Você está na página 1de 63

INSTITUTE OF PHYSICS PUBLISHING REPORTS ON PROGRESS IN PHYSICS

Rep. Prog. Phys. 69 (2006) 443505 doi:10.1088/0034-4885/69/2/R04


Femtosecond x-ray science
T Pfeifer, C Spielmann and G Gerber
Physikalisches Institut, Universit at W urzburg, Am Hubland, 97074 W urzburg, Germany
Received 11 November 2005
Published 18 January 2006
Online at stacks.iop.org/RoPP/69/443
Abstract
We present the advances in x-ray femtosecond pulse generation and the most recent discoveries
in the eld of ultrashort (femtosecond) x-ray science. Nowadays x-rays show their potential
not only when it comes to resolving atomic spatial scale but also the inherent temporal scale
of quantum dynamics in atoms, molecules and solids. We discuss ultrafast x-ray sources that
are currently used to generate femtosecond duration pulses of soft and hard x-ray radiation.
Several techniques of x-ray pulse characterization are presented along with a method to control
the shape of coherent soft x-rays. A large number of experiments using femtosecond x-ray
pulses have been conducted recently and we review some of them. The eld of ultrafast x-ray
science draws its strength from the large variety of different sources of femtosecond duration
x-ray pulses that are complementary rather than competing.
(Some gures in this article are in colour only in the electronic version)
0034-4885/06/020443+63$90.00 2006 IOP Publishing Ltd Printed in the UK 443
444 T Pfeifer et al
Contents
Page
1. Introduction 445
2. Femtosecond x-ray sources 446
2.1. High-harmonic generation 448
2.1.1. Single particle response 449
2.1.2. Propagation effects 457
2.2. Laser-induced plasma 465
2.3. Ultrafast x-ray tube 469
2.4. Accelerator-based sources 469
2.4.1. Electron-bunch slicing 469
2.4.2. Electron-bunch compression 470
2.4.3. Free-electron laser 471
3. Pulse characterization and control 471
3.1. Direct measurement (streak cameras) 473
3.2. Correlation techniques 474
3.2.1. Autocorrelation 475
3.2.2. X-FROG measurements 475
3.2.3. Crosscorrelation 476
3.2.4. Ponderomotive streaking 478
3.2.5. Laser-induced absorption gating 478
3.3. Control of pulse parameters 479
4. Experiments 481
4.1. Gas-phase 483
4.1.1. Probing of molecular dynamics 483
4.1.2. Auger lifetime 485
4.1.3. Adaptive control with spectrally shaped soft x-rays 486
4.2. Clusters 487
4.3. Surfaces 489
4.3.1. Surface electron dynamics 489
4.3.2. Surface photochemistry 490
4.4. Solid state 491
4.4.1. Lattice dynamics 492
4.4.2. Nonthermal melting 495
4.4.3. Solid-state inner-shell electron dynamics 496
4.4.4. Electronlattice interaction dynamics 497
4.5. Dense plasmas 498
4.6. Nonlinear x-ray optics 498
5. Conclusion 499
Acknowledgments 500
References 501
Femtosecond x-ray science 445
1. Introduction
The past 20th century has witnessed tremendous progress of science in many different elds
and disciplines. Without question, the discovery of the x-rays [1] has been an important if not
the most important driving force to push forward our understanding and knowledge in all kinds
of scientic areas. Applications range from structure analysis in solid-state as well as atomic
physics and molecular chemistry via imaging applications in medicine and the life sciences to
the discovery of the basic building blocks of life, in particular the DNA and more generally the
structure of proteins and other macromolecules.

Uber eine neue Art von Strahlen (About a


newkind of rays) was the title that WilhelmConrad Roentgen gave to his seminal report about
the x-rays that he discovered in 1895 at the University of W urzurg. This discovery opened
up the door to a new era of science where it should be and indeed is possible to examine the
microscopic structural details of molecules, liquids and solids.
Today at the beginning of the 21st century, we are once again in the situation where a new
kind of rays begins to shed light on unexplored regions of science. Having access to the spatial
resolution of molecular/crystal structure and electron orbitals is only one side of the coin to
be illuminated by x-rays. It took about one century to ip the coin to the other side showing
the temporal resolution of the atomic and molecular motion, making it possible to monitor
the dynamics of molecules (such as rotation, vibration and dissociation) and the dynamics of
electrons on their natural time scale, which is in the femtosecond to picosecond range and even
much shorter for electron dynamics. Early x-ray generation devices and techniques such as
x-ray tubes, electrical discharges or even the rst synchrotron sources have not been able to
deliver x-ray pulses that had a duration of less than several nanoseconds and thus could not be
used to gather both types of information, spatial and temporal, simultaneously.
Interestingly enough, insight into the temporal dynamics of quantum systems was rst
gained by using probes of much larger wavelength than the one of x-rays: infrared (IR),
visible and ultraviolet (UV) laser pulses. In this respect, the invention of the laser in 1960
started a development in which the accessible time scale for monitoring dynamic processes
could be steadily decreased to shorter and shorter events. Today, the fundamental limit of
these ultrashort femtosecond laser pulses has almost been reached [2, 3], which is the single
optical cycle, lasting 1 fs in the near UV to several femtoseconds in the IR spectral region.
Having at hand the possibility of monitoring molecular and electronic motion, it is of course
not possible to use lasers to directly image molecular structure at atomic resolution due to the
large wavelength of the laser photons.
The solution to this problem seems easy but it took several decades to become feasible
in practice: combine x-rays and lasers to take advantage of both the short wavelength and the
temporal coherence properties to create ultrashort (femtosecond) pulses of x-ray radiation.
This can be done in many ways and different experimental approaches have been carried out
in the past.
This paper reviews the development in the eld of femtosecond x-ray science. It covers
both the development of sources/instrumentation and experimental applications of ultrafast
x-rays to monitor and control quantum dynamical processes. To monitor fast events in time,
two approaches can be taken: in one of them, the detector has to be fast enough to resolve the
details of the dynamical processes (just as the opening time of a photo camera has to be short
enough in order not to blur the picture). The other approach is to use a short pulse of radiation
to illuminate the dynamical process (working in the same way as a strobe ash for resolving
fast mechanical motion by eye or another slow detector). At the moment, the latter approach
(using ultrashort, femtosecond duration pulses) is the more common one. Usually, an intense
femtosecond laser (pump) pulse interacts with a sample in order to start a dynamical process for
446 T Pfeifer et al
S
y
n
c
h
r
o
t
r
o
n
l
i
g
h
t

s
o
u
r
c
e
s
L
a
s
e
r
-
i
n
d
u
c
e
d
p
l
a
s
m
a

g
e
n
e
r
a
t
i
o
n
Lasers
High-harmonic
generation
F
r
e
e
-
e
l
e
c
t
r
o
n
l
a
s
e
r
s

E
l
e
c
t
r
o
n

b
u
n
c
h

s
l
i
c
i
n
g
/
c
o
m
p
r
e
s
s
i
o
n
a
n
d

p
r
o
p
o
s
e
d

f
r
e
e
-
e
l
e
c
t
r
o
n

l
a
s
e
r
s
1 m
100 nm
10 nm
1 nm
1
1 ns 1 ps 1 fs
1 eV
10 0 eV
1000 eV
1 kkeV
10 kkeV
w
a
v
e
l
e
n
g
t
h
p
h
o
t
o
n

e
n
e
r
g
y
pulse duration
Figure 1. Temporal and wavelength/photon-energy ranges covered by ultrafast x-ray sources. The
different sources of pulsed x-rays cover different ranges of photon energy and pulse duration in
the x-ray spectrum. It is thus not possible to identify an optimal source of ultrafast x-rays; the
different techniques are rather complementary. The dashed line in the lower right of the graph
denotes the single-optical-cycle limit for electromagnetic radiation. High-harmonic generation
driven by few-cycle laser pulses is currently the most efcient method to produce x-ray pulses that
are closest to this limit.
instance by populating an excited state of the system, which is then analysed by the ultrashort
x-ray (probe) pulse, which arrives at a selectable time delay afterwards and delivers structural
information about the system. In some cases, for instance to study inner-shell dynamics [4, 5],
it is necessary to use an x-ray pump pulse.
This review is structured as follows: in section 2, the most important sources of
femtosecond-duration x-ray pulses will be presented. Section 3 will outline ways of measuring
and characterizing the temporal shape and duration of these short pulses. Experimental results
in the eld of ultrafast x-ray science will be discussed in section 4, before we conclude this
review in section 5.
2. Femtosecond x-ray sources
Presently, a variety of methods is available to produce femtosecond-duration ashes of x-ray
radiation. In particular, there are many different sources to cover the large extent of the
electromagnetic spectrum in the frequency region which we refer to as x-rays. In addition,
each of these sources has its own characteristics in terms of the minimum achievable pulse
duration or spatial and temporal coherence properties (gure 1). Virtually for any of these
ultrafast x-ray sources, femtosecond laser pulses have to be employed in some way. One
method is the direct conversion of the laser pulse into the x-ray region by a nonlinear optical
process called high-harmonic generation (section 2.1). In another method, the laser pulse
generates a hot and dense plasma emitting incoherent x-rays (laser-induced plasma x-rays,
section 2.2). A third way of ultrashort x-ray production is to use the laser pulse in order to
create an ultrashort electron pulse, which is accelerated to kiloelectronvolt energies by a dc
electric eld (ultrafast x-ray tube, section 2.3) or to gigaelectronvolt energies at large scale
electron accelerator facilities (synchrotrons, free-electron lasers, section 2.4) before x-rays are
emitted as bremsstrahlung radiation. Another common aspect of all ultrafast x-ray sources is
Femtosecond x-ray science 447
Photon Energy [eV]
1 10 100 1000 10000
A
v
e
r
a
g
e

B
r
i
l
l
i
a
n
c
e

p
h
o
t
o
n
s
/
(
s

m
m
2

m
r
a
d
2

0
.
1
%

B
W
)
10
6
10
9
10
12
10
15
10
18
10
21
10
24
Photon Energy [eV]
1 10 100 1000 10000
P
e
a
k

