer and a
more traditional frequency chain or interval divider, which were
subsequently compared in the optical domain [79, 97, 99]. The best result in
this case was an uncertainty limit as low as 5 × 10-16 [79]. In general, the
limit in these cases was imposed by the short-term instability and associated
long averaging times [79, 99] or by the uncertainty of the optical standard
that was measured [97]. Significantly improved short-term instability can be
obtained when the femtosecond laser-based synthesizer is referenced to an
optical standard. In an earlier test, two such devices were compared and
agreement was found at the level of 4 × 10-17 [98]. More recently, this same
test was repeated with four systems, two of which employed octave-
spanning spectra generated in nonlinear microstructure fiber and two which
generated broad spectra directly from the laser [100]. With improved short-
term instability and longer averaging times, a fractional frequency
uncertainty limit of near 1 × 10-19 was achieved. Using the “transfer
oscillator” technique already discussed above [94], Stenger et al. tested a
femtosecond laser comb by measuring the ratio of the frequency of a cw
laser to its second harmonic with an uncertainty of 7 × 10-19.
3.4 Alternatives to Ti:sapphire
In the few years since the introduction of femtosecond lasers into field of
frequency metrology, the palette of available sources has already begun to
broaden beyond that of Ti:sapphire lasers (see Figure 9-7). The reliability of
the mature Ti:sapphire laser has made it the natural place to begin this
exciting field, and it is likely that femtosecond laser-based synthesizers
employing Ti:sapphire will continue to be used in many applications.
However, the present size and cost of the pump laser for Ti:sapphire-based
systems has motivated the search for alternative systems. In the future, it
would not be surprising to find that femtosecond laser-based synthesizers
and optical clocks will be widely used in science and technology and will be
commercially available in compact packages similar to today’s microwave
synthesizers and clocks. The availability of robust, low-priced femtosecond-
laser synthesizers will be particularly important for applications in air- or
space-borne platforms or widespread applications (such as communications
systems) for which cost and rugged packaging are of greatest importance.
To date, some of the promising femtosecond lasers that have produced
octave-spanning spectra include diode-pumped Cr:LiSAF [82], a fiber-laser-
pumped Cr:Forsterite [101], and an Er-doped fiber laser [83, 85, 102]. Each
of these has some advantages and disadvantages relative to Ti:sapphire. For
example, both Cr:LiSAF and Cr:Forsterite have more convenient and
compact pumping schemes either directly with diode lasers or with a Yb-
doped fiber laser. One of the trade-offs here is that these laser hosts are not
as broadband and efficient as Ti:sapphire and tend to have worse thermal
properties. An Er-fiber laser-based comb generator has a number of
advantages over Ti:Sapphire. It can be much more compact, robust, lighter,
and power-efficient than a bulk optic solid-state laser system, and would
require less alignment. Additionally, it can be easily integrated into a
telecommunication system in the important 1300–1600 nm regime.
However, at this point, the Er-fiber-based systems only operate at repetition
252 Chapter 9
rates of 50–100 MHz and can have excess noise that is not fully understood
[85].
g h
2004
f i
2002 e
c d
2000
Year
b
1998
1996 a
1994
500 1000 2000 4000 8000 16000

log2[wavelength] (nm)
Figure 9-7. Spectral extent of some sources used as combs in frequency metrology. (a)
Electro-optic modulator-based comb developed by Kourogi et al. [103]; (b) Ti:sapphire
femtosecond laser comb [13, 14]; (c) Octave-spanning femtosecond laser comb generated
using microstructure fiber [15, 16]; (d) Broadband spectrum generated directly from a
Ti:sapphire laser [69]; (e) Cr:LiSAF femtosecond laser plus microstructure fiber [82]; (f)
Octave-spanning comb generated with femtosecond Cr:forsterite and nonlinear optical fiber
[104]; (g) Octave-spanning comb generated with Er-fiber laser and nonlinear optical fiber
[83]; (h) Offset-frequency-free comb near 3400 nm generated via difference-frequency
generation [93]; (i) Tunable frequency comb generated via difference-frequency generation
between synchronized Ti:sapphire lasers [105].
If wavelength coverage is a concern, one can take advantage of nonlinear

frequency conversion external to the femtosecond laser itself. This has been
demonstrated for 2ν-to-3ν with Cr:forsterite and fiber laser sources [85,
106]. Another interesting option is to use DFG between two extremes of a
Ti:sapphire laser comb to generate a frequency comb further in the IR [92,
93]. While this provides extra wavelength tunability for the femtosecond
source, it also is a means of generating a frequency comb that is independent
of the offset frequency f0. As also shown in Figure 9-7, wide tunability
between 7 and 10 microns can also be accomplished with independent
Ti:sapphire lasers that are synchronized and potentially phase-locked [105].
4. SIGNAL TRANSMISSION AND CROSS-LINKING
With the advent of optical atomic clocks and the associated superior
short-term frequency stability, transfer of signals linked to such
clock/frequency standards over an appreciable distance with a minimal loss
of stability has become an active subject for research [107, 108]. The
stability of the best microwave and optical frequency standards, for times
less than ~ one week, can now exceed the capabilities of the traditional
transmission systems (i.e., GPS, Two-Way Time transfer [109]) used to
distribute these time/frequency reference signals to remote sites. The
instabilities in transmission channels can contribute a significant fraction of
the overall uncertainty in the comparison of high-performance standards that
are not co-located. While improvement in the transfer process over large-
scale signal paths remains challenging, researchers have started
experimenting with optical fibers as effective means for local networks of
dissemination or comparison of time and frequency standards, both in the
microwave and optical domains. The attractiveness of this approach lies in
the fact that an environmentally isolated fiber can be considerably more
stable than open-air paths, especially at short time scales. Furthermore, the
same advantages enjoyed by communication systems in optical fiber (e.g.,
low loss, scalability, etc.) can be realized in a time/frequency distribution
system. Active stabilization of fiber-optic channels for the distribution of
reference frequencies can also be employed to improve the stability of the
transmitted standard. Besides the obvious benefit of more precise
time/frequency transfer, an actively stabilized optical fiber network can play
a critical role in the implementation of long-baseline coherent interferometry
or ultralow timing jitter in particle-accelerator-based novel light sources.
An rf signal can be transferred in an optical fiber network by amplitude
modulating a suitable optical carrier (for example at 1.55 µm) used for
transmission and then recovering the modulation frequency at the remote
end as the transferred reference signal. However, the instability of this rf
modulation-based frequency transfer protocol seems to rise to a few parts in
1013 at 1-s over a 6.9 km long fiber linking JILA to NIST. Jet Propulsion
Laboratory colleagues have demonstrated the rf transfer instability of parts
in 1014 at 1 s for a 16 km-long fiber under active noise cancellation control
[110]. In comparison, direct transfer of a cw laser-based optical frequency
standard through the same 6.9 km fiber suffered an instability of a few parts
in 1014 at 1 s. This instability is further reduced to 3 × 10-15 at 1 s after active
optical-phase-noise cancellation is implemented [107]. A femtosecond laser
located at the remote end can be phase locked to this incoming cw laser
254 Chapter 9
carrier, essentially redistributing the optical frequency to the microwave

domain via the mode-locked laser’s repetition frequency.
With the development of precision femtosecond comb technology, it
appears natural to use a mode-locked source to transfer both optical and
microwave reference signals at the highest level of stability. Of course, the
first logical step is to extend the precision optical-comb spectral coverage
into the 1.5 µm wavelength region, where compact, reliable, and efficient
mode-locked lasers exist for fiber distribution networks. This step has been
accomplished by tight synchronization between the repetition rates and
coherent phase locking of the optical carriers of the 1.5 µm mode-locked
laser sources and a Ti:Sapphire-based femtosecond frequency comb, which
is used as the clockwork for an optical atomic clock based on the molecular
iodine transition. A phase-coherent link between mode-locked lasers requires
two distinct conditions to be met [63, 111]. First, the comb spacing of the
1550 nm source (fr,1550) must be stabilized to the optical clock’s comb
spacing (fr,775). Second, the two combs’ offset frequencies (f0,775 and f0,1550)
must be phase locked together. This latter step requires spectral overlap
between the two combs. The optical comb of the 1550 nm source is
frequency doubled and compared against the Ti:sapphire comb at a mutually
accessible spectral region to generate a heterodyne beat.
We have investigated a number of different types of passively mode-
locked lasers in the 1550 nm region, including an erbium/ytterbium-doped
waveguide laser [112], an erbium-doped fiber laser [108, 113], and a mode-
locked laser diode (MLLD) [75, 114]. Each laser offers at least two control
parameters to obtain simultaneous time synchronization and carrier phase-
locking. For example, stabilization through the cavity length is a common
feature for all these lasers. The carrier-envelope-offset frequency (f0,1550) can
be tuned in both waveguide and fiber lasers through control of the intensity
of their pump lasers. A MLLD, on the other hand, can be tuned with both the
injection current and the reverse bias voltage on the saturable absorber.
Among the three mode-locked laser sources, the waveguide laser achieved
the lowest residual timing jitter with a record-low, root-mean-square relative
timing jitter of 14.4 fs integrated from 10 Hz to 375 MHz (the Nyquist
frequency), owing to the laser’s high-Q cavity and overall gain dynamics.
Although the MLLD has a larger rms timing jitter of ~22 fs within the
bandwidth of 1 Hz – 100 MHz, it does offer the advantage of a compact size,
robust operation, and completely electrical control, along with the potential
in device improvements to lower the jitter at high frequencies.
Transfer of optical frequency standards is then implemented by using the
stabilized mode-locked laser sources at 1.5 µm that are transmitted through
an optical fiber network. We have explored the dual transfer process for both
the optical carrier frequency and the rf signal represented by the laser’s
repetition frequency and to compare the transfer of such an rf signal with

that of a modulated optical carrier. When the 1.5 µm mode-locked laser is
phase coherently connected to a Ti:sapphire mode-locked laser serving as
the clockwork for an optical atomic clock, both the optical carrier frequency
and the repetition frequency of the 1.5 µm mode-locked laser are connected
to a single optical frequency standard. Therefore a single fiber link allows a
user at the remote end to either simply use a fast photodiode to recover the rf
reference signal or establish a more elaborate, direct connection to the
optical carrier. Simultaneous existence of both the optical phase and the
pulse-repetition-rate information would also enable novel measurement
capabilities on material dispersion and length determination. The
experimental results confirm that with the use of a mode-locked laser at 1.5
µm, the instability of the transfer process for the repetition frequency (rf
reference) over the 6.9 km long fiber is nearly the same as that of optical
carrier transfer of a cw laser discussed earlier, and is almost an order of
magnitude better than that of transfer of a rf-modulated cw laser carrier.
Furthermore, it is clear that a pulsed transfer process preserves the stability
of the optical carrier as in the cw optical transfer process. Preliminary results
on pulsed transfer under active noise cancellation indicate rf transfer
instability reaching a few parts in 1015 at 1 s. We are currently exploring the
utility of such ultrastable frequency transfer for time-domain experiments
where the goal is to deliver extremely low-jitter timing signals throughout a
fiber network, permitting ultrafast lasers located at remote areas be
synchronized together at a precision level of a few femtoseconds within a
multi-megahertz bandwidth.
ACKNOWLEDGEMENTS
The work presented in this chapter would not have been possible without
the valuable contributions of many people. In particular, we would like to
thank Albrecht Bartels, Jim Bergquist, Martin Boyd, Sebastien Bize, Lisheng
Chen, Steven Cundiff, Anne Curtis, Seth Foreman, Rich Fox, John Hall,
Kevin Holman, Darren Hudson, Tetsuya Ido, Erich Ippen, Euguene Ivanov,
David Jones, Jason Jones, Franz Kärtner, Thomas Loftus, Andrew Ludlow,
Long-Sheng Ma, Adela Marian, Kevin Moll, Oliver Mücke, Nathan
Newbury, Chris Oates, Jin-Long Peng, Tanya Ramond, Matthew Stowe,
Thomas Udem, Kurt Vogel, Guido Wilpers, Robert Windeler, and Dave
Wineland.
256 Chapter 9
REFERENCES
[1] H. Dehmelt, Rev. Mod. Phys. 62, 525-530 (1990).
[2] H. G. Dehmelt, IEEE Trans. Instrum. Meas. 31, 83-87 (1982).
[3] J. L. Hall, Science 202, 147-156 (1978); J. L. Hall, M. Zhu, and P. Buch, J.
Opt. Soc. Am. B 6, 2194-2205 (1989).
[4] T. W. Hänsch, I. S. Shahin, and A. L. Schawlow, Nature-Physical Science
235, 63 (1972).
[5] S. N. Bagayev, Y. D. Kolomnikov, V. N. Lisitsyn, and V. P. Chebotayev,
IEEE J. Quant. Electr. 4, 868 (1968).
[6] V. P. Chebotayev, V. G. Goldort, V. M. Klementyev, M. V. Nikitin, B. A.
Timchenko, and V. F. Zakharyash, Appl. Phys. B 29, 63-65 (1982).
[7] D. J. Wineland, Science 226, 395-400 (1984); D. J. Wineland, J. C.
Bergquist, W. M. Itano, F. Diedrich, and C. S. Weimer, in The Hydrogen
Atom, edited by G. F. Bassani, M. Inguscio and T. W. Hänsch (Springer-
Verlag, Berlin, 1989), p. 123-133.
[8] A. Clairon, A. Vanlerberghe, C. Salomon, M. Ouhayoun, and C. J. Borde,
Opt. Commun. 35, 368-372 (1980).
[9] C. O. Weiss, G. Kramer, B. Lipphardt, and E. Garcia, IEEE J. Quantum
Electron. 24, 1970-1972 (1988).
Lett. 76, 18-21 (1996).
[11] J. E. Bernard, A. A. Madej, L. Marmet, B. G. Whitford, K. J. Siemsen, and
S. Cundy, Phys. Rev. Lett. 82, 3228-3231 (1999).
(1990).
883 (1999).
3568-3571 (1999).
84, 5102-5105 (2000).
[17] T. Udem, R. Holzwarth, and T. W. Hänsch, Nature 416, 233-237 (2002).
(2001).
[19] J. L. Hall, J. Ye, S. A. Diddams, L.-S. Ma, S. T. Cundiff, and D. J. Jones,
IEEE J. Quantum Electron. 37, 1482-1492 (2001).
[20] L. Hollberg, C. W. Oates, E. A. Curtis, E. N. Ivanov, S. A. Diddams, T.
Udem, H. G. Robinson, J. C. Bergquist, R. J. Rafac, W. M. Itano, R. E.
Drullinger, and D. J. Wineland, IEEE J. Quantum Electron. 37, 1502-1513
(2001).
[21] S. T. Cundiff and J. Ye, Rev. Mod. Phys. 75, 325-342 (2003).
[22] J. Ye, H. Schnatz, and L. W. Hollberg, IEEE J. Sel. Top. Quantum
Electron. 9, 1041-1058 (2003).
[23] J. Jespersen and J. Fitz-Randolph, From Sundials to Atomic Clocks:

Understanding Time and Frequency (NIST, U.S. Department of
Commerce, Gaithersburg, 1999).
[24] D. A. Jennings, K. M. Evenson, and D. J. E. Knight, Proc. IEEE 74, 168-
179 (1986); C. O. Weiss, G. Kramer, B. Lipphardt, and H. Schnatz,
Frequency Measurement and Control: Advanced Techniques and Future
Trends 79, 215-247 (2000).
[25] O. Acef, J. J. Zondy, M. Abed, D. G. Rovera, A. H. Gerard, A. Clairon, P.
Laurent, Y. Millerioux, and P. Juncar, Opt. Commun. 97, 29-34 (1993).
[26] J. Levine, Rev. Sci. Instrum. 70, 2567-2596 (1999); J. Levine, Reports on
Progress in Physics 65, 1119-1164 (2002).
[27] J. Ye, Opt. Lett. 29, 1153-1155 (2004).
and A. Clairon, Phys. Rev. Lett. 84, 5496-5499 (2000); C. Schwob, L.
Jozefowski, B. de Beauvoir, L. Hilico, F. Nez, L. Julien, F. Biraben, O.
Acef, and A. Clairon, Phys. Rev. Lett. 82, 4960-4963 (1999); B. de
Beauvoir, C. Schwob, O. Acef, L. Jozefowski, L. Hilico, F. Nez, L. Julien,
A. Clairon, and F. Biraben, Euro. Phys. J. D 12, 61-93 (2000).
[29] S. G. Karshenboim, Can. J. Phys. 78, 639-678 (2000).
[30] J. D. Prestage, R. L. Tjoelker, and L. Maleki, Phys. Rev. Lett. 74, 3511-
3514 (1995).
[31] H. Marion, F. P. Dos Santos, M. Abgrall, S. Zhang, Y. Sortais, S. Bize, I.
Maksimovic, D. Calonico, J. Grunert, C. Mandache, P. Lemonde, G.
Santarelli, P. Laurent, A. Clairon, and C. Salomon, Phys. Rev. Lett. 90
(2003); S. Bize, S. A. Diddams, U. Tanaka, C. E. Tanner, W. H. Oskay, R.
E. Drullinger, T. E. Parker, T. P. Heavner, S. R. Jefferts, L. Hollberg, W.
M. Itano, and J. C. Bergquist, Phys. Rev. Lett. 90, 150802 (2003).
[32] A. Brillet and J. L. Hall, Phys. Rev. Lett. 42, 549-552 (1979); D. Hils and J.
L. Hall, Phys. Rev. Lett. 64, 1697-1700 (1990); V. A. Kostelecky and C. D.
Lane, Phys. Rev. D 6011, art. no.-116010 (1999); D. Bear, R. E. Stoner, R.
L. Walsworth, V. A. Kostelecky, and C. D. Lane, Phys. Rev. Lett. 85,
5038-5041 (2000).
[33] G. Gabrielse, N. S. Bowden, P. Oxley, A. Speck, C. H. Storry, J. N. Tan,
M. Wessels, D. Grzonka, W. Oelert, G. Schepers, T. Sefzick, J. Walz, H.
Pittner, T. W. Hänsch, and E. A. Hessels, Phys. Rev. Lett. 89, 233401
(2002); M. Amoretti, C. Amsler, G. Bonomi, A. Bouchta, P. Bowe, C.
Carraro, C. L. Cesar, M. Charlton, M. J. T. Collier, M. Doser, V. Filippini,
K. S. Fine, A. Fontana, M. C. Fujiwara, R. Funakoshi, P. Genova, J. S.
Hangst, R. S. Hayano, M. H. Holzscheiter, L. V. Jorgensen, V.
Lagomarsino, R. Landua, D. Lindelof, E. L. Rizzini, M. Macri, N. Madsen,
G. Manuzio, M. Marchesotti, P. Montagna, H. Pruys, C. Regenfus, P.
Riedler, J. Rochet, A. Rotondi, G. Rouleau, G. Testera, A. Variola, T. L.
Watson, and D. P. van der Werf, Nature 419, 456-459 (2002).
[34] J. Walz, P. Fendel, M. Herrmann, M. Konig, A. Pahl, H. Pittner, B. Schatz,
and T. W. Hänsch, J. Phys, B 36, 649-654 (2003); J. Walz, H. Pittner, M.
Herrmann, P. Fendel, B. Henrich, and T. W. Hänsch, Appl. Phys. B 77,
713-717 (2003).
258 Chapter 9
[35] R. Bluhm, V. A. Kostelecky, and N. Russell, Phys. Rev. Lett. 82, 2254-
2257 (1999).
[36] A. Javan, Annals of the New York Academy of Sciences 168, 715 (1970).
[37] P. Gill, Proc. 6th Symp. on Freq. Standards and Metrology, World
Scientific, Singapore (2002).
[38] D. W. Allan, Proc. IEEE 54, 221 (1966).
[39] W. M. Itano, J. C. Bergquist, J. J. Bollinger, J. M. Gilligan, D. J. Heinzen,
F. L. Moore, M. G. Raizen, and D. J. Wineland, Phys. Rev. A 47, 3554-
3570 (1993); C. Santarelli, P. Laurent, P. Lemonde, A. Clairon, A. G.
Mann, S. Chang, A. N. Luiten, and C. Salomon, Phys. Rev. Lett. 82, 4619-
4622 (1999).
[40] J. J. Bollinger, J. D. Prestage, W. M. Itano, and D. J. Wineland, Phys. Rev.
Lett. 54, 1000-1003 (1985); A. A. Madej and J. E. Bernard, Frequency
Measurement and Control: Advanced Techniques and Future Trends 79,
153-194 (2000).
82, 3799-3802 (1999).
[42] R. J. Rafac, B. C. Young, J. A. Beall, W. M. Itano, D. J. Wineland, and J.
C. Bergquist, Phys. Rev. Lett. 85, 2462-2465 (2000).
[43] D. J. Wineland, M. Barrett, J. Britton, J. Chiaverini, B. DeMarco, W. M.
Itano, B. Jelenkovic, C. Langer, D. Leibfried, V. Meyer, T. Rosenband, and
T. Schatz, Philos. Trans. R. Soc. Lond. Ser. A 361, 1349-1361 (2003).
[44] M. D. Barrett, B. DeMarco, T. Schaetz, V. Meyer, D. Leibfried, J. Britton,
J. Chiaverini, W. M. Itano, B. Jelenkovic, J. D. Jost, C. Langer, T.
Rosenband, and D. J. Wineland, Phys. Rev. A 68 (2003).
[45] C. W. Oates, E. A. Curtis, and L. Hollberg, Opt. Lett. 25, 1603-1605
(2000); T. Udem, S. A. Diddams, K. R. Vogel, C. W. Oates, E. A. Curtis,
W. D. Lee, W. M. Itano, R. E. Drullinger, J. C. Bergquist, and L. Hollberg,
Phys. Rev. Lett. 86, 4996-4999 (2001).
[46] G. Wilpers, T. Binnewies, C. Degenhardt, U. Sterr, J. Helmcke, and F.
Riehle, Phys. Rev. Lett. 89 (2002).
[47] E. A. Curtis, C. W. Oates, and L. Hollberg, J. Opt. Soc. Am. B 20, 977-984
(2003).
[48] M. Takamoto and H. Katori, Phys. Rev. Lett. 91 (2003).
[49] X. Y. Xu, T. H. Loftus, J. W. Dunn, C. H. Greene, J. L. Hall, A. Gallagher,
and J. Ye, Phys. Rev. Lett. 90 (2003).
[50] H. Katori, M. Takamoto, V. G. Pal'chikov, and V. D. Ovsiannikov, Phys.
Rev. Lett. 91 (2003).
[51] T. Mukaiyama, H. Katori, T. Ido, Y. Li, and M. Kuwata-Gonokami, Phys.
Rev. Lett. 90 (2003).
[52] J. L. Hall, C. J. Borde, and K. Uehara, Phys. Rev. Lett. 37, 1339-1342
(1976); V. P. Chebotayev, Radio Science 14, 573-584 (1979); A. Clairon,
B. Dahmani, A. Filimon, and J. Rutman, IEEE Trans. Instrum. Meas. 34,
265-268 (1985).
[53] M. A. Gubin, D. A. Tyurikov, A. S. Shelkovnikov, E. V. Kovalchuk, G.
Kramer, and B. Lipphardt, IEEE J. Quantum Electron. 31, 2177-2182
(1995); J. von Zanthier, J. Abel, T. Becker, M. Fries, E. Peik, H. Walther,
R. Holzwarth, J. Reichert, T. Udem, T. W. Hänsch, A. Y. Nevsky, M. N.
Skvortsov, and S. N. Bagayev, Opt. Commun. 166, 57-63 (1999); B. Frech,
L. F. Constantin, A. Amy-Klein, O. Phavorin, C. Daussy, C. Chardonnet,
and M. Murtz, Appl. Phys. B 67, 217-221 (1998); A. Shelkovnikov, C.

Grain, C. T. Nguyen, R. J. Butcher, A. Amy-Klein, and C. Chardonnet,
Appl. Phys. B 73, 93-98 (2001).
[54] J. Ye, L.-S. Ma, and J. L. Hall, IEEE Trans. Instrum. Meas. 46, 178-182
(1997); J. Ye, L.-S. Ma, and J. L. Hall, J. Opt. Soc. Am. B 15, 6-15 (1998).
[55] J. Ye, L. Robertsson, S. Picard, L.-S. Ma, and J. L. Hall, IEEE Trans.
Instrum. Meas. 48, 544-549 (1999); A. Y. Nevsky, R. Holzwarth, J.
Reichert, T. Udem, T. W. Hänsch, J. von Zanthier, H. Walther, H. Schnatz,
F. Riehle, P. V. Pokasov, M. N. Skvortsov, and S. N. Bagayev, Opt.
Commun. 192, 263-272 (2001); F. L. Hong, J. Ishikawa, K. Sugiyama, A.
Onae, H. Matsumoto, J. Ye, and J. L. Hall, IEEE Trans. Instrum. Meas. 52,
240-244 (2003).
[56] W. Y. Cheng, L. S. Chen, T. H. Yoon, J. L. Hall, and J. Ye, Opt. Lett. 27,
571-573 (2002).
[57] R. J. Jones, W. Y. Cheng, K. W. Holman, L.-S. Chen, J. L. Hall, and J. Ye,
Appl. Phys. B 74, 597 (2002); A. Goncharov, A. Amy-Klein, O. Lopez, F.
du Burck, and C. Chardonnet, Appl. Phys. B 78, 725-731 (2004).
[58] C. Salomon, D. Hils, and J. L. Hall, J. Opt. Soc. Am. B 5, 1576-1587
(1988).
850 (1978).
Wineland, Science 293, 825-828 (2001).
[61] J. Ye, L.-S. Ma, and J. L. Hall, Phys. Rev. Lett. 87, 270801 (2001).
[62] T. H. Yoon, A. Marian, J. L. Hall, and J. Ye, Phys. Rev. A 63, 011402
(2000); A. Poppe, R. Holzwarth, A. Apolonski, G. Tempea, C. Spielmann,
T. W. Hänsch, and F. Krausz, Appl. Phys. B 72, 977-977 (2001); R.
Kienberger, M. Hentschel, M. Uiberacker, C. Spielmann, M. Kitzler, A.
Scrinzi, M. Wieland, T. Westerwalbesloh, U. Kleineberg, U. Heinzmann,
M. Drescher, and F. Krausz, Science 297, 1144-1148 (2002).
[63] R. K. Shelton, L.-S. Ma, H. C. Kapteyn, M. M. Murnane, J. L. Hall, and J.
Ye, Science 293, 1286-1289 (2001).
[64] A. Baltuška, T. Udem, M. Uiberacker, M. Hentschel, E. Goulielmakis, C.
Gohle, R. Holzwarth, V. S. Yakoviev, A. Scrinzi, T. W. Hänsch, and F.
Krausz, Nature 421, 611-615 (2003).
[65] P. F. Moulton, J. Opt. Soc. Am. B 3, 125-133 (1986); D. E. Spence, P. N.
Kean, and W. Sibbett, Opt. Lett. 16, 42-44 (1991).
M. Murnane, Opt. Lett. 18, 977-979 (1993); R. Szipocs, K. Ferencz, C.
Spielmann, and F. Krausz, Opt. Lett. 19, 201-203 (1994).
[67] J. K. Ranka, R. S. Windeler, and A. J. Stentz, Opt. Lett. 25, 25-27 (2000);
T. A. Birks, W. J. Wadsworth, and P. S. Russell, Opt. Lett. 25, 1415-1417
(2000).
[68] E. N. Ivanov, S. A. Diddams, and L. Hollberg, IEEE J. Sel. Top. Quantum
Electron. 9, 1059-1065 (2003).
260 Chapter 9

Opt. Lett. 26, 373-375 (2001).
[70] A. Bartels and H. Kurz, in XIII International Conference on Ultrafast
Phenomena, edited by M. M. Murnane, N. F. Scherer, D. J. D. Miller and
A. M. Weiner (OSA, Vancouver, BC, 2002), Vol. 72, p. 173-174.
(2003).
1842-1844 (2002).
[73] S. A. Diddams, A. Bartels, T. M. Ramond, C. W. Oates, S. Bize, E. A.
Curtis, J. C. Bergquist, and L. Hollberg, IEEE J. Sel. Top. Quantum
Electron. 9, 1072-1080 (2003).
28, 851-853 (2003).
[75] K. W. Holman, D. J. Jones, J. Ye, and E. P. Ippen, Opt. Lett. 28, 2405-2407
(2003).
[76] N. R. Newbury, B. R. Washburn, K. L. Corwin, and R. S. Windeler, Opt.
Lett. 28, 944-946 (2003).
[77] K. L. Corwin, N. R. Newbury, J. M. Dudley, S. Coen, S. A. Diddams, B. R.
Washburn, K. Weber, and R. S. Windeler, Appl. Phys. B 77, 269-277
(2003).
[78] J. M. Dudley and S. Coen, Opt. Lett. 27, 1180-1182 (2002); R. Holzwarth,
M. Zimmermann, T. Udem, and T. W. Hänsch, IEEE J. Quantum Electron.
37, 1493-1501 (2001).
[80] J. D. Jost, J. L. Hall, and J. Ye, Opt. Expr. 10, 515 (2002); L. S. Chen and J.
Ye, Chem. Phys. Lett. 381, 777-783 (2003).
Keller, Appl. Phys. B 69, 327 (1999).
[82] R. Holzwarth, M. Zimmermann, T. Udem, T. W. Hänsch, P. Russbuldt, K.
Gabel, R. Poprawe, J. C. Knight, W. J. Wadsworth, and P. S. J. Russell,
Opt. Lett. 26, 1376-1378 (2001).
[83] F. Tauser, A. Leitenstorfer, and W. Zinth, Optics Expr. 11, 594-600 (2003);
B. R. Washburn, S. A. Diddams, N. R. Newbury, J. W. Nicholson, M. F.
Yan, and C. G. Jorgensen, Opt. Lett. 29, 250-252 (2004).
[85] F. L. Hong, K. Minoshima, A. Onae, H. Inaba, H. Takada, A. Hirai, and H.
Matsumoto, Opt. Lett. 28, 1516-1518 (2003).
[86] K. W. Holman, R. J. Jones, A. Marian, S. T. Cundiff, and J. Ye, IEEE J.
Sel. Top. Quantum Electron. 9, 1018-1024 (2003).
172, 59-68 (1999).
Lett. 27, 1436-1438 (2002).
[89] A. Bartels, S. A. Diddams, C. W. Oates, G. Wilpers, W. H. Oskay, J. C.
Bergquist, and L. Hollberg, in Conference on Lasers and Electo-optics, San
Francisco, 2004), postdeadline presentation.
[90] A. Bartels, C. W. Oates, L. Hollberg, and S. A. Diddams, Opt. Lett. 29,