B
r
i
l
l
i
a
n
c
e

p
h
o
t
o
n
s
/
(
s
h
o
t

m
m
2

m
r
a
d
2

0
.
1
%

B
W
)
10
17
10
20
10
23
10
26
10
29
10
32
Ti:S laser
Undulator Bessy II
X-ray FEL
(proposed)
Laser plasma
x-ray
line source
Laser pumped
x-ray laser
High harmonic
soft x-rays
Discharge pumped
x-ray laser
Laser pl asma
x-ray
line source
Laser pumped
x-ray laser
X-ray FEL
(proposed)
Ti: S laser
High harmonic
sof t x-rays
Undulator Bessy II
(a) (b)
Figure 2. Comparison of (a) average and (b) peak brilliance of different light sources. As can
be seen, the most promising x-ray light source in terms of both average and peak brilliance are
the proposed x-ray free-electron lasers (FELs) that are currently under construction. The rescaling
from average to peak brilliance can be understood in terms of two parameters: pulse duration and
repetition rate. This is why the undulator source falls back behind the high-harmonic sources, since
the high average ux at synchrotron undulators typically comes with high repetition rate (MHz)
and long (ns) pulse duration as compared with high-harmonic sources where repetition rates are on
the order of kilohertz and pulse durations on the femtosecond to attosecond time-scale.
the fact that accelerated electrons are used in the conversion process. This acceleration step
is most evident in the cases of the ultrafast x-ray tube or the synchrotron/free-electron laser
based sources. For laser-induced plasma x-rays, a hot electron plasma is produced which
emits black-body radiation or characteristic lines (during hot-electron plasma interaction with
coreshell bound electronic states in the atoms) at x-ray photon energies. Laser-driven high-
harmonic generation uses acceleration of electrons on time-scales that are on the order of and
smaller than the optical cycle of the driving laser eld itself. Currently this latter technique
gives rise to the shortest ashes of (coherent soft x-ray) light which are on the order of a few
hundreds of attoseconds.
The number of photons produced in each pulse also differs for the various sources. Figure 2
shows the brightness of radiation that is generated by different x-ray light sources. In general,
the number of photons (intensity) of the generated x-ray light per pulse depends on three crucial
factors: (1) the density of accelerated electrons, (2) the number of bremsstrahlung emission
acts and (3) the mutually spatial and temporal coherence of the electrons that contribute to the
generation of x-ray light. The more the electrons and the emission acts, the more photons will
be produced. However, the intensity scales differently with electron density and the number
of emission acts for the coherent and the incoherent generation regime. If spatial coherence
prevails between electrons across the x-ray pulse beam (all electrons across the beam oscillate
in the same phase), the intensity depends on the density squared, since all electric dipoles
add up coherently (intensity is proportional to the square of the electric eld). If in addition
temporal coherence is present, each subsequent bremsstrahlung emission act contributes an
equal amount of eld strength that adds coherently to the already generated x-ray eld. This
leads to the x-ray intensity depending quadratically on the number of acceleration acts in
the temporally coherent case. For incoherent generation, the dependence is only linear since
averaging over the phase occurs. From a different point of view, to obtain high peak intensities
it is benecial to have high spatial and temporal coherence. Coherent radiation can be focused
on very small spot sizes and have short pulse durations, both of which give rise to high intensity.
448 T Pfeifer et al
Both the temporal and spatial coherence are largely absent in the cases of x-ray tubes
or laser-induced plasma x-ray generation. High-harmonic generation provides a maximum
amount of both the temporal and spatial coherence. The newly developed free-electron x-ray
lasers will also feature particular spatial and temporal coherence properties, whereas the typical
insertion devices in beamlines (wigglers, undulators) exhibit only limited spatial coherence.
In the following, we will discuss different sources of ultrafast x-ray radiation. We will
start out with femtosecond-laser driven high-harmonic generation, continue with laser-induced
plasma sources, before we discuss ultrafast x-ray tubes, the large-facility synchrotron sources
and the x-ray free-electron lasers. A recent review of pulsed x-ray sources can also be
found in [6].
2.1. High-harmonic generation
High-harmonic generation describes the process in which laser light at a given frequency
is converted to integer multiples of this fundamental frequency during the highly nonlinear
interaction with a conversion medium. The rst report of high-harmonic generation dates back
to the late 1980s [7, 8]. A plateau of equally intense harmonics of high order was observed,
which was not immediately understood. Applying perturbation theory one would expect a
rapid decrease in efciency with increasing harmonic order: in the framework of n-photon
excitation, the excitation probability decreases exponentially with n. However, since there
was a large number of equally strong harmonic peaks, another mechanism had to be found to
explain the results. In 1993 Corkum [9] and Kulander et al [10] published a quasiclassical
theory, whichreproducedthe plateaubehaviour of harmonic emissionfoundinthe experiments.
According to these works, the electron cannot be treated as a bound particle in the high electric
elds at work in the experiments, which had so far been assumed. In fact, the electron is
ionized (freed from the binding force of the nucleus) when the absolute electric eld of the
laser is close to its peak during an optical cycle and is driven away from the parent ion. Since
the laser electric eld changes its sign about a quarter of a period later, the electron will slow
down, stop at a position far from the ion and start to re-accelerate towards it. When it returns
to the ion, it can possess a signicant amount of kinetic energy, much larger than the photon
energy. This energy plus the ionization potential will be transferred into photon energy as
soon as the electron recombines with its parent ion, which gives rise to the very high harmonic
orders observed in the experiment. It is thus three steps which make up the model: ionization,
propagation in the laser eld and recombination. The model has therefore been named three-
step model or also simple-mans model due to its striking simplicity. As this three-step
processand therefore also high-energy photon emissionusually occurs every half-cycle of
the laser eld it is immediately clear that the spectrum of the produced radiation has to consist
of peaks at odd integer multiples of the laser frequency. Inspired by Kulanders and Corkums
idea, a fully quantum-mechanical treatment of the three-step model has also been found shortly
after its introduction [11, 12] which conrmed the validity of the classical approximation. It
is a very intriguing aspect of nonlinear physics that highly nonlinear effects (as the process
of high-harmonic generation indeed is) can be treated with high accuracy considering simple
models.
Typical high-harmonic spectra are shown in gure 3. As was mentioned, the harmonic
intensity decreases rapidly for low harmonic orders (the so-called perturbative region). For
higher orders, a plateau of equally intense harmonics is found, which extends up to the cut-
off harmonic order. In the following, the mechanism of high-harmonic generation will be
discussed in more detail (reviews can also be found in [14, 15]). First of all the single particle
response is considered in section 2.1.1, since it is this microscopic process outlined above
Femtosecond x-ray science 449
plateau cut-off
perturbative regime
40 80 120 160 200 240 280
0.0
0.2
0.4
0.6
0.8
1.0
s
p
e
c
t
r
a
l

i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
.
]
3 5 7 9 11 13
10 11 12 13 14
0.0
0.2
0.4
0.6
0.8
1.0
83 79 75 71 67 63 59 55
wavelength [nm] wavelength [nm]
(b) (a)
harmonic order harmonic order
Figure 3. Typical spectra produced in high harmonic generation. (a) A full harmonic spectrum
generated in xenon gas shows the characteristic spectral shape (from [13]). For low orders,
the harmonic intensity is rapidly decreasing. A plateau is visible for the higher orders which
is terminated at the cut-off, the highest harmonics that are generated. The spectral position of
the plateau depends on both, the gas species used and the intensity of the driving light. Another
measurement in neon reveals very high harmonic orders (b). Lower orders are less intense due to
the zirconium lter used to separate fundamental and harmonic light.
which generates the harmonic frequencies in the rst place. High-harmonic generation is not
a process involving only one single atom but many of them which are coherently stimulated
by the laser. For that reason, the importance of propagation effects, including phase-matching
and distortions, is addressed in section 2.1.2. High-harmonic generation can be employed to
produce the shortest man-made electromagnetic pulses. The duration of these pulses is on the
order of hundreds of attoseconds, far shorter than the optical cycle of the driving laser light.
2.1.1. Single particle response. As mentioned, the process of high-harmonic generation
can be broken up into three steps: ionization, propagation and recombination. Each of these
fundamental processes will now be discussed separately.
Ionizationstep I. If the intensity of light interacting with matter is steadily increased,
the electric eld of the electromagnetic wave, E(t ) = E
0
cos(t ), at some point becomes
comparable to the intratomic eld strength (i.e. the average eld strength sensed by the
electron). The electron can escape the binding potential of the atom by tunnelling. The
electric eld of the laser will produce a potential, e

E(t ) r, in addition to the Coulomb potential


of the ion:
V( r, t ) =
e
2
4
0
r
+ e

E(t ) r. (1)
As can be seen in gure 4(b)) the Coulomb potential is signicantly distorted for high enough
laser eld strengths. Abarrier is created for the bound electronic state, which can be overcome
by tunnelling, upon which the electron can be considered free, affected only by the electric
eld of the laser (note the proximity of the asymptote to the real potential at the appearance
position of the electron in gure 4(b)). This process of tunnel ionization has been treated in
detail in the literature. The rst notion by Keldysh [16] dates back to 1965. He calculated an
analytical formula for the ionization rate, w, of the hydrogen atom exposed to a strong electric
450 T Pfeifer et al
(a) (b) (c)
Figure 4. Different ionization scenarios. If the ionization potential is low compared with the
frequency of the light and large compared with the electric eld of the laser (all in atomic units)
the absorption of multiple photons is the dominant ionization mechanism (multiphoton ionization
(a)). If the electric eld strength is increased, the Coulomb potential is signicantly modied. If
the frequency of light is low enough such that the electron can respond to this changing potential,
ionization can be understood as the tunnelling of the electron through a static potential wall (tunnel
ionization (b)). If the electric eld is yet higher, the barrier is completely suppressed and the
electron will be classically ripped off the ion (barrier-suppressed ionization (c)).
eld, E, in the quasi-static limit:
w
K
=

6I
p
4 h
_
eE h
m
1/2
I
3/2
p
_
1/2
exp
_

2mI
3/2
p
e hE
_
1
m
2
I
p
5e
2
E
2
_
_
, (2)
where e and m are the unit charge and the electron mass, respectively, and I
p
denotes the
ionization potential.
Much later, in 1986, Ammosov, Delone and Krainov published a generalized analytical
theory [17] extending Keldyshs approach to arbitrary atoms and initial electronic states. Their
calculated ionization rate, which agreed very well with experimental ndings, is now known
as the ADK ionization rate. It reads (atomic units, h = m = e = 1, are used here)
w
ADK
=
_
3E
(2I
p
)
3/2
[C
n

l
[
2
f (l, m)I
p
_
2(2I
p
)
3/2
E
_
((2Z/

2I
p
)[m[1)
exp
_

2(2I
p
)
3/2
3E
_
,
(3)
with the (time dependent) electric eld strength, E, the ionization potential, I
p
, the ion
charge, Z (once the electron is detached) and l and m represent the angular momentum and
magnetic quantum number, respectively. Further, the factor f (l, m) is given by
f (l, m) =
(2l + 1)(l + [m[)!
2
[m[
([m[)!(l [m[)!
,
and the constant C
n

l
is on the order of 2, but more precisely
[C
n

l
[
2
=
2
2n

(n

+ l

+ 1)(n

)
.
The effective principal and angular momentumquantumnumber are given by n

= Z(2I
p
)
1/2
and l

= n

1, respectively.
The work of Keldysh contained another important nding. He determined for which
range of laser eld strengths, E
0
, and angular frequency, together dening the so-called
ponderomotive potential, U
p
in combination with a particular ionization potential, I
p
, the
tunnelling description is valid. He introduced a parameter
=
_
I
p
2U
p
, U
p
= e
2
E
2
0
/(4m
e