1081-1083 (2004).
[91] J. Ye, J. L. Hall, and S. A. Diddams, Opt. Lett. 25, 1675 (2000).
[92] M. Zimmermann, C. Gohle, R. Holzwarth, T. Udem, and T. W. Hänsch,
Opt. Lett. 29, 310-312 (2004).
[93] O. D. Mücke, O. Kuzucu, N. C. Wong, E. P. Ippen, F. X. Kaertner, S. M.
Foreman, D. J. Jones, L.-S. Ma, J. L. Hall, and J. Ye, CLEO 2004
postdeadline presentation and Opt. Lett., in press (2004).
073601 (2002).
[95] H. R. Telle, B. Lipphardt, and J. Stenger, Appl. Phys. B 74, 1-6 (2002).
[96] R. Holzwarth, Ph.D Thesis, Ludwig-Maximilians-Universität (2001).
[98] S. A. Diddams, L. Hollberg, L.-S. Ma, and L. Robertsson, Opt. Lett. 27, 58-
60 (2002).
[99] L.-S. Ma, L. Robertsson, S. Picard, M. Zucco, Z. Y. Bi, S. H. Wu, and R. S.
Windeler, Opt. Lett. 29, 641-643 (2004).
[100] L.-S. Ma, Z. Y. Bi, A. Bartels, L. Robertsson, M. Zucco, R. S. Windeler, G.
Wilpers, C. Oates, L. Hollberg, and S. A. Diddams, Science 303, 1843-
1845 (2004).
[101] K. L. Corwin, I. Thomann, T. Dennis, R. W. Fox, W. Swann, E. A. Curtis,
C. W. Oates, G. Wilpers, A. Bartels, S. L. Gilbert, L. Hollberg, N. R.
Newbury, S. A. Diddams, J. W. Nicholson, and M. F. Yan, Opt. Lett. 29,
397-399 (2004).
[102] H. Hundertmark, D. Wandt, C. Fallnich, N. Haverkamp, and H. R. Telle,
Opt. Expr. 12, 770-775 (2004).
2693-2701 (1993); K. Imai, M. Kourogi, and M. Ohtsu, IEEE J. Quantum
Electron. 34, 54-60 (1998); K. Imai, B. Widiyatmoko, M. Kourogi, and M.
Ohtsu, IEEE J. Quantum Electron. 35, 559-564 (1999).
[104] I. Thomann, L. Hollberg, S. A. Diddams, and R. Equall, Appl. Optics 42,
1661-1666 (2003).
[105] S. M. Foreman, D. J. Jones, and J. Ye, Opt. Lett. 28, 370-372 (2003).
[106] I. Thomann, A. Bartels, K. L. Corwin, N. R. Newbury, L. Hollberg, S. A.
Diddams, J. W. Nicholson, and M. F. Yan, Opt. Lett. 28, 1368-1370
(2003).
[107] J. Ye, J. L. Peng, R. J. Jones, K. W. Holman, J. L. Hall, D. J. Jones, S. A.
Diddams, J. Kitching, S. Bize, J. C. Bergquist, L. W. Hollberg, L.
Robertsson, and L.-S. Ma, J. Opt. Soc. Am. B 20, 1459-1467 (2003).
[108] K. W. Holman, D. J. Jones, D. D. Hudson, and J. Ye, Opt. Lett. 29, 1554-
1556 (2004).
[109] J. Levine, IEEE Trans. Ultrason. Ferroelectr. Freq. Control 46, 392-398
(1999).
[110] M. Calhoun, R. Sydnor, and W. Diener, (Jet Propulsion Laboratory,
Pasadena, CA, 2002), p. 1-11.
[111] R. K. Shelton, S. M. Foreman, L.-S. Ma, J. L. Hall, H. C. Kapteyn, M. M.
Murnane, M. Notcutt, and J. Ye, Opt. Lett. 27, 312-314 (2002).
262 Chapter 9
[112] J. B. Schlager, B. E. Callicoatt, R. P. Mirin, N. A. Sanford, D. J. Jones, and

J. Ye, Opt. Lett. 28, 2411-2413 (2003).
[113] J. Rauschenberger, T. M. Fortier, D. J. Jones, J. Ye, and S. T. Cundiff, Opt.
Expr. 10, 1404-1410 (2002).
[114] D. J. Jones, K. W. Holman, M. Notcutt, J. Ye, J. Chandalia, L. A. Jiang, E.
P. Ippen, and H. Yokoyama, Opt. Lett. 28, 813-815 (2003).
Chapter 10
GENERATION AND MEASUREMENT OF
INTENSE PHASE-CONTROLLED FEW-CYCLE
LASER PULSES
Andrius Baltuška1,2, Gerhard G. Paulus2,3, Fabrizio Lindner2, Reinhard

Kienberger1,2, and Ferenc Krausz1,2
1
Institut für Photonik, Technische Universität Wien
2
3
Department of Physics, Texas A&M University
Abstract: Intense, ultrashort waveforms of light can be produced with a predetermined

electromagnetic field. These waveforms are essential in many applications of
extreme nonlinear optics, most prominently in laser-driven sources of high-
energy attosecond radiation. Field reproducibility in each laser shot requires a
stable carrier-envelope phase. We analyze different schemes of phase-stable
amplification and identify constraints limiting the precision. We describe a
phase-stabilized system based on a 20 fs multipass Ti:sapphire amplifier
supplemented with a fiber compression stage for producing few-cycle pulses.
The amplifier introduces only a slow phase drift and, therefore, can be seeded
by a standard phase-stabilized oscillator. The phase stability of the 5 fs, 400 µJ
pulses is verified by high harmonic generation, in which different carrier-
envelope phases produce distinctly different XUV spectra. The carrier-
envelope phase (with a ±π ambiguity) is calibrated from a series of spectra.
The calibration allows full characterization of the electric field. The estimated
precision of the phase control is better than π/5, which reduces the timing jitter
between the driving laser pulse and the XUV bursts to ~250 as and enables the
generation of stable, isolated attosecond pulses. We demonstrate a more robust
phase measurement based on the detection of electron emission from
photoionized atoms in opposite directions. This method determines the carrier-
envelope phase with the π/10 accuracy and without inversion ambiguity. Using
the photoionization technique, we demonstrate the Gouy effect for focused
few-cycle pulses. This result is of critical importance for any phase-dependent
strong-field applications of ultrashort laser pulses.
Key words: carrier-envelope phase, ultrashort pulses, Gouy phase shift, high harmonic
generation, above threshold ionization, phase-stabilized amplifiers
264 Chapter 10
1. INTRODUCTION
The applications of ultrafast laser science demand an increasing degree of

control over many parameters of intense ultrashort laser pulses. Some forms
of control, such as techniques for adjusting spatial and temporal intensity
pulse profiles and the sweep of carrier frequency (chirp) across the electric
field, are already well established [1]. There are also well-developed
methods for pulse diagnostics, which can fully characterize the envelope of
the pulse and its chirp, that are an integral part of any control scheme [2, 3].
New demands, such as control over the time-dependent polarization state of
the pulse field on the scale of a single optical cycle [4], emerge in the
context of strong-field applications sensitive to the strength of individual
light-wave cycles within the laser pulse rather than to the intensity envelope.
As broadband laser systems approach the single-cycle regime [5-7], the
electric-field strength of the pulses varies substantially from peak to peak of
every optical half cycle. Therefore, it becomes necessary to precisely control
the evolution of the electric field underneath the pulse envelope for
ultrashort wave forms of light used in field-sensitive experiments. The
parameter that determines the offset of the most prominent field peak with
respect to the pulse envelope is the carrier-envelope phase. Change of the
carrier-envelope phase represents the slippage of the carrier wave through
the pulse envelope, which takes place in conventional mode-locked laser
systems [8] i.e., those without carrier-envelope-phase stabilization.
The development of techniques suitable for carrier-envelope-phase
stabilization of mode-locked oscillators [9-11] and demonstration of such
laser systems [12, 13] were significant steps toward accomplishing full
control over the temporal and spectral properties of femtosecond laser
pulses. Maintaining the value of the carrier-envelope phase and fixed
repetition frequency manifests itself in the frequency domain as a perfectly
defined and steady comb of laser modes across the laser bandwidth and
thereby provides an invaluable tool for frequency metrology [9, 10, 14-18].
Driven by the demand for simple, precise, and robust instrumentation for
frequency measurements, the technology of carrier-envelope-phase control
of mode-locked femtosecond lasers has rapidly matured [17, 18] and is now
commercially available [19]. In comparison with laser oscillators operating
at megahertz-to-gigahertz repetition rates, phase stabilization of laser
amplifiers is significantly complicated by their low, typically 1–10 kHz,
repetition rate. Whereas it is possible to trace the carrier-envelope-phase
offset in the output of an oscillator and compensate it by active stabilization
of the laser cavity [12, 13], the output phase of a standard chirped-pulse-
amplifier (CPA) is random [20-22]. Therefore, the information on phase
excursions monitored behind the amplifier is insufficient for the phase
10. GENERATION AND MEASUREMENT OF INTENSE PHASE- 265
CONTROLLED FEW-CYCLE LASER PULSES
stabilization of the output pulse train. Other techniques, possibly in

combination with the use of a phase-stabilized seed oscillator, are required.
Several different strategies for carrier-envelope stabilization of amplified
laser pulses will be discussed.
The significance of the carrier-envelope-phase control of intense
ultrashort laser pulses, and the motivation for pursuing it, can be readily
grasped from the numerical example presented in Figure 10-1. This
simulation compares the XUV/soft x-ray emission from a target of noble gas
exposed to three different driving optical fields depicted in Figure 10-1(a)–
(c). The generation of the soft x-ray radiation [23, 24] includes the following
steps: (1) ionization occurring in the vicinity of the field peaks, (2)
acceleration of the detached electrons by the laser electric field, and (3)
recombination of the electron into its ground state following a recollision
with the parent ion. The sum of the binding and the kinetic energy of the
electron is thus released as a high-energy photon. The highest-energy soft x-
ray quanta, which are emitted at the cutoff frequency, appear after
approximately two-thirds of the optical cycle following the moment of
ionization. This release of a photon nearly coincides with the first zero
crossing of the electric field behind the strongest field peak.
Figure 10-1(d)–(i) shows the corresponding temporal structure of the soft
x-ray radiation emitted around 65 and 130 eV, respectively. The latter
photon energy represents the cutoff obtainable for a neon gas target with a
length of 2 mm and a pressure of 100 mbar. Pulse intensity was chosen to
reach the cutoff frequency of 130 eV. It is obvious from Figures 10-1(d) and
(g) that for a 20-femtosecond pulse, typical of many currently available laser
amplifiers, the x-ray bursts are emitted repeatedly over many successive
optical cycles. For this driving laser pulse, the threshold of field strength
[horizontal bars in Figures 10-1(a)–(c)] required to generate cutoff-
frequency x-ray photons can be overcome by multiple field peaks because of
their comparable magnitude. Consequently, regardless of the carrier-
envelope-phase setting, the emitted x-rays consist of a train of attosecond
spikes.
266 Chapter 10
∆τ=20 fs ∆τ=5 fs ∆τ=5 fs

Electric field GV/m
Electric field GV/m
Electric field GV/m

50 50 50
0 0 0
-50 -50 -50

(a) (b) (c)
-30 -15 0 15 30 -5 0 5 -5 0 5
Time [fs] Time [fs] Time [fs]
XUV intensity [arb. units]

1.0 1.0 1.0
ωphoton=65 eV ωphoton=65 eV ωphoton=65 eV
0.5 0.5 0.5
(d) (e) (f)

0.0 0.0 0.0
-30 -15 0 15 30 -5 0 5 -5 0 5

1.0 1.0 1.0
ωphoton=130 eV ωphoton=130 eV ωphoton=130 eV
0.5 0.5 0.5
(g) (h) (i)

0.0 0.0 0.0
-30 -15 0 15 30 -5 0 5 -5 0 5
Figure 10-1. Numerical simulation of laser-driven soft x-ray emission from noble gas atoms.
(a) Electric field of a 20 fs laser pulse. (b) Electric field of a 5 fs laser pulse optimized for
generation of a single subfemtosecond x-ray pulse. (c) Electric field of a 5 fs pulse producing
the two highest energy x-ray pulses. (d) – (f) Time-domain structures of x-ray radiation
emitted in a 10 eV bandwidth at half the cutoff energy. (g) – (i) Time-domain structures of x-
ray radiation emitted in a 10 eV bandwidth at the cutoff energy. The lower x-ray peak yield in
(d) and (g) in comparison to (e), (f), (h), and (i) is the consequence of higher ionization by
more numerous field peaks of the 20 fs pulse. In total, gas concentration loss due to ionization
was 6.7% and 1.7% for the 20 and 5 fs pulses, respectively. Horizontal bars in (a)–(c) show
the peak intensity threshold required to yield cutoff-frequency x-ray radiation.
The situation changes dramatically for a 5 fs driving pulse [Figure 10-

1(b) and (c)]. Whereas the emission at half the cutoff energy also comprises
a train of spikes [Figure 10-1(e)–(f)], the soft x-rays at the cutoff frequency
[Figure 10-1(h) and (i)] are extremely sensitive to the arrangement of the
optical cycles in the driving laser field. A π/2 change in the carrier-envelope
phase determines whether a single half cycle or two field peaks fulfill the
threshold requirement that, in turn, leads to the generation of an isolated
subfemtosecond pulse [Figure 10-1(h)] or a pair of bursts [Figure 10-1(i)].
The case study presented in Figure 10-1 underscores several compelling

reasons for seeking carrier-envelope-phase control of amplified laser pulses
in conjunction with the generation of subfemtosecond pulses [25] and their
use in the emerging field of attosecond metrology [26, 27] and spectroscopy
[28]. First, pulse-to-pulse fluctuations of the carrier-envelope phase lead to
an unstable temporal profile of the soft x-ray pulses, including duration
variations and, for some phases, strong prepulses. Second, a variable carrier-
envelope phase results in large intensity instability of the soft x-ray bursts,
which severely limits their utility. Third, carrier-envelope-phase shifts cause
timing jitter between the soft x-ray burst and the envelope peak of the
driving laser field. This timing jitter further reduces the temporal resolution
of excite-probe experiments that use the laser and the soft x-ray pulses,
respectively, as excitation and probe [28] and vice versa. The implications of
carrier-envelope-phase variations on the higher-order harmonic generation
(extending into the XUV/soft x-ray range) have been studied in a number of
papers [6, 23, 24, 29-31]. The lack of carrier-envelope-phase control has
been identified as a cause of an excessive spectral noise observed in x-ray
spectra [32, 33]. Similar critical dependence on the value of the carrier-
envelope phase is expected in other nonlinear optical interactions that exhibit
a threshold with respect to the field strength of the incident laser pulse [6]. In
particular, the first clear experimental evidence of a carrier-envelope-phase-
dependent effect was obtained in an above-threshold ionization measurement
in gas [34]. In this scheme, the angular yield of photoelectron emission was
found to correlate with the carrier-envelope phase of the circularly polarized
driving pulses. Once carrier-envelope-phase-stabilized pulses at the required
intensity level become available, it should be possible to steer photoelectrons
from gas [34] or solid [35] targets in a predefined way by adjusting the phase
of light. In principle, carrier-envelope-phase control of laser pulses would
open the way to synthesizing intense wave forms of light (i.e., very brief
high-peak-intensity electromagnetic fields exactly reproducible in every
laser shot) that could be used to control the motion of charged particles with
ultimate precision. Next to phase stabilization, an essential part of carrier-
envelope phase control is the measurement of the actual phase value.
Standard techniques of pulse characterization [2, 3] are insensitive to carrier-
envelope phase since its value has no influence on the shape of the pulse
envelope. Information on the relative pulse-to-pulse phase drift can be
obtained by various methods involving interference of spectrally overlapping
harmonic orders [10-13, 20, 22, 36, 37]. However, such techniques do not
permit phase calibration. Instead, the use of high-order nonlinear responses
has been suggested for this purpose [6, 30, 38, 39]. In the context of the
numerical example presented in Figure 10-1, we have seen that phase
268 Chapter 10
calibration is needed to ascertain the presence of isolated attosecond pulses,

especially if such pulses are intended for time-resolved attosecond
applications.
This chapter investigates both the problem of carrier-envelope-phase
stabilization of amplified ultrashort laser pulses and methods of phase
calibration. Several possible schemes of carrier-envelope-phase control
capable of operating at kilohertz repetition rates are addressed. We look into
the origin of phase distortion added throughout pulse amplification and beam
passage through peripheral components, such as dispersive pulse
compressors, and stages of nonlinear spectral broadening. We then explain
in detail the operation of a carrier-envelope-phase-stabilized multipass
Ti:sapphire 5 fs amplifier [40]. Next, we focus on measuring the actual value
of the carrier-envelope phase. To perform such phase calibration, we
develop a method relying on the spectrum of XUV emission from a noble
gas. The robustness of this approach is confirmed both experimentally and
by numerical simulations. Whereas phase determination based on the
spectral XUV measurement is very practical in conjunction with the
generation of isolated subfemtosecond pulses in that spectral range, the
information provided by this technique is still insufficient for a complete
characterization of the electric field of the driving laser pulse. Therefore, we
also present an accurate, unambiguous carrier-envelope measurement based
on directional observation of photoelectron emission. We demonstrate that a
dual time-of-flight photoelectron detector can be used to calibrate the
carrier-envelope phase and visualize the Gouy phase shift of a few-cycle
light field in a focused or defocused beam.
2. CARRIER-ENVELOPE PHASE OF A MODE-

LOCKED PULSE TRAIN AND A SINGLE
ATTOSECOND PULSE
The temporal evolution of a linearly polarized electric field of a single

pulse of light can be expressed
E L (t ) = AL (t ) cos[ω L (t )t + φ ] . (1)
This description includes three physical quantities: the amplitude AL (t ) , the

frequency of field oscillations ω L (t ) = ω 0 + β (t ) ,4 and the carrier-envelope
phase, φ. The latter parameter determines the timing of the carrier-wave
4
The notation used in this chapter corresponds to that of other chapters through the following
relationships: ωn = 2πνn; ωr = 2πfr; ωCE = 2πf0; ∆φ = ∆φce
oscillations with respect to the amplitude envelope AL (t ) , whereas β(t)

stands for a possible chirp, i.e., a carrier frequency sweep across the pulse.
The relationship between the carrier-wave and the envelope changes as a
result of pulse propagation through a dispersive medium. Therefore,
Equation (1) is only relevant at a specific location. Also, referring to a
carrier-envelope–phase-stabilized pulse train, we mean that at a given point
in space every pulse has the same parameters in Equation (1) and imply that
the phase evolution with propagation is identical for all these pulses.
For applications of nonlinear optics that are sensitive to the strength of
individual field peaks, a change in φ makes a physically measurable
difference only if the amplitude is subject to significant variation within the
∂AL (t ) 2π
light period T0 = 2π/ ω0, i.e., ≅ AL (t ) . Evidently, this condition
∂t T0
is fulfilled for 5 fs pulses [Figure 10-1(b) and (c)], where the XUV emission
depends on φ, as opposed to 20 fs pulses [Figure 10-1(a)] where the strength
variation between neighboring peaks is substantially smaller. Nevertheless,
the carrier-envelope phase preserves a clear meaning regardless of the pulse
duration. This can be easily understood by invoking the frequency-domain
representation of a mode-locked laser spectrum [9-13, 17, 18, 41]. The
frequency of each individual laser mode is
ω n = nω r + ω CE , (2)
where n is the mode number, ω r is the repetition rate of the laser, and ω CE is
a frequency shift from an exact integer multiple of ω r . Equation (2)
describes the comb of frequencies across the mode-locked bandwidth that is
of interest in frequency-domain metrology [17, 42]. The physical origin of
the frequency mismatch ω CE lies in the difference between the group delay
(the cavity round-trip time of the envelope) and the phase delay (the round
trip of the carrier wave). Consequently, the carrier-envelope phase (i.e., the
position of the carrier-wave oscillation with respect to the envelope that is
advancing at a different velocity) is shifted with each successive laser pulse
by
ω CE
∆φ = 2π . (3)
ωr
The phase slippage [8] ∆φ can be compensated by managing the cavity

dispersion [15, 36, 43], thereby forcing all emitted pulses to carry the same
270 Chapter 10
value of φ. Reduction of the repetition rate (e.g., by selecting every mth pulse
with a pulse picker in front of a chirped-pulse-amplifier) leads to an m-fold
increase of the mode density but does not scale the frequency mode
mismatch, which remains mod ω r m {ω CE } . In fact, the generation of new
frequency modes is the result of applying a modulation that is nonlinear in
the time domain (the fast switching of the pulse picker). This nonlinearity
does not destroy the properties of the frequency comb as all the modes
remain rigorously locked to each other. The same reasoning applies to the
case of a single-isolated laser pulse ( ωr = 0 ) that corresponds to a
continuously filled spectrum, I L (ω ) . In this treatment, φ can be interpreted
as a frequency-independent offset of the spectral phase φ L (ω ) . This
relationship is immediately apparent from the following Fourier-transform
link between the time- and frequency-domain electric fields
I L (ω ) exp[i(φ L (ω ) − ωt + φ )] dω + c. c. .
1
E L (t ) = (4)
2π ∫
Here the intensity spectrum I L (ω ) and the phase φ L (ω ) fully determine the
quantities AL (t ) and ω L (t ) [2] but do not specify the carrier-wave offset.
The spectral manifestation of φ, given by Equation (4), is particularly
valuable for describing various single-shot measurements of the carrier-
envelope-phase throughout the chapter.
3. MEASUREMENT OF PHASE VARIATIONS
In this section, we address practical methods for characterizing carrier-

envelope phase changes that are vital for implementing phase-stabilizing
feedback loops.
3.1 Detecting carrier-envelope drift of oscillator pulses
The standard tracking technique is employed in several carrier-envelope–

phase-controlled oscillators [12, 13, 16-18] and will be discussed later in this
section. It relies on detecting the frequency beat from the laser modes of the
frequency comb that are separated by a spectral octave, i.e., they differ in
their frequency by a harmonic order. The lower-frequency mode is then
frequency doubled and the two modes are brought into interference with
each other, producing a beat at the carrier-envelope frequency
ωCE = 2(nω r + ω CE ) − (2nω r + ω CE ) . (5)
This scheme requires a very broadband, octave-spanning spectrum,

which is unavailable from a typical mode-locked oscillator. To gain the
necessary bandwidth, the laser spectrum can be broadened in a medium with
third-order nonlinearity. Using a photonic crystal fiber (PCF) [44], this task
can be accomplished even with nJ pulses. The device incorporating the
spectral broadening, frequency doubling, and beat-signal generation is called
a nonlinear interferometer or ν-to-2ν setup. In state-of-the-art broadband
mode-locked oscillators, it has already become possible to derive the ν and
the 2ν, frequency components directly from the laser spectrum, thus
obviating the need for extracavity spectral broadening [45]. For further
details on ν-to-2ν interferometry, see Chapter 1.
3.2 Detecting carrier-envelope drift of amplified pulses
There are three significant difference in the phase-drift of amplified laser

pulses compared with an oscillator. First, the repetition rate of the amplifier
must be lowered, typically by 5 orders of magnitude, to 1–10 kHz, which
complicates accurate detection of ω r and ω CE . Second, the availability of
high-energy pulses simplifies the problem of producing an octave-spanning
continuum. Third, laser amplifiers exhibit a considerably higher peak
intensity noise as compared to “quiet” Kerr-lens–mode-locked oscillators
[46]. Consequently, the excursions of a quasi-monochromatic ν-to-2ν signal
detected behind the amplifier do not necessarily indicate a change in carrier-
envelope phase but might reflect pulse energy fluctuations.
Polarizer
(b)
VND Spectrometer
SHG
Epulse<1 µJ
WLG (sapphire)
Figure 10-2. Setup for phase characterization of amplified pulses by nonlinear spectral
interferometry. VND: variable neutral density filter; SHG: second harmonic generator; WLG:
white light generator.
Initial amplifier phase tracking schemes [7, 20, 40] took advantage of
abundant pulse energy and used different types of single-shot nonlinear
spectral interferometry. The objective was to use broadband detection of the
ν-to-2ν beat as opposed to the narrowband scheme acceptable in the case of
272 Chapter 10
an oscillator. The essence of this approach can be described as follows. After

generating a sufficiently broadband spectrum, I WL (ω ) , the pulse is sent to a
second-harmonic-generation (SHG) crystal, resulting in an up-shifted field
E SHG (ω ) ∝ exp(i 2ϕ )∫ χ (ω : ω ′, ω − ω ′) I WL (ω ′) I WL (ω − ω ′)
2
× exp[i(φWL (ω ′) + φWL (ω − ω ′) )]dω ′ (6)

= I SHG (ω ) exp[i(φ SHG (ω ) + 2ϕ )],
where φWL (ω ) is the spectral phase of the white light and χ is the second-
2
order susceptibility. The interference between the white-light and SHG

pulses, separated by the time delay τ 0 , yields a spectrum
S (ω ) = (1 − a) I WL (ω ) + aI SHG (ω )
, (7)
+ 2 a(1 − a) I WL (ω ) I SHG (ω ) cos(φ SHG (ω ) − φWL (ω ) + ωτ 0 + ϕ )
where the coefficient a stands for the polarizer transmission for the
polarization of the SHG light. For a sufficiently large pulse separation τ 0 ,
corresponding to many spectral fringes in the interferogram S (ω ) , the
argument of the cosine in Equation (7) can be recovered using the standard
algorithm of spectral interferometry [47]. The phase of the newly obtained
complex spectral function then directly yields the differential phase of the
two fields, mod 2π {φ SHG (ω ) − φWL (ω ) + ϕ }. In principle, such an interference
measurement can be used not only to characterize the pulse-to-pulse changes
of φ, but also to find the actual value of φ. In practice, however, this is not
feasible because with current methods, φ SHG (ω ) and φWL (ω ) can only be
characterized with the precision of an arbitrary constant [2]. Only by fully
accounting for the evolution of the spectral phase in the SHG crystal, which
includes both linear and nonlinear pulse propagation, would it be possible to
retrieve the value of φ in this scheme. Therefore, the ν-to-2ν interferometer
can only measure the full carrier-envelope phase after calibration by an
independent external experiment, as shown in Section 7. Otherwise, this
device reflects the phase jump in the jth laser pulse with respect to another
(e.g., 0th) pulse arriving earlier from the same pulse train and chosen as
reference, i.e., the interferometer measures δφj = φj − φ0. The difference
between the method outlined above and Fourier-transform-spectral-
interferometry based on the use of parametric waves, as reported in
Reference [22], is that, in the latter case, one measures the interference
between the frequency-doubled idler and the signal pulses. The advantage is
that the white-light continuum injected into the parametric device does not
have to span an octave and the interference is observed close to the spectral
peak of the signal rather than in the continuum wing.
In this work, however, we opted for a simpler setup, as shown in [Figure
10-2], which is similar to the scheme published in Reference [20]. For
further simplification, we replaced white-light generation in a hollow
waveguide with bulk continuum generation in a 2 mm of sapphire. Because
of the dispersion of the sapphire and of the 1 mm BBO (β-BaB2O4)
frequency-doubling crystal, the white-light and SHG are sufficiently
separated in time to employ spectral interferometry. The spectral
interferogram is measured by a fiber-coupled spectrometer behind a
polarizer cube that selects the common polarization of the beams. With a 2
MHz acquisition card, single-shot ν-to-2ν interferograms can be detected for
every 3rd laser shot at a 1 kHz repetition rate. In the ν-to-2ν unit shown in
Figure 10-2, the timing jitter, τ = τ 0 + δτ , has an immediate impact on the
fringe pattern of the measured interferogram, as will be shown in Section
3.3. However, for the collinear arrangement, depicted in Figure 10-2, the
problem of timing jitter is negligible as the interfering beams follow the
same path.
3.3 Measuring the phase difference by linear

interferometry
We have presented interferometric methods that employ nonlinear

frequency conversion and reveal variations in the value of the carrier-
envelope phase. Here we discuss the technique of linear spectral
interferometry [48, 49], which measures the phase difference between the
test and reference arms of an interferometer. It is instructive to compare the
relevance of this method to the characterization of carrier-envelope-phase
drift with the measurement of pulse envelope and chirp. If the intensity
spectrum I L ( ω ) and the phase φ L (ω ) are determined in front of an
interferometer and the phase accumulated inside the interferometer φ ( ω ) is
recovered by spectral interferometry, then the phase of the outgoing pulse is
known and equals φ L (ω ) + φ (ω ) . Therefore, once φ L (ω ) is found with a
nonlinear optical technique (e.g., autocorrelation or frequency-resolved
optical gating [2]), complete phase characterization can be performed with
linear methods provided a reference beam derived after the nonlinear
measurement is available. In essence, we have just stated the principle of the
pulse measurement by TADPOLE (Temporal Analysis by Dispersing a Pair
Of Light E-fields) [2].
Linear spectral interferometry can be used in exactly the same way to
study carrier-envelope-phase deviations occurring behind a nonlinear ν-to-
274 Chapter 10
2ν interferometer. To do so, a small fraction of the reference beam can be

split behind the point where the changes of φ are tracked with the ν-to-2ν
unit and sent to propagate through free space (air). For practical purposes,
fluctuations of air density are the only cause of carrier-envelope-phase jitter
of the reference pulses. Therefore, we consider these pulses carrier-
envelope-phase stable. The test beam is directed into an optical arrangement
(e.g., an amplifier) and accumulates a phase difference φ (ω ) with respect to
the reference. As a consequence of beam-pointing instability during
propagation through elements with angular dispersion and other linear and/or
nonlinear dispersion changes [50], the test pulse can pick up a frequency-
dependent phase offset
δ (ω ) = δ (ω 0 ) + δ ′(ω 0 )(ω − ω 0 ) + δ ′′(ω 0 )(ω − ω 0 )2 +

1
2
1
6
K. (8)
We assume that δ (ω ) corresponds to a very minute dispersion modification

that does not involve reshaping of the pulse envelope. Otherwise, in case
AL (t ) is modified, carrier-envelope-phase control would become irrelevant.
Indeed, a tiny (<1 µ m) change in the propagation length within dispersive
material already leads to a sizable leap in the carrier-envelope phase [8],
which we call δφ . Consequently, we can justifiably disregard higher-order
terms, starting with the parabolic phase distortion. In fact, δφ represents the
sum of frequency-independent terms in Equation (8). The linear shift with
frequency term in Equation (8) is next in significance. It simply represents
the time shift due to a modified propagation length and, therefore, has no
bearing on the carrier-envelope dynamics. Thus we will only pay attention to
the phase jump, δφ , in the characterization of the phase drift of the test pulse
and treat the test-reference timing offset separately.
As was pointed out in Section 3.2, a temporal jitter between the
interfering pulses causes fringe rearrangement in the spectrogram S (ω ) . This
timing variation is inevitable, especially if the reference beam is not
interferometrically stabilized. For large laser systems, the path of the
reference beam can be several meters long and fluctuate by several
wavelengths. Thus, the ability of spectral interferometry (linear or nonlinear)
to discriminate between δφ and the timing jitter needs to be scrutinized.
Figure 10-3 gives an overview of the properties of spectral
interferometry. Figure 10-3(a) shows a typical interferogram and the power
spectrum of its Fourier transform (inset) used in phase retrieval. In Figure
10-3(b), we examine the impact on the interferogram of a pure time shift
( δτ ) and a pure phase jump ( δφ ), depicted by the solid and the dashed
curves, respectively. Figure 10-3(c) presents the corresponding phases after
subtracting the phase slope ωτ 0 , which is due to the delay between the test
and reference pulses. The shape of the intensity spectrum I L (ω ) [solid curve
in Figure 10-3(a)] and differential phase φ (ω ) [dotted curve in Figure 10-
3(c)] are selected arbitrarily. The values of δτ and δφ are chosen so that the
interferograms in Figure 10-3(b) closely resemble each other. Nevertheless,
there is a fundamental difference because the timing variation changes the
fringe period. This change can be clearly observed in an interferogram
consisting of a large number of fringes. Therefore, a charged coupled device
(CCD) camera-based spectrograph (which covers the relevant spectral width
and has a sufficiently high resolution) makes it possible to separate the
contributions of δφ and δτ as long as Fourier-transform–spectral-
interferometry noise criteria [49] are met. We implemented the time jitter
correction numerically in our Fourier-transform–spectral-interferometry
code by extracting a linear phase obtained by back-transformation into the
frequency domain.
(a) FT φ(ω)+ωτ0
Intensity
Intensity
-τ0 0 τ0
Time
(b)
Intensity
φ(ω)+ωτ0+δϕ
φ(ω)+ω(τ0+δτ)
(c) φ(ω)+ωδτ
φ(ω)+δϕ
Phase
δϕ
φ(ω)
δτ
0
0 ω0
Frequency
Figure 10-3. Discrimination between phase and timing jitter in Fourier-transform spectral
interferometry. (a) Intensity spectrum (solid curve) and spectral interferogram (dotted curve).
(b) Spectral interferograms modified by a timing shift (solid curve) and a frequency-
independent phase shift (dashed curve). (The fringe period changes in the case of timing
shift.) (c) Differential spectral phases after time-delay slope subtraction. Dotted and solid and
dashed curves correspond to interferograms in (a) and (b), respectively. Inset shows the power
276 Chapter 10
spectrum of the Fourier transform to the time domain. Dashed contour corresponds to a
possible filter required in the phase extraction to isolate the temporal peak at τ0.
The immediate implication of the timing-problem solution is the ability

to trace carrier-envelope-phase jumps with respect to a reference beam
propagating in free space. We believe this method can be particularly useful
for extending carrier-envelope-phase stabilization in a multistage laser
system, in which the earliest stage is phase stabilized and subsequent stages
are expected to introduce only slow phase drift.
4. PHASE JITTER OF THE WHITE-LIGHT

CONTINUUM
The ν-to-2ν phase characterization methods presented in Section 3 (and

all phase-stabilized laser systems reported by 2002 [12, 13, 18]) rely on
white-light generation to create a phase-coherent, spectrally broadened
replica of the input laser pulse. The phase noise added in the process of
white-light generation is an intrinsic problem of carrier-envelope-phase
stabilization loops [18]. In this section, we outline the reason for the phase
coherence between the white light and the initial pulse and then study the
influence of intensity fluctuations on phase noise.
4.1 Phase lock between the input pulse and the white-
light continuum
Generally, white-light generation consists of several nonlinear wave-

mixing processes that lead to generation of new frequency components.
Therefore, it is not immediately clear whether white-light generation can be
relied on for producing a frequency-broadened, but phase-offset-repeating,
replica of the laser pulse. The complete propagation equation for pulse self-
action (assuming the perturbative regime of nonlinear optics [6]) is derived
directly from Maxwell’s equation [6, 51] to be
[∇ 2
]
+ k 2 (ω ) E (r , ω ) =
ω 2 nl
ε 0c 2
P (r, ω ) , (9)
where E (r , ω ) and P nl (r, ω ) denote the electric field and generated

nonlinear polarization, respectively. The particular form of P nl (r, ω ) that
describes both the delayed Raman response (i.e., slow nonlinearity) and the
parametric four-wave mixing based on fast-electronic Kerr nonlinearity, also

known as the process of self-phase-modulation [51] is
P nl (r, ω ) = ∫∫ χ (ω : ω1 ,ω 2 , ω1 + ω 2 − ω )
3
× E (r, ω1 ) E (r, ω 2 ) E * (r, ω1 + ω 2 − ω )dω1 dω 2

= exp[i(ϕ + ϕ − ϕ )]
× ∫∫ χ (ω : ω1 , ω 2 , ω1 + ω 2 − ω ) I (r, ω1 ) I (r, ω 2 ) I (r, ω1 + ω 2 − ω )
3
× exp[i(φ (r ,ω1 ) + φ (r, ω 2 ) − φ (r, ω1 + ω 2 − ω ) )]dω1 dω 2 ,

(10)
where χ
3
represents the effective third-order susceptibility and the
conjugation symbol marks emission [52]. The propagation-dependent
change of dispersion is included in φ (r , ω ) . Here we assume that all mixing
frequencies are positive [53]. The appearance of the conjugation operator in
the convolution integral in Equation (10) accounts for the preservation of the
initial phase offset φ in the output field. After solving Equation (9) over the
interaction length of the white-light-generation medium r0 → r1 , the
resultant output pulse will generally carry a different value of the carrier-
envelope phase [6, 54] than the input pulse (i.e., φWL ≠ φ), which is also the
case for any type of dispersive linear or nonlinear pulse propagation.
Nevertheless, the important criterion for an accurate carrier-phase-phase drift
determination is the ability of the white-light continuum to repeat (without
distortion) the input phase offset. This ability is the consequence of the rule
for phase summation in Equation (10). After plugging Equation (10) into
Equation (9), φ cancels out, i.e., it has no effect on φ (r1 , ω ) of the output
field regardless of the microscopic origin of the nonlinearity. The presence
of a resonant or nonresonant fifth-order parametric wave mixing
( EE * EE * E ) [55] does not lead to a φ-dependent propagation either.
However, the solution for pulse propagation does become φ-dependent if
second-harmonic generation and/or third-harmonic generation cannot be
neglected. To account for these contributions, P nl (r, ω ) should also include
the terms describing the three-wave mixing of second-harmonic generation
and the four-wave mixing of third-harmonic generation, with their phase
summation rules given by the field products EE and EEE , respectively.
We would like to underscore two points. First, the phase shift added by
both the nonresonant and/or resonant white-light generation does not depend
on the carrier-envelope phase of the input pulse. Therefore, the techniques of
spectral broadening can be readily applied to schemes for measuring the
278 Chapter 10
carrier-envelope-phase drift. Second, mixing in of undesired mechanisms of

frequency conversion, such as second-harmonic generation (or, in a broader
sense, sum-frequency generation) and third-harmonic generation, can distort
the frequency-independent phase of the white-light pulse even if there are no
intensity fluctuations of the input pulse. In typical experimental conditions,
however, second- and third-harmonic generation do not play a significant
role. For instance, non-phase-matched second-harmonic radiation is often
observed in the cladding of conventional single-mode fibers, but it is
probably not sufficiently strong to cause a noticeable pulse-to-pulse phase
jitter. For bulk-material chirping (typically performed in sapphire or CaF2),
the contributions of second- and third-harmonic generation are negligible.
4.2 Phase noise resulting from intensity fluctuations
In Section 4.1, we dismissed the interplay between different types of

nonlinearity as a concern for added phase jitter in white-light generation. In
practice, a much more important phase distortion is the variation of the
intensity-induced phase shift [56]. As the result of intensity-dependent
nonlinear interaction, the phase accumulated throughout the white-light
generation can vary from pulse to pulse and compromise the fidelity of
carrier-envelope-phase-drift characterization. In the case of ν-to-2ν
interferometers suited for oscillators, these problems can arise from
mechanical vibrations of the PCF, which affect the amount of light coupled
in. In contrast, ν-to-2ν interferometers for amplified pulses can employ bulk
white-light generation (cf. Section 3.2), which is almost insensitive to beam-
pointing fluctuations. Typical studies of the phase stability of white-light
generation consist of interference measurements of independently produced
white-light continua [57, 58]. The influence of white-light-generation phase
noise on the operation of phase-stabilization loops has also been investigated
by measuring the carrier-envelope drift with two independent PCF-based ν-
to-2ν interferometers [18, 59].
Here, we check the phase jitter of white-light generation in bulk sapphire
against increased intensity noise of amplified pulses [46]. The white-light-
generation stage shown in Figure 10-2 was supplemented by the linear
interferometer depicted in Figure 10-4. This addition allowed us to measure
the phase shift of the white light with respect to the replica of the injected
pulse. The intensity of the beam was varied with a neutral density filter
while single-shot–white-light interferograms and laser spectra were being
recorded by a dual-channel spectrometer. The synchronous registration of
the laser spectra is necessary to calibrate the pulse energy in each laser shot.
We used a multipass Ti:sapphire amplifier with an rms intensity noise of
1%. The optimized setting of the neutral density filter corresponds to a pulse
energy of 0.7 µJ in front of the sapphire and a focal-spot diameter of ~ 30

µm. These parameters allowed us to generate a bandwidth adequate for ν-to-
2ν measurement. For these conditions, the rms phase jitter retrieved from the
laser–white-light interferogram was ~100 mrad.
Spectrometer Adjustable
delay
CCD 1 CCD 2
BS
VND
BS BS WLG
Figure 10-4. Single-shot spectral interferometer for measurement of carrier-envelope-phase

jitter arising from white-light generation. VND: variable neutral density filter and BS:
dielectric beamsplitter.
Next, we recorded phase statistics for various settings of the neutral

density filter. The resultant phase evolution, plotted as a function of incident
pulse energy, is presented in Figure 10-5. The data cover several distinct
regimes, indicated in Figure 10-5 by vertical sections. First, we notice that
there is an abrupt phase change that corresponds to the increasing strength of
self-focusing and filament formation. This phase change can be explained by
an alteration in the group velocity of light with the increase of the intensity-
dependent part of the refractive index of sapphire. After the markedly noisy
region of the onset of filamentation, the phase shows a certain “rebound”
that probably indicates an increase in the filament length. The rebound is
followed by an interval of stable filamentation that is relatively broad in
terms of pulse energy — about ±8% around the energy set point of the ν-to-
2ν setup. At higher pulse energies, we observe a collapse of the filament and
multiple filamentation. The phase noise and the downward phase trend
somewhat resemble the situation before the formation of a single stable
filament. Measurement at even higher energies could not be performed with
a multifilament beam because of the breakdown of the interference pattern
into several modes.
In the regime of a stable filament, the phase exhibits a quasi-linear
change with energy. The rate of this variation, given by the straight line in
Figure 10-5, is 12 mrad/nJ. This rate corresponds to an 84 mrad phase shift
for a 1% jump in the pulse energy.
The phase evolution covers only the spectral overlap region between the
white light and the input pulse, i.e., approximately a 60 nm bandwidth.
280 Chapter 10
Therefore, the results of our measurement correspond to the best-case

scenario since phase excursions in the blue (2ν) spectral region cannot be
visualized. Additionally, the second-harmonic generation of the red-shifted
part of the continuum also causes an intensity-dependent phase shift and,
consequently, is likely to worsen the precision of the ν-to-2ν carrier-
envelope-phase-drift characterization.
5π Increasing Stable single

self-focusing filament
Onset of Breakdown of
Nonlinear phase shift
filament single filament
conditions of
f-to-2f measurement
-5π
200 400 600 800 1000

Pulse energy [nJ]
Figure 10-5. Intensity-phase coupling of the white-light generation in bulk sapphire. Solid
dots depict phase changes extracted from single-shot interferograms. Straight line shows a
linear fit marking the slope of phase intensity-dependent variation in the vicinity of the pulse
energy used for ν-to-2ν characterization.
The issues of the phase properties of white-light generation presented in

this section are important for identifying the limitations of carrier-envelope-
phase-control schemes and for evaluating the applicability of white-light
generation as a tool for extending a mode-locked frequency comb [58] and
creating phase-coherent frequency-shifted pulse replicas [21].
5. CONCEPTS OF PHASE-CONTROLLED
AMPLIFICATION
In this section, we suggest several practical designs of carrier-envelope

phase-stabilized amplifiers. The corresponding schematics are sketched in
Figure 10-6 and can employ both CPA [60] and optical parametric
amplification (OPA) [61, 62]. Two types of carrier-envelope phase noise or
drift in amplifiers must be considered. The first kind of phase noise is
associated with the fluctuations of the intensity-dependent nonlinear phase
contribution. It can vary randomly from pulse to pulse as the result of pump
energy fluctuations and, therefore, cannot be compensated with active phase-
stabilization loops. To minimize this problem, the value of the B-integral
[63] ψ (t ) = k (ω 0 )n2 ∫ I (t , z )dz , i.e., the accumulated nonlinear phase, or
self-phase modulation in the amplifier, should be on the order of π . This
condition is fulfilled in typical CPA amplifiers. Recent measurements of
time-domain interferometric cross-correlation between seed-oscillator pulses
and their amplified replicas have shown that the phase jitter at the output of a
multipass Ti:sapphire amplifier lies below 0.1 rad [64]. In fact, with pump
energy fluctuations being the chief contributor to the jitter of the nonlinear
phase, the use of low-noise diode-pumped pump lasers should make it
possible to phase stabilize amplifiers operating at higher B-integral values.
The second kind of carrier-envelope drift is caused by beam-pointing
instability that leads to path-length fluctuations in dispersive elements [50]
of the amplifier. Such mechanical perturbations occur on a relatively slow
timescale, with characteristic time constants of 1 second or longer.
Therefore, the resulting carrier-envelope phase modulation could be easily
traced and precompensated.
The schemes of carrier-envelope-phase control amplification, discussed
here, can be divided into three categories: (1) all-optical phase self-
stabilization [Figure 10-6(a)]; (2) passive triggering on a seed pulse with an
appropriate carrier-envelope-phase value [Figure 10-6(b)]; and (3) use of
carrier-envelope–phase-stabilized seed with a subloop for phase correction
of the amplifier output [Figure 10-6(c)–(e)].
Recently it has been shown that passive carrier-envelope-phase
stabilization can be obtained in an appropriately configured OPA [21]. This
is achieved by seeding the OPA with a white-light pulse generated by the
OPA pump pulse. As discussed in Section 4, both the seed and the pump
pulses are phase coherent and, in the process of difference-frequency
generation, the phases of these two pulses cancel. Therefore, the amplified
idler pulse, which is a result of difference-frequency generation between the
pump and the seed, is free of carrier-envelope drift even when the OPA is
pumped with random-carrier-envelope-phase pulses. This scheme, shown in
Figure 10-6(a), can be used as a phase-stable extension of a conventional,
free-running CPA. Subsequent amplification can be carried out by
broadband chirped-pulse OPA into a TW–PW regime [62]. The drawback of
this approach is that the red-shifted output wavelength of the idler wave does
not permit postamplification to be carried out in the laser gain medium used
282 Chapter 10
to generate the pump beam. Consequently, implementation of all-optical

phase stabilization in high-energy systems is not straightforward and
demands development of dedicated amplification stages.
Other designs presented in Figure 10-6(b)–(e) are intended as upgrades of
existing amplified systems. Figure 10-6(b) represents the simplest possible
realization of phase control based on the 105 difference in the repetition rates
of the seed oscillator and a kHz amplifier. The carrier-envelope phase-
evolution of the oscillator pulse train is quasi-periodic [8, 13]. The rate of the
pulse-to-pulse phase slippage ∆φ can be adjusted by manipulating the
cavity dispersion. With a ν-to-2ν interferometer, one can monitor the
evolution of the ω CE beat and employ this signal to trigger a pulse picker to
select an oscillator pulse with a desired phase [13]. To ensure proper
amplification, the pulse picker must be triggered shortly after the flash of the
kHz pump laser and within the lifetime of population inversion, which
leaves room to select an “appropriate” seed pulse from thousands of
oscillator shots. The key drawback of this scheme is the relatively poor
precision with which such a selection can be performed, since there could be
several additional laser shots between the triggering pulse and the one that is
actually sent to the amplifier. In this situation, variations of the quasi-
periodic phase change would introduce a variable carrier-envelope-phase
shift of the selected pulse. Another difficulty in this method is the lack of an
obvious mechanism to correct phase fluctuations arising in the amplifier.
(a) optional post amplifier

CE-
Epulse>10 µJ OPA stabilized
pump OPA output
WLG seed
(b)
Epulse~5 nJ 3 nJ
Oscillator PP Ampl-
ifier
Trigerring
2 nJ signal
NLI Comparator
(c) Epulse~60 nJ
cw 3 nJ
Oscillator ampl- PP Ampl-
ifier ifier
2 nJ
Cavity
GD <0.1 µJ
control NLI
Linear
PLL Σ FTSI
(d) Epulse~60 nJ
kHz rep. rate 3 nJ
Ampl-
Oscillator PP ifier
2 nJ
Cavity MHz rep. rate <0.1 µJ
GD NLI
control Linear
FTSI
PLL Σ
(e)
Epulse=5 nJ 3 nJ
Oscillator PP Ampl-
ifier
~1 µJ
Cavity
GD 2 nJ NLI
control NLI
Nonlinear
FTSI
PLL Σ
Figure 10-6. Multiple concepts of phase-stable pulse amplification. (a) All-optical (passive)
phase stabilization with optical parametric amplification (OPA). (b) Phase-selective seed by a
free-running oscillator. (c)–(e) Schemes employing a phase-stabilized seed oscillator with a
secondary phase-correction loop behind the amplifier based on linear (c), (d) and nonlinear (e)
phase-drift detection. Concepts (c) and (d) require high-energy seed that can be attained with
a cw-amplified or a long-cavity oscillator (c) and a cavity-dumped oscillator (d). PP: pulse
picker; PLL: phase-locked loop; NLI: nonlinear ν-to-2ν interferometer; and FTSI: Fourier-
transform spectral interferometry.
284 Chapter 10
The schemes drawn in Figure 10-6(c) and (d) are based on the use of a
carrier-envelope-controlled oscillator [12, 13]. They explore the possibility
of sending a reference (pilot) beam for the measurement of the phase offset
in the amplifier. Information on the slow phase drift behind the amplifier is
obtained with the linear spectral-interferometry described in Section 3.3. It
can then be plugged into the phase-stabilization loop of the oscillator. The
core of the phase-stabilizing electronics is the phase-lock loop (PLL in
Figure 10-6) that forces the beat signal ω CE to oscillate in phase with an
external local oscillator via a feedback to the intracavity dispersion. The
inverse of the phase offset measured by spectral interferometry at the
amplifier output can then be combined with the phase of the local oscillator
signal ( 1 4 ω r ), thus precompensating this offset already in the oscillator.
The use of the reference beam for (a potentially single-shot) spectral
interferometry imposes the demand for high-seed pulse energy. Pulses of
tens of nanojoules can be obtained by intermediate cw amplification of the
seed pulse train before ν-to-2ν characterization, by employing long-cavity
oscillators [13, 65] or using a cavity-dumped oscillator [66]. This type of
laser has a dual output. It emits a pulse train through the output coupler at
the full repetition rate, which can be conveniently used to drive the ν-to-2ν
interferometer. At the same time, the pulses ejected by the intracavity pulse
picker are directly synchronized with the amplifier and satisfy the needs for
a kHz seed and a spectral interferometry reference.
The remaining concept for phase control of an amplifier [Figure 10-6(e)]
offers the most compatibility with existing CPA systems and is well suited
for retrofitting them. This scheme requires implementation of a standard
phase-lock loop of the seed oscillator, which consists of a ν-to-
2ν interferometer for the oscillator, electronics, and an additional secondary
feedback loop. The feedback loop consists of a high-energy ν-to-2ν
interferometer and computer for the spectral interferometry. Neither the
oscillator nor the amplifier requires significant modification. We have
followed this blueprint to upgrade a standard multipass Ti:sapphire CPA
system [46] for carrier-envelope–phase-stabilized operation.
6. PHASE-STABILIZED 5 fs, 0.1 TW-AMPLIFIED

SYSTEM
In our high-power laser setup [Figure 10-6(e)], we have incorporated a

commercially available phase-stabilized 10 fs laser system (FS 800, Menlo
Systems GmbH). It consists of a dispersive-mirror-controlled Kerr-lens–
mode-locked Ti:sapphire oscillator (Femtosource Compact Pro, Femtolasers
GmbH), a ν-to-2ν interferometer, and phase-locking electronics that drive an
acousto-optic modulator to control the intensity of the cw pump laser (Verdi

V, Coherent). About 50% of the output of the oscillator is diverted into 2 cm
of PCF for detection of the carrier-envelope beat frequency. The reference
signal (i.e., the local oscillator) for the phase-locked loop is derived by
dividing the repetition rate of the oscillator (80 MHz) by a factor of 4. As
can be seen from Equation (3), selecting ω r or its (sub)harmonics as the
reference signal is the best choice to control the value of ω CE . Before
activating the phase lock loop, we set the intracavity carrier-envelope-phase
shift ∆φ to approximately 2π(m + 1/4), where m is an integer. This rough
adjustment of ∆φ is performed by changing the optical path length through a
pair of thin intracavity wedges of fused silica. At this stage, φ is
approximately reproduced in every fourth round trip in the laser cavity,
resulting in a quasi-periodic modulation of the ν-to-2ν beat signal at ω r 4 ,
i.e., 20 MHz. Next, to enhance the accuracy of the phase reproducibility,
ω CE and ωr / 4 are phase locked. Phase locking is achieved by controlling
∆φ via nonlinear effects in the Ti:sapphire crystal through the variation of
the pump intensity [8]. For this purpose, the cw pump laser is passed through
the acousto-optic modulator [13, 50], which varies the power of the
transmitted beam according to the phase error [Figure 10-6(c)–(e)]. As a
result, every fourth pulse in the 80 MHz pulse train carries the same carrier-
envelope phase, giving rise to an accurate optical field reproduction.
In 2004, we employed a multipass Ti:sapphire chirped-pulse amplifier
(Femtopower, Femtolasers GmbH). The Q-switched diode-pumped pump
laser (Corona, Coherent Inc.) of the amplifier and the pulse picker (Pockels
cell) in the amplifier were synchronized with exactly every 80,000th pulse
from the oscillator to ensure that only pulses with identical φ are amplified
(at a 1 kHz repetition rate). The nanojoule pulses are temporally stretched by
a glass block before their energy is boosted to ≈ 1 millijoule in 9 passes
through a 40 mm Ti:sapphire crystal. After amplification, the duration of the
pulses is reduced to ≈20 fs in a refractive pulse compressor consisting of two
pairs of Brewster-angled prisms. To shorten the pulse even further, the
output of the amplifier is injected into a 1 m hollow-core glass fiber filled
with neon gas and subsequently passed through a set of ultrabroadband
chirped mirrors. To detect carrier-envelope-phase variations, a small fraction
(< 1%) of the energy of the 20 fs pulses is split off from the main beam
directly behind the amplifier and fed into the second ν-to-2ν interferometer
sketched in Figure 10-2 and discussed in Section 3.2. The parameters of the
resultant output pulses are summarized in Figure 10-7(a). The insets in
Figure 10-7(a) depict the measured spectrum and second-order
interferometric autocorrelation of the compressed pulses. An iterative pulse
retrieval from these data yields a near-bandwidth-limited pulse with an
286 Chapter 10
amplitude envelope AL (t ) , as shown by the thin-dashed line in Figure 10-

7(a). It also yields a pulse duration [full width at half maximum of AL (t ) 2 ]
of τL = 5.4 ± 0.5 fs.
Experimental results obtained with the second ν-to-2ν interferometer in
different phase-stabilization regimes of the amplifier are summarized in
Figure 10-7(b). In the case of a phase-unlocked seed oscillator, the
interference pattern is completely averaged out within approximately 200
laser shots [black curve, left inset of Figure 10-7(b)]. This finding
underscores the importance of a single-shot monitoring capability for the φ
drift. In contrast, seeding the amplifier system with phase-locked pulses
preserves a clear interference pattern for CCD exposure times as long as
several seconds. The only additional stabilization required is the tracing of
ω CE at the oscillator output. The temporal evolution of the carrier-envelope-
phase variation in this mode, which we term simple phase lock, is
exemplified in Figure 10-7(b) by the dotted curve. These data reveal that the
carrier-envelope phase of the high-energy 20 fs pulses remains virtually
stable for periods of up to several seconds. This is remarkable considering
the sensitivity of the carrier-envelope phase to path-length change in any
solid optical medium [8]. Beam direction instabilities in the microradian
range can produce path-length changes of this order of magnitude in the
prism compressor and are likely to provide a significant contribution to the
phase drifts on a longer time scale. However, compared with other
femtosecond amplifiers that employ long-pulse–diffraction-grating-based
stretchers, our system presents a clear advantage in terms of its angular beam
stability. This stability results from employing bulk-material pulse chirping.
The slow drift of φ shown by the dotted curve in Figure 10-7(b) can be
readily tracked by computer analysis of the ν-to-2ν fringe pattern [right inset
in Figure 10-7(b)]. A voltage proportional to the change of φ is generated by
a personal computer that performs the ν-to-2ν Fourier-transform spectral
interferometry analysis. As shown in Figure 10-6(e), this signal is combined
with the fast-changing feedback from the first ν-to-2ν unit that monitors the
phase evolution of the seed oscillator. The absence of fast phase changes in
the amplifier allows us to accumulate the interferograms measured in the
second nonlinear interferometer [Figure 10-2] for about 20–30 ms (i.e., 20–
30 laser shots). The implementation of this signal integration smoothens the
operation of the feedback loop by preventing accidental overshoots of the
carrier-envelope-phase control. Once stabilized at a certain value of φ, the
phase behind the amplifier can be controlled with the Fourier-transform–
spectral-interferometry computer by adding an offset value to the feedback
signal. The simultaneous use of both feedback loops allows us to stabilize
the phase for extended periods (>> 1 minute). The solid curve in Figure 10-
7(b) shows the typical phase drift measured in the second ν-to-2ν device
with both feedbacks active. The device exhibits a rms jitter of ~75 mrad.
This figure, however, should not be confused with the actual precision with
which the carrier-envelope phase can be maintained behind the amplifier.
Since the phase characterization in the nonlinear interferometer is influenced
by the device noise (cf. discussion in Section 3), the feedback loop would
not be able to discriminate this measurement noise from the real phase jitter
in the amplifier. Therefore, an external (out-of-loop) experiment is required
to confirm the carrier-envelope phase stability. It will be presented in
Section 7.
Delay [fs] Wavelength [nm]
(a) -20 -10 0 10 20 1000 800 600
Measured 8 1
Electric field strength [arb. units]
Retrieved
Intensity
Intensity
1
0 0
1.5 2.0
Photon energy [eV]
AL(t) AL(t)cos(ωL(t))
-1
-15 -10 -5 0 5 10
Time [fs]
(b) Unlocked 1 1 f-to-2f interferogram