2
), (4)
Femtosecond x-ray science 451
which is nowcalled the Keldysh parameter. The unit charge is denoted by e, m
e
is the electron
mass and E
0
is the amplitude of the laser electric eld oscillating at frequency, . By the use
of this quantity we are able to determine whether the atom is ionized in the tunnel ( _1) or
the multiphoton regime ( 1). To understand this in a qualitative way we can imagine the
laser-distorted Coulomb potential to oscillate with the laser frequency. For higher and higher
frequency (larger and larger ), the quasi-static approximation breaks down and the electron
does not have enough time to accommodate to the fast changes in the potential. Its motion will
be governed by an average over many cycles of the laser eld rather than tunnelling in a single
cycle, since the tunnelling timedepending on E
0
and I
p
is larger than the optical period.
The steady nonlinear interaction with the laser eld (absorption of many photons) will nally
lead to an electronic state with an energy larger than zero, thus a free electron (gure 4(a)).
In the opposite limiting case, the eld of the laser can get high enough to fully suppress the
barrier (gure 4(c)). The electron is then classically ripped off the atom. This ionization
scenario is called the barrier-suppression ionization [18]. It will now be discussed why the
tunnel-ionization regime is best suited to high-harmonic generation.
PropagationStep II. After the electron is ionized, it can approximately be regarded as a
free electron whose motion is just governed by the laser eld. To understand the motion of
the electron, we rst regard an initially free classical electron interacting with a laser eld
E(t ) = E
0
cos(t ) and calculate its velocity
v(t ) =
_
t
0

e
m
E(t
/
) dt
/
+ v
0
=
E
0
e
m
sin(t ) + v
0
, (5)
where vector arrows are omitted due to the one-dimensional motion of the electron. If we
consider only electrons possessing a zero drift velocity, v
0
= 0, their average kinetic energy,

E
kin
, denes the ponderomotive potential, U
p
, mentioned above (equation (4)). Note that U
p
is proportional to E
2
0
, which makes the average kinetic energy of the electrons grow linearly
with laser intensity. It can also be written as
U
p
[eV] = 0.97 10
13
I[Wcm
2
]
2
[m
2
]. (6)
Let us now consider an electron which is initially bound to an atom at x = 0 to appear in
the continuum (i.e. to be ionized) at time zero with initial velocity, v
0
= 0. It is ionized at an
arbitrary phase of the electric eld E(t ) = E
0
cos(t + ). Velocity v(t ) and position x(t )
can then be calculated to be
v(t ) =
_
t
0

e
m
E(t
/
) dt
/
=
E
0
e
m
(sin(t + ) sin()), (7)
x(t ) =
_
t
0
v(t
/
) dt
/
=
E
0
e
m
2
(cos(t + ) cos()) + sin()t. (8)
The time-independent term in the velocity can be understood as drift velocity. If this term is
zero, the electron will oscillate around a xed reference position. If it is nonzero, the reference
position will be moving in time. The maximum kinetic energy of an electron can also be
extracted from equation (7) to be 8U
p
. The amplitude, E
0
e/m
2
= a
0
, of the oscillatory
position is sometimes referred to as the ponderomotive radius, a
0
. Typical values (at laser
intensities 10
14
Wcm
2
) are on the order of some nanometres, thus much larger than the
atomic radius, demonstrating the validity to treat the electron as a freely moving particle in
the laser eld. Note that the ponderomotive radius is limited to a quarter of the wavelength
for electrons moving close to the speed of light. These relativistically moving electrons, of
course, need to be treated differently from above and do not play a role in the high-harmonic
452 T Pfeifer et al
0 5 10 15 20
0
10
20
30
40
50
time [fs]
=0
=20
=50
=100
0.0 0.5 1.0 1.5 2.0
-2
-1
0
1
2
3
4
5
6
time [fs]
=0
=20
=50
=100
e
l
e
c
t
r
o
n
-
a
t
o
m

d
i
s
t
a
n
c
e
[
p
o
n
d
e
r
o
m
o
t
i
v
e

r
a
d
i
i

a
0
]
(a) (b)
Figure 5. Electron trajectories in a laser electric eld. Plotted is the distance to the atom versus
time for a unidirectional motion in a linearly polarized laser eld. (a) is an enlarged view of (b)
for early times. Electrons that ionize at different phases of the ac-eld travel different paths. Only
some of them reencounter the atom, leading to possible recombination. The absolute value of the
slope of the electron trajectory at the zero crossings is the velocity and thus a measure for the kinetic
energy of the electron. Its highest value is reached at an ionizing phase of the laser close to 20

.
generation process considered here. The laser intensities (at 800 nm wavelength) needed
to accelerate electrons to relativistic intensities are beyond 10
17
Wcm
2
, whereas the laser
intensities typically used for high-harmonic generation are below 10
16
Wcm
2
. If we plot the
electron trajectory versus time, we observe different paths for electrons that become ionized at
different phases of the ac electric eld (gure 5). In particular, only some electron trajectories
return to the origin at x = 0, where the electron can recombine with the parent ion. In addition,
only the rst two encounters of the electron with the ion lead to signicant photon emission.
The reason for this is the quantum-mechanical nature of the electron, which suffers from
dispersion (spreading of the wavefunction, thus the overlap with the atom becomes smaller)
as soon as it is not bound to a potential. This will be treated in more detail below.
RecombinationStep III. When an electron recombines with an ion, typically a photon is
emitted carrying away the excess energy of the electron. In particular, the electrons accelerated
in the laser eld emit a high harmonic photon with energy
h = E
kin
+ I
p
, (9)
where E
kin
is the kinetic energy of the electron acquired in the laser eld and I
p
is the ionization
potential. As can be seen in gure 5, the slope of the trajectory at the intersection with
x = 0 assumes different values, corresponding to different velocities and thus different kinetic
energies. In particular, there exists a maximum kinetic energy that the electron can have
when it re-encounters its parent ion. This maximum kinetic energy therefore corresponds to
the maximum photon energy which can be generated in the process. By solving the equation
x(t ) = 0 numerically and calculating the kinetic energy for various phases, , we can calculate
this maximum kinetic energy to be 3.17 U
p
(gure 6). It occurs at a phase of 18

. Electrons
that are ionized at this moment of time produce harmonic photons of the highest energy. We
expect the high-harmonic soft x-ray spectrum to vanish (to be cut off) at photon energies
higher than
h
c
= 3.17U
p
+ I
p
, (10)
which is indeed the case in experiments. This formula is commonly called the cut-off lawfor
high-harmonic generation. It is another example for the remarkable agreement of the three-step
Femtosecond x-ray science 453
0 20 40 60 80
0
1
2
3
0.0 0.1 0.2 0.3 0.4 0.5
k
i
n
e
t
i
c

e
n
e
r
g
y

[
U
p
]
cos[t+] electric field phase []
cos[t+] electric field phase [ rad]
18
3.17 U
p
Figure 6. Kinetic energy of the electrons at the moment of re-encounter for various phases, , of
the driving laser eld. A maximum exists at a phase of 18

.
model [9] with experimental data. Before the idea of the semi-classical three-step model was
published, Kulander and co-workers empirically introduced this cut-off law formula [19].
It can now also be understood why the tunnelling regime supports the generation of high-
harmonic radiation. The highest photon energies are produced when electrons ionize at a
phase of 18