Intensity
Intensity
CE phase drift, ϕ - ϕ0
+
π
_ 0 0
500 520 540 500 520 540
2 Wavelength [nm] Wavelength [nm]
-π_2 (oscillator
Simple phase-lock
loop only)
Combined phase-lock
(oscillator+amplifier loop)
0 5 10 15
Time [s]
Figure 10-7. Performance overview of the phase-stabilized amplified system. (a) Pulse
properties behind the hollow-fiber–chirped-mirror compressor: retrieved pulse envelope
(dashed curves) and a possible electric field (solid curve) of the compressed pulse. Insets
show measured intensity spectrum (right) and measured and retrieved second-order
interferometric autocorrelation (left). (b) Carrier-envelope phase dynamics with oscillator-
only phase stabilization (dotted curve) and combined oscillator-amplifier phase stabilization
(solid curve). Insets show ν-to-2ν interference spectrograms recorded with the nonlinear
interferometer at the amplifier output [Figure 10-6(e)].
288 Chapter 10
We turn now to the carrier-envelope-phase stability of pulses passing

through a hollow waveguide used to obtain the spectral broadening, as
shown in the right inset of Figure 10-7(a). We placed the second ν-to-2ν
interferometer behind the amplifier rather than at the output of the hollow-
fiber-chirped-mirror compressor to obtain better intensity stability of the 20
fs pulses (around 1% rms). The intensity noise of the white light emerging
from the fiber is higher and varies from 1.6% rms in the spectral region of
the input laser pulse to 4% at 650 nm. At the same time, the spectral breadth
of this white-light is insufficient to support ν-to-2ν detection under optimal
conditions for generating 5 fs pulses. To assess the detrimental effect of
hollow-fiber white-light generation on carrier-envelope-phase stability, we
repeated the experiments described in Section 4 (Figure 10-4). The resultant
pulse-to-pulse rms phase noise was found to range from 80 mrad to 140
mrad depending on the pressure of neon in the capillary. We also conducted
intensity-dependence measurements of the phase offset within a ±15%
energy range around the mean-energy value (700 µJ) injected into the fiber.
This study revealed a 3.4 mrad/µJ slope of the phase within the region of
spectral overlap between the laser pulse and the white light. Unfortunately,
with spectral-interferometry, it was impossible to assess the phase jitter
across the whole spectrum. Nevertheless, one cannot expect dramatic phase
excursions in the spectral wings since they would have an immediate impact
on the pulse profile. Such envelope fluctuations of the compressed pulse
have not been observed in our system. The spectral-interferometry
measurements justify the use of ν-to-2ν detection in front of the hollow
fiber.
7. CONTROL OF LIGHT FIELD OSCILLATIONS
The reference phase in the combined feedback loop of our high-power

phase-controlled laser system can be adjusted within a range of 2π,
providing a convenient means of setting φ to an arbitrary value.
Nevertheless, the techniques we have implemented so far merely enable us
to keep φ constant and vary it by a known amount, whereas the actual value
of φ remains unknown. As opposed to the ν-to-2ν technique, strong-field
phenomena driven by few-cycle pulses have been predicted to be sensitive to
φ and potentially could serve for its determination. The processes predicted
to be suitable for this purpose include high-order harmonic emission [30, 31,
67] and above-threshold ionization [20, 34, 38, 39, 68]. We chose high-order
harmonic generation for our investigations because it has a relatively simple
experimental implementation and relies on linearly polarized light, which is
readily available in our setup. In addition, a clear intuitive insight into this
process had been developed in several excellent models [23, 24, 69] that
have been tested and confirmed by a large number of numerical studies and
experiments [29-31, 70].
The relevant aspects of the process can be summarized as follows. The
highest-energy XUV or soft x-ray photons are emitted around the zero
transition(s) of the laser electric field near the peak of the envelope. This is
implicit in semiclassical models [6, 23, 24, 69] and was recently verified in
an attosecond experiment [27]. The energy of the emitted photon is
essentially determined by the intensity of the laser half-cycle preceding the
emission. For a few-cycle pulse with φ ≈ 0 or π [the “cosine” wave, shown in
Figure 10-1(b)], there is only one most intense half cycle that implies the
emission of an isolated x-ray burst at the highest photon energies
[Figure 1(h)]. The presence of such an isolated feature in the time domain
corresponds to a smooth spectrum. In stark contrast, an identical pulse with a
“sine” carrier [φ ≈ ± π/2, Figure 10-1(c)] exhibits two most intense half
cycles, giving rise to a pair of soft x-ray bursts separated by T0/2 in time
[Figure 10-1(i)]. In the frequency domain, this temporal structure implies a
modulated (quasi-harmonic) spectrum up to the highest photon energies.
To check the validity of this simple intuitive picture, we developed a
computer code simulating propagation of an intense few-cycle laser pulse
through an ionizing medium. The program solves Maxwell’s equations in
three space dimensions and calculates the emergence of high-order
harmonics using a refined version [6] of the quantum theory of Lewenstein
et al. [24, 71]. Figure 10-8 summarizes the results of the simulations relevant
to the experiments described below. In the simulation, the laser pulse
duration is τL = 5 fs, the pulse energy is 0.2 mJ, the beam diameter is 2wL =
122 µm, the medium is neon, the pressure is 100 mbar, and the length is
2 mm. Figure 10-8(a)–(d) displays the computed spectral distribution of few-
cycle-driven coherent soft x-ray emission from neon in the cutoff range for
four different values of the carrier-envelope phase of the driving laser pulse.
Figure 10-8(e) and (f) shows the temporal intensity profile of cutoff x-rays
transmitted through a Gaussian filter [solid contour in Figure 10-8(a)] for φ =
0 and φ = π/2 (solid curves) along with the instantaneous intensity of the
driving laser [i.e., E L (t ) 2 ], depicted by dashed curves. Detailed numerical
simulations show that the smoothest and most-modulated spectral-cutoff
features actually correspond to the values of ~20° and ~110°, respectively,
rather than to the intuitive values of 0° and 90°. However, this discrepancy is
insignificant in our experiments, because it lies within the uncertainty with
which we can identify different carrier-envelope phase values.
The results of our simulations, together with previous numerical studies
[6, 30, 31, 72], support our intuitive analysis and show a strong spectral
290 Chapter 10
dependence between the temporal structure of the x-ray bursts and the
spectral shape of the cutoff region. The width of the smooth spectral
continuum, ∆(hω)cont, that corresponds to an isolated x-ray burst depends on
the ratio of instantaneous intensity E L (t ) 2 of the strongest half cycle and its
neighboring peak. The corresponding difference of instantaneous intensity is
marked by ∆ in Figure 10-8(e). Because of the linear dependence between
the highest-emitted x-ray photon energy and the peak laser intensity, the
continuum bandwidth can be expressed as:
hω cutoff − I p
∆ (hω ) cont = ∆,
(E L (t )
2
)
max
(11)
where I p is the ionization potential. Table 10-1 lists the instantaneous

intensity ratio and the width of the continuum (calculated at 125 eV) for
several values of τL/T0, which is the number of oscillation cycles within the
laser pulse duration.
(a) ϕ=0
Soft-X-ray spectral intensity [arb. units]
(b) ϕ=π/6
(c) ϕ=π/3
(d) ϕ=π/2
120 125 130

Photon energy [eV]
Soft-X-ray intensity [arb. units]
(e) 2
(EL)max ϕ =0 EL(t) [10 W/cm ]
1
∆
2
15
0
(f) ϕ=π/2
2
0
-2 -1 0 1 2 3
Time [fs]
Figure 10-8. Numerical simulations of few-cycle-driven coherent soft x-ray emission from
ionizing atoms. (a)–(d) Cutoff-range spectra are shown for different carrier-envelope-phase
settings of the driving laser pulse. (e)–(f) The solid curves depict the temporal intensity profile
of the cutoff harmonic radiation filtered through a Gaussian bandpass filter with a full width
at half maximum of 7 eV [solid contour in (a), whereas the dashed curves plot EL(t)2]. These
results support the intuitive analysis that the field carrying φ = 0 is predicted to produce a
single soft x-ray burst [filtered in the cutoff (e)]. Deviation of φ from zero gradually
suppresses the magnitude of the main burst and gives rise to a satellite spike. The latter
becomes most prominent for φ → π/2 (f). In the frequency domain, the isolated pulse
emerging for φ = 0 implies a continuous spectrum shown in (a) that becomes increasingly
modulated with the appearance of the second burst for φ → π/2 (b)–(d).
292 Chapter 10
Table 10-1. Expected spectral width of the cutoff continuum for a φ = 0 wave as a function of
pulse duration.
τ L T0 † ∆ (E L ( t )2 )max ‡
∆(hω )cont @125 eV *
2 0.16 20 eV
3 0.07 8.8 eV
4 0.04 5.0 eV
5 0.03 3.8 eV
† Number of oscillation cycles within the laser pulse duration
‡ Instantaneous intensity ratio
* width of continuum.
According to Table 10-1, τL/T0 ≤ 2.5 is required to produce a continuum

with a sizeable (>10 eV) bandwidth in the 100 eV range. For few-cycle
pulses that meet with this criterion, the appearance of the predicted
continuum [Figure 10-8(a)] is robust over a broad range of parameters as
long as the propagation length does not exceed half the coherence length [6].
Experimentally, this condition is readily fulfilled by operating at a gas
density equal to or below that resulting in the maximum cutoff yield. We
generated high-order harmonics by gently focusing 5 fs, 0.2 mJ laser pulses
into a 2 mm neon gas medium. The cycle-averaged peak intensity was
estimated to be 7×1014 W/cm2. The neon gas was supplied in a thin-walled
metal tube [32] with a backing pressure of 160 mbar. However, the pressure
in the interaction region was somewhat lower because of the gas expansion
into the surrounding vacuum chamber. Thin zirconium foils were installed in
the pathway of the emitted soft x-ray radiation to block the laser light and
low-order harmonics. The high-energy part of the spectrum (above 80 eV)
was spectrally analyzed with a simple spectrometer consisting of a 10,000-
line/mm transmission grating and a backside-illuminated soft x-ray CCD
camera (Roper Scientific).
Figure 10-9 shows a series of soft x-ray spectra generated for different
values of the carrier-envelope phase of the 5 fs pump pulses. For settings of
φ = φ0, the harmonics near the cutoff (hω > 120 eV) completely merge into a
continuum. With a gradual change of the phase, the continuous spectral
distribution of the cutoff radiation begins to break up into discrete harmonic
peaks, with the maximum modulation depth appearing for settings of φ = φ0
± π/2. This behavior is in agreement with the intuitive picture and the
computer simulations under the assumption of φ0 = 0 + nπ, where n=0.1. The
residual ambiguity in the determination of φ relates to the inversion
symmetry of the interaction with the atomic gas medium. A π-shift in φ
results in no change of the light wave other than reversing the direction of
the electromagnetic field vectors. In an isotropic medium, such a phase flip

cannot lead to any measurable physical consequences.
(a)
ϕ=ϕ0-π/2
(b)
ϕ=ϕ0+0
Soft-X-ray spectral intensity [arb. units]
(c) ϕ=ϕ0+π/3
(d) ϕ=ϕ0+π/2
(e) unlocked
90 100 110 120 130

Photon energy [eV]
Figure10- 9. Measured spectral intensity of soft x-ray emission from ionizing atoms driven by
5 fs pulses. (a)–(d) Data obtained with phase-stabilized pulses for different carrier-envelope-
phase settings are shown. (e) Spectrum measured without carrier-envelope-phase stabilization.
X-ray CCD exposure was 0.5 s in all cases.
The presence of distinct spectral behavior in the cutoff region allows us

to calibrate the actual value of the carrier-envelope phase. A continuum with
a bandwidth of ≈16 eV is clearly visible in Figure 10-9(b) and is in fair
agreement with the value of ∆(hω)cont = 20 eV predicted by our simple
considerations for 5 fs, 750 nm (τL/T0 = 2) light pulses (Table 10-1). On the
contrary, the x-ray spectra corresponding to the phase settings shifted by
294 Chapter 10
±π/2 [Figure 10-9(a) and (d)] display a pronounced modulation surviving

into the cutoff. Therefore, the carrier-envelope phase can be evaluated with
the highest possible accuracy by recording pairs of soft x-ray harmonic
spectra at φ1 = φ and φ2 = φ + π/2, where φ is varied in small steps. In this
method, φ = 0 (or, equivalently, π) can be identified from a pair of spectra
that exhibit the smallest and the largest modulation depth in the cutoff range,
respectively. Applying this procedure to the data in Figure 10-9, we
conclude that the phase setting φ0 ≈ 0. The current amplitude and phase
stability of our source permits identification of the carrier-envelope-phase
value with an accuracy better than π/5. The obtained value of φ is
subsequently used to calibrate the interference pattern recorded in the second
ν-to-2ν interferometer. After performing this calibration, we can
subsequently change φ (with accuracy better than π/10) using the
interferometer.
The soft x-ray experiments with phase-stabilized pulses also show that
the harmonic peaks shift with φ. This effect was predicted by previous [31,
73] and current simulations. It clearly manifests itself in the spectra shown in
Figure 10-9. This shift indicates that the peaks visible in the few-cycle-
driven harmonic spectra do not, strictly speaking, represent genuine
harmonics, because the laser frequency is kept unchanged. Thus its
harmonics are expected to remain invariant as well. The shift of these
“quasi-harmonics” is a direct consequence of the variation of the field
amplitude within T0 and the confinement of emission within a couple of
oscillation cycles. This carrier-envelope-phase-induced shift completely
smears the harmonic structure of the near-cutoff soft x-rays generated by our
5 fs pulses in the absence of phase stabilization [Figure 10-9(e)].
To gain further insight into the effect of spectral shift and smearing, we
recorded x-ray spectra obtained with longer laser pulses of about 10 fs. The
energy of the pulses sent to the target was increased to reach the same cutoff
frequency as in the data in Figure 10-9. The spectra we obtained are
displayed in Figure 10-10 and exhibit several differences with respect to the
5 fs case. First, there is no pronounced continuum in the cutoff region that
would permit identification of the φ0 phase offset, as is expected for this
pulse width in accordance with Table 10-1. Second, there are clear spectral
peaks at photon energies up to 110 eV that survive into the phase-unlocked
regime (Figure 10-10, bottom panel). The blowup in Figure 10-10(b) shows
that the frequencies of these peaks do not vary with phase and, consequently,
represent true odd-number harmonic orders of the laser field. However, the
situation is dramatically changed toward the cutoff, because the spectral
peaks seen there [Figure 10-10(c)] shift their frequency by the full laser
photon energy. Similar behavior was observed by another group of
researchers who detected single-shot x-ray spectra with a carrier-envelope-
phase-unlocked laser [74]. Apparently, a quasi-periodic spectral pattern is

the result of complicated interference among x-rays bursts generated by a
laser field, the amplitude of which changes significantly within an optical
cycle [72]. The “genuine” harmonics at low frequencies are produced by
many participating peaks of the laser field, whereas the “spurious”
harmonics near the cutoff are formed by very few field peaks. In this regard,
the anomaly in the energy interval of 95–105 eV in Figure 10-10 is
noteworthy. This “intermediate” region features quasi-harmonic peaks with a
period equal to the energy of one laser photon. This is remarkable, since
such periodicity corresponds to laser field peaks separated by T0 rather than
by the usual T0/2 (i.e., by the time spacing between the nearest half cycles).
We believe the reason for this transient behavior is the involvement of three
neighboring half cycles capable of generating x-rays at that frequency, a
condition that cannot be fulfilled at both lower- and higher-photon energies.
(a) ϕ=ϕ0+0 (b) (c)

Soft X-ray spectral intensity [arb. units]
ϕ=ϕ0+π/4
ϕ=ϕ0+π/2
ϕ=ϕ0+3/2π
ϕ=ϕ0+0.9π
Unlocked
90 100 110 120 130 140 82 84 86 88 90 110 112 114 116 118
Photon energy [eV] Photon energy [eV] Photon energy [eV]
Figure 10-10. Measured spectral intensity of soft x-ray emission from ionizing atoms driven
by 10 fs pulses. (a) Full-range spectra. (b) and (c) Close-ups on spectral regions with
distinctly different spectral dependencies on the carrier-envelope phase.
Both the experiments with 5 and 10 fs pulses underscore the benefits that
carrier-envelope-phase stabilization brings to the study of coherent laser-
driven x-ray sources.
296 Chapter 10
8. CARRIER-ENVELOPE-PHASE MEASUREMENT
WITHOUT AMBIGUITY
Carrier-envelope-phase effects (discussed above in the context of soft x-

ray emission) are also expected to be observed in other strong-field
experiments. The ability to measure and precisely stabilize the carrier-
envelope phase is crucial for many applications. Clear evidence of phase
effects has been seen in photoionization without phase control for few-cycle
pulses [34, 75]; in soft x-ray generation with phase-stabilized pulses [40] and
in the single-shot regime [74, 76]; and in photoemission from surfaces with
nonamplified few-cycle pulses [77]. Phase-dependent effects in
semiconductors have been predicted [78]. In fact, carrier-envelope-phase
effects due to the field asymmetry of few-cycle pulses can be easily
understood when compared with coherent control by an asymmetric two-
color light field [79]. Nonlinear photoionization provides a straightforward
approach to looking for signatures of the carrier-envelope phase. Particular
emphasis is given to noninversion symmetric emission of photoelectrons,
i.e., a different count rate for electrons emitted to the right and to the left. A
number of theoretical investigations have been devoted to the problem in the
past years [38, 68, 80-82]. Most of them concentrate on total yields of
photoelectrons. Here we present experimental results of nonlinear
photoionization from a gas target [75, 83] with the intense phase-stabilized,
few-cycle light source described in Section 6. The spatially asymmetric field
of few-cycle laser pulses leads to different photoionization rates for opposite
directions. By detecting this difference, it is possible to measure the carrier-
envelope phase of intense few-cycle laser pulses without the field-inversion
ambiguity present in the x-ray emission experiment on the centro-symmetric
system (described in Section 7).
As in the case of x-ray emission where the interaction with most
prominent half cycles of a 5 fs laser pulse causes the generation of the most
energetic x-ray photons, the highest-peak-strength half cycles are responsible
for the release of photoelectrons with the largest drift momenta during
nonlinear photoionization. The results of a classical calculation illustrating
the difference for cosinelike and sinelike light waves are depicted in
Figure 10-11 and give a qualitative explanation of how this effect can be
utilized for carrier-envelope-phase calibration.
In our experiments, xenon atoms were ionized with 5 fs laser pulses with
peak intensities of nearly 1014 W/cm2. Absorption of eight photons is
necessary to overcome the ionization threshold. Higher-order ionization
processes, known as above-threshold ionization [84], are also observed at
correspondingly high photoelectron energies.
Figure 10-11. Electric field (short-dashed curve) and envelope (long-dashed curve) of a 5 fs
laser pulse for CE phase 0 (left panel) and –π/2 (right panel). The drift momentum of high-
energy photoelectrons as a function of the ionization time t0 is shown in black (emission to the
right) and gray (emission to the left). It was calculated with a classical model. Note that
highest momentum does not necessarily coincide with highest electron yield: In the right
panel the ionization probability at times labeled with 1 is considerably lower than those
labeled with 2 because of lower field strength at t0.
To study the above-threshold-ionization phenomena, we employed the

time-of-flight (TOF) spectrometer schematically depicted in Figure 10-12. In
this apparatus, two opposing electrically and magnetically shielded TOF
detectors are mounted in an ultrahigh vacuum apparatus. Xenon atoms are
fed in through a nozzle from the top and are ionized in the focus of a few-
cycle laser beam. The lens shown in the sketch is a concave mirror with a
focal length of 250 mm. The pulse energy was attenuated to 20 µJ. The laser
polarization is linear and parallel to the flight tubes. The laser field changes
sign while propagating through the focus. Slits with a width of 250 µ m are
used to block the electrons emitted outside the focal region. The position of
the slits can be adjusted. A photodiode and microchannel plates detect the
laser pulses and photoelectrons, respectively. The laser repetition rate is 1
kHz and 50 electrons are typically recorded per pulse at each microchannel
plate. A pair of glass wedges (apex angle 2.8°) is used to optimize dispersion
and adjust the carrier-envelope phase. Therefore, a ∆s translation of one
wedge introduces a ∆x = ∆s ⋅ tan(2.8°) change in the total glass thickness
seen by the laser beam. For our laser pulses centered at the wavelength of
760 nm, an addition of 52 µ m of fused silica changed the carrier-envelope-
phase value by 2π without affecting the pulse duration.
298 Chapter 10
Figure 10-12. The “stereo above-threshold-ionization’’ spectrometer. MCP: microchannel

plates and PD: photodiode.
Figure 10-13(a) shows a series of above-threshold-ionization spectra

corresponding to different carrier-envelope phases. Simply by inspecting the
high-energy part (>20 eV) of the spectra, one can verify that a carrier-
envelope phase shift by ∆φ = π corresponds to a change from the electron
spectrum emitted into the left detection channel to the spectrum emitted to
the right one. Furthermore the above-threshold-ionization spectrum is
invariant under a 2π phase shift. For high electron energies, the difference
between the left and right electron yield reaches up to an order of magnitude
and can be controlled by adjusting the carrier-envelope phase. This phase-
determined yield discrimination constitutes a new kind of coherent control
and exhibits, to our knowledge, the highest contrast observed so far.
Figure 10-13(b) displays the ratio of photoemission to the left and to the
right for different spectral regions. The ratio of high- and low-energy
electrons oscillates with a periodicity of ∆φ in excellent agreement with the
periodicity expected from glass dispersion. Besides the much higher contrast
for high-energy electrons, a phase shift of both sets of data is also apparent.
To determine the carrier-envelope phase, we have to establish the connection
between the phase of the field and the above-threshold-ionization signal. We
do this by using high-energy (>20 eV) electrons. One reason for choosing
this spectral region is its higher sensitivity to the carrier-envelope phase.
The sensitivity is similar to that in the spectral cutoff of x-ray radiation
discussed in Section 7. More importantly, the dominant overall features in
this spectral region of above-threshold ionization are known to be rather
independent of specific properties of the atomic species and of laser
intensity.
Figure 10-13. (a) Photoelectron spectra for different CE phases controlled by fine movement
of one of the wedges. ∆x indicates the glass added hereby. The black curves correspond to
emission to the right (positive direction), the gray ones to the opposite direction. The insets
show the corresponding real time variation of the electric field, as deduced from the phase
assignment shown in Figure 10-14. Only without phase stabilization were identical spectra
measured to both left and right as expected. (b) Left-right ratio of the total electron yield
(solid black curve) and high-energy electrons (dashed gray curve) as a function of glass
thickness ∆x added or subtracted by moving one of the wedges. ∆x = 0 corresponds to optimal
dispersion compensation, i.e., the shortest pulses. Maximal left/right ratio for the total yield
does not coincide with that for high-energy electrons. Note the different scales for low- and
high-energy electrons. The upper horizontal scale indicates the CE envelope phase of the
pulse, as deduced from the comparison with theory shown in Figure 10-14. (c) Fine-increment
phase scan indicating long-term CE phase stabilization drift of the laser setup. The
measurement was performed over ≥10 min by slowly scanning the wedges forth (triangles
pointing to the right) and back (triangles pointing to the left) using a fine increment of the
glass thickness. The mismatch at the start and the end indicates a phase drift of ~50 mrad/min.
300 Chapter 10
Figure 10-14 shows a comparison between measurements and classical

calculations. The latter take into account only the high-energy electrons. In
the experimental and theoretical data, the ratio of emission to the left and
right depends in a characteristic way on electron energy and the carrier-
envelope phase. These dependences lead to the inclined stripes in the false-
color representations. Another common feature is that the asymmetry of left-
right emission increases with the strengthening of the electron energy.
Unlike the asymmetry contrast, these characteristic features do not depend
on the pulse shape or duration. The sharp edges in the calculation originate
from classical energy conservation. This artifact is removed by energy-time
uncertainty in quantum mechanics. Matching the characteristic features of
the theoretical and experimental data leads to an unambiguous determination
of the carrier-envelope phase with an estimated error of π/10.
Figure 10-14. Determination of the CE phase. Grayscale representation (logarithmic scale) of

the left-right ratio of photoelectrons as a function of the electron energy and the CE phase.
Higher values (white) indicate dominant left emission, lower values (black) dominant right
emission. For the theoretical part (bottom panel), only rescattered electrons were taken into
account. The parameters used are the same as in the experiment. This comparison allows
precise carrier-envelope phase determination. Within the convention of Equation (1), the
electric field is now uniquely characterized.
The calculation used to determine the carrier-envelope phase is based on

the insight that at high intensities above-threshold ionization can be
understood in terms of quasi-static field ionization at some time t0 and
subsequent classical evolution of the electron in the laser field [23, 85]. For a
review, see [86]. A considerable fraction of the photoelectrons generated

when atoms or molecules are exposed to intense laser pulses do not escape
directly. Rather, they return to the ion core within times of typically less than
an optical cycle and in precise synchronization with the electric field of the
laser pulse. The characteristics of returning electrons can be observed in the
kinetic energy spectra of photoelectrons. The underlying mechanism is
elastic scattering of the electrons when they return to the ion core at t = t1
and are subsequently accelerated by the laser field. Consequently,
photoelectrons with high kinetic energy are observed. In the photoelectron
spectra, they manifest themselves in the high-energy plateau on the
exponentially decreasing count rate at low electron energies [87].
The kinetic energy spectrum of the plateau electrons is sensitive to the
carrier-envelope phase for several reasons. First, high-energy electrons
returning to the ion core can be created only in subfemtosecond time
intervals close to peaks of the electric field of the laser pulse (see Figure 10-
11). However, the probability that they will tunnel through the atomic
potential at t0 depends exponentially on the field strength EL(t0) [cf. Equation
(1)]. As few-cycle pulses are involved, tunneling is likely only for those very
few optical half-cycles close to the pulse maximum. Generally, the highest
kinetic energies are reached for electrons returning to the core when the
electric field becomes nearly zero, i.e., EL(t1)=0. For few-cycle pulses, the
field amplitude EL also needs to be as high as possible for t > t1 to allow
efficient acceleration after rescattering. Since the start time t0 and return time
t1 differ by almost one optical cycle, both the highest possible field strength
at t0 and the highest possible amplitude after t1, are difficult to meet and
result in a strong dependence of photoionization on the carrier-envelope
phase. Number, strength, and timing of the wave packets lead to distinctive
structures in the above-threshold-ionization spectra. Their analysis therefore
provides detailed information about the key processes of attosecond science.
Quantum mechanical calculations are in very good qualitative agreement
with this classical treatment [82].
Having established that above-threshold ionization can be used for
precise determination of the carrier-envelope phase, we provide an example
of the sensitivity of the instrument. Figure 10-13(c) shows a measurement
performed with fine ∆φ increments. These results reveal small phase drifts
that we attribute to the laser system and its stabilization scheme. The phase
drifts are of the order of 50 mrad/min, measured for more than 10 minutes.
Since the stereo above-threshold-ionization phase meter is not only able to
measure the carrier-envelope phase, but also to detect small phase drifts, it
clearly has the potential to be used in a feedback loop. In addition, the target
gas pressure is so low (< 10-4 mbar) that the above-threshold-ionization
302 Chapter 10
experiment does not affect the laser beam. Thus a stereo above-threshold-
ionization phase meter can be placed anywhere along a laser beam line.
9. GOUY PHASE SHIFT FOR FEW-CYCLE LASER

PULSES
So far we have considered carrier-envelope stabilization of amplified

few-cycle pulses and their applications. We have also addressed the question
of determining the carrier-envelope phase value. An important issue for
nonlinear processes driven with phase-stabilized few-cycle pulses is that
they usually take place in the focus of the laser beam. However, an
electromagnetic beam propagating through a focus experiences an additional
π phase shift with respect to a plane wave. This can be explained by
introducing the phenomenon predicted by Kirchhoff and experimentally
observed by Gouy [88], any spherical wave radiating from a source that is
small compared with the wavelength advances in phase by π/2 within a short
distance of the source. This phase, also referred to as Gouy phase, is a
general property of a focused wave, including sound waves [89]. Intuitive
explanations of this phase anomaly have been proposed [90]. Recently,
direct observations of the polarity change of focused single-cycle terahertz
pulses have been reported [91]. For applications of carrier-envelope-stable
few-cycle light fields, phase matching of higher-order harmonic generation
(see Section 7) is governed by the atomic response and the Gouy phase shift
of the focused fundamental laser beam [92]. Moreover, the Gouy phase
strongly affects the spatial variation of the carrier-envelope phase of
ultrashort pulses around the beam focus.
In principle, the Gouy phase shift of a TEM00 wave can be described by a
simple formula depending on the focusing geometry and the wavelength,
 z 
φ = − arctan ,

(12)
 z R (λ ) 
where the beam is traveling in the +z direction and zR is the Rayleigh

distance (dependent on the wavelength λ). Because of the large bandwidth of
few-cycle pulses, their different spectral components experience the Gouy
phase shift on different spatial scales. In particular, the blue colors undergo a
steeper phase change than the red colors. It has been shown theoretically
[38, 93] that this leads to a difference of phase and group velocities in the
vicinity of the focus and, therefore, to carrier-envelope-phase slippage. For
this reason, the CE phase variation cannot be generally described by
Equation (12). Indeed, the details of the phase change in the focus depend on
the spatial profile of the laser beam and on the focusing geometry. In all
cases, an overall π phase shift is expected between two symmetric positions
far away from a spherical focus. Because of the continuous phase change of
the pulse propagating through the focal region, the corresponding electric
field of light takes on all possible phase variations. For example, a focused
“+cos” E-field will be a sinelike wave in the focal region and will reemerge
as a “-cos” field beyond the focus. This variation, of course, can affect the
outcome of a phase-sensitive light–matter interaction that is not confined to a
length much shorter than the Rayleigh range. Thus, precise control of the
spatial variation of the carrier-envelope phase within the whole focal region
is crucial to any phase-dependent experiment.
We present the experimental determination of the evolution of the
carrier-envelope phase in the focus of few-cycle laser pulses. Together with
the determination of the carrier-envelope phase presented in Section 8, this
result constitutes a full and unambiguous characterization of the electric field
of laser pulses in space and time within the paraxial approximation (i.e.,
neglecting the small longitudinal component of the electric field). We
studied the spatial variation of the carrier-envelope phase using the same
technique that was presented in Section 8, i.e., by simultaneously detecting
above-threshold-ionization photoelectron spectra in opposite directions
(Figure 10-12). The pulse energy of carrier-envelope-controlled 5 fs pulses
was attenuated to 20 µJ, and the beam focused with a f/30 geometry into a
low-density xenon gas jet. Under these focusing conditions, the electric field
is expected to undergo a π phase shift over a few millimeters. To reveal the
the Gouy phase, one has to selectively detect the electrons generated at a
well-defined position of the focus. A pair of moveable slits perpendicular to
the beam (z direction) and to the polarization axes allows the entire focal
region to be scanned. The slit width is 250 µ m, well below the Rayleigh
range (≈1 mm). To achieve optimum spatial resolution in the z direction, the
slits are placed at a distance of only 1 mm from the beam (Figure 10-12).
With this setup, the angular distribution of the emitted photoelectrons does
not affect the phase resolution, which is estimated to be ≈0.1 rad.
Figure 10-15 shows the asymmetry of the electron count rate (left/right
ratio) as a function of the glass thickness introduced by moving the fused-
silica wedges in front of the above-threshold-ionization spectrometer. We
made the measurement by moving the pair of slits to a distance of ≈2 mm
before (dashed line) and after (solid line) the focus. The phase shift of π
between the two curves is a direct measurement of the Gouy phase shift in
the focus.
304 Chapter 10
Figure 10-15. Variation of glass thickness in the beam path to change the carrier-envelope
phase of the pulses. The left/right ratio of the electron yield exhibits clear oscillations with a
periodicity consistent with glass dispersion at the wavelength of the laser. The measurement
was performed before (dashed line) and after (solid line) the focus. The π phase shift is due to
the Gouy carrier-envelope phase shift resulting from the passage through the focus.
The electric-field polarity reversal observed by measuring the phase

value in front of and behind the focus does not describe the details of the
phase slippage in the focal region itself. Since many experiments take place
over an extended area of the focus (e.g., higher-order harmonic generation in
the XUV), it is essential to characterize the phase variation precisely and, in
particular, to recognize possible anomalies in the behavior of the carrier-
envelope phase evolution. Indeed, the possible presence of an anomalous
phase change in the focus has been suggested as an explanation of the
observed enhancement of the harmonic-generation efficiency by using
truncated Bessel beams [94].
To analyze the phase variation within the focal range, we acquired
electron spectra at several positions by moving the pair of slits. Figure 10-16
shows the detected left/right asymmetry as a function of the electron energy
and of the carrier-envelope phase for a few z positions. Different approaches
are possible to retrieve the phase variation. The integrated electron yield
provides a clear phase indication and can be used to show the overall phase
shift in the focus (Figure 10-15). However, scanning the focal range implies
measuring at constantly changing intensities. Since it is predicted that the
maximum of the integrated left/right asymmetry should occur at a phase that
depends on the intensity [81], it is not easy to decouple the Gouy phase shift
from the possible intensity-dependent phase shift. A better approach relies
on the spectral information in Figure 10-16. Light and dark shades
correspond to spectral regions of strong asymmetry. Since these colors are
easily distinguishable, they can be used to follow the phase evolution. The
measurements were made by approaching the focus from the outer part and
by moving the slits alternatively before and after the focus. With this
procedure, measurements of symmetric positions around the focus are
consecutive, thus reducing detrimental effects from possible long-term phase
drifts.
Figure 10-16. Left/right asymmetry maps (logarithmic scale) for different longitudinal
positions as a function of the electron energy and glass thickness introduced. Lighter shades
indicate dominant left emission; darker shades— dominant right emission. The maps (a) –(f )
correspond, respectively, to the positions: z=-1.75 mm, z=-1.0 mm, z=-0.25 mm, z=+0.25 mm,
z=+1.0 mm, z=+1.75 mm (positive values represent positions after the focus). The phase
difference is determined by evaluating the shift of the characteristic structures (indicated by
the dashed lines) of the asymmetry pattern. The extension of the electron yield to higher
energies in the central part of the focus is due to the higher intensity.
Figure 10-16(a) and (f) corresponds to the outer part of the focal range.
The strong asymmetry in the high-energy part of the spectra [above-
threshold-ionization plateau (dashed area)] changes sign while passing
through the focus, confirming the π phase shift already discussed (see Figure
10-15). More interestingly, Figure 10-16(b)–(e) corresponds to positions in
the central part of the focus. The asymmetry in the plateau is partly smeared
out, but another clear asymmetric area appears in the low-energy part of the
306 Chapter 10
spectra [20–25 eV (dashed area)], just before the above-threshold-ionization

plateau. Its nature is not entirely understood, but, because of being definitely
a carrier-envelope phase effect, it can be exploited to retrieve the phase
variation in the focus. By moving in the direction of beam propagation, i.e.,
from before to after the focus [Figures 10-16(a)–(f)], the characteristic
pattern moves toward the left side of the graphs, i.e., toward decreasing glass
thickness. This pattern is in full agreement with the intuitive physical origin
of the Gouy carrier-envelope-phase shift. Indeed, in a dispersive medium
like glass, the phase velocity exceeds the group velocity. The same situation
occurs in the propagation of the ultrashort pulse in the laser focus [93]. Thus,
it is not surprising that to observe similar features, less glass is needed after
the focus with respect to a symmetric position before the focus. The Gouy
carrier-envelope-phase shift determination can be performed simply by
numerically evaluating the shift of the asymmetry pattern; converting the
glass-thickness shift obtained (µ m) into phase difference (rad) is then
straightforward.
Figure 10-17 shows the retrieved carrier-envelope phase shift for the
positions of Figure 10-16 and for several others not shown there. For
comparison, the Gouy phase of a Gaussian beam [Equation (12)] with an
f/30 focusing geometry is also shown (solid line). Note that the carrier-
envelope-phase shift is not expected to follow the Gouy phase, the latter
being a property of cw lasers. The beam in our experiments is a few-cycle
pulse-truncated Bessel beam. The use of a spherical focusing mirror in our
setup inevitably introduces a slight astigmatism.
Figure 10-17. Retrieved carrier-envelope Gouy phase shift as a function of the propagation
distance in the focus. The solid line is the Gouy phase of a cw Gaussian beam, shown for
comparison. In the outer part of the focus, the electron count rate rapidly decreases, making
detection of additional experimental points difficult.
The pulses undergo the π phase shift within a few Rayleigh distances.
Because of the rapid decrease of electron yield at lower intensities, the
measurements were stopped at a distance of ≈2 mm before and after the
focus. This prevented observation of the expected area of constant carrier-
envelope phase in the outer part of the focus. However, the region of interest
for all experiments is entirely covered, and the estimated error for the
experimental data is relatively low (≤0.1 rad). The phase changes smoothly
with a constant slope and does not exhibit any wiggles or irregularities,
which is particularly important for experiments.
10. CONCLUSIONS AND OUTLOOK
We demonstrated the generation of intense few-cycle light pulses with a

reproducible temporal evolution of the electromagnetic field. With these
light wave forms, we explored the sensitivity of microscopic atomic currents
to the timing of light-field oscillations with respect to the pulse peak. The
short-wavelength (XUV, soft x-ray) radiation emitted by the atomic currents
allowed determination of this timing with sub-T0/10, i.e., sub-250
attoseconds, accuracy. The generation of intense few-cycle light with a
reproducible and known wave form is expected to benefit the emerging field
of attosecond physics in several ways.
The high-frequency atomic currents, employed as a light-waveform
detector in the current experiments, are also sources of subfemtosecond
XUV/soft x-ray pulses [25, 26]. Therefore, the measurement and
stabilization of the carrier-envelope phase presented in this research directly
opens the door to generating single, isolated subfemtosecond pulses that are
ideally suited for most spectroscopic applications. Ultrashort wave packets
with controlled field oscillations will also benefit the techniques that have
been introduced for attosecond metrology [26, 27] and spectroscopy [28].
They are indispensable to measuring the shape and chirp of subfemtosecond
XUV or x-ray pulses [95] and to observing inner-shell atomic processes on
an attosecond timescale [28]. In addition, control over the temporal structure
of electron wave packets liberated by optical-field ionization also relies on
synthesized light wave forms. Few-cycle light with controlled field
oscillations will be able to launch an electron wave packet with well-
controlled subfemtosecond temporal structure into the continuum and
precisely steer its subsequent motion. One intriguing scenario is to make the
electron recollide with its parent ion and use this recollision to excite and/or
probe molecular [96], atomic, or possibly even nuclear dynamics. Finally,
yet another potential application area of high-intensity wave forms is
electron acceleration in the forward direction to ultrarelativistic energies in a
highly controlled manner.
308 Chapter 10
ACKNOWLEDGEMENTS
We gratefully acknowledge the contributions of M. Uiberacker,

E. Goulielmakis, M. Lezius, Th. Udem, R. Holzwarth, V.S. Yakovlev, and
A. Scrinzi to this work. This research has been supported by the Austrian
Science Fund (Grants No. F016, No. Z63, and No. P15382), the European
ATTO network, and The Welch Foundation (Grant No. A-1562).
REFERENCES
[1] A. M. Weiner, Rev. Sci. Instrum. 71, 1929-1960 (2000).
[2] R. Trebino, K. W. DeLong, D. N. Fittinghoff, J. N. Sweetser, M. A.
Krumbugel, B. A. Richman, and D. J. Kane, Rev. Sci. Instrum. 68, 3277-
3295 (1997).
[3] C. Iaconis and I. A. Walmsley, Opt. Lett. 23, 792-794 (1998).
[4] E. Constant, Ph.D Thesis, University of Sherbrooke (1997).
[5] G. Steinmeyer, D. H. Sutter, L. Gallmann, N. Matuschek, and U. Keller,
Science 286, 1507-1512 (1999); B. Schenkel, J. Biegert, U. Keller, C.
Vozzi, M. Nisoli, G. Sansone, S. Stagira, S. De Silvestri, and O. Svelto,
Opt. Lett. 28, 1987-1989 (2003).
Opt. Lett. 21, 2008-2010 (1996).
[9] T. Udem, Ph.D Thesis, Ludwigs-Maximilians Universität (1997).
172, 59-68 (1999).
Keller, Appl. Phys. B 69, 327 (1999).
(2000).
883 (1999); T. Udem, R. Holzwarth, and T. W. Hansch, Nature 416, 233-
237 (2002); M. Zimmermann, C. Gohle, R. Holzwarth, T. Udem, and T. W.
Hansch, Opt. Lett. 29, 310-312 (2004).
84, 5102-5105 (2000).
(2001).
[19] MenloSystem-GmbH (http://www.menlosystem.com)
[20] M. Kakehata, H. Takada, Y. Kobayashi, K. Torizuka, Y. Fujihira, T.
Homma, and H. Takahashi, Opt. Lett. 26, 1436-1438 (2001).
[21] A. Baltuška, T. Fuji, and T. Kobayashi, Phys. Rev. Lett. 88, art. no.-133901
(2002).
[23] P. B. Corkum, Phys. Rev. Lett. 71, 1994-1997 (1993).
[24] M. Lewenstein, P. Balcou, M. Y. Ivanov, A. Lhuillier, and P. B. Corkum,
Phys. Rev. A 49, 2117-2132 (1994).
[25] M. Drescher, M. Hentschel, R. Kienberger, G. Tempea, C. Spielmann, G.
A. Reider, P. B. Corkum, and F. Krausz, Science 291, 1923-1927 (2001); P.
M. Paul, E. S. Toma, P. Breger, G. Mullot, F. Auge, P. Balcou, H. G.
Muller, and P. Agostini, Science 292, 1689-1692 (2001).
[26] M. Hentschel, R. Kienberger, C. Spielmann, G. A. Reider, N. Milosevic, T.
Brabec, P. Corkum, U. Heinzmann, M. Drescher, and F. Krausz, Nature
414, 509-513 (2001).
[27] R. Kienberger, M. Hentschel, M. Uiberacker, C. Spielmann, M. Kitzler, A.
Scrinzi, M. Wieland, T. Westerwalbesloh, U. Kleineberg, U. Heinzmann,
M. Drescher, and F. Krausz, Science 297, 1144-1148 (2002).
[28] M. Drescher, M. Hentschel, R. Kienberger, M. Uiberacker, V. Yakovlev,
A. Scrinizi, T. Westerwalbesloh, U. Kleineberg, U. Heinzmann, and F.
Krausz, Nature 419, 803-807 (2002).
[29] I. P. Christov, M. M. Murnane, and H. C. Kapteyn, Phys. Rev. Lett. 78,
1251-1254 (1997).
[30] A. de Bohan, P. Antoine, D. B. Milosevic, and B. Piraux, Phys. Rev. Lett.
81, 1837-1840 (1998).
[31] G. Tempea, M. Geissler, and T. Brabec, J. Opt. Soc. Am. B 16, 669-673
(1999).
[32] C. Spielmann, N. H. Burnett, S. Sartania, R. Koppitsch, M. Schnurer, C.
Kan, M. Lenzner, P. Wobrauschek, and F. Krausz, Science 278, 661-664
(1997).
H. C. Kapteyn, Phys. Rev. Lett. 83, 2187-2190 (1999).
[35] C. Lemell, X. M. Tong, F. Krausz, and J. Burgdorfer, Phys. Rev. Lett. 90, -
(2003); V. S. Yakovlev, P. Dombi, G. Tempea, C. Lemell, J. Burgdorfer, T.
Udem, and A. Apolonski, Appl. Phys. B 76, 329-332 (2003).
Hansch, and F. Krausz, Appl. Phys. B 72, 977-977 (2001).
[37] Y. Kobayashi and K. Torizuka, Opt. Lett. 25, 856-858 (2000).
310 Chapter 10
[38] P. Dietrich, F. Krausz, and P. B. Corkum, Opt. Lett. 25, 16-18 (2000).
[39] I. P. Christov, Opt. Lett. 24, 1425-1427 (1999).
Gohle, R. Holzwarth, V. S. Yakoviev, A. Scrinzi, T. W. Hansch, and F.
Krausz, Nature 421, 611-615 (2003).
[41] J. Eckstein, Ph.D Thesis, Stanford University (1978).
[42] S. T. Cundiff and J. Ye, Rev. Mod. Phys. 75, 325-342 (2003).
1547-1549 (1997); J. K. Ranka, R. S. Windeler, and A. J. Stentz, Opt. Lett.
25, 25-27 (2000); J. K. Ranka, R. S. Windeler, and A. J. Stentz, Opt. Lett.
25, 796-798 (2000).
Opt. Lett. 26, 373-375 (2001); A. Bartels and H. Kurz, Opt. Lett. 27, 1839-
1841 (2002); T. M. Fortier, D. J. Jones, and S. T. Cundiff, Opt. Lett. 28,
2198-2200 (2003).
[46] S. Sartania, Z. Cheng, M. Lenzner, G. Tempea, C. Spielmann, F. Krausz,
and K. Ferencz, Opt. Lett. 22, 1562-1564 (1997).
[47] M. Takeda, H. Ina, and S. Kobayashi, J Opt Soc Am 72, 156-160 (1982).
[48] C. Dorrer and F. Salin, J. Opt. Soc. Am. B 15, 2331-2337 (1998).
[49] C. Dorrer, J. Opt. Soc. Am. B 16, 1160-1168 (1999); C. Dorrer, N. Belabas,
J. P. Likforman, and L. Joffre, Appl. Phys. B 70, S99-S107 (2000).
[50] F. W. Helbing, G. Steinmeyer, U. Keller, R. S. Windeler, J. Stenger, and H.
R. Telle, Opt. Lett. 27, 194-196 (2002).
[51] G. P. Agrawal, Nonlinear Fiber Optics (Academic Press, San Diego, 1995);
P. V. Mamyshev and S. V. Chernikov, Opt. Lett. 15, 1076-1078 (1990); J.
Botineau and R. H. Stolen, J Opt Soc Am 72, 1592-1596 (1982).
[52] N. Bloembergen, Nonlinear Optics (World Scientific, Singapore, 1996).
[53] S. M. Gallagher, A. W. Albrecht, T. D. Hybl, B. L. Landin, B. Rajaram,
and D. M. Jonas, J. Opt. Soc. Am. B 15, 2338-2345 (1998).
[54] T. Brabec and F. Krausz, Phys. Rev. Lett. 78, 3282-3285 (1997).
[55] S. Mukamel, Principles of Nonlinear Optical Spectroscopy (Oxford
University Press, New York, 1995).
[57] X. Gu, L. Xu, M. Kimmel, E. Zeek, P. O'Shea, A. P. Shreenath, R. Trebino,
and R. S. Windeler, Opt. Lett. 27, 1174-1176 (2002); P. Baum, S.
Lochbrunner, J. Piel, and E. Riedle, Opt. Lett. 28, 185-187 (2003).
[58] M. Bellini and T. W. Hänsch, Opt. Lett. 25, 1049-1151 (2000).
447 (2002); T. M. Fortier, D. J. Jones, J. Ye, S. T. Cundiff, and R. S.
Windeler, Opt. Lett. 27, 1436-1438 (2002).
[60] D. Strickland and G. Mourou, Opt. Commun. 56, 219-221 (1985); M. D.
Perry and G. Mourou, Science 264, 917-924 (1994).
[61] A. Dubietis, G. Jonusauskas, and A. Piskarskas, Opt. Commun. 88, 437-

440 (1992); I. N. Ross, P. Matousek, M. Towrie, A. J. Langley, and J. L.
Collier, Opt. Commun. 144, 125-133 (1997).
[62] I. N. Ross, J. L. Collier, P. Matousek, C. N. Danson, D. Neely, R. M. Allot,
D. A. Pepler, C. Hernandez-Gomez, and K. Osvay, Appl. Opt. 39, 2422-
2427 (2000).
[63] A. E. Siegman, Lasers (University Science Books, Mill Valley, California,
1986).
[64] C. Corsi and M. Bellini, Appl. Phys. B 78, 31-34 (2004).
[65] S. H. Cho, F. X. Kartner, U. Morgner, E. P. Ippen, J. G. Fujimoto, J. E.
Cunningham, and W. H. Knox, Opt. Lett. 26, 560-562 (2001).
[66] M. S. Pshenichnikov, W. P. Deboeij, and D. A. Wiersma, Opt. Lett. 19,
572-574 (1994); G. N. Gibson, R. Klank, F. Gibson, and B. E. Bouma, Opt.
Lett. 21, 1055-1057 (1996); A. Baltuska, Z. Wei, M. S. Pshenichnikov, D.
A. Wiersma, and R. Szipocs, Appl. Phys. B 65, 175-188 (1997).
[67] F. Krausz, T. Brabec, M. Schnürer, and C. Spielmann, Opt. Photon. News.
9, 46-50 (1998).
[68] E. Cormier and P. Lambropoulos, Euro. Phys. J. D 2, 15-20 (1998).
[69] K. J. Schafer, B. Yang, L. F. Dimauro, and K. C. Kulander, Phys. Rev.
Lett. 70, 1599-1602 (1993).
[70] C. Kan, N. H. Burnett, C. E. Capjack, and R. Rankin, Phys. Rev. Lett. 79,
2971-2974 (1997).
[71] N. Milosevic, A. Scrinzi, and T. Brabec, Phys. Rev. Lett. 88, 093905
(2002).
[72] V. S. Yakovlev and A. Scrinzi, Phys. Rev. Lett. 91, 153901 (2003).
[73] E. Priori, G. Cerullo, M. Nisoli, S. Stagira, S. De Silvestri, P. Villoresi, L.
Poletto, and P. Ceccherini, Phys. Rev. A 61, 063801 (2000).
[74] M. Nisoli, G. Sansone, S. Stagira, S. De Silvestri, C. Vozzi, M. Pascolini,
L. Poletto, P. Villoresi, and G. Tondello, Phys. Rev. Lett. 91, 213905
(2003).
[75] G. G. Paulus, F. Lindner, H. Walther, A. Baltuska, E. Goulielmakis, M.
Lezius, and F. Krausz, Phys. Rev. Lett. 91, 253004 (2003).
[76] G. Sansone, C. Vozzi, S. Stagira, M. Pascolini, L. Poletto, P. Villoresi, G.
Tondello, S. De Silvestri, and M. Nisoli, Phys. Rev. Lett. 92, 113904
(2004).
[77] A. Apolonski, P. Dombi, G. G. Paulus, M. Kakehata, R. Holzwarth, T.
Udem, C. Lemell, K. Torizuka, J. Burgdorfer, T. W. Hansch, and F.
Krausz, Phys. Rev. Lett. 92, 073902 (2004).
[78] O. D. Mücke, T. Tritschler, M. Wegener, U. Morgner, and F. X. Kartner,
Phys. Rev. Lett. 89, 127401 (2002).
[79] A. N. Chudinov, Y. E. Kapitzky, A. A. Shulginov, and B. Y. Zel'dovich,
Opt Quant Electron 23, 1055-1060 (1991); D. W. Schumacher, F. Weihe,
H. G. Muller, and P. H. Bucksbaum, Phys. Rev. Lett. 73, 1344-1347
(1994); M. J. J. Vrakking and S. Stolte, Chem. Phys. Lett. 271, 209-215
312 Chapter 10
(1997); A. Haché, Y. Kostoulas, R. Atanasov, J. L. P. Hughes, J. E. Sipe,

and H. M. van Driel, Phys. Rev. Lett. 78, 306-309 (1997); T. Kanai and H.
Sakai, J. Chem. Phys. 115, 5492-5497 (2001).
[80] I. P. Christov, Appl. Phys. B 70, 459-462 (2000); R. M. Potvliege, N. J.
Kylstra, and C. J. Joachain, J. Phys. B 33, L743-L748 (2000).
[81] S. Chelkowski and A. D. Bandrauk, Phys. Rev. A 65, - (2002).
[82] D. B. Miloševic, G. G. Paulus, and W. Becker, Opt. Expr. 11, 1418-1429
(2003).
[83] F. Lindner, G. G. Paulus, H. Walther, A. Baltuska, E. Goulielmakis, M.
Lezius, and F. Krausz, Phys. Rev. Lett. 92, - (2004).
[84] P. Agostini, F. Fabre, G. Mainfray, G. Petite, and N. K. Rahman, Phys.
Rev. Lett. 42, 1127-1130 (1979).
[85] K. C. Kulander, K. J. Schafer, and J. L. Krause, in Superintense Laser Atom
Physics, edited by B. Pireaux, A. L’Huillier and J. Rzazewski (Plenum,
New York, 1994), Vol. B 316, p. 95.
[86] W. Becker, F. Grasbon, R. Kopold, D. B. Milosevic, G. G. Paulus, and H.
Walther, Adv. Atom. Mol. Opt. Phy. 48, 35-98 (2002).
[87] G. G. Paulus, W. Nicklich, H. L. Xu, P. Lambropoulos, and H. Walther,
Phys. Rev. Lett. 72, 2851-2854 (1994); G. G. Paulus, W. Becker, W.
Nicklich, and H. Walther, J. Phys. B 27, L703-L708 (1994).
[88] L. G. Guoy, Compt. Rendue Acad. Sci. Paris 10, 1251-1253 (1890).
[89] N. C. R. Holme, B. C. Daly, M. T. Myaing, and T. B. Norris, App. Phys.
Lett. 83, 392-394 (2003).
[90] R. W. Boyd, J. Opt. Soc. Am. 70, 877-880 (1980); S. M. Feng and H. G.
Winful, Opt. Lett. 26, 485-487 (2001).
[91] A. B. Ruffin, J. V. Rudd, J. F. Whitaker, S. Feng, and H. G. Winful, Phys.
Rev. Lett. 83, 3410-3413 (1999); P. Kuzel, M. A. Khazan, and J. Kroupa, J.
Opt. Soc. Am. B 16, 1795-1800 (1999); R. W. McGowan, R. A. Cheville,
and D. Grischkowsky, App. Phys. Lett. 76, 670-672 (2000).
[92] P. Balcou and A. L'Huillier, Phys. Rev. A 47, 1447-1459 (1993); M.
Lewenstein, P. Salières, and A. L'Huillier, Phys. Rev. A 52, 4747-4754
(1995); F. Lindner, W. Stremme, M. G. Schätzel, F. Grasbon, G. G. Paulus,
H. Walther, R. Hartmann, and L. Struder, Phys. Rev. A 68, 013814 (2003).
[93] Z. L. Horváth, J. Vinkó, Z. Bor, and D. von der Linde, “Acceleration of
femtosecond pulses to superluminal velocities by Gouy phase shift,” Appl.
Phys. B. 63, pp. 481-484, 1996.
[94] M. Nisoli, E. Priori, G. Sansone, S. Stagira, G. Cerullo, S. De Silvestri, C.
Altucci, R. Bruzzese, C. de Lisio, P. Villoresi, L. Poletto, M. Pascolini, and
G. Tondello, Phys. Rev. Lett. 88, 033902 (2002).
[95] J. Itatani, F. Quere, G. L. Yudin, M. Y. Ivanov, F. Krausz, and P. B.
Corkum, Phys. Rev. Lett. 88, 173903 (2002); M. Kitzler, N. Milosevic, A.
Scrinzi, F. Krausz, and T. Brabec, Phys. Rev. Lett. 88, 173904 (2002); R.
Kienberger, E. Goulielmakis, M. Uiberacker, A. Baltuska, V. Yakovlev, F.
Bammer, A. Scrinzi, Th.Westerwalbesloh, U. Kleineberg, U. Heinzmann,
M. Drescher, and K. F., Nature 427, 817-821 (2004).
[96] H. Niikura, F. Legare, R. Hasbani, A. D. Bandrauk, M. Y. Ivanov, D. M.