, which is close to the peak of the electric eld in the optical wave. In the
tunnel-ionization regime, the ionization rate given by equations (2) and (3) increases with the
electric eld. Therefore, many ionized electrons will contribute to high-harmonic generation.
On the other hand, in the multiphoton-ionization regime electrons are continuously produced,
depending only on the intensity of the eld and not on its phase. It is also clear considering the
cut-off formula, equation (10), that in order to generate the highest harmonic orders we need
to provide a large ponderomotive potential, U
p
. This is also fullled in the tunnelling regime
of ionization, where the Keldysh parameter dened in equation (4) has to be much smaller
than unity. It is clear from the dependence of the high-harmonic process on the phase of the
laser eld that harmonic radiation is intrinsically coherent to the fundamental eld.
In gure 7 the three-step model is summarized. What has not been discussed so far is the
fact that harmonic peaks exist at odd integer multiples of the fundamental frequency, f . We
can understand this considering the temporal structure of the high-harmonic emission. The
three-step process repeats every half-cycle T/2 of the laser eld. The Fourier transform will
thus be discrete, with a separation corresponding to 1/(T/2) = 2f , which is what we observe
in the spectrum. If the conversion medium exhibits a broken inversion symmetry, harmonic
emission will not occur in the same way at every half-cycle but at every full cycle of the laser
eld. In the harmonic spectrum this gives rise to a harmonic peak spacing of one times the
laser frequency; thus odd and even harmonic orders are produced. Another way to break
inversion symmetry is to drive a plasma at very high intensities, where the magnetic eld of
the laser becomes high enough to break the inversion symmetry. This occurs in the process of
nonlinear Thompson scattering [20]. In this case the laser intensities are on the order of more
than 10
18
Wcm
2
and the free-electron motion is not harmonic anymore as described by the
classical propagation equation (8). It is strongly inuenced by the relativistic mass increase
of the electron and the high magnetic elds. This leads to the well-known nonlinear gure of
eight-motion of the electron, which again radiates harmonics at even and odd frequencies. Yet
another possibility to generate even harmonics is the use of few-optical-cycle duration laser
pulses. In this regime, where the carrier-envelope (absolute) phase of the laser pulse becomes
important, the situation for the atom is also different for successive negative and positive
half-cycles, and even harmonic orders and coherent continuum radiation can be generated.
454 T Pfeifer et al
tunnel ionization
acceleration in
the laser field
E
c
~ I
p
+ 3.17U
p
h
recombination
and photoemission
E
t
1
2
3
x
x
x
step 1 step 2
step 3
Figure 7. Summary of the three-step model. The electron is ionized in step 1 at some particular
phase of the electric eld. It is then driven away from the parent ion in the laser eld (step 2). After
sign reversal of the ac-eld, the electron stops far from the atom, possibly returns and recombines
to emit a photon carrying the kinetic energy of the electron plus its ionization potential (step 3).
The kinetic energy of the returning electron can be as high as 3.17U
p
, dening the so-called cut-off
photon energy in the harmonic spectrum.
It should be stated here that high-harmonic generation represents only one class of strong-
eld nonlinear processes. Other important processes are nonsequential double ionization and
above-threshold ionization (ATI). Any of these processes can be understood in the framework
of the three-step model that was introduced above.
In nonsequential double ionization [21,22], a characteristic intensity dependence of doubly
ionized atomic ions was reported that followed the intensity dependence of single ionization.
A knee was obtained in the intensity dependence exactly at the point where single ionization
saturated, indicating that both processes are coupled. One possible explanation for the coupling
can again be given in terms of the three-step model [9]. The rst electron is tunnel ionized
and accelerated in the laser eld. When it returns to the ion, the high kinetic energy can be
sufcient to knock out a second electron.
In ATI [23, 24] high-energy photo electrons are detected at integer multiples of the photon
energy. Again, the photo-electron spectra show a characteristic plateau structure, reminiscent
of the plateau of photon energies obtained in high-harmonic generation. The high kinetic
energy and the peaked structure of the electrons can again be explained by the three-step
model [9]: the tunnel-ionized electron that is accelerated to high kinetic energies by the laser
eld can scatter off its parent-ion. This scattering event changes the motion of the electron
in such a way that it can reach very high energies after further acceleration by the laser
eld.
The common aspect in all these effects is the interaction of a tunnel-ionized free electron,
accelerated in the laser eld, with its ion core. Only the particular kind of interaction with
the core differentiates these processes. Elastic or inelastic scattering of the electron with the
ion core leads to above-threshold ionization or nonsequential double ionization, respectively.
Recombination of the electron with the core produces high-harmonic radiation.
Strong-eld process Electron-core interaction
High-harmonic generation Recombination
Nonsequential double ionization Inelastic scattering
Above-threshold ionization Elastic scattering
Femtosecond x-ray science 455
The electron always has to revisit the core for any of these processes to occur. Therefore
the term rescattering scenario is nowadays frequently used to describe these strong-eld
effects.
Quantum-mechanical description. We now turn towards the quantum-mechanical
formulation of the three-step model, which has been found by Lewenstein et al [11, 12].
The source of additional frequencies to be generated besides the fundamental incoming laser
frequency is the nonlinear dipole oscillation of the medium. We thus have to calculate the
dipole response for the special case of high-harmonic generation. The Schr odinger equation
in this case is given by (in atomic units)
i

t
[( r, t )) =
_

1
2

2
+ V( r) + r

E cos(t )
_
[( r, t )) . (11)
We need to calculate the time-dependent dipole moment,
(t ) = ( r, t )[ r[( r, t )), (12)
from which we can then extract the harmonic spectrum by Fourier transformation. The
calculation presented in [11] expands the time-dependent electron wavefunction (where only
one electron is considered to be responsible for harmonic generation, which is often referred
to as single-active electron approximation (SAE) in the literature [10, 25]) in terms of the
bound electron ground state of the atom and the continuum states [ v) where v stands for the
kinetic momentum
[( r, t )) = e
iI
p
t
_
a(t ) [0) +
_
b( v, t )[ v) d
3
v
_
, (13)
with time-dependent ground-state, a(t ), and continuum-state amplitudes, b( v, t ). By doing
so, we do not take into account excited bound states of the electron. Since the continuumstates
are dened to be solutions to the free electron Schr odinger equation,
E
kin
[ v) =
1
2

2
[ v), (14)
the potential of the nucleus is neglected as soon as the electron is ionized. Using the denition of
the time-dependent electron wavefunction in equation (13), the time-dependent dipole moment
can be calculated to be
(t ) = i
_
t
0
dt
/
_
d
3
pE cos(t
/
) (

d( p

A(t
/
))
. ,, .
a
exp(iS( p, t, t
/
))
. ,, .
b

( p

A(t
/
)))
. ,, .
c
+c.c.
(15)
where

A(t ) is the vector potential of the laser eld. The canonical momentum, p, is given by
p = v +

A(t ). (16)
S denotes the so-called quasi-classical action and is written as
S( p, t, t
/
) =
_
t
t
/
dt
//
_
( p

A(t
//
))
2
2
+ I
p
_
. (17)
It contains the phase advance of the electron during the time it spends in the continuum.
The atomic potential only enters as a constant ionization potential, I
p
. The expression

d( v)
in equation (15) stands for the transition probability from the bound electronic state [0) to a
456 T Pfeifer et al
continuumstate v[ (describing ionization), where the complex conjugate describes the inverse
process, i.e. recombination of the free electron to the ground state,

d( v) = v[ r[0), (18)

( v) = 0[ r[ v). (19)
This helps us to nd a very intuitive interpretation of the formula of the time-dependent dipole
moment (t ) in equation (15). We can now straightforwardly identify the different parts of
the formula:
a: ionization of the ground state at time t
/
,
b: propagation in the continuum in the time interval t t
/
,
c: recombination to the ground state at time t .
Thus, the classical three-step model discussed before is contained in the quantum-
mechanical description as well. It also resembles Feynmans path integral description of
quantum-mechanical processes [26]. To calculate (t ), we do not need to calculate the integral
over all p. We only need to consider the p for which the action becomes stationary,

p
S( p, t, t
/
) = 0 (20)
This relation can again be interpreted in terms of the classical three-step model. Since

p
S( p, t, t
/
) = x(t ) x(t
/
), (21)
the stationary phase condition equation (20) yields the information that we only need to account
for those electron trajectories that return at time t to the same point they left at time t
/
where they
were ionized. Most importantly, the quantum-mechanical treatment also yields the classical
cut-off lawequation (10) up to a small correction. By Fourier transforming the time-dependent
dipole moment, the harmonic spectra can be calculated and analysed; the cut-off photon energy
can now be found to be
h
c
= 3.17U
p
+ f
_
I
p
U
p
_
I
p
, (22)
where f (x) is a slowly varying function on the order of 1, which assumes the values
f (0) 1.32 at x = 0 (U
p
I
p
) and f (3) 1.25. The physical origin of this correction
lies in purely quantum-mechanical effects such as tunnelling and the spreading of the electron
wavepacket in the continuum that have not been included in the purely classical treatment.
These effects enable the electron to collect more energy on its trajectory than the amount
predicted by the classical equations of motion.
According to the formula for the dipole moment equation (15), different electron
trajectories contained in the integral acquire different phases,
at
= S(p, t, t
/
) (called the
atomic dipole phases), during their propagation in the continuum. The shape of the electronic
wavepacket at the moment of recombination will be governed by the interference between
these separate quantum paths. In particular, different trajectories leading to the same photon
energy (having the same kinetic energy at the time of recombination) will interfere with each
other. Re-examination of the classically calculated kinetic energy of the electron (which agrees
very well with the quantum-mechanical result) at the moment of return to the nucleus reveals
that each kinetic energy in the plateau region of an electron can be produced by two distinct
particular phases of the electric eld at the moment of ionization (see gure 8). Therefore,
there are two electron trajectories which are most important for the generation of a particular
photon energy. Since one of them spends a longer time in the continuum, we call it the long
trajectory (the one ionizing at a smaller phase of the electric eld), the other one is called the
short trajectory. These trajectories will interfere constructively or destructively, depending
Femtosecond x-ray science 457
atom
e
-
e
-
-400 -200 0 200 400 600
-1.0
-0.5
0.0
0.5
1.0
e
l
e
c
t
r
i
c

f
i
e
l
d
phase (deg)
e
-
e
-
0 20 40 60 80 100
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
E
k
i
n

[
U
p
]
phase [deg]
long
trajectory
short
trajectory
short
trajectory
long
trajectory
(a) (b)
Figure 8. Atomic dipole phase. For the generation of a particular photon energy in high-harmonic
generation, two electron trajectories (a) contribute that ionize at different phases of the electric eld
(b). They spend different times in the continuum, giving rise to different phases of their nonlinear
atomic dipole contribution. One of these phases is linearly proportional to the intensity of the laser
eld, while the intensity dependence of the other one can be neglected. Interference between these
trajectories is one way to control the high-harmonic generation process.
on their relative dipole phase. Since the electric eld enters the dipole phase, it can be used to
modify this interference. Hence, one possibility to control the harmonic generation process is
to manipulate the atomic dipole phase with a shaped laser eld [27, 28].
It can be shown by careful analysis that only one of the two trajectories mentioned (the
long trajectory) exhibits a phase that is linearly proportional to the laser intensity, I, while the
phase of the other (short) one is almost constant [29, 30]:
d
at,l
dI
= a,
d
at,s
dI
0, (23)
where
at,l
and
at,s
denote the atomic dipole phases of the long and the short trajectory,
respectively, and a is some constant on the order of 26 10
14
cm
2
W
1
.
The effect of this intensity-dependent nonlinear dipole phase becomes evident in
modications of the spectral line shape of high-harmonics [26]. In addition, the spatial
properties of harmonic generation in the near-eld region [31] and the beam prole [32]
are altered as a consequence of the intensity dependence.
This dependence is also responsible for the linear chirp (quadratic phase) imposed on
the high harmonics. Since the laser pulse intensity at the peak of the laser pulse can be
approximated to vary quadratically with respect to time, the same temporal dependence will
be imparted on the temporal high-harmonic phase [33, 34].
In conclusion, high-harmonic generation is an example of how the coherence properties
of the laser can be transferred to the electronic wavefunction. Another recent demonstration of
coherence transfer from laser light to electrons concentrated on the direct measurement of the
momentum wavefunction of free electrons [35]. It was shown that the free electrons created
by phase-coherent double pulse excitation exhibit a clear signature of wavepacket interference
in the continuum.
2.1.2. Propagation effects. We will now turn towards discussing phase-matching issues in
the special case of high-harmonic generation. The equation for the wavevector mismatch for
production of the mth harmonic order can then be written as
k = mk(
f
) k(m
f
), (24)
458 T Pfeifer et al
with
f
denoting the fundamental laser frequency. Since we will only consider the collinear
harmonic generation geometry, we wrote the formula in the scalar form. The wavevector
dependence on is governed by a variety of physical effects. First of all, since the generation
medium used for high-harmonic generation is usually a gas which exhibits dispersion, the
refractive index depends on . This modies the wavevector by the amount k
disp
. Second,
very high intensities are employed to drive the harmonic-generation process. This inevitably
leads to the generation of free electronsthe majority of electrons do not recombine in the
third step of the three-step model. The generated plasma gives rise to a collective plasma
resonance that modies the refractive index and creates another contribution, k
plasma
, to the
wavevector. In addition to these contributions, the focusing and propagation of the laser beam
itself give rise to a geometrical modication, k
geom
, of the wavevector. In mathematical
terms, we can write
k() = k
vac
() + k
disp
() + k
plasma
() + k
geom
(), (25)
where k
vac
= 2/c is the wavevector in free space. If we consider the wavevector mismatch,
which is the quantity of interest for the harmonic intensity we may write
k = k
disp
+ k
plasma
+ k
geom
, (26)
with any of the k given by mk(
f
) k(m
f
). The vacuum contribution k
vac
vanishes
due to
k
vac
= m

f
c

m
f
c
= 0, (27)
thus phase matching is always accomplished in vacuum.
Let us now discuss each of the other contributions in more detail.
Neutral dispersion. Any conversion medium used for high-harmonic generation exhibits
dispersion, which is a refractive index n() that depends on the frequency (or wavelength )
of the light. Since the wavevector in turn depends on the refractive index
k
disp
() = (n() 1)

c
, (28)
the wavevector mismatch,
k
disp
() = mk
disp
(
f
) k
disp
(m
f
) = (n(
f
) n(m
f
))
m
f
c
, (29)
thus depends solely on the difference of refractive indices n() at =
f
and = m
f
.
Dispersion is caused by resonances of the medium. Therefore, the refractive index depends on
the spectral position of absorption lines or bands in the medium. If we are not too close to a
resonance, the wavelength dependence of the refractive index can in general be approximated
by the Sellmeier equations with coefcients that are specic for a particular medium. For a
certain medium, a particular set of constants is valid only over a limited spectral range. The
applicability of the Sellmeier equation is usually limited to sub-UV photon energies and can
thus only be used for the fundamental laser pulse. In general, the refractive index is larger
than 1 in the visible or the near-infrared (where the fundamental laser light is located) and
smaller than 1 in the XUV, where the harmonics are generated. Therefore, we obtain
k
disp
> 0. (30)
There are however ways to manipulate and control the refractive index of the medium by
mixing of different gases or the application of controlling light elds [36], which is a common
technique closely related to electromagnetically induced transparency (EIT) [37].
Femtosecond x-ray science 459
Plasma dispersion. High-harmonics are generated when the laser electric eld is high enough
to generate free electrons that are accelerated in the eld and recombine. As we discussed in
section 2.1.1, only a very small part of electrons really recombine or interact with the core
to emit a harmonic photon. The vast majority of electrons miss the core and become free
for a long time compared with the duration of the laser pulse (several ns fs). The plasma
resonance frequency is given by

p
=
_
e
2
N
e

0
m
e
, (31)
where e is the unit charge, N
e
the free-electron density,
0
is the dielectric constant and m
e
is
the electron mass. This resonance leads to a polarizability of the plasma that in turn causes a
refractive index of the form
n
plasma
() =
_
1
_

_
2
=
_
1
_
N
e
N
c
()
_
, (32)
where
N
c
=

0
m
e

2
e
2
(33)
is the critical plasma density. It is the free-electron density at which the plasma becomes
completely absorbing for electromagnetic radiation of frequency . Typically, in high-
harmonic-generation experiments the plasma densities generated are very small compared
with the critical density (N
c
= 1.75 10
21
cm
3
) of the fundamental 800 nm laser radiation
(and even more so for the high-harmonics), we can linearly approximate the refractive index,
which gives
n
plasma
() . 1
1
2
_

_
2
. (34)
The plasma contribution to the wavevector then yields
k
plasma
() = (n
plasma
() 1)

c
=

2
p
2c
. (35)
For the wavevector mismatch that means
k
plasma
= mk
plasma
(
f
) k
plasma
(m
f
) =

2
p
(1 m
2
)
2mc
f
. (36)
We are left with a negative contribution of the plasma to the phase mismatch,
k
plasma
< 0. (37)
Geometric dispersion. Now let us discuss the geometrical contribution to the wavevector.
This term is only present if the electromagnetic light wave is conned to a small region in
space. For a plane wave, this contribution would be zero. Since we need to generate high
light intensity in order to drive the process of high-harmonic generation, we always have to
take this term into account. There are basically two geometries that play a role in practice: the
region around the waist of a focused laser beam and propagation in a waveguide. The initially
collimated gaussian laser beammeaning the radial intensity distribution is gaussian, which
is also known as TEM
00
modeis sent through a lens, which creates a converging beam. The
radius, w, of this beam is continuously decreasing down to the waist size:
w
0
=
f
w
, (38)
460 T Pfeifer et al
detector
gas-filled capillary
laser pulse
gas jet
Figure 9. High-harmonic generation can be conducted in the free focusing and the waveguide
geometry. Aspecic geometric phase is imparted onto the femtosecond laser pulse as it propagates
through a focus or through a capillary.
where f is the focal length of the lens and w denotes the radius of the beam at the lens. This
formula is derived from the more general formula for the propagation of a gaussian beam [38]:
w(z) = w
0
_
1 +
_
z
z
R
_
2
, (39)
where w(z) is the radius of the beam at some point z along the propagation direction (optical
axis). The Rayleigh length is given by
z
R
=
w
2
0

. (40)
A focusing beam with beam size w(z) transports the same integrated electromagnetic power
everywhere along the optical axis. Thus its intensity increases with 1/w(z)
2
. However, due
to equation (39), the focal spot size, w
0
, is only sustained over approximately one Rayleigh
length, after which the beam size increases approximately linearly with z.
To overcome this problem and sustain high intensity over an extended propagation length,
hollow waveguides (capillaries) can be used for high-harmonic generation (gure 9). When
a laser pulse is focused into a capillary having about the same radius, a, as w
0
(optimally
w
0
= 0.64a), the beam radius is bound to be constant over the length of the capillary, due
to (partial) reection of the light at the boundaries. The capillary geometry also affects the
wavevector, due to the particular boundary conditions at the capillary walls. The smaller
the diameter of the waveguide compared with the wavelength of the guided light, the larger is
the modication of the wavevector.
Free space propagation. In analogy to the temporal phase, (t ), of a short laser pulse, we
can dene the spatial phase of a laser beam ( r):
E( r) e
i

k r
+ c.c. = e
i( r)
+ c.c. (41)
Also analogous to the denition of the instantaneous frequency (t ), we can dene the local
wavevector:

k( r) =

( r). (42)
Femtosecond x-ray science 461
By solving the source-free (vacuum) wave equation for a focusing (converging) beam, we
obtain an additional spatial phase,
geom
=
foc
, along the optical axis (z-direction) that is
equal to

foc
(z) = arctan
_
z
w
2
0
_
= arctan
_
2z
b
_
, (43)
with the confocal parameter, b = 2z
R
. This is also called the Guoy phase shift [38]. It means
that the beamacquires a phase of (as compared with a plane wave) during its passage through
the focus. For a small region around the focus (z _b), we may write, using equation (42),
k
foc
(z) =
d
foc
(z)
dz
.
2
b
. (44)
Since the harmonics are generated with the same confocal parameter, b, of the driving beam,
this leads to k
foc
(
f
) = k
foc
(m
f
) and thus
k
foc
= qk
foc
(
f
) k
foc
(m
f
) =
2(m1)
b
. (45)
For this contribution, we can therefore write
k
foc
> 0. (46)
It should be noted that k
foc
can be varied by placing the focus at different positions on the
optical axis relative to the conversion medium.
Guidedbeampropagation. If harmonics are generatedina guidedgeometry[3942], the laser
light is focused on the capillary and propagates inside. The solution to Maxwells equations
taking into account the boundary conditions of the hollow waveguide results in a system of
waveguide modes that is capable of transmitting electromagnetic radiation. These modes are
called the EH
nl
modes [43]. Each of these modes exhibits a particular dispersion relation of
k
geom
() = k
cap
():
k
cap
() =
u
2
nl
c
2a
2

, (47)
with the capillary inner radius a and u
nl
being the lth zero of the Bessel function J
n1
(u
nl
) = 0.
This is valid as long as the capillary is much larger than the wavelength. The wavevector
mismatch contribution of the waveguide now becomes
k
cap
= mk
cap
(
f
) k
cap
(m
f
) =
u
2
nl
c(1 m
2
)
2ma
2

f
. (48)
Obviously, the capillary wavevector contribution has the same form as the one resulting from
the plasma. Even the sign
k
cap
< 0 (49)
is the same.
Synthesis. To generate high-harmonic radiation at the maximumpossible efciency, we have
to ensure phase-matching conditions, taking into account all given contributions to the phase
mismatch. Since neutral dispersion and plasma dispersion are always present, resulting in
k
disp
and k
plasma
, we have to consider only two cases, i.e. the two different geometries
with their contributions, k
geom
: focusing on a jet or cell in a free space geometry (k
foc
) or
propagating in a hollow capillary lled with conversion medium as a waveguide (k
cap
). For
both cases, the focusing and guiding geometry, we will now summarize the relevant issues.
462 T Pfeifer et al
The total phase mismatch for the focusing geometry can be written as
k = k
disp
. ,, .
+ k
plasma
. ,, .
+ k
foc
. ,, .
>0 <0 > 0
=
, .. ,
(n(
f
) n(m
f
))

f
c
+
, .. ,

2
p
(1 m
2
)
2qc
f
+
, .. ,
2(m1)
b
.
(50)
Since we needtoachieve k = 0 for a most efcient conversionof laser light into mthharmonic
radiation, the positive contribution of neutral dispersion and the focusing geometry have to be
balanced by the plasma dispersion. The above expression contains only the formula for k
foc
at the position of the waist. However this term in general depends on z, the distance from the
focus on the optical axis, as can be seen by looking at equations (43) and (42). In addition
to this geometric contribution, we need to keep in mind that the intensity also changes along
the propagation direction z as the beam is focused. Since the atomic dipole phase depends
on intensity with equation (23) for one particular trajectory, this also affects the wavevector
of the harmonic radiation [31, 44]. This behaviour leads to different spatial distributions of
harmonic intensity in the generated beam depending on whether the gas jet is placed before or
behind the focus [26, 31]. The beam shape itself can be used in order to optimize the harmonic
production process in the free focusing case [45].
The density of the gas can also be varied in order to arrive at perfect phase matching.
This can be done either by increasing the backing pressure of the gas jet or by increasing the
pressure in the gas cell.
The plasma density depends on the gas species (ionization potential) and can be controlled
by means of the intensity or the duration of the laser pulse. Note, however, that a change
in intensity will naturally cause a change in the maximum harmonic order (cut-off law
equations (10) and (22)) that can be produced.
For the case of a hollowwaveguide lled with the conversion medium, the total wavevector
mismatch reads
k = k
disp
. ,, .
+ k
plasma
. ,, .
+ k
cap
. ,, .
>0 >0 >0
=
, .. ,
(n(
f
) n(m
f
))