Villeneuve, and P. B. Corkum, Nature 417, 917-922 (2002); H. Niikura, F.
Legare, R. Hasbani, M. Y. Ivanov, D. M. Villeneuve, and P. B. Corkum,
Nature 421, 826-829 (2003).
Chapter 11
QUANTUM CONTROL OF HIGH-ORDER
HARMONIC GENERATION
Applied Attosecond Science
Emily Gibson, Ivan Christov, Margaret M. Murnane, and Henry C. Kapteyn

JILA, University of Colorado and National Institute of Standards and Technology;
Department of Physics, University of Colorado; and NSF Engineering Research Center for
Extreme-Ultraviolet Science and Technology
Abstract: We discuss the physical processes involved in the generation and optimization
of extreme ultraviolet and soft x-ray light though the process of high-order
harmonic generation. We show that by manipulating the sub-optical-cycle
attosecond dynamics of this process using optimized waveguide structures and
pulse shapes, we can control the energy of the emitted photons, the phase
matching of the conversion process, and the spatial and temporal coherence of
the light. High-order harmonic generation is a useful source of short
wavelength light with ultrashort time duration. Thus, optimization and
manipulation of high-order harmonic generation demonstrates control of
electron dynamics on attosecond time scales.
Key words: high harmonic generation, phase matching, attosecond
Ultrafast laser technology makes it possible to generate extremely high

field intensities above 1018 W/cm2 or, alternatively, to probe extremely short
time durations corresponding to only a few femtoseconds [1]. One of the
most prominent applications of very high-power, ultrashort-pulse lasers has
been to study the regime of “nonperturbative” light-matter interactions [2].
At intensities above ~1014 W/cm2, the magnitude of the electric field of the
laser radiation is comparable to the field binding an electron to an atom. In
this regime, the strong electric field can easily ionize atoms. Once the atom
is ionized, the subsequent motion of the free electron is controlled by the
oscillating laser field, and the electron can easily reach kinetic energies
many times that of the original electron binding energy.
One of the most significant consequences of this intense laser-matter
interaction is the process of high harmonic generation (HHG) [3]. In HHG,
11. QUANTUM CONTROL OF HIGH-ORDER HARMONIC 315
GENERATION
the ionized electron recollides and then recombines with its parent ion,
releasing a high energy photon. The energy and phase of the emitted high-
harmonic light depend on the detailed electron trajectory in the laser field—
dynamics that occur on a fraction of an optical cycle, or attosecond,
timescale. Thus, the properties of HHG are sensitive to the field evolution of
the driving laser [4]; in contrast, traditional nonlinear processes, such as
second-harmonic generation, are insensitive to many aspects of the driving
field such as the carrier-envelope offset (CEO). Detailed studies of the
dependence of high-harmonic generation on the time-history of the driving
laser field represent the first results from the field of attosecond science [5,
6], an area of research that has received considerable attention in the early
2000s [7]. Consequences of the attosecond dynamics of HHG include the
ability to manipulate electron dynamics with attosecond precision [8-10], the
phase matching of the conversion process (attosecond engineering) [11, 12],
and the ability to generate pulses of light with subfemtosecond duration. The
strong dependence on the laser field also means that HHG is sensitive to the
CEO of the driving laser pulse [13-15]. By combining pulse shaping
techniques with recently developed methods to control the CEO of an
ultrafast pulse by locking the mode frequencies to a cw reference [16], it
becomes possible to control the complete electric field of the pulse in the
time domain with attosecond precision. This allows us to access the fastest
time scales that are possible with modern laser technology.
1. THE PHYSICS OF HIGH-ORDER-HARMONIC

GENERATION
High-order-harmonic generation results when an intense laser field

interacts with a gas or material. An intuitive model of the process at the
atomic level was developed by Corkum, Kulander, and others [4, 17] and is
sometimes referred to as the “three-step” model. In the first step, the strong
electric field of the laser suppresses the Coulomb barrier binding an electron
to the atom, freeing the valence electron either by tunneling or “over-the-
barrier” ionization. The freed electron is then accelerated by the field. Since
the laser field is oscillating, the electron can, with some probability, return to
its parent ion and recombine, emitting a high-energy photon. This process
occurs for many atoms driven coherently over several laser cycles, resulting
in emission of higher-order odd harmonics in a coherent, low-divergence
beam. The three-step model accurately predicts that the highest photon
energy that can result from HHG occurs when an electron is ionized at a
316 Chapter 11
phase of 18 degrees after the peak of the laser cycle. This cutoff photon
energy is then predicted to be:
Emax = I p + 3.2U p , (1)
where Ip is the ionization potential of the atom and Up is the ponderomotive

energy given by
U p = e 2 E 2 / 4mω 2 ∝ Iλ 2 , (2)
where e, E, m, ω, I, and λ are the electron charge, field amplitude, electron

mass, laser frequency, intensity, and wavelength, respectively. The highest
photon energy therefore scales linearly with the laser intensity.
The “recollision” energies associated with the intensities required to
field-ionize atoms can be as high as several hundred electron volts [18, 19].
Thus, the emitted high-harmonic photons can correspond to the combined
energy of several hundred of the incident photons. The linear relationship
between the cutoff and the incident intensity also presents very attractive
scaling as compared to, for example, EUV (extreme ultraviolet) laser
schemes where the power requirements scale as the photon energy to the 3rd–
5th power [20]. However, to take advantage of this favorable scaling, several
challenges must be overcome. The HHG process necessarily ionizes the gas,
generating free-electron plasma. The dispersion of the plasma causes a
mismatch in the phase velocities of the fundamental and harmonic light,
significantly reducing the amount of harmonic signal produced. Also, the
plasma can defocus the laser beam, decreasing the peak intensity, and
thereby limiting the maximum harmonic energy [21]. To date, the highest
energy HHG emission has been achieved using short-duration laser pulses
and noble gases with high ionization potentials that can reduce the amount of
ionization for a given peak intensity [18, 19]. This emission can extend from
the ultraviolet up to the soft x-ray region of the spectrum (>500 eV), making
it an ideal source for pure and applied science such as photoemission
spectroscopy [22], plasma interferometry [23], or metrology for optics for
EUV lithography.
GENERATION
2. “SINGLE-ATOM” EFFECTS IN HIGH-ORDER-

HARMONIC GENERATION: MANIPULATION
AND COHERENT CONTROL
A simple classical picture of HHG as given by the three-step model

accurately predicts the range of photon energies that can be obtained for a
given incident laser intensity. However, electrons are not “classical” objects,
and a more accurate picture of the HHG process includes the wave nature of
the oscillating electron. A fully quantum, analytical theory of HHG has been
derived by Lewenstein et al. [4] and has been extremely successful in
describing both the general characteristics of HHG, such as the photon
energy cutoff, as well as more specific characteristics such as the divergence
properties of the generated beam [24] and the specific spectral characteristics
of the emission [5]. Qualitatively, the spectral characteristics of HHG
emission are a result of the fact that the ionized electron has an associated
deBroglie wavelength corresponding to the kinetic energy acquired in the
laser field, λ = h p , where h is Planck’s constant and p is the electron
momentum. The emission also has a phase related to the total phase the
electron accumulates during its free trajectory, approximately [24]
ϕ ≈ qωt f − τ sU p , (3)
where q is the harmonic order, ω is the laser frequency, and τ s = t f − ti is

the time difference between when the electron is ionized and when it
recombines with the ion. The phase of the emitted x-rays is therefore not
simply related to the phase of the driving laser but also includes an “intrinsic
phase” component that can vary rapidly with intensity. The intrinsic phase
has consequences for both the spatial and spectral emission characteristics.
Spatially, this intrinsic phase can result in a complex spatial profile of the
harmonic emission when generated using a beam that is converging toward a
focus. This is in contrast to the near-Gaussian profile for HHG generated by
a beam diverging from a focus [24, 25] or the ~100% spatial coherence of
light generated using a waveguide configuration [26]. In the spectral domain,
this time-varying intrinsic phase results in large frequency shifts and spectral
broadening or narrowing of the harmonic emission [5].
From a nonlinear optics point of view, HHG is unique as a purely
electronic nonlinearity with a finite time response. In many ways, HHG is
simply an extension of nonlinear optical processes such as second-harmonic
generation (SHG). In SHG, an electron driven in an “anharmonic” potential,
318 Chapter 11
which is created by the crystalline configuration of the material, moves in a

nonsinusoidal path that results in emission of radiation at the second-
harmonic frequency. This is the same for HHG, with the “anharmonic”
potential consisting of free space with the very strong perturbation of an
ionic core localized in a small area. The fact that the electron takes a fraction
of an optical cycle to return to the core means that the atomic emission
depends on the time history of the driving field over the preceding interval of
a few hundred attoseconds. This finite response time makes it possible to
manipulate the atomic response through quantum control techniques by
manipulating the laser pulse shape in time. The first studies relating to
quantum control of high-harmonic generation were simple studies of the
effect of the chirp of the driving laser pulse on the spectrum of high-
harmonic generation [5]. HHG emission generated near the peak of the
driving laser pulse exhibits an intrinsic-phase-induced negative chirp since
ϕ∝ ˜ Ilaser ; i.e., the harmonic emission spectrum varies from blue to red in
time. Any chirp imprinted on the laser pulse itself adds to the intrinsic chirp,
resulting in an asymmetry in the spectral linewidth of the emission with
respect to chirp. These experiments demonstrated a direct spectral-temporal
coupling that is a result of the attosecond time response of the HHG
nonlinearity.
This phenomenon can be exploited in a dramatic way through a
technique called intra-atomic phase matching (IAPM) by temporally shaping
the driving laser pulse [8, 10]. A simple picture of IAPM is given in Figure
11-1. In HHG, the radiation is emitted as a series of attosecond-time-scale
bursts, twice each cycle of the laser field. However, for a particular spectral
component of each burst corresponding to an individual harmonic order, the
phase is not fixed to the driving laser but varies on a cycle-by-cycle basis as
the light intensity increases and decreases, changing the value of the intrinsic
phase. This effect results in a partial destructive interference that reduces the
intensity of the emitted harmonics.
In IAPM, the intensity of a selected harmonic is optimized by adjusting
the shape of the laser pulse so that a nonlinear chirp is added to the light
pulse. This pulse shaping shifts the position of the peaks in the laser field on
a suboptical cycle basis; i.e., on timescales ≤100 attoseconds. The harmonic
bursts from each half-cycle of the laser field can then add constructively,
optimizing the harmonic emission intensity. IAPM thus makes it possible to
“re-phase” the harmonic emission from an atom for a selected harmonic
order. Furthermore, this optimization can in some circumstances be
extremely selective, with the pulse shape chosen to optimize one harmonic
order while suppressing the neighboring harmonics. IAPM thus represents a
useful manipulation of attosecond time-scale electron dynamics and
demonstrates the ability to do attosecond engineering. Another interesting
GENERATION
aspect of the IAPM process is that although it involves precise manipulation

of a laser pulse and an electron wave function, it was not predicted
beforehand but instead was discovered by applying learning algorithm
techniques to optimize the HHG emission by manipulation of the laser pulse
shape [27]. The physics of this process was subsequently explained and
reproduced by implementing a learning algorithm in conjunction with the
three-step model theory of HHG. This computational model reproduced the
experiment and made it possible to deduce the mechanism behind IAPM
[10]. Optimization of high-harmonic emission to generate nearly
monochromatic radiation (i.e., primarily a single harmonic) is desirable for
many applications in ultrafast spectroscopy using EUV radiation. Other
recent, closely related work has shown that the spectral and temporal
characteristics of high-harmonic radiation are sensitive to the carrier-
envelope offset of the driving laser pulse [15].
Figure 11-1. Illustration of intra-atomic phase matching. When the optimized pulse shape is
used, a particular harmonic constructively adds over each half cycle (from [9]).
3. PHASE MATCHING OF HIGH-HARMONIC

GENERATION
Although the single-atom effect of high-order harmonic generation

presents intriguing physics, applications of HHG radiation as a “table-top”
coherent soft x-ray source depend upon optimizing the total flux generated
from an ensemble of atoms. In nonlinear optical processes, the conversion
efficiency from the fundamental to the harmonic field is enhanced by phase
320 Chapter 11
matching. In its simplest manifestation, phase matching will occur when the
driving laser and the harmonic signal travel with the same phase velocity, so
that signal generated throughout the conversion region adds constructively.
In HHG, since the ionization level is changing throughout the pulse and the
increasing density of free electrons has a large effect on the phase velocity of
the driving laser, phase matching can only be obtained in limited time
windows in the pulse. Enhancement of the flux from HHG is possible by
guiding the driving laser with a hollow-core, gas-filled waveguide [28, 29].
The laser light is guided by glancing incidence reflection, allowing
propagation over an extended interaction length with a well-defined intensity
and phase profile. The fundamental and harmonic light propagate through
the waveguide with phase velocities determined by the dispersion of the
neutral gas, the plasma, and the waveguide at the two different frequencies.
The difference in the wave vectors of the fundamental and harmonic light
results in a phase mismatch
q112 λ 2π (1 − η )
∆k = + Pη N atm re (qλ − λ q ) − ∆δ , (4)
4π a 2
λ
where λ, q, a, u11 , η, P, Natm, and re are the fundamental wavelength,

harmonic order, waveguide radius, the first zero of the Bessel function J0,
ionization fraction, gas pressure in atmospheres, number density at 1 atm,
and classical electron radius, respectively; ∆δ = δ (λ ) − δ (λ / q) is the
difference in the index of refraction at the fundamental and harmonic
frequencies of gas at 1 atm. Here, the small contribution from the nonlinear
refractive index is neglected. The waveguide allows control over the
experimental parameters, such as gas pressure and radius, to optimize phase
matching.
For low levels of ionization, the pressure can be adjusted so that the
waveguide and plasma dispersion balance the dispersion due to the neutral
atoms, which is of opposite sign, and phase matching can be achieved (∆k =
0). Since phase matching is accomplished by adjusting the phase velocity of
the fundamental and the harmonic radiation to travel at a phase velocity ~c,
the bandwidth of this phase matching is very broad, encompassing many
harmonic orders. However, for higher intensities, the changing ionization
during the pulse means that there is a narrow time window for phase
matching so that only the harmonics generated in that window are enhanced.
When phase matched, the harmonic signal increases quadratically with
interaction length until it is limited by background absorption of the gas.
This method of phase matching has limitations. When a large enough
fraction of the gas is ionized, the plasma dispersion is greater than the
GENERATION
neutral dispersion for any pressure. Thus phase matching is no longer

possible, and the phase velocity of the fundamental light exceeds that of the
generated harmonic signal. The critical ionization fraction is [30]
−1
 N atm re λ 2 
η cr = 1 +  . (5)
 2π∆δ 

Figure 11-2 plots the pressure required to achieve phase matching as a

function of the ionization fraction.
No Phase-
Matching
∆k = 0
Figure 11-2. Plot of pressure for phase matching as a function of normalized ionization
fraction (η/ηcr). Beyond critical ionization, phase matching is no longer possible.
For low intensities, and therefore lower harmonic orders, where the
ionization fraction is small, phase matching is primarily achieved by
balancing the waveguide dispersion with the neutral gas dispersion. At the
higher intensities required to generate higher energy harmonics, however,
the plasma term becomes significant and higher neutral gas pressures are
needed to compensate. Beyond the critical ionization, phase matching is not
possible [13, 31]. The optimum pressure also depends on the diameter of the
waveguide; for the case of “plane-wave” propagation without a waveguide
in a uniform-density gas, phase matching will occur when the fractional
ionization equals the critical ionization [30]. However, this requires either
very high pulse energy or high gas pressure and results in spatially varying
phase matching that can create a complex spatial mode of the harmonic
emission. In contrast, the use of the hollow waveguide allows for better
phase matching, resulting in build-up of an EUV beam with a spatial mode
322 Chapter 11
of extremely high coherence [26, 32]. In 2004, stable beams of EUV light
with full spatial coherence had only been generated using the hollow
waveguide geometry.
Another consequence of the time-varying–phase-matching conditions is
sensitivity to the CEO of the pulse, both for few-cycle pulses and also in the
case of light pulses with durations of tens of femtoseconds [13, 15, 33]. At
the peak of the pulse, the ionization fraction is increasing significantly in a
series of steps over each half cycle, so that the phase-matching conditions
vary rapidly with time. The ionization as a function of time for a particular
driving laser pulse can be calculated using the Ammosov-Delone-Krainov
(ADK) tunneling ionization rates [34]. Figure 11-3 shows the fractional
ionization of argon gas for two different values of the CEO. The amount of
ionization created at each half-cycle is dependent on the CEO; however, the
final level of ionization is the same for both pulses.
1.0
0.10
0.8
Ionization Fraction
0.08

0.6
0.06
0.4
0.04
0.02
0.2
0.00 0.0
-20 -10 0 10 20
Time (fs)
Figure 11-3. Calculation of the fractional ionization of argon using ADK ionization rates for a
20 fs pulse with peak intensity of 2.2 x 1014 W/cm2 for a cosine and sine pulse.
For a given harmonic order, tuning the pressure can allow phase
matching at different half cycles of the pulse. Durfee et al. observed
evidence of the CEO effect on phase-matching [13]. Figure 11-4(a) shows
the calculation of the pressure dependence of the flux of the 29th harmonic
for one value of the CEO for a 20 fs pulse. The different peaks correspond to
optimal phase matching at different ionization “steps.” The exact position of
these peaks depends on the intensity, the pressure, and the CEO of the pulse.
For a fixed gas pressure, the positions of the peaks move around as the
phase-matching conditions change with the CEO. Figure 11-4(b) shows the
experimentally measured 29th order harmonic signal. In the pressure region
corresponding to phase matching at the peak of the pulse, where the amount
of ionization is dependent on the CEO, the signal shows a strongly
GENERATION
oscillating behavior, consistent with the effect of a rapidly varying CEO.

This fluctuation is only observed at higher pressures because, at lower
pressures, exact phase matching between the fundamental and the harmonic
is never achieved, and thus the relative ionization and CEO are averaged
along the length of the waveguide. Rapidly increasing levels of ionization,
together with a stable carrier-envelope offset, may make it possible to phase
match only the emission from a particular single half cycle, giving a single,
enhanced attosecond-duration pulse.
Figure 11-4. (a) Calculation of the flux of the 29th harmonic for a 20 fs laser pulse and peak
intensity of 2.2 x 1014 W/cm2 as a function of argon pressure using a 3 cm waveguide. (b)
Experimentally measured flux for the same conditions as (a) (from [29]).
4. QUASI-PHASE MATCHING OF HIGH-

HARMONIC GENERATION
Implementing high-order-harmonic generation in a waveguide geometry

has many advantages. However, normal phase matching in a straight
waveguide is limited to relatively low levels of ionization below the critical
ionization fraction (typically 0.5%, 1% and 5% in He, Ne and Ar, for
example). This restricts the enhancement of HHG emission to photon
324 Chapter 11
energies < 80 eV for a 20 fs pulse. At higher photon energies and ionization

levels, the plasma contribution becomes much greater than the neutral gas
contribution, making phase matching impossible. In the absence of phase
matching, the harmonic emission builds up periodically over a coherence
length and then interferes with out-of-phase light generated in the next
section of the nonlinear medium, drastically reducing the total flux.
In recent work, we showed that the inability to phase match in high
ionization levels can be overcome for the first time by applying quasi-phase
-matching (QPM) techniques to the HHG process. The ability to quasi-phase
match HHG significantly extends the range of photon energies where it is
possible to generate flux efficiently. The process of QPM was first proposed
in 1962, shortly after the birth of nonlinear optics. Armstrong et al. [35]
proposed a “phase-corrective scheme,” whereby the phase mismatch in a
nonlinear optical process is periodically corrected by introducing a
periodicity in the medium itself of the coherence length. For example, quasi-
phase matching of second-harmonic light can be achieved by periodically
reversing the crystal orientation, and therefore the polarity of the nonlinear
response, so that a 180-degree phase shift is introduced every coherence
length. This reverses the destructive interference, allowing continual build-
up of the harmonic signal. Practical implementation of this concept awaited
the development of crystal-poling techniques in the mid-1990s [36].
Unfortunately, the generation of coherent light at EUV wavelengths must
take place in a gas or on a surface, making the use of standard QPM
techniques impossible. Fortunately, the extreme nonlinear nature of HHG
makes alternate approaches possible. For example, quasi-phase matching of
the frequency conversion process can be achieved by restricting HHG to
regions where the signal will be in phase and add constructively. Some ideas
for quasi-phase matching of HHG include using a modulated density gas or
plasma to periodically change the nonlinear susceptibility [37]. Another
proposal suggests using counter-propagating light to modulate the intensity
of the driving laser [38]. Since both the intrinsic phase and the amplitude of
the harmonic emission vary quite dramatically with laser intensity, even
modest periodic modulations of the laser intensity can result in a phase
modulation of the generated harmonic light. Experimentally, quasi-phase
matching of HHG was first implemented using a technique proposed by
Christov et al. [11]. In this technique, a hollow-core waveguide is modified
to have a periodically changing inner diameter to modulate the driving laser
intensity. Figure 11-5(a) shows a calculation of the signal of the 95th
harmonic as a function of propagation length in a straight waveguide using a
15 fs pulse with a peak intensity of 7 x 1014 W/cm2 and 1 Torr of argon. The
calculation uses three-dimensional propagation models, and the harmonic
generation signal is calculated using a quasi-classical approximation for the
GENERATION
dipole moment. Figure 11-5(b) shows the same calculation using a hollow-
core waveguide with a 0.5 mm period sinusoidal corrugation that changes
the laser intensity by 5%. The modulated waveguide causes a dramatic
enhancement in the final signal (note the different scales of the two plots).
We can understand the basic physics for enhancement of the HHG signal
by using classical nonlinear optics theory for quasi-phase-matching. In a
simplified model of harmonic generation, the field of harmonic order q, after
propagating a distance L in a nonlinear medium, is related to the phase
mismatch, ∆k , by
L
E q ∝ ∫ Eωn ( z ) d ( z )e −i∆kz dz , (6)
0
where Eω is the fundamental field, n is the effective order of the nonlinear

process, d(z) is the nonlinear coefficient, and ∆k is the phase mismatch
calculated in Equation (4). In our case, the nonlinear coefficient can be
expressed as a general periodic function of z with period Λ [36]
∑D e
∞
d (z) = m
iK m z
, (7)
m = −∞
where K m = 2πm / Λ is the effective wave vector of QPM, Λ is the

modulation period of the quasi-phase matching, and m is the order of the
QPM process. Equations (6) and (7) show that the harmonic signal will be
enhanced when K m ≈ ∆k . The signal enhancement is greatest for m = 1, but
QPM still enhances the generated HHG signal for higher values of m.
326 Chapter 11
Figure 11-5. (a) Calculation of the signal of the 95th harmonic for a straight waveguide as a
function of propagation distance. (b) Calculation for a 0.5 mm period modulated waveguide
(from [11]).
The first set of experiments demonstrating quasi-phase matching of HHG

showed a dramatic extension of the harmonic spectrum using modulated
waveguides [12]. In this experiment, 25 fs duration pulses from a high
repetition rate (2–5 kHz, 1 mJ/pulse) Ti:sapphire laser system operating at
760 nm [39] were focused into 150 µ m diameter modulated-hollow-core
waveguides filled with various gases. The modulated waveguides used in
this experiment were produced using precision glass-blowing techniques
starting with a straight hollow-core fiber (also 150 µm in diameter). The
modulations were approximately sinusoidal, with a radial depth on the order
of 10 microns, corresponding to a 13% modulation of the fiber radius. The
initial experiments used waveguides with modulation periods varying from
1–0.5 mm. Figure 11-6 shows the experimentally measured HHG spectra
from He for three different modulated-waveguide periodicities. As the
modulation period is reduced from 1 mm to 0.75 mm to 0.5 mm, the highest
observable harmonic energy increases from 112 eV to 175 eV.
GENERATION
Figure 11-6. Experimental harmonic spectra from 111 Torr He with a driving pulse of 25 fs
and peak intensity ~5 x 1014 W/cm2 for different modulated-waveguide periodicities (from
[12]).
The significance of QPM is that it permits phase matching of HHG at

higher ionization levels and hence higher photon energies than was
previously possible. This effect becomes apparent from a calculation of the
coherence length that results from the mismatch in the phase velocity of the
fundamental and harmonic because of ionization. At high ionization levels,
the phase mismatch is predominantly due to the plasma dispersion and is
given by
qne e 2 λ
∆k plasma ≈ , (8)
4πmeε o c 2
where λ is the laser wavelength and ne is the electron density. For example,
for fully ionized argon at a pressure of 1 Torr, ne = 3.5 x 1016 cm-3, giving ∆k
~7550 m-1 for the 95th harmonic order (150 eV). Therefore, the coherence
length, Lc, given by Lc = π/∆k, is ~0.4 mm. Thus, very substantial levels of
ionization can be compensated for by using QPM with modulation periods in
the range of 1–0.25 mm that can be readily manufactured with glass-blowing
techniques.
In initial experiments, the ionization level in helium was still relatively
low (~1%) and only ~4% in argon. However, more recent experiments have
demonstrated QPM in fully ionized gas. The results of more recent
experiments performed at higher laser intensities of 1.6 x 1015 W/cm2, in 9
328 Chapter 11
Torr of Ne gas, using 0.25 mm-modulated and straight waveguides, are

shown in Figure 11-7 [40]. At these intensities, a simple calculation of the
cutoff harmonic energy, given in Equation (1), yields an expected cutoff of
330 eV. However, the observable harmonic emission from the straight
waveguide [Figure 11-7 (lower curve)] only extends to around 225 eV. In
contrast, the harmonic emission from the 0.25 mm-modulated waveguide
(upper curve) is brighter and extends to significantly higher energies, where
the carbon edge at 284 eV is clearly visible.
Modulated
fiber (0.25mm)
Carbon K-edge
Straight
fiber
Figure 11-7. Comparison of experimental harmonic spectra for neon for straight and
modulated fibers [40].
Using the expression in Equation (8), which is valid at high ionization

levels, we calculate ∆k ~90,000 m-1 for the 183rd harmonic order at the C
edge (284 eV) for ~60% ionized Ne. The effective wave vector for first-order
QPM for a modulation period of 0.25 mm is K1 ~25,000 m-1. Again, the
harmonic signal will be enhanced when K m ≈ ∆k . For the harmonics in Ne
near the C edge, the large phase mismatch of ~90,000 m-1 can be
compensated for using third or fifth order (m = 3,5) QPM. For these cutoff
harmonics, the phase mismatch above 225 eV is large enough that the signal
can only be observed using a modulated waveguide and is not seen in a
straight waveguide.
Based on conservative estimates of detection efficiency, filter
transmission, and measurements of grating efficiency, we determined a
minimum flux of between 106–108 photons/sec in a 10% bandwidth at the C
edge. This flux may be sufficient for biological imaging applications, and a
number of improvements should further increase it.
The data of Figure 11-7 illustrate that the highest observed harmonic
orders are not limited by the Ip + 3.17 Up relation but by ionization-induced
phase mismatch. The modulated waveguide compensates for this phase
GENERATION
mismatch, making it possible to observe higher harmonics. By the peak of

the laser pulse, calculations indicate that Ne should be ~70–90% ionized,
which is well above the ionization level at which it has been previously
possible to demonstrate phase-matching enhancement. Although past work
observed HHG from He in the water window, this process was not phase
matched and produced significantly lower signal levels [18, 41].
In the future, since the highest HHG photon energy scales linearly with
laser intensity, using very reasonable laser parameters (i.e., pulse durations
of 10 fs, intensities of 5x1015 W/cm2, and waveguides with 0.1 mm
periodicity), it will be possible to generate high-order quasi-phase-matched
light at photon energies beyond 1 keV.
5. CONCLUSION
Recent advances in laser technology and in our understanding of electron

dynamics in strong fields have resulted in enormous progress in the quest to
develop practical coherent light sources at wavelengths much shorter than
visible light. The use of quantum control techniques makes it possible to
generate light with a spectrum that can be engineered for specific
applications and to allow for the generation of attosecond-duration pulses.
Using quasi-phase-matching techniques, there is a clear path to pursue for
generating flux sufficient to implement a table-top soft x-ray microscope and
to reach multi-keV energies.
ACKNOWLEDGEMENTS
The authors gratefully acknowledge funding for this work by the

National Science Foundation and the Department of Energy.
REFERENCES
[1] S. Backus, C. G. Durfee, M. M. Murnane, and H. C. Kapteyn, Rev. Sci.
Instrum. 69, 1207-1223 (1998).
[3] A. McPherson, G. Gibson, H. Jara, U. Johann, T. S. Luk, I. A. McIntyre, K.
Boyer, and C. K. Rhodes, J. Opt. Soc. Am. B 4, 595-601 (1987); M. Ferray,
330 Chapter 11
A. Lhuillier, X. F. Li, L. A. Lompre, G. Mainfray, and C. Manus, J. Phys.

B. 21, L31-L35 (1988).
[4] M. Lewenstein, P. Balcou, M. Y. Ivanov, A. L'Huillier, and P. B. Corkum,
Phys. Rev. A 49, 2117-2132 (1994).
[5] Z. Chang, A. Rundquist, H. Wang, I. Christov, H. C. Kapteyn, and M. M.
Murnane, Phys. Rev. A 58, R30-R33 (1998).
[6] M. Bellini, C. Lynga, A. Tozzi, M. B. Gaarde, T. W. Hansch, A. L'Huillier,
and C. G. Wahlstrom, Phys. Rev. Lett. 81, 297-300 (1998); H. J. Shin, D.
G. Lee, Y. H. Cha, J. H. Kim, K. H. Hong, and C. H. Nam, Phys. Rev. A
63, 053407 (2001).
[7] P. B. Corkum, N. H. Burnett, and M. Y. Ivanov, Opt. Lett. 19, 1870-1872
(1994); P. M. Paul, E. S. Toma, P. Breger, G. Mullot, F. Auge, P. Balcou,
H. G. Muller, and P. Agostini, Science 292, 1689-1692 (2001); M.
Hentschel, R. Kienberger, C. Spielmann, G. A. Reider, N. Milosevic, T.
Brabec, P. Corkum, U. Heinzmann, M. Drescher, and F. Krausz, Nature
414, 509-513 (2001); R. Kienberger, E. Goulielmakis, M. Uiberacker, A.
Baltuska, V. Yakovlev, F. Bammer, A. Scrinzi, T. Westerwalbesloh, U.
Kleineberg, U. Heinzmann, M. Drescher, and F. Krausz, Nature 427, 817-
821 (2004); I. P. Christov, M. M. Murnane, and H. C. Kapteyn, Phys. Rev.
Lett. 78, 1251-1254 (1997).
[8] R. Bartels, S. Backus, E. Zeek, L. Misoguti, G. Vdovin, I. P. Christov, M.
M. Murnane, and H. C. Kapteyn, Nature 406, 164-166 (2000).
[9] R. Bartels, S. Backus, I. Christov, H. Kapteyn, and M. Murnane, Chem.
Phys. 267, 277-289 (2001).
[10] I. P. Christov, R. Bartels, H. C. Kapteyn, and M. M. Murnane, Phys. Rev.
Lett. 86, 5458-5461 (2001).
[11] I. P. Christov, H. C. Kapteyn, and M. M. Murnane, Optics Expr. 7, 362-367
(2000).
[12] A. Paul, R. A. Bartels, R. Tobey, H. Green, S. Weiman, I. P. Christov, M.
M. Murnane, H. C. Kapteyn, and S. Backus, Nature 421, 51-54 (2003).
[13] C. G. Durfee, A. Rundquist, S. Backus, Z. Chang, C. Herne, H. C. Kapteyn,
and M. M. Murnane, J. Nonlinear Opt. Phys. Mater. 8, 211-234 (1999).
Gohle, R. Holzwarth, V. S. Yakoviev, A. Scrinzi, T. W. Hansch, and F.
Krausz, Nature 421, 611-615 (2003).
Hall, and S. T. Cundiff, Science 288, 635-639 (2000); S. T. Cundiff and J.
Ye, Rev. Mod. Phys. 75, 325-342 (2003).
[17] P. B. Corkum, Phys. Rev. Lett. 71, 1994-1997 (1993); K. C. Kulander, K. J.
Schafer, and J. L. Krause, in Superintense Laser Atom Physics, edited by B.
Pireaux, A. L’Huillier and J. Rzazewski (Plenum, New York, 1994), Vol. B
316, p. 95.
GENERATION
[18] Z. H. Chang, A. Rundquist, H. W. Wang, M. M. Murnane, and H. C.

Kapteyn, Phys. Rev. Lett. 79, 2967-2970 (1997).
[19] C. Spielmann, N. H. Burnett, S. Sartania, R. Koppitsch, M. Schnurer, C.
Kan, M. Lenzner, P. Wobrauschek, and F. Krausz, Science 278, 661-664
(1997).
[20] H. C. Kapteyn, L. B. Dasilva, and R. W. Falcone, Proc. IEEE 80, 342-347
(1992).
[21] A. L'Huillier, M. Lewenstein, P. Salieres, P. Balcou, M. Y. Ivanov, J.
Larsson, and C. G. Wahlstrom, Phys. Rev. A 48, R3433-R3436 (1993).
[22] R. Haight and D. R. Peale, Rev. Sci. Instrum. 65, 1853-1857 (1994); M.
Bauer, C. Lei, K. Read, R. Tobey, J. Gland, M. M. Murnane, and H. C.
Kapteyn, Phys. Rev. Lett. 8702, art. no.-025501 (2001).
[23] D. Descamps, C. Lynga, J. Norin, A. L'Huillier, C. G. Wahlstrom, J. F.
Hergott, H. Merdji, P. Salieres, M. Bellini, and T. W. Hansch, Opt. Lett.
25, 135-137 (2000); J. J. Rocca, C. H. Moreno, M. C. Marconi, and K.
Kanizay, Opt. Lett. 24, 420-422 (1999).
[24] P. Balcou, P. Salieres, A. Lhuillier, and M. Lewenstein, Phys. Rev. A 55,
3204-3210 (1997).
[25] P. Salieres, T. Ditmire, M. D. Perry, A. Lhuillier, and M. Lewenstein, J.
Phys. B. 29, 4771-4786 (1996); J. Peatross, M. V. Fedorov, and K. C.
Kulander, J. Opt. Soc. Am. B 12, 863-870 (1995).
[26] A. R. Libertun, X. Zhang, A. Paul, E. Gagnon, T. Popmintchev, S. Backus,
M. M. Murnane, H. C. Kapteyn, and I. P. Christov, App. Phys. Lett. 84,
3903-3905 (2004).
[27] R. S. Judson and H. Rabitz, Phys. Rev. Lett. 68, 1500-1503 (1992).
[28] A. Rundquist, C. G. Durfee, Z. H. Chang, C. Herne, S. Backus, M. M.
Murnane, and H. C. Kapteyn, Science 280, 1412-1415 (1998).
H. C. Kapteyn, Phys. Rev. Lett. 83, 2187-2190 (1999).
[30] A. Rundquist, Ph.D Thesis, Washington State University (1998).
[31] E. Constant, D. Garzella, P. Breger, E. Mevel, C. Dorrer, C. Le Blanc, F.
Salin, and P. Agostini, Phys. Rev. Lett. 82, 1668-1671 (1999).
[32] R. A. Bartels, A. Paul, H. Green, H. C. Kapteyn, M. M. Murnane, S.
Backus, I. P. Christov, Y. W. Liu, D. Attwood, and C. Jacobsen, Science
297, 376-378 (2002).
[33] G. Sansone, C. Vozzi, S. Stagira, M. Pascolini, L. Poletto, P. Villoresi, G.
Tondello, S. De Silvestri, and M. Nisoli, Phys. Rev. Lett. 92, 113904
(2004).
[34] D. N. B. Ammosov M.V., Krainov V.P., Soviet Physics JETP 64, 1191
(1986).
[35] J. A. Armstrong, N. Bloembergen, J. Ducuing, and P. S. Pershan, Phys.
Rev. A 127, 1918 (1962).
332 Chapter 11
[36] M. M. Fejer, G. A. Magel, D. H. Jundt, and R. L. Byer, IEEE J. Quantum

Electron. 28, 2631-2654 (1992).
[37] P. L. Shkolnikov, A. Lago, and A. E. Kaplan, Phys. Rev. A 50, R4461-
R4464 (1994).
[38] S. L. Voronov, I. Kohl, J. B. Madsen, J. Simmons, N. Terry, J. Titensor, Q.
Wang, and J. Peatross, Phys. Rev. Lett. 87, 133902 (2001).
[39] S. Backus, R. Bartels, S. Thompson, R. Dollinger, H. C. Kapteyn, and M.
[40] E. A. Gibson, A. Paul, N. Wagner, R. Tobey, D. Gaudiosi, S. Backus, I. P.
Christov, A. Aquila, E. M. Gullikson, D. T. Attwood, M. M. Murnane, and
H. C. Kapteyn, Science 302, 95-98 (2003).
[41] C. Spielmann, L. Xu, and F. Krausz, Appl. Optics 36, 2523-2525 (1997).
Chapter 12
APPLICATIONS OF ULTRAFAST LASERS
Sensors of extreme sensitivity
Jean-Claude Diels1, R. Jason Jones2 and Ladan Arissian1

1
Department of Physics and Astronomy, University of New Mexico
2
JILA, National Institute of Standards and Technology and the University of Colorado
Abstract: We discuss implementations of mode-locked ring lasers, their stabilization via

passive optical cavities, and their applications to the development of ultra-
sensitive sensors.
Key words: mode-locked laser, ring laser, sensors, optical parametric oscillator, optical
cavity stabilization
The frequency of an electromagnetic wave is a quantity that lends itself

to the most precise measurement of all physical observables. Stabilized cw
lasers have led to accurate and precise length and time measurements, with
an accuracy reaching one part in 1015 [1]. By transposing the stability of a
single frequency laser to more than 1,000,000 equidistant frequencies, the
mode-locked laser opens the way to new metrological applications. Some of
these applications (optical frequency measurements, atomic clocks, extended
combs leading to attosecond pulses, nonlinear optics, and attosecond
metrology) have been covered in the previous chapters. The applications
discussed in this chapter relate to the remarkable phase relationship that
exists across the million equidistant modes. As a consequence of that phase
relationship, it is possible to make phase measurements with an
unprecedented accuracy. Unlike standard phase measurements that are made
by measuring the arccos of an interference fringe (essentially an intensity
measurement), the phase measurements presented here always convert the
phase into a frequency. We will discuss in Sections 1-4 how mode-locked
combs lead to practical sensing applications such as measurements of
displacements, fields, indices of refraction, and elongation. The sensitivity
limit of these applications can be extended through stabilization. The
334 Chapter 12
stabilization technique outlined in Section 5 can lead to extremely sensitive

measurements of dispersion, taking advantage of the phase relation between
modes across a very large bandwidth (Section 6). Finally, future prospects of
stabilizing a ring laser for all sensing applications are discussed in Section 7.
1. MODE LOCKING
The purpose of this introduction to “mode locking” is to bring the reader

back to the real world of the typical mode-locked laser, which is quite
different from the highly sophisticated sources discussed in this book. The
expression mode locking suggests the concept of equidistant, longitudinal
modes of a laser cavity emitting in phase. This was, in fact, the early
interpretation given for cw lasers emitting a train of pulses at the cavity
repetition rate. As detailed in the previous chapters, this frequency
description of mode locking is equivalent to having a continuous single
frequency carrier, sampled at equal time intervals by an envelope function,
in the time domain. Unless sophisticated stabilization techniques (described
in earlier chapters) are used, an ordinary mode-locked laser does not fit the
above description.
Each cavity mirror is subject to vibrational motions of its support. A
typical mechanical resonance is around 100 Hz and has motion amplitude
ranging from 0.1 to 1 µ m. Because of this motion, the position of the
longitudinal modes of the cavity is not fixed in time. The intracavity laser
pulse follows this change in time, because it is Doppler-shifted at each
reflection from the vibrating mirror. For this reason, one can no longer talk
about an output pulse train made up of identical pulses. The difference
between the properties of the radiation from an ultrastable “frequency comb”
as opposed to the ordinary “mode-locked laser” can be established in a
coherence measurement. Coherence can be measured with a Mach-Zehnder
interferometer, as sketched in Figure 12-1. In the case of a single pulse, no
coherence can be measured for an optical delay of the interferometer
exceeding the pulse duration. In the case of a pulse train, as the delay of the
Mach Zehnder is being scanned, interferences may occur for delays equal to
an integer multiple m of the pulse spacing. In a standard mode-locked laser,
the fringe visibility decays rapidly to zero with increasing m. This is because
the carrier frequency is Doppler-shifted at each round trip, and the pulse
spacing is not constant because of mirror motion. In the frequency domain,
there is no longer a perfect comb of delta-function modes in phase.
12. APPLICATIONS OF ULTRAFAST LASERS 335
(a)
Delay of
fs pulses 1 s??
Interferences?
(b)
Beat
fs pulses note
A
time
D
B
Figure 12-1. Coherence measured by a Mach-Zehnder interferometer. (a) In the case of a

regular pulse train, by providing the near perfect frequency comb as discussed in previous
chapters, interferences could be observed even in a delay of the interferometer approaching 1
s, or 14 orders of magnitude larger than the pulse duration. (b) The coherence of an unknown
source A can be measured by optical beating with a reference source B, provided both sources
have the same repetition rate. An optical delay is required to ensure that the pulse of each
train interfere at detection.
A Mach-Zehnder interferometer is obviously not a practical instrument

for measuring the coherence of a pulse train over many interpulse spacings.
As described in the previous chapters, it is possible to have a pulse train
coherent over delays 14 orders of magnitude larger than the pulse duration.
Another method to measure the coherence of source A is to compare it with a
perfectly coherent source B with the same repetition rate [Figure 12-1(b)]. If
two sources of the same repetition rate but different carrier frequencies are
made to interfere on a detector, a beat note at the difference between the two
carrier frequencies will be observed [2]. The inverse of the bandwidth of this
beat note is the coherence time of source A. A convolution of the bandwidths
of each source is involved if source B is not perfectly coherent.
An example of a beat note between two pulse trains of the same
repetition rate, but different carrier frequency, is shown in Figure 12-2.
336 Chapter 12
0.8
(a)
0.0
volts
-0.8
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10
seconds
(b)
0.5
0.4
Amplitude
0.3
FWHM ~ 3Hz
0.2 Coherence time 0.3 s
0.1
0.0
200 250 300 350 400 450 500
Frequency (Hz)
Figure 12-2. (a) Beat note recorded between two femtosecond pulse trains of the same
repetition rate. (b) Fourier transform of the same recording, showing a bandwidth of only 3
Hz, or 0.8 Hz broader than the sampling time limit.
It may seem like a “tour de force” to achieve such a degree of

stabilization as to record a coherence time of 0.3 s for a comb of fs pulses.
Indeed, this bandwidth, obtained for an unstabilized mode-locked laser, is 3
orders of magnitude narrower than that of a single mode of a frequency
comb stabilized to a reference cavity [3]. The measurement for Figure 12-2,
however, was made with an unstabilized laser. The “trick” in this
measurement is that the two beams being interfered have individually low
coherence, but the fluctuations in repetition rate and carrier frequency of the
two lasers are correlated. The two beams are the outputs corresponding to a
clockwise and a counterclockwise circulating pulse in a ring cavity.
Therefore, the same fluctuations in the laser apply to the two laser beams
since both share the same geometrical mode of the cavity. The properties of
such a mode-locked bidirectional ring laser are discussed in Section 2.
2. GROUP AND PHASE VELOCITIES IN RING

LASERS
Round-trip i
ϕ0
Envelope
crossing
∆ϕ
Round-trip i+1
Moving interference
pattern (standing wave
if the two carrier frequencies are equal)
Figure 12-3. Topological representation of a mode-locked ring laser. The pulse envelopes
overlap at the same location of the cavity at each round trip.
A topological representation of a mode-locked ring laser is given in

Figure 12-3. Two femtosecond pulses circulate in opposite sense in the
cavity, crossing at each round trip at the same point in the cavity. In general,
a nonlinear element, such as a saturable absorber, forces the pulses to meet at
a fixed point. The average group velocity is the same for the counter-
circulating beams. The phase velocity, however, may differ for the two
directions of circulation. Therefore, the carrier frequency for the two
directions may or may not be identical. If the two carrier frequencies are
equal, a standing wave is formed in the cavity at the pulse crossing. If the
frequency of the clockwise circulating pulse is slightly different from that of
the counterclockwise pulse, the interference pattern of the two waves moves
under the crossing envelopes (top left, Figure 12-3). Therefore the phase of
the interference pattern with respect to the fixed envelope changes at each
round trip. It is that change of carrier-to-envelope phase, at each roundtrip,
that is being measured by the beat note between the two outputs. At each
round-trip, the clockwise and counterclockwise pulses come to the same
position (because the group velocity is the same for both directions). Let us
assume that at a given round-trip index, i, the interference pattern has the
338 Chapter 12
phase φ0 with respect to the envelope. At the next round-trip, i + 1 , the

phase of the interference with respect to the envelope is φ0 + ∆φ.
Even though the cavity perimeter fluctuates by as much as a wavelength
during the measurement and the comb mode is imperfect, the interference
pattern has a bandwidth of only of the order of a few Hz. In fact, in this
measurement, the interference pattern is not determined by the total
fluctuation of the cavity but by the fluctuation in difference in clockwise vs
counterclockwise perimeters during each single roundtrip. A difference in
perimeter ∆P between the clockwise and counterclockwise circulating beams
causes a beat note ∆ν given by [4]:
∆ν ∆P
= , (1)
ν P
where P is the perimeter of the ring laser and ν the optical frequency. A beat
note ∆ν of 3 Hz corresponds to a perimeter difference ∆P of P x 3 Hz/ν ≈
10-14 m. A mirror position fluctuation of 10-14 m during a round-trip time of
10 ns corresponds to a motion of 0.01 µm in a mechanical resonance period
of 10 ms. To further reduce the bandwidth of the beat note, it is necessary to
suppress the fluctuations that cause a difference in time of arrival that, in
turn, will cause a difference in perimeter between the two circulating pulses
during any round trip. Stabilization of the cavity perimeter may achieve the
goal of reducing the beat note bandwidth. The type of stabilization required
will be discussed in Section 7, which follows a discussion of the applications
of the ring laser.
3. RING LASERS AS SENSORS
The detection of the beat note between the two outputs of a mode-locked
ring laser leads to numerous sensing applications of exceptional sensitivity.
The “sensor” consists of a mode-locked ring laser operating in a
bidirectional mode and a detector located at an equal optical path distance
from the intracavity pulse crossing point, as sketched in Figure 12-4. To
avoid gain competition and have equal energy for the two counter-
circulating intracavity pulses, the amplifying medium should be located 1/4
cavity perimeter away from the pulse crossing point.
Synchronized
S excitation
Detector
Amplitude
coupling
Gain medium
∆ν ∆P
=
ν P
∆ν = ∆ϕ /τRT
Figure 12-4. The general configuration of the sensor is that of a mode-locked ring laser and a
detector located at equal optical path from the pulse crossing point, via the clockwise or
counterclockwise direction of circulation. The gain medium is located at 1/4 cavity perimeter
from the pulse crossing point. S is the sample of which the characteristics have to be
determined. An eventual excitation of the sample is applied exactly at the cavity round-trip
time.
There are three types of detection that can be performed: (1)

Measurements of nonreciprocal effects, in which the “sample” can be the
laser itself (rotation sensing), a flowing fluid (motion measurement by
Fresnel drag), a material with a high Verdet constant (magnetic field
measurement), or a resonant atomic vapor (intracavity phase spectroscopy);
(2) changes of cavity optical length externally synchronized such as electro-
optic effect, displacement due to phonons, electric field measurements, or a
nonlinear medium (measurement of nonlinear index); and (3) monitoring of
continuous small motions
The first type of response exists also with cw lasers, but, with mode-
locked ring lasers, the sensitivity is not limited by a dead band. The next two
types of response are unique to mode-locked ring lasers, for which the
detection is not limited to nonreciprocal effects. While case (2) above
requires a signal that can be modulated at the frequency of the cavity
repetition rate, a continuous or slow drift can also be detected [case (3)] by
sending the two intracavity pulses through separate pathways using electro-
optical means [5].
340 Chapter 12
3.1 Case (1): nonreciprocal effects
Without the need for adding any intracavity element, the device of Figure
12-4 can be used for rotation sensing. The rotational beat note response ∆νR
is related to the rotation of the laser (angular velocity ΩR) around its axis by:
4A
∆ν R = ΩR , (2)
Pλ
where A is the ring area. A large area to perimeter (A/P) ratio is desirable to
have an optimum sensitivity to rotation. For applications where rotation
sensing is not desired, a figure-of-8 laser with a net-zero area is the preferred
geometry. Since the femtosecond ring laser acts as laser gyroscope without a
dead band, the sensitivity to rotation is only limited by the bandwidth of the
beat note, itself directly related to the stability of the resonator. Without
active stabilization of the cavity, the 3 Hz bandwidth of the beat note of the
Ti:sapphire ring laser shown in Figure 12-2 corresponds to a sensitivity to
rotation of 0.2 degrees/hour, which compares favorably with the best
navigational optical gyroscopes. Stabilization of the cavity perimeter would
improve this figure by at least three orders of magnitude.
Rotation is not the only effect that causes a differential change in carrier
frequency for the counter-circulating pulses. The beat note is extremely
sensitive to air flows through the Fresnel Drag effect. The first evidence of
nonreciprocal response in a ring mode-locked dye laser was a measurement
of air flow through the Fresnel drag [6]. Very small air currents (air
velocities of the order of 1 mm/s) can also account for the observed beat note
bandwidth in Figure 12-2.
Magnetic field detection requires the insertion of a material with a high
Verdet constant in the ring cavity, as with sample S sandwiched between two
quarter-wave plates in Figure 12-4. The clockwise and counter-clockwise
circulating pulses become, respectively, right and left circular polarized in
the sample. In presence of a magnetic field, there is a difference in optical
path ∆P between the two senses of circulation, resulting in a beat note
proportional to the magnetic field. This effect can be particularly large in the
presence of resonance. Optimal magnetic field sensitivity requires the use of
a material for which the index change due to a magnetic field is maximal.
Atomic vapors, such as used in potassium magnetometers [7], have a large
resonant Faraday rotation, hence a large magnetic field-induced change in
index for circularly polarized light. These vapor magnetometers have been
used successfully in aerial mapping of the earth’s magnetic field and to
locate minerals, in particular magnetic iron deposits and nickel deposits [8].
The beat note response of the ring laser is significantly more sensitive to
magnetic field than the extracavity measurement performed in the potassium

magnetometer [7] and should approach the sensitivity of the superconducting
weak link detector or SQUID [9]. The latter has the disadvantage that it
provides only microscopic information and has a limited dynamic range.
To reach a better sensitivity than the superconducting magnetometer, it is
necessary to stabilize the laser to the atomic line selected as magnetic field
sensor. Stabilizing a ring laser that produces a continuous train of ultrashort
pulses is a very challenging task, which will be discussed in Section 7.
Intracavity spectroscopy is the last example of nonreciprocal
measurement that, unlike the previous ones, requires the full mode comb
structure of the mode-locked laser. In the case of the atomic vapor used in
magnetometers, the level structure is a Λ , of which each branch is near
resonance with one of the counter-circulating pulses. The two counter-
circulating pulses are thus probing the near resonant dispersion of the sample
S. Doppler-resolved near-resonant spectroscopy can be performed by
recording the beat note while linearly scanning the carrier frequency of the
mode-locked train in one direction. This experiment applied to samarium led
to the first observation of the dispersion of the dark resonance [10].
3.2 Case (2): reciprocal effects that can be synchronized

to the cavity repetition rate
In Figure 12-4, an arrow at the sample S indicates a possibility of

changing some property of a cavity element with the same periodicity as the
time of arrival of one of the intracavity pulses at that location. The
synchronization can be externally applied (through a detector monitoring the
pulse rate) or self-induced, as in the measurement of a nonlinear index. If the
sample S in Figure 12-4 has an intensity dependent index of refraction n(I),
there will be a contribution to the beat note given by:
∆φ 2πd
∆ν = ≈ (n(I + ) − n(I −)) , (3)
τ RT λ
where I+ and I- are the beam intensities at the sample for the clockwise and
counterclockwise circulating pulses, respectively. Equation 3 is valid only in
the approximation that the sample thickness d is sufficiently small so that the
intensities of each beam can be assumed to be constant in the sample. The
difference n(I + ) − n(I −) can be varied by translating the sample along the
beam axis between two lenses or spherical mirrors [11]. An example of
measurement of the nonlinear index of lithium niobate with an optical
parametric oscillator ring laser can be found in Reference [12].
342 Chapter 12
The best sensitivity in these measurements can be achieved when all

other contributions to the beat note can be eliminated, in particular the
nonreciprocal effects listed above. The topology least sensitive to all
nonreciprocal effects is the linear limit of an elongated ring. Instead of a
clockwise and counterclockwise circulating pulse, in this “linear ring” two
pulses circulate simultaneously, meeting exactly in the middle of the
resonator. As with the ring laser sensor, a beat note is measured between the
output pulse trains corresponding to each of these pulses. Unlike the case of
a ring laser, the phase shift is the same for either of the intracavity
circulating pulses because they travel through the same optical elements in
the same order. There is therefore no bias beat note. To measure the electro-
optic coefficient of a sample S inserted in this cavity, an electric field is
applied at the cavity round-trip rate (1/2 of the pulse rate). To observe a beat
note between the two interfering pulse trains requires a voltage only in the
milliVolt range to be applied to the sample [13], leading to the most
sensitive and accurate method to determine an electro-optic coefficient.
Figure 12-2 indicates that the sensitivity to the position of a reflecting
surface is of the order of 0.001 Å. One immediate application is the study of
displacements associated with ultrasound phonons, which can be produced
in a coating synchronously with the cavity repetition rate. Other applications,
such as monitoring the growth rate of a crystal or optical scanning
nanoscopy, require a similar sensitivity to slow-moving displacements.
3.3 Case (3): reciprocal effects, slow motions
Either in the ring laser or in its linear limit, an electro-optic device can be
inserted in the cavity to split the path of two circulating pulses, sending one
through the sample to be analyzed and the other through a reference path [5],
which can be stabilized to a reference cavity. Unlike the previous
measurement, a stable cavity is not a sufficient attribute to perform accurate
measurements. As will be seen in Section 5 on stabilization, the reference
cavity may drift; hence there would be an ambiguity in the measurement: is
the change in beat note reflecting an elongation of the sample or of the
reference cavity? Thus this type of measurement requires accuracy (long-
term stability) in addition to short-term stability through the reference cavity.
The solution is to use properties of atomic transition to accurately fix a
reference repetition rate and carrier frequency [14].
4. INTRACAVITY-PUMPED OPTICAL
PARAMETRIC OSCILLATOR AS A MODE-
LOCKED RING LASER
Ti-Sapphire
OPO crystal
D
L2
MQW
Figure 12-5. Illustration of the intracavity optical parametric oscillator (OPO) pumped by the
Ti:sapphire laser. Four LaKL21 prisms are incorporated in the pump cavity to compensate for
the group-velocity dispersion (GVD) from the Ti:sapphire crystal, the Periodically Poled
Lithium Niobate (PPLN) crystal, and other intracavity elements such as lenses and mirrors.
This four-prisms configuration was necessitated by the desire to have large GVD
compensation (needed because of the large positive GVD of LiNbO3) and a reasonably short
cavity length (1/2 of the perimeter of the OPO cavity). The main control of the GVD
compensation is the prism spacing L2. Two quantum wells (MQW) of AlGaAs on top of a
mirror structure are used in the cavity as a saturable absorber to mode lock the laser.
Most measurements with ring laser sensors have been performed with
either dye lasers [6] or Ti:sapphire lasers [4, 11, 13] mode locked with
flowing saturable absorbers. Such systems are appropriate for demonstration
purposes, but the future is for all-solid-state compact systems that can be
readily stabilized. Some initial results on the most promising system, an
intracavity-pumped Optical Parametric Oscillator (OPO) ring, are presented
here. A synchronously pumped OPO offers the possibility of decoupling
relative phase and repetition rates of the oscillating signals. The repetition
rate of the OPO signal is equal to that of the pump laser and can therefore be
controlled through stabilization of the mode spacing of the pump laser. The
carrier frequency of the OPO is directly determined by the OPO cavity
length. The position of the crossing point of the two circulating pulses in the
OPO is simply determined by the timing of two successive pump pulses sent
344 Chapter 12
in opposite directions. It is essential that the OPO crystal be part of the pump
cavity. Only such an intracavity-pumped configuration can ensure that the
OPO gain volumes for either direction are directly superimposed (because
they share a common mode of the pump cavity). The operation of such a ring
OPO has been demonstrated [15]. The cavity configuration of such an OPO-
pumped intracavity by a linear Ti:sapphire laser is sketched in Figure 12-5.
The Ti:sapphire laser radiation consists of 200 fs pulses centered at 785
nm, with a repetition rate of 95MHz. The OPO crystal is a 3 mm Brewster-
cut PPLN crystal (HC Photonics, Taiwan) with a period of 19.4 mm (quasi-
phase matching for the signal near 1.36 mm) that is temperature stabilized at
408 K to prevent photorefractive damage. The cavity mirrors and crystal are
located away from the two crossing points of the ring.
0 .3 5
(a.u.)
0 .3 0
Amplitude
(a.u.)
0 .2 5
Amplitude
0 .2 0
- 0 .0 4 - 0 .0 2 0 .0 0 0 .0 2 0 .0 4
T Time
i m e (s)
(s )
0 .0 8
68.64
Amplitude (a.u.)
0 .0 6
0 .0 4
Amplitude (a.u.)
0 .0 2
0 .0 0
0 5 0 1 0 0 1 5 0 2 0 0
B e aBeat
t f r frequency
e q u e n c (Hz)
y (H z )
Figure 12-6. Beat frequency (a) taken by a sensitive InGaAs detector (D in Figure12-5) and
(b) its Fourier transform.
The observed beat frequency is due to the intensity difference of the

intracavity beams. The pump in one direction is attenuated by depletion
through the crystal and transmission through four antireflection-coated
surfaces and one reflective coating before coming back to pump into the
other direction. The intensity difference, combined with the nonlinear
refractive index of the lithium niobate crystal, results in a differential phase
shift between the counter-circulating OPO beams—hence a beat note. The
narrow bandwidth of the beat note in Figure 12-6 demonstrates that indeed,
intracavity pumping ensures that the gain for the two directions originates
from the same point in the OPO crystal. The absence of any moving part and
the fundamental absence of a dead band make the intracavity-pumped OPO

ring the most promising sensor based on nonreciprocal effects in ring
cavities. The beat note bandwidth results from cavity length fluctuations and
can be eliminated by active stabilization. A reduced beat note bandwidth
implies a better sensitivity (resolution) in the response of the ring laser as a
sensor for any of the applications listed in Section 3.
5. STABILIZATION OF THE FREQUENCY COMB
The bidirectional ring laser produces a pair of highly correlated pulse

trains that allow one to minimize the effects of perturbations on the
femtosecond comb and make precision measurements of nonreciprocal
effects inside the laser cavity. The absolute noise on each pulse train still has
an impact—albeit greatly diminished—on the precision of these
measurements. Further reduction of the linewidth of the signal—a beat note
in the case of the ring laser—requires active stabilization of the femtosecond
laser cavity. This section is devoted to the stabilization technique applied to
a linear cavity and its application. In Section 7, this technique is extended to
the ring laser.
Stabilization and control of mode-locked lasers has been demanded by
novel applications using the femtosecond comb. Like their single frequency
counterparts, mode-locked lasers are perturbed by mechanical and acoustic
vibrations that shift and broaden the comb structure. However, because of
the higher peak intensities of the intracavity pulses and the mode-locking
dynamics, these systems are more susceptible to noise induced by pulse
energy fluctuations. As discussed in previous chapters, the most common
technique for stabilization of the femtosecond laser uses a combination of
two phase-locked loops to lock both degrees of freedom (the carrier
frequency and the repetition rate) of the femtosecond frequency comb. Some
techniques choose to do all measurements in the rf domain, substituting the
offset frequency for that of the average frequency. The most straightforward
implementation of such techniques requires octave bandwidth spanning.
In the technique described here, optical frequencies are directly used to
avoid the multiplication error involved with rf-source measurements. Beat-
note detections with a high signal-to-noise ratio and at least one highly stable
frequency reference are required. In this method, the average frequency and
repetition rate of the femtosecond laser is locked to a stable Fabry-Perot
reference cavity. With proper stabilization, the pulse train inherits the
stability of the reference cavity and precision measurements of phase shifts
are spectrally resolved across a broad bandwidth. In the time domain, one
can think of the reference cavity as a "reference pulse" source for time scales
346 Chapter 12
up to the lifetime of the cavity. Although mechanical cavities are susceptible