f
c
+
, .. ,

2
p
(1 m
2
)
2mc
f
+
, .. ,
u
2
nl
c(1 m
2
)
2ma
2

f
(51)
Now, the geometric contribution, k
cap
, has the opposite sign as in the focusing geometry.
This means that only neutral gas dispersion has to balance both the plasma and the waveguide
contribution. Free parameters in this case are plasma density (controlled by the intensity and
duration of the laser pulse) and the density of the neutral medium. In practice, we are often
bound to work at a given laser pulse duration and intensity to produce harmonics up to a certain
harmonic photon energy. Thus varying the neutral density is the easiest way to experimentally
accomplish phase matching. In gure 10 the harmonic spectrum is plotted versus pressure in
the capillary. Obviously, the experimental nding matches the simulation based on the above
formula quite well. In particular, note the shift of the optimum pressure to higher values with
increasing harmonic order m, which is visible in both experiment and simulation results. This
is due to the inherent dependence of the wavevector mismatch equation (51) on the harmonic
order m. The refractive indices of the gas in the soft x-ray region used for the simulation were
taken from the literature [46].
The high-harmonic capillary output has also been shown to have extraordinary spatial
coherence properties [47]. This is because the bre acts as a spatial lter for the driving laser
Femtosecond x-ray science 463
20 40 60 80 100 120 140
50
45
40
35
pressure [mbar]
w
a
v
e
l
e
n
g
t
h

[
n
m
]
20 40 60 80 100 120 140
42
40
38
36
34
32
30
28
pressure [mbar]
1.0
0
0.5
(a) (b)
H21
H19
H17
H21
H23
H25
H27
s
p
e
c
t
r
a
l

i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
.
]
Figure 10. Phase-matched generation of high-order harmonics in a capillary. The gas pressure
(krypton gas was used here) can be employed as a means to control the wavevector mismatch,
k, in the high-harmonic generation process. At a particular pressure, perfect phase matching
is realized and the harmonic conversion efciency is maximized. The simulation (b) agrees with
the experimental nding (a). In particular, the shift of the optimum pressure, which is due to the
wavelength dependence of k, is reproduced in the experiment. Since refractive index values were
not available for simulation in the experimental wavelength region, the simulation was carried out
for higher harmonic orders.
pulse. If high-harmonics are generated at phase-matching conditions (correct gas pressure),
the good spatial properties of the driving laser are transferred to the harmonic beam. In the
unguided free-space focusing case, a good spatial prole of the harmonic radiation can also
be obtained, which is however strongly dependent on the position of the focus relative to the
conversion medium and the fundamental pulses spatial quality.
For both geometries, the bre and focus geometry, the plasma contribution represents the
ultimate limit for phase-matching for multi-cycle pulses. This will be shown in the experiments
on high-harmonic generation from water droplets (section 2.2). For the case of few-cycle
pulses, nonadiabatic self-phase matching [48, 49] allows to circumvent the plasma limit and
enable generation of high harmonics of 1 keV photon energy [50, 51]. As can be seen from
equation (36), k
plasma
becomes very large for increasing harmonic order m. It is therefore
crucial to always keep the plasma density at low values. On the other hand, we need to
apply high intensities to generate very high harmonic orders. This apparent contradiction
of requirements can be resolved by using very short driving laser pulses [5257]. Plasma
generation is governed by the rate equation (3), stating that for a given electric eld strength a
given number of ions is created per unit time. Therefore, the shorter pulse can reach a higher
peak intensity (compared with a longer pulse) for the same amount of plasma being created.
This is particularly important for the extreme case, when the long pulse has fully ionized the
medium by the time it reaches its peak intensity. This intensity cannot be used for harmonic
generation since the process relies on the availability of nonionized (bound) electrons. The
shorter pulse on the other hand reaches the same intensity in a shorter time such that the medium
does not have enough time to become fully ionized (assuming ionization with a rate depending
predominantly on instantaneous electric eld or intensity). The remaining bound electronic
population in the short pulse case can thus generate very high harmonics corresponding to the
high peak intensity of the pulse (gure 11).
For the sake of completeness, it should be mentioned at this point that the guiding of
laser pulses can also be achieved by the self-focusing effect [58]. If the laser is suitably
focused, the intensity-dependent index of refraction of the gas and the free-electron plasma
can counterbalance each other. In this case, the laser beam propagates over long distances at
464 T Pfeifer et al
-60 -40 -20 0 20 40 60
0.0
0.2
0.4
0.6
0.8
1.0
-0.1
0.0
0.1
i
o
n
i
z
a
t
i
o
n

p
r
o
b
a
b
i
l
i
t
y
time [fs]
-60 -40 -20 0 20 40 60
0.0
0.2
0.4
0.6
0.8
1.0
-0.1
0.0
0.1
e
l
e
c
t
r
i
c

f
i
e
l
d

[
a
.
u
.
]
time [fs]
(a) (b)
Figure 11. Plasma build-up in a long 25 fs FWHM(a) and a short 8 fs FWHM(b) laser pulse of the
same peak electric eld. The probability of ionization reaches 1 for the long pulse before the peak
electric eld is reached. Therefore, no harmonic photons corresponding to the highest intensity can
be produced in this case. For the shorter pulse, the electron has about 20% probability to survive
in the atom when the peak electric eld is reached. This is the reason why very short pulses are
benecial for the production of highest harmonic photon energies.
a conned beam diameter without an externally applied guiding structure. Thus, an effective
wave-guide scenario is created, and harmonic generation can take place over a long interaction
length [59, 60].
Absorption. A limiting factor to phase-matched high-harmonic generation is reabsorption of
the generated harmonic light in the conversion medium. The high-harmonic photon energies
are typically in the soft x-ray spectral region (for 800 nm driving laser pulses), where light is
extremely well absorbed by matter. This is because the outermost electrons in many materials
are bound with energies corresponding to 10100 eV. This results in large photoionization
cross-sections, , of any system in this spectral range, explaining the strong absorption. The
absorption length, L
a
(the distance after which the intensity of light propagating in an absorbing
medium drops to 1/e), is given by
L
a
= , (52)
where denotes the particle density. As the driving laser pulse propagates through
the conversion medium, it continuously generates new harmonic light that adds to the
copropagating harmonic radiation generated earlier. This earlier-generated light however is
affected by absorption. By summing over all contributions to the mth-harmonic radiation at
the point where the laser pulse leaves the conversion medium, we nd for the mth-harmonic
yield [61]
I
m

_
L
0
A
m
(z) exp
_

L z
2L
a
_
exp(i
m
(z)) dz

2
, (53)
with the harmonic amplitude, A
m
(z), of the single-particle response and phase
m
(z), at the
exit of the conversion medium of length, L. If we consider A
m
(z) does not depend on the
position z along the optical axisas is approximately the case in the waveguide or loose
focusing geometry where the laser intensity stays roughly constantequation (53) becomes
I
m

2
A
2
m
4L
2
a
1 + 4
2
(L
2
a
/L
2
c
)
_
1 + exp
_

L
L
a
_
2 cos
_
L
L
c
_
exp
_

L
2L
a
__
, (54)
where againL
c
= /k is the coherence lengththat canbe calculatedfromthe total wavevector
mismatch, k, introduced above. It can be seen that unlike perfect phase matching without
Femtosecond x-ray science 465
0 2 4 6 8 10 12 14
0.0
0.2
0.4
0.6
0.8
1.0
1.2
h
i
g
h
-
h
a
r
m
o
n
i
c

y
i
e
l
d
medium length [in units of L
a
]
no absorption
L
c
>> L
a
L
c
= 10 L
a
L
c
= 5 L
a
L
c
= L
a
Figure 12. The effect of reabsorption in the conversion medium during high-harmonic generation.
If absorption is not included in the consideration, the harmonic yield increases quadratically with
the medium length (- - - -). If absorption is present (absorption length L
a
) there is a limit ( )
to the maximum harmonic yield even in the case of perfect phase matching, L
c
L. For smaller
coherence lengths, L
c
, the maximum achievable harmonic yield decreases.
absorption (where the converted signal increases with L
2
) equation (54) converges to a nite
value for the case of L . In gure 12, equation (54) is evaluated for different ratios of
L
c
/L
a
. In order to generate about half the asymptotic high-harmonic yield obtained for a long
coherence and propagation length (indicated as dotted line in gure 12), we have to fulll the
conditions [61]:
L > 3L
a
, (55)
L
c
> 5L
a
. (56)
In this case, we generate close to the maximum amount of harmonic radiation that is possible
given the absorption of the gas. This absorption limit has been attained in a number of
experiments [6163].
At the end of this section, we would like to note that it has been possible to use high-
harmonic radiation as a seed for an x-ray laser [64]. In the future, this approach could possibly
lead to the generation of very high-intensity soft-x-ray pulses by making use of the excellent
spatial and coherence properties of x-rays produced by high-harmonic generation of laser light.
2.2. Laser-induced plasma
When a highly intense femtosecond laser pulse interacts with high-density material (solid-
state), a hot and dense plasma is created. Electrons are accelerated to very high kinetic
energies. The energy transfer mechanismfromlaser light intensity to the electrons is mediated
by collisions of the electrons with the ions they left behind and other electrons in the plasma.
This mechanism is also known as inverse bremsstrahlung absorption, since electrons in this
case do not emit but absorb photons when colliding with other particles. These high kinetic
energies acquired by the electrons can then be released as bremsstrahlung emission in the x-ray
when the electrons scatter off the ionic cores. The time it takes the very hot plasma to cool down
(by expanding) is typically in the sub-picosecond regime. Therefore, the emitted x-ray pulses
have femtosecond time durations. The rst laser-induced plasma femtosecond x-ray source
466 T Pfeifer et al
Figure 13. Spectrum of a laser-induced plasma source. The surrounding thin lines indicate the
95%condence interval of the measurement. Laser pulses were focused (intensity 10
17
Wcm
2
)
onto a copper target at a 2 kHz repetition rate. The spectral resolution was 170 eV. (Reprinted gure
with permission from [67]. Copyright (2003) by the Optical Society of America.)
was demonstrated by Murnane et al [65] in 1991. The emission spectrumoriginally resembles
the Planckian distribution for black-body radiation at a certain high temperature given by the
hot electrons. The production of very hard x-ray line emission on a femtosecond time-scale
is also possible: when the laser intensity is high enough to produce electron energies in the
relativistic regime these can in turn create inner-shell vacancies in atoms by inelastic electron
electron scattering. These short-lived core-holes then lead to the emission of characteristic line
radiation [66] when recombination takes place. The duration of these monochromatic x-ray
pulses in this case is dependent on the maximumdepth the energetic electrons can penetrate into
the material. These laser-induced plasma sources are presently unique in that they can deliver
very high photon energies in very short (a few hundreds of femtoseconds) pulses, perfectly
synchronized to a high-intensity femtosecond laser pulse, which is available as a pump pulse.
The source volume of the x-ray radiation is typically very small (not so much in the case of
plasma heating, more for the relativistic plasma regime), enhancing the brightness of the source.
Adrawback is the very lownumber of photons that can be used (on target) for experiments. This
is due to the fact that the generated plasma radiation, as opposed to for instance laser-produced
high-harmonic radiation (section 2.1), is spatially incoherent and has almost no directionality
(emission into 4 solid angle) (gure 13). X-ray optics have to be employed to collect a large
solid angle which is focused onto the sample object. High-repetition rate (multi-kilohertz) laser
systems delivering high pulse energies (multi-millijoules) are benecial [67] to produce high
average x-ray ux in order to keep integration times at acceptable levels. For more information
on laser-induced plasma x-rays please refer to [68, 69].
In addition to solid-state materials, liquid-phase targets can also be employed for ultrafast
x-raygenerationbylaser inducedplasma emission. If droplets of liquidnoble-gases or similarly
volatile substances are used, these sources overcome the problemof debris, which occurs when
laser-ablated solid state material escapes fromthe interaction region and covers optical surfaces
Femtosecond x-ray science 467
in the beam-path of the laser (such as the focusing lens or vacuum chamber window). Along
these lines, droplets have been shown to provide an excellent conversion medium [70, 71].
Another advantage is the automatic replacement of the material with each laser shot.
Microdroplets can even be used to monitor the transition between two different regimes
of laser-driven ultrafast x-ray generation, namely high harmonic generation (section 2.1) and
the just discussed laser-induced plasma emission. A microdroplet serves as a high-density
conversion medium for nonlinear lasermatter interaction, but it can also be forced to expand
and thus to decrease its density to arbitrary levels. This way, it is possible to study the density of
the conversion medium as an optimization parameter for efcient conversion of laser light into
soft coherent or hard incoherent x-rays. An intense enough prepulse can be used to drive the
microdroplet into explosion, allowing the second laser pulse to interact with a target medium
of variable density between the condensed and gaseous phases.
In our experiments [72], we used a water droplet jet of droplet size 20 m. Since the
repetition rate of the droplet jet was 1 MHz compared with the repetition rate of our laser at
1 kHz, each laser pulse encountered a fresh droplet. The laser we used was a regeneratively
amplied Ti : sapphire laser system, delivering 80 fs, 800 nm central wavelength, 0.8 mJ laser
pulses at an intensity of 5 10
14
Wcm
2
at the interaction region after focusing with a 20 cm
focal length lens. When a single intense laser pulse is used for excitation, emission of bright
frequency converted light is observed from the droplet under a variety of observation angles.
When this light is analysed by an XUV spectrometer, we can clearly identify plasma emission
lines of highly charged oxygen states, ranging up to O
5+
. However, no high-harmonic lines
at integer multiples of the fundamental (1.5 eV) laser photon energy could be measured. The
frequency conversion mechanism in this case is thus the laser-induced plasma generation
process.
If the intensity of the laser pulse is split into a double pulse of 2 10
14
Wcm
2
each, and
the time delay between the pulses is chosen larger than 1 ns, no plasma line emission occurs
and high-harmonic generation takes place with harmonic lines at odd integer multiples of the
laser frequency. The highest harmonic order to be observed was the 27th, which is the highest
one ever recorded to date in an aqueous conversion medium to the best of our knowledge.
Since the temporal delay seemed to play a crucial role in the process, we set up an optical
delay line to create a variable delay between our pump pulse (exploding the droplet) and the
driver pulse (generation harmonics froman expanded water droplet). This pumpdrive scheme
can be considered a general proposal to enhance the high-harmonic conversion efciency by
suitably preparing the target medium in a state that most efciently produces high-harmonic
radiation [73]. In our case, variation of the pumpdrive delay resulted in high harmonics
actually being produced at all, since for an unprepared droplet irradiated by a single pulse
harmonic generation could not be observed. This shows the potential of sample preparation
by a pumpdrive approach to possibly increase the conversion efciency by many orders of
magnitude.
The results from our experiment (gure 14) can be interpreted in the following way: the
rst laser pulse produces a conned overcritical plasma inside the droplet (due to a small region
of high intensity caused by additional focusing at the curved droplet surface) that creates a
high-pressure region within the droplet. The laser pulse is thus completely absorbed by the
overcritical plasma and no harmonics are produced since the laser pulse experiences severe
distortion in the plasma region and high-density water prevents produced harmonics from
escaping the droplet. The droplet expands immediately after the interaction with the rst
pulse. Most of the expanding droplet, however, will be composed of neutral water clusters and
molecules instead of ions (which were only efciently created in the conned plasma region).
For small time delays, the high density of the neutral expanding target material will be enough
468 T Pfeifer et al
17
19
21
23
25
h
a
r
m
o
n
i
c

o
r
d
e
r
driver pulse delay [ps]
w
a
v
e
l
e
n
g
t
h

[
n
m
]
0 0.5 >1.0
intensity [arb. units]
Figure 14. Transient XUV-emission spectrumacquired in the pumpdrive experiment. Atransition
occurs at 650 ps froma regime where only plasma luminescence is detected (in second diffraction
order) into a different one with high-harmonic generation being the dominant contribution to the
spectrum.
to create overcritial plasma density again upon interaction with the second (driver) laser pulse.
As a result, for short time delays the same as for a single laser pulse happens, where any
laser energy only feeds into plasma build-up and no harmonics can be observed. Another
possible explanation for the absence of high-harmonic generation has also been worked out in
theory [74]. For large enough time-delays, the density of the droplet is low enough to permit
propagation of the driver pulse without creating the critical plasma density. Spatio-temporal
distortion of the pulse is now very low, and the density of the water cluster is sufciently low
to allow for substantial transmission in the harmonic spectral region.
Apart from demonstrating high-harmonic generation in water droplets and clusters, this
study points out a way towards efcient high-harmonic generation by the application of pump
drive schemes. Using this approach, it has been possible to determine and to explore the nal
limit of high-density high-harmonic generation: plasma breakdown. Although a higher and
higher target pressure in the case of perfect phase-matching generally leads to a rapid increase
of conversion efciency (scaling quadratically with density), we have to nd a means of
overcoming the plasma threshold for high-efciency high-harmonic generation in the future.
Recently, another way of producing ultrafast x-ray radiation fromlaser-generated plasmas
was pioneered. Rousse et al [75] used a laser pulse focused to relativistic intensities
(3 10
18
Wcm
2
) into a He gas jet. The intense laser pulse generates a wakeeld by its
ponderomotive force, the resulting plasma wave of which can break and accelerate electrons
(in the propagation direction of the laser pulse) to high kinetic energies. These energies are
typically on the order of mega-electron-volts and can attain hundreds of mega-electron-volts
even for relatively small millimetre-size interaction lengths compared with several metres
of length needed for conventional particle accelerators. These laser-accelerated electrons
can undergo betatron oscillations caused by transverse elds in the plasma and emit highly
directional hard x-ray synchrotron radiation up to several kilo-electron-volts. This scheme is
Femtosecond x-ray science 469
thus similar to the production of x-ray radiation in large-scale synchrotron facilities by the use
of wiggler insertion devices to deect the electron beam. Since the x-ray radiation in this case
is generated by a highly directional electron beam, its divergence is very small (100 mrad [75])
and directed along the original laser path.
2.3. Ultrafast x-ray tube
The very rst technique to generate x-rays, as employed by Roentgen himself, is to use
x-ray tubes. An evacuated tube contains a cathode and an anode at an electrostatic potential
difference, U, on the order of a fewkilovolts. The cathode is heated by an additional voltage in
order to emit electrons according to the Richardson equation. These electrons are accelerated
from cathode to anode, where they will lose all their energy in collisions with the anode
materials atoms. Bremsstrahlung radiation is emitted, which has a typical high-energy cutoff
at Ue (e is the unit charge). Since electrons are continuously emitted from the cathode, the
x-ray radiation produced is also continuous.
In order to create ultrashort ashes of x-rays with this setup, another mechanism for
electron release was used in the group of Rentzepis [7678]. Instead of thermally ejecting
electrons from the cathode, a femtosecond laser pulse can be employed [79, 80] to create
photoelectrons just as long as the laser pulse lasts. This electron bunch produces an ultrashort
x-ray burst by the time it interacts with the anode. Again, as in the case with laser-driven
plasma x-rays (section 2.2), the duration of the x-ray pulse is governed by the penetration
depth of the electrons inside the anode material. Care has also to be taken in order not
to release too many electrons from the cathode since the Coulomb repulsion also increases
the duration of the electron pulse on its way to the anode. This sets an upper limit on the
photon number achievable with this technique, when sub-picosecond x-ray pulses are to be
generated. It was experimentally shown that increasing the acceleration voltage between
cathode and anode decreases the pulse duration and could eventually enable femtosecond
x-ray pulse generation [81]. A number of experiments on structural dynamics of solids and
liquids with nanosecond and picosecond resolution have already been carried out using this
approach [82, 83].
2.4. Accelerator-based sources
In the following, electron-beamx-ray sources will be discussed, which also feature good spatial
coherence. However, the pulse length of standard synchrotron pulses is on the order of 100 ps.
With the ongoing progress in the eld of high-energy free-electron lasers, pulse durations of
10100 fs are expected in the near future. An advantage of these sources is the high photon
energy that is achievable, which can be up to more than 10 keV. This enables applications such
as time-resolved structure determination in crystals. X-ray free-electron lasers are expected to
deliver enough x-ray photons in one shot to resolve the structure even of single molecules.
2.4.1. Electron-bunch slicing. For many scientic applications, in particular the monitoring
of atomic and electronic dynamics, the typical pulse duration of synchrotron sources poses a
fundamental limit. To overcome these limitations in the future, interesting new techniques to
shorten the synchrotron x-ray pulses were developed and implemented experimentally. The
so-called slicing scheme was experimentally realized by Schoenlein et al [84]: before the
electron pulse passes through the undulator or bend magnet to produce x-ray radiation, it
interacts with an intense femtosecond laser pulse in another undulator section (see gure 15).
This leads to two femtosecond duration electron subpulses of lower and higher energy electrons
470 T Pfeifer et al
Figure 15. Schematic setup of femtosecond electron-bunch slicing. An intense femtosecond laser
pulse interacts with an electron bunch in a wiggler device. The laserelectron interaction generates
a femtosecond slice of electrons with slightly higher and lower kinetic energies than the main bunch
energy. These femtosecond electron pulse slices (higher and lower energy) can be separated from
the main bunch by using bend magnets. The photons produced by each of these pulses in another
bend magnet or wiggler/undulator will then be separated in angle as well, allowing to single out
a femtosecond x-ray pulse. (Reprinted gure with permission from [84]. Copyright (2000) by
AAAS.)
in the main electron bunch, which can be extracted. This femtosecond subpulse of electrons
will generate a femtosecond pulse of x-rays along its passage through the bend magnet or
undulator. These ultrashort x-ray pulses were already used for experimental applications.
Direct evidence of ultrafast disordering in laser-perturbed InSb on a sub-picosecond time scale
could be observed for the rst time by time-resolved x-ray diffraction [85].
Another way to shorten the duration of synchrotron light pulses is to use ultrafast shutters.
It was shown that laser-induced acoustic pulses in crystals can modulate and switch the x-ray
transmission properties by making use of the Borrmann effect. This technique could potentially
be employed to generate subpicosecond pulses [86].
Common to both of these shortening approaches just discussed is the fact that a shorter
pulse is only cut out of the longer synchrotron pulse, leading to a decrease in the photon
number per shot. For that reason, tremendous effort is put into the development and
construction of free-electron lasers, which would supply both ultrashort pulses of hard x-rays
and large photon numbers per shot.
Adifferent route to ultrashort x-ray pulse production by combining lasers and high-energy
electronbeams has alsobeendevelopedbySchoenleinet al [87]. Whenanintense femtosecond
laser pulse scatters off a high-energy electron beam at an incidence angle of 90

, the forward
scattered photons (photons that after scattering travel in the same direction as the electrons)
can be promoted to several kilo-electron-volts photon energy due to momentum transfer from
electrons to photons. In addition, the emission of these photons is highly directed.
2.4.2. Electron-bunch compression. Instead of only using a short temporal slice of an
originally longer electron pulse to generate x-rays, there are possibilities of compressing
the original electron pulses to ultrashort time-scales prior to x-ray generation. In an
experiment currently performed at Stanfords Linear Accelerator Center (SLAC), magnetic
deecting chicanes are used to perform the electron-pulse analogue of optical laser pulse
compression [88]. If a femtosecond laser pulse is used to create a femtosecond electron pulse
in an electron injector section of an accelerator, electron pulse broadening typically arises due
to an energy spread of the electrons in the bunch. By having the slower electrons in the bunch
run a shorter distance as compared with the faster ones the magnetic chicane recompresses
the electron bunch to sub-picosecond durations. This compressed electron bunch accordingly
Femtosecond x-ray science 471
produces sub-picosecond x-ray pulses after sending it through a bend magnet or on insertion
device such as an undulator or a wiggler. At the Subpicosecond-pulse source (SPPS) [89]
at SLAC, it has been possible to produce x-ray pulses as short as 80 fs and to use them
in measurements of ultrafast crystalline-to-liquid phase transitions [90]. The disadvantage
of this approach lies in the fact that the produced ultrashort x-ray pulse is not inherently
locked and synchronized to a femtosecond laser pulse, which has to be used for a pumpprobe
experiment unless the x-ray intensity is high enough to induce a measurable two-photon optical
nonlinearity by itself. The problem of temporal jitter arises between the x-ray pulse and an
articially (technically) synchronized femtosecond laser pulse. Recently it has been possible
to measure the jitter of a linear-accelerator-based x-ray source with respect to a femtosecond
laser with an accuracy of 60 fs [91].
2.4.3. Free-electron laser. The free-electron laser (FEL) working principle is an extension
of the undulator technique. In an undulator, the electron beam is deected by a periodic
magnetic structure along the propagation path where the modulation period is chosen such
that the bremsstrahlung emitted by each electron constructively interferes with the radiation
produced by one modulation period further downstream. The radiation produced by different
electrons, however, is not coherent and leads to a linear growth of x-ray radiation with electron
density. In the proposed x-ray FEL sources, the electron density in the beam is increased (for
instance by compressing the electron bunch prior to its entry into the undulator), such that the
more intense radiation produced by the electrons interacts with the electron cloud itself. In this
case, microbunches are formed in the electron beam (gure 16) which now lead to a coherent
addition of the radiation produced by each electron. We therefore obtain an increase in x-ray
ux that is proportional to the electron density squared.
If the microbunches have to form spontaneously this is called self-amplied spontaneous
emission (SASE) [93]. If the microbunching is created externally by interaction with a laser
eld, for example, the term seeded free-electron laser is used [94, 95].
Recentlythe rst experiments onnonlinear interactionof VUVlight froma SASEFEL[96]
with xenon atoms and clusters were carried out [97] and will be described in section 4.2. The
very high charge states of xenon ions detected in the experiment led to fundamentally new
insights into lightmatter interaction [98].
3. Pulse characterization and control
It has been and still is a big challenge in the eld of ultrafast x-ray science to obtain reliable
measurements of the shape of the x-ray pulse that is produced by a certain generation technique.
There is only one direct measurement technique, which is able to acquire a snapshot of the
x-ray pulse intensity as a function of time with a resolution of some hundreds of femtoseconds:
the streak camera, which is essentially no more than a combination of a fast photodiode
and an oscilloscope. Almost any other technique (except the autocorrelation method) that is
available today relies on a crosscorrelation of the x-ray pulse with an optical femtosecond laser
pulse, raising two important implications: rst, the measurement process is a nonlinear optical
process, and second, we need a laser pulse that is perfectly (or at least sufciently, depending
on the desired temporal resolution) synchronized to the pulsed x-ray source. The fact that the
process is an optical nonlinear interaction of light elds requires us to use high intensities of
the pulses involved. Since optically-nonlinear polarizabilities rapidly decrease for photons of
higher and higher energy, we are usually bound to use highly intense laser pulses (instead of
highly intense x-ray pulses, where it is also much harder to produce intensities comparable
472 T Pfeifer et al
a)
b) c) d)
Figure 16. Proposed x-ray free-electron-laser working principle (SASE) with simulation results.
(a) Schematic setup: a high-density high-energy electron beam enters an undulator device from
the left-hand side. If the electron density is sufciently high, enough electromagnetic radiation is
generated in order to affect the electron beam itself. This leads to the longitudinal formation of
structure (microbunches) in the electron beam (shown in (b), (c), and (d) at positions close to the
beginning, the centre, and the end of the undulator, respectively) that will coherently emit radiation.
(Reprinted gure with permission from [92].)
to the ones achievable with lasers). The fact that the laser pulse has to be synchronized to
the x-ray pulse is certainly of great concern if the x-ray pulse production proceeds without
direct employment of an optical laser. For instance, for measuring the pulse duration of a
free-electron laser (section 2.4.3) we need to actively synchronize and lock an optical laser to
the electron bunches that produce the x-ray pulses. This synchronization is also essential if
we wish to perform experiments to record the temporal behaviour of systems in a pumpprobe
approach. Usually, the femtosecond pulse then serves as a pump-pulse (trigger) to create
sufcient population of an excited state or a coherent superposition of such states in a sample
system and the x-ray pulse is used to monitor the ensuing structural or chemical changes.
Measuring or monitoring the pulse width is essential in order to evaluate the meaning of the
recorded data fromsuch experiments: when we observe a smooth/slowpumpprobe behaviour,
is it due to the fact that the dynamics is really slowor does it merely mean that the pulse duration
of the pump or the probe or the temporal stability of the pump and probe pulse with respect to
each other are not sufcient? Monitoring the pulse width is also helpful to improve the x-ray
source in terms of obtaining shorter and more stable x-ray pulses.
Additionally, as soon as we can produce ultrashort pulses of radiation to passively monitor
the evolution of a system under study, we are immediately thinking of using these pulses in
order to actively control its dynamics. The progress in the eld of femtosecond laser pulse
shaping [99102] and coherent/optimal control [103105] showed that it is nowadays possible
Femtosecond x-ray science 473
to use specically tailored, shaped light elds in order to drive molecular dynamics along
certain pathways in molecular conguration space. Using shaped intense x-ray light elds,
one day we will be able to control the quantum dynamics of electrons, even in low-lying states
of atoms and molecules.
This section is structured as follows: streak cameras will be presented in section 3.1 as a
way to directly measure the intensity of the x-ray pulse versus time. Afterwards in section 3.2,
we will turn to the discussion of correlation techniques, where intense femtosecond laser pulses
or the x-ray pulse itself (autocorrelation) are used to acquire information about the temporal
shape of the x-ray pulses. Finally, we will present the rst steps towards controlling the shape
of coherent x-ray light elds in section 3.3.
3.1. Direct measurement (streak cameras)
A common approach to measure a short temporal phenomenon is to convert it to an electrical
signal that can be recorded for instance with an oscilloscope. To measure the duration of a
light pulse (how long is the light switched on), fast photodetectors such as photodiodes or
photomultipliers can be employed. However, this approach is only feasible down to pulse
durations that are longer or at least on the order of the response time of the detector. The
response time is typically given by the temporal duration of the detected pulse of photoelectrons
produced in the detector for an innitesimally short pulse of incident light. The photoelectrons
thus convert the photonic signal into an electric signal that can be monitored or recorded by
means of a (fast) oscilloscope. The electronic response time of these devices is not only
governed mainly by the size, geometry and material of the photosensitive material but also by
the intensity of the incident light. The intensity dependence can be understood from the fact
that many photoelectrons create a signicant space charge and thus repel each other enough
to lengthen the electronic pulse duration. Typically, the shortest response times that are easily
attainable fromfast photodiodes and photomultipliers are larger than tens of picoseconds. This
makes these devices unsuitable for the detection of femtosecond laser or x-ray pulses.
In the early 1970s, a method of recording very short ashes of light directly in the time
domain [106, 107] with picosecond resolution was found. The idea was to convert the light
pulse into an electron beam pulse by photoemission and ramp up a transverse