to drifts on long time scales ( >10 s), such cavities can provide an extremely
high stability reference (σ(1s) ~3 x 10-16 [1]) because of a combination of
narrow cavity resonances and a linear response to the interrogating laser.
5.1 Locking femtosecond pulses to stable reference

cavities
Earlier stabilization experiments with mode-locked lasers often used

short Fabry-Perot etalons to stabilize the average position of the comb,
directly employing the techniques developed for single frequency
lasers [16]. More recent experiments have used Fabry-Perot cavities as
“mode filters,” transmitting every 20th comb component, such that any
individual mode of the comb can be unambiguously identified with a wave
meter [17, 18].
Constructive interference of the femtosecond pulse with the intra-
reference-cavity pulse is achieved when the pulse envelopes and optical
phases overlap. Satisfying these two conditions will get to the resonance and
energy build-up in the cavity [19], where each pulse in the cavity is a life-
time average of the previous pulses. Complete transmission is identical to
locking the repetition rate and average frequency of the femtosecond pulses
to the reference cavity.
Close-up of such an experiment [20] is shown in Figure 12-7. The
reference cavity is a long (62.5 cm) Fabry-Perot resonator made of a solid
block of ultralow-expansion quartz (ULE) with high reflectivity mirrors of
the same material optically contacted on both ends. The cavity is placed in a
vacuum chamber to isolate it from thermal and acoustic noise and, if needed,
to control the pressure. The vacuum chamber usually keeps the pressure
inside the reference cavity below 15 mTorr. The laser cavity length is twice
that of the reference cavity, so that every other mode can be transmitted.
Kerr-lens mode locking provides pulses around 40 fs. The two degrees of
freedom are controlled with two piezoelectric transducers (PZT). A change
in the cavity length (PZT1) results in mode shifting and tilting of the end
mirror (PZT2) selects different frequency components and controls the group
velocity. These adjustments can be made nearly orthogonal by proper gain
adjustment on Servo1 and Servo2. An acousto-optic modulator is used for
fast (up to 150 kHz) carrier-frequency corrections without changing the
repetition rate of the laser. The error signal is obtained by Pound-Drever-
Hall discrimination [21]. The laser beam is phase modulated at 10.7 MHz
and detection of the cavity resonance is made at the modulated frequency.
Detectors PD1 and PD2 collect marginal reflected light from the reference
cavity that is spatially expanded with a grating.
PZT1
Ti:S laser
PZT2
Isolator 1st order beam
To main
AOM experiment
BS
BS
Reference Cavity PM
L2 L1
Grating
In from
L.O.
PD2 PD1
Servo1, Servo2,
νo frR
fm= 10.7MHz
To PM
0 .6
0.6
-
Voltage
Voltage
0 .0
0.0
-0.6
-0 .6
+
Frequency
Frequency
Figure 12-7. Schematic of experimental setup. The laser output, protected from feedback by a
Faraday isolator, is sent through an acousto-optic modulator that controls average frequency
before being sent through a phase modulator and mode matched to the reference cavity.
AOM: acousto-optic modulator; PM: phase modulator; B: beam splitter; PD1 and PD2:
photodiodes 1 and 2; and PZT1–3: piezo-electric transducers.
The error signals obtained in this way are the composite signals from all
longitudinal modes detected within the spectral regions seen at PD1 and
PD2. One detector can provide the error signal to lock the position of the
comb about some average frequency by adjusting AOM and PZT1 (i.e.,
controlling the cavity length). The difference between the error signals from
PD1 and PD2 determines fluctuations in the laser repetition rate that can be
stabilized by PZT2, which selects the average group velocity [18]. The
sensitivity of this discriminator is equal to the slope of the error signals
multiplied by the number of modes between them. This technique provides
an extremely sensitive discriminator for the repetition rate by effectively
locking a high harmonic of it to the cavity. The reference cavity sees the
348 Chapter 12
stabilized modes of frequency ν m = mf r + f 0 + fVCO , where fr is the laser

repetition frequency and f0 the offset frequency of the comb. The frequency
fVCO is provided by the voltage-controlled oscillator (VCO) used to drive the
AOM. Note that the sum f 0 + fVCO is stabilized in this technique.
120
Frequency- 385.2845017 THz (kHz)
100
80 drift rate = 1.48 kHz/second
60 standard deviation = 707 Hz
40
20
0
-20
-40
-60
-80
-100
06 06 06 06 06 06 06 06 06
:38 :38 :39 :39 :39 :40 :40 :40 :40
:38 :55 :12 :29 :46 :03 :20 :37 :54
Time Stamp
Figure 12-8. Measurement of optical frequencies of the femtosecond comb. One hundred
consecutive measurements are made showing the slow, but easily measurable, drift of the
reference cavity. Fractional Allan deviation from these counts is σ (1s) ~3 x 10-12.
5.2 Characterization of femtosecond comb stability
The stability of a femtosecond laser cavity can be confirmed by

measuring any single optical frequency of the comb. A frequency mode νm =
mfr + f0 is determined by measuring the repetition rate fr and the offset
frequency f0. Here the offset frequency of the comb was detected by beat
note measurement [22] of a frequency mode with its second harmonic in a
“ν–to–2ν interferometer," using microstructure fiber. The nonzero offset
frequency of the femtosecond-stabilized comb is because of the finite
dispersive phase shifts of the reference cavity mirrors. Figure 12-8 shows the
linear drift for mode number 3,212,199 of the femtosecond comb. The value
of this drift can be greatly improved through temperature control of the
vacuum chamber walls. (Drift rates well below 100 Hz/s are usually
obtainable even without temperature stabilization.) The beat note of the 5th
harmonic of the repetition rate (120 MHz) with the 6th harmonic of an
accurate 100 MHz frequency reference source provides the repetition rate
value. The 1-s Allan deviation for these counts is ~3 x 10-12. This number is
limited by the stability of the low-noise frequency reference (100 MHz).
6. DISPERSION MEASUREMENT APPLICATIONS
A reference cavity and a laser have mutual dependency: each one can be
a source and reference for the other. As mentioned before, the nonzero offset
frequency is due to the phase change upon reflection from the two cavity
mirrors. In the frequency domain, this phase shift is usually frequency
dependent, i.e., the reference cavity mode spacing changes with frequency.
In the time domain, the pulse’s frequency component goes through different
depths in the mirror coating; thus each frequency component “sees” a
different length of the reference cavity, known as an "external cavity."
The stabilized femtosecond laser can act as a frequency comb to
characterize the dispersion of the external cavity mirrors. This ruler can
measure the dispersion of gases and low-density materials inside the external
cavity over a broad bandwidth. Furthermore, the high peak-intensity build-
up in the cavity provides a tool to study nonlinear phase shifts. An
experimental setup is presented in the following section.
6.1 Cavity characterization
A tunable cw Ti:sapphire laser is added to the setup and locked to a

single mode of the reference cavity while femtosecond laser stabilization has
been achieved, as shown in Figure 12-9. The equally spaced femtosecond
comb provides an absolute frequency grid against which the unequally
spaced reference cavity modes can be measured. The beat frequency
between the cw laser and the mth mode of the femtosecond comb, ∆(m),
gives a direct measurement of the deviation of the reference cavity
frequencies from the equally spaced modes of the femtosecond comb.
Figure 12-10 shows the results of this measurement when the external
reference cavity was in a vacuum. Each data point for ∆(m) corresponds to
approximately 100 measured readings. The cavity’s free spectra range (FSR)
as a function of frequency, σ ( f ) , is calculated from the relation
σ ( f ) = σ 0 (1 + ∂∆ ∂f ) and is indicated by the dashed lines and right side
ordinate. Notice that the maximum deviation (∆ ~300 kHz) of the FSR is
less than the width of the cavity resonance (~1 MHz). This explains why it is
possible to align the entire bandwidth of the femtosecond laser to the modes
of the external cavity. The mirrors are standard high-reflecting dielectric
350 Chapter 12
mirrors. Higher quality mirrors with lower GVD increase the bandwidth of
the transmitted laser and improve locking stability of femtosecond lasers to
high-finesse reference cavities.
Mode-locked laser
fm= 10.7
MHz
CW laser
fm= 16.1 MHz
KLM laser
grating m-1 m m+1
CW laser
Frequency
BS
PBS
Error signals
Figure 12-9. Experimental setup to measure longitudinal modes of reference cavity. PBS:
polarizing beamsplitter and BS: non-polarizing beamsplitter.
200
Beat Frequency (kHz)
5
Cavity FSR - fr (Hz)
4
100
0
2
1
-100
0
360 365 370 375 380 385
Frequency (THz)
Figure 12-10. Measured deviation (∆) of the reference cavity longitudinal modes under
vacuum from the equidistant positions of the frequency comb (left ordinate; squares) and the
calculated mode spacing of the cavity vs frequency (right ordinate; dashed curve).
6.2 Atmospheric dispersion
The same measurement is performed when the external cavity is in

atmospheric pressure (Figure 12-11). The dispersion of air may result in a
beat note higher than the cavity linewidth (~1 MHz) in some spectral
regions, indicating regions of the femtosecond comb that are not coupled
into the cavity. From this data, one can calculate the round-trip GVD inside
the cavity. The inverse of the pulse round-trip time in the cavity is equal to
the FSR and will be frequency dependent in the presence of GVD. The pulse
round-trip time is
∂Φ n
τn = , (4)
∂Ω
where Φ n is the total round-trip phase shift inside the cavity. The change in
∂σ ∂ 1 1 ∂τ 1 ∂ 2Φ n
= =− 2 n =− 2 , (5)
∂Ω ∂Ω τ n τ n ∂Ω τ n ∂Ω 2
and therefore the round-trip GVD can be related to the measured FSR by
1 ∂σ
Φ n′ = − . (6)
σ 2 ∂Ω
The calculated round-trip GVD for the cavity at atmospheric pressure is

≈ 20 fs2. Assuming the calculated GVD is entirely due to the passage of the
pulse through the atmosphere in the 62.5-cm long reference cavity, Equation
(6) gives k ′′ ≈ 0.16 fs2/cm at 385 THz (780 nm). This value is in good
agreement with that calculated for dry air at atmospheric pressure based on
the measured index of refraction [23]. The accuracy of this measurement
technique and the long interaction length available inside a resonant cavity
make this system suitable for pursuing precise measurements of linear and
nonlinear dispersion in different media and characterizing phase shifts from
dielectric mirrors.
352 Chapter 12
9 900
8 800
Cavity FSR- fr (Hz)

7 700
Beat Frequency (MHz)
6 600
5 500
4 400
3 300
2 200
1 100
0 0
-1 -100
-2 -200
355 360 365 370 375 380 385 390 395 400 405
Frequency (THz)
Figure 12-11. Measured deviation of the reference cavity longitudinal modes at atmospheric
pressure from the equidistant positions of the frequency comb (left ordinate; data points) and
the calculated mode spacing of the cavity vs frequency (right ordinate; dashed curve).
7. RING LASER STABILIZATION
Applying the previous techniques to the stabilization of a bidirectional

ring laser is not straightforward. Since the sensor applications involve
intracavity measurements, the laser itself, rather than an extracavity beam, is
shifted in frequency by an AO modulator. A basic challenge is to find two
orthogonal cavity controls to act independently on the pulse repetition rate
and the carrier frequency. In the case of the Ti:sapphire ring laser, one can
control the carrier frequency through the pump intensity and the repetition
rate through the cavity length. These two controls are however not perfectly
independent, as opposed to the case of an OPO, where the repetition rate of
the OPO is controlled by the pump laser and the OPO signal carrier
frequency is determined by the ring perimeter. An example of intracavity-
pumped OPO geometry is sketched in Figure 12-12.
SERVO-LOOP FOR OPO

00 00 00 00 00 00 00
grating
CARRIER FREQUENCY
+
0000000
SERVO-LOOP FOR
REPETITION RATE
-
10.6 MHz
EO
BS
M2
M1
DBS
M3
C
M4
Figure 12-12. Sketch of a possible ring OPO geometry and the stabilization loop.
The beams of the pump laser and the OPO signal are combined by a
dichroic beam splitter (DBS). Both beams are given the same phase
modulation at 10.5 MHz by an electro-optic crystal. The reflection of the
reference cavity (1/2 the cavity length of the pump) is dispersed by a grating.
The difference between the two error signals taken at the edges of the pump
spectrum are used to control the cavity length (mirror M1) of the pump laser
and hence the repetition rate of the system. The wavelength of the OPO
signal is unaffected, since it is solely determined by the OPO cavity
perimeter. An error signal is thus derived from the dispersion of the grating
at the signal wavelength and mixed with the modulation. The correction
signal is sent simultaneously to the piezo drivers of mirrors M3 and M4,
located symmetrically with respect to the pulse crossing point C. It is
essential to symmetrically apply the correction for the signal cavity
simultaneously to the two circulating pulses to prevent a broadening of the
beat-note bandwidth by the motion of the cavity length-correcting mirrors.
As mentioned earlier, for applications that require long-term stability or
accuracy (monitoring or slow drifts in the picometer range), the OPO
frequency and the pulse repetition rate have to be locked to some atomic
transition. Because of the general tunability of an OPO, there is a large
choice of atomic systems available. One approach [14] is to use a Λ level
structure that has a two-photon resonance at the carrier frequency of one of
the OPO pulses and a ground-level splitting resonant with the pulse
repetition rate. The main challenge is to find an error signal (fluorescence,
dispersion, absorption) that has a well-defined peak at the proper optical
354 Chapter 12
frequency and repetition rate. This problem has been investigated

theoretically for the case of rubidium vapor in Reference [14].
REFERENCES
82, 3799-3802 (1999).
Ye, Science 293, 1286-1289 (2001).
[3] R. J. Jones and J. C. Diels, Phys. Rev. Lett. 86, 3288-3291 (2001).
[4] S. Diddams, B. Atherton, and J. C. Diels, Appl. Phys. B 63, 473-480
(1996).
[5] J. C. Diels, M. Bohn, J. Jones, and T. T. Dang, (United States Patent, Filed
April 27 2000., 2000), Vol. Patent Pending, (UNM Docket No. UNM-490).
[6] M. L. Dennis, J. C. M. Diels, and M. Lai, Opt. Lett. 16, 529-531 (1991).
[7] P. J. Hood and H. Ward, Airborne Geophysical Methods (Academic Press,
1969).
[8] D. J. Teskey, P. J. Hood, L. W. Morley, R. A. Gibb, P. Sawatzky, M.
Bower, and E. E. Ready, Can. J. Earth Sci. 30, 243-260 (1993).
[9] P. J. Hood, Northern Miner, 1-14 (1991).
[10] S. A. Diddams, J. C. Diels, and B. Atherton, Phys. Rev. A 58, 2252-2264
(1998).
[11] M. Bohn, Ph.D Thesis, University of New Mexico (1998).
[12] X. M. Meng, J. C. Diels, D. Kuehlke, R. Batchko, and R. Byer, Opt. Lett.
26, 265-267 (2001).
[13] M. J. Bohn, J. C. Diels, and R. K. Jain, Opt. Lett. 22, 642-644 (1997).
[14] L. Arissian, J. Jones, and J. C. Diels, J. Mod. Opt. 49, 2517-2533 (2002).
[15] X. M. Meng, Ph.D Thesis, University of New Mexico (2003).
[16] A. I. Ferguson and R. A. Taylor, Opt. Commun. 41, 271-276 (1982); E.
Kruger, Rev. Sci. Instrum. 66, 4806-4812 (1995).
3568-3571 (1999).
172, 59-68 (1999).
[19] R. J. Jones and J. Ye, Opt. Lett. 27, 1848-1850 (2002).
[20] R. J. Jones, J. C. Diels, J. Jasapara, and W. Rudolph, Opt. Commun. 175,
409-418 (2000).
[21] R. W. P. Drever, J. L. Hall, F. V. Kowalski, J. Hough, G. M. Ford, A. J.
Munley, and H. Ward, Appl. Phys. B 31, 97-105 (1983).
[23] Handbook of Chemistry and Physics (The Chemical Rubber Co.,
Cleveland, OH, 1985/1986).
Appendix
AUTHOR ADDRESSES
Arissian, Ladan: Department of Physics and Astronomy, 800 Yale Blvd. NE,
University of New Mexico, Albuquerque, NM 87131
Baltuška, Andrius: Institut für Photonik, Technische Universität Wien,

Gusshausstrasse 27, A-1040 Wien, Austria and Max-Planck-Institut für
Quantenoptik, 85748 Garching, Germany
Bartels, Albrecht: Time and Frequency Division, National Institute of

Standards and Technology, 325 Broadway M.S. 847, Boulder, CO 80305
Cundiff, Steven T.: JILA, National Institute of Standards and Technology

and University of Colorado, University of Colorado, Boulder, CO
80309-0440
Christov, Ivan: JILA and Department of Physics, University of Colorado

and NIST, and NSF Engineering Research Center for Extreme-
Ultraviolet Science and Technology, Boulder, CO 80309-0440
Diddams, Scott A.: Time and Frequency Division, National Institute of

Standards and Technology, 325 Broadway, Boulder, CO 80305
Diels, Jean-Claude: Department of Physics and Astronomy, 800 Yale

Blvd. NE, University of New Mexico, Albuquerque, NM 87131
Fischer, Marc: Max-Planck-Institut für Quantenoptik, 85748 Garching,

Germany
356 Appendix
Gaeta, Alexander L.: School of Applied and Engineering Physics, Cornell

University, Ithaca, NY 14853
Gibson Emily: JILA and Department of Physics, University of Colorado and

NIST, and NSF Engineering Research Center for Extreme-Ultraviolet
Science and Technology, Boulder, CO 80309-0440
Hall, John L.: JILA, National Institute of Standards and Technology and
University of Colorado, University of Colorado, Boulder, CO 80309-
0440
Hänsch, Theodor W.: Max-Planck-Institut für Quantenoptik, 85748

Garching, Germany
Haverkamp, Nils: Physikalisch-Technische Bundesanstalt, Bundesallee

100, D-38116 Braunschweig, Germany
Hollberg, Leo: Time and Frequency Division, National Institute of

Standards and Technology, 325 Broadway, Boulder, CO 80305
Holzwarth, Ronald: Max-Planck-Institut für Quantenoptik, 85748 Garching,

Germany
Ippen, Erich: Department of Electrical Engineering and Computer Science,

Research Laboratory of Electronics, Massachusetts Institute of
Technology, Cambridge, MA, 02139-4307
Jones, R. Jason: JILA, National Institute of Standards and Technology and

0440
Kapteyn Henry C.: JILA and Department of Physics, University of Colorado

and NIST, and NSF Engineering Research Center for Extreme-
Kärtner, Franz: Department of Electrical Engineering and Computer

Science, Research Laboratory of Electronics, Massachusetts Institute of
Technology, Cambridge, MA, 02139-4307
Keller, Ursula: Institut für Quantenelektronik, ETH Zürich Hönggerberg

CH-8093 Zürich, Switzerland
Author addresses 357
Kienberger, Reinhard: Institut für Photonik, Technische Universität Wien,

Kobayashi, Takayoshi: Department of Physics, Faculty of Science,

University of Tokyo, 7-3-1 Hongo, Bunkyo, Tokyo 113-0033, Japan
Kolachevsky, Nikolai: Max-Planck-Institut für Quantenoptik, 85748

Garching, Germany
Krausz, Ferenc: Institut für Photonik, Technische Universität Wien,

Lindner, Fabrizio: Max-Planck-Institut für Quantenoptik, 85748 Garching,

Germany
Lipphardt, Burghard: Physikalisch-Technische Bundesanstalt,

Bundesallee 100, D-38116 Braunschweig, Germany
Murnane, Margaret M.: JILA and Department of Physics, University of

Colorado and NIST, and NSF Engineering Research Center for Extreme-
Paulus, Gerhard G.: Max-Planck-Institut für Quantenoptik, 85748 Garching,

Germany and Department of Physics, Texas A&M University, College
Station, Texas 77843-4242, USA
Schnatz, Harald:Physikalisch-Technische Bundesanstalt, Bundesallee 100,

D-38116 Braunschweig, Germany
Steinmeyer, Günter: Max-Born-Institut für Nichtlineare Optik und

Kurzzeitspektroskopie, Max-Born-Str.2a, D-12489 Berlin, Germany
Stenger, Jörn:Physikalisch-Technische Bundesanstalt, Bundesallee 100, D-

38116 Braunschweig, Germany
Udem, Thomas: Max-Planck-Institut für Quantenoptik, 85748 Garching,

Germany
Weiss, Carl-Otto: Physikalisch-Technische Bundesanstalt, Bundesallee

100, D-38116 Braunschweig, Germany
358 Appendix
Windeler, Robert S.: OFS Laboratories, 600 Mountain Avenue, Murray Hill,
NJ 07974
Ye, Jun: JILA, National Institute of Standards and Technology and

0440
Zimmermann, Marcus: Max-Planck-Institut für Quantenoptik, 85748

Garching, Germany
INDEX
above threshold ionization, 296 dispersion-compensating mirrors, 44,

absolute optical frequency, 17, 24, 64, 82, 137, 160
204 extreme nonlinear optics, 48, 264,
attosecond, 268, 315 314
carrier-envelope offset frequency. femtosecond comb. See frequency
See offset frequency comb
carrier-envelope phase, 56, 265 femtosecond lasers. See mode-locked
coherence, 22, 38 lasers
definition, 14, 113 fine structure constant, 179, 192
dynamics, 74, 123 FRAC. See frequency resolved
evolution, 16, 113 autocorrelation
noise, 120, 278 frequency chains, 8, 24, 200
stabilization frequency comb, 204, 240
amplifiers, 271, 280 applications, 24, 338
oscillators, 66, 72, 126, 270 definition, 13, 177
CEO frequency. See offset frequency dynamics, 112
chirp, 159 electro-optic, 3, 19, 185, 204
coherence time, 23 generator, 19-22
combs. See frequency combs history, 1–10
cross-correlation, 22, 117 infrared, 41, 252
dispersion, 113 line spacing. See repetition
fiber, 103 frequency
management, 58 offset frequency, 16
solitons, 75 control, 22, 126, 244
plasma, 316 stability, 348
dispersion map, 59 stabilization, 112, 345
dispersion measurement, 349 verification, 25, 213, 249
360 Appendix
frequency division, 3, 202 mode-locked lasers, cont.

frequency multiplicaton. See phase locking, 40
frequency chains ring, 337
frequency ratio, 120, 216, 249 as sensors, 338
frequency-resolved autocorrelation, stabilization, 352
163 spectrum, 15, 115, 207
frequency transfer, 253 synchronization, 39
frequency-dependent spatial mode, ti:sapphire, 3, 15–21, 31, 38, 42,
67 57–64, 78, 88, 97, 120, 125,
fundamental constants, drift, 192, 134, 182, 183, 204, 239, 246
230 nonlinear Schrödinger equation, 57
global positioning system, 229 NOPA. See optical parametric
Gouy phase shift, 268, 302–306 amplification; noncollinear
group velocity matching, 146 octave spanning lasers, 63
high harmonic generation, 265, 314 prismless, 70
carrier-envelope phase with prisms, 64
dependence, 267 octave-spanning spectrum, 17, 118,
carrier-envelope phase sensitivity, 183, 241
289 offset frequency, 16, 114, 177
cutoff, 265 stabilization, 126, 244
phase matching, 320 OPA. See optical parameteric
quasi-phase-matching, 324 amplification
high repetition rate oscillator, 79, 240 optical atomic clocks, 27, 225–252
basic design, 80 basic components, 227
broadband, 88 frequency references, 237
Cr:forsterite, 86, 251 ion, 233
intra atomic phase-matching (IAPM), molecular, 26, 236
318 neutral atom, 234
Kerr-lens–mode-locked, 15, 54, 206 history, 230
KLM. See Kerr-lens–mode-locked uses, 229
microstructure fiber, 6, 20, 63, 97– optical cavity, 42–45
111, 184, 244 amplification, 44
dispersion, 103 reference, 346
fabrication, 98 optical coherence, 23
properties, 102 optical frequency combs. See
types, 100 frequency combs
mode locking, 14, 334 optical frequency metrology, 24, 179,
mode-locked lasers, 12–23 209
Cr:forsterite, 251 Ca, 210
Cr:LiSAF, 251 cesium D1, 180
diode, 254 H, 190
Er-doped fiber, 251 Hg+, 194
Index 361
optical frequency metrology, cont. self-amplitude modulation, 90

Yb+, 194, 212 self-focusing, 57
optical frequency synthesis, 29, 181– self-phase modulation, 20, 183, 205
185, 238 self-referencing, 18, 79, 117, 182,
noise, 242 208, 244
optical local oscillator, 206, 238 sensors, 338 –342
optical parametric amplification, 133- shock, 74
171 spectral broadening. See
carrier-envelope phase supercontinuum generation
stabilization, 171 spectral interferometry, 272
noncollinear, 135, 140 spectroscopy, 30
principles, 138 atomic, 30
optical parametric oscillators, 155, hydrogen, 184
343 rubidium, 31
phase matching hyperfine structure, 34
high harmonic generation 319– molecular, 33
329 supercontinuum generation, 20, 104,
parametric amplification, 136 151, 184, 205, 244
photodetectors, 241 noise, 107
photonic crystal fiber. See three-step model, 265, 315
microstructure fiber tilted pulse, 157
pulse front matching, 155 timing
pulse synthesis, 40 jitter, 38
QIC. See quantum interference synchronization, 39
control transmission, 253 –255
quantum coherence, 31, 46 ultrashort pulse, 14
quantum interference control, 46 propagation, 105, 113
recollision, 265, 316 white light generation. See
repetition frequency, 16, 115, 177, supercontinuum generation
207, 240
stabilization, 28, 39, 182, 246, 346

Frequency Comb

Uploaded by

Copyright:

Available Formats

Frequency Comb

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Frequency Comb

Uploaded by

Copyright:

Available Formats

Femtosecond Optical Frequency Comb:

Principle, Operation, and Applications

Jun Ye and Steven T. Cundiff

History of Optical Comb Development ..................................................1

4.5 Coherent control via quantum interference between

Femtosecond Laser Development .........................................................54

Gigahertz Femtosecond Lasers.............................................................78

Microstructure Fiber and White-Light Generation.............................97

5.3 Continuum instability and noise ........................................107

Optical Comb Dynamics and Stabilization ........................................112

Femtosecond Noncollinear Parametric Amplification and

10. Conclusions and Outlook..........................................................170

Optical Frequency Measurement .......................................................176

Optical Frequency Measurement Using Frequency

Femtosecond Lasers for Optical Clocks and Low Noise

3.1 Considerations in designing a femtosecond comb for

Generation And Measurement of Intense Phase-Controlled Few-

Quantum Control of High-Order Harmonic Generation ..................314

Applications of Ultrafast Lasers .........................................................333

3.2 Case (2): reciprocal effects that can be synchronized to

John L. Hall,1 and Theodor W. Hänsch,2

In the past five years, progress in laser stabilization, optical frequency

The historical development of pulsed and continuous wave (cw) lasers

landmark experiment to demonstrate Fourier’s reality to us: stable pulse

the time delays to be equalized. Miraculously, stable white interference

comb configuration that had a stabilized carrier-envelope-offset frequency.

team built up one of the best rf-to-optical harmonic frequency “chains,”

designed so they measure differences or temporal changes and hence are

[16] D. J. Jones, S. A. Diddams, M. S. Taubman, S. T. Cundiff, L. S. Ma, and J. L.

Jun Ye and Steven T. Cundiff

Key words: frequency comb, carrier-envelope phase, spectroscopy, optical frequency

Mode-locked lasers generate “ultrashort” optical pulses by establishing a

is a frequency-domain concept, mode-locked lasers and their applications are

new capabilities have brought novel approaches to precision atomic and

1. TIME- AND FREQUENCY-DOMAIN PICTURES

Understanding the connection between the time-domain and frequency-

1.1 Introduction to mode-locked lasers

Mode-locked lasers generate short optical pulses by establishing a fixed-

1.2 Frequency spectrum of a mode-locked laser

To understand how frequency-domain techniques can be used to control

descriptions [23]. To start, we ignore the carrier-envelope phase and assume

1.3 Determining absolute optical frequencies with

Armed with the understanding of the frequency spectrum of a mode-

νn=nfr - f0 x2 2νn=2nfr -2 f0 ν2n=2nfr - f0

Self-referencing is not the only means of determining the absolute optical

1.4 Femtosecond optical-frequency comb generator

A frequency comb generator produces a spectrum that consists of a series

known absolute frequencies. There are several possible implementations,

Figure 1-3. Schematic of a femtosecond comb generator. AOM: acousto-optic modulator;

The heart of the comb generator is a KLM Ti:sapphire laser. A small

the two arms must be matched in order to achieve temporal overlap,

1.5 Time- and frequency-domain characterizations of f0

Carrier-envelope phase coherence is critical for all of the time-domain

femtosecond laser has been properly phase stabilized, experimental results

2. PRECISION OPTICAL FREQUENCY

It is useful to review the historical development of optical frequency

2.1 Measurement of absolute optical frequency

The dramatic simplification of a complex optical frequency chain to a

of a femtosecond comb-based–frequency-measurement system at a level of a

There has been an explosion of absolute frequency measurements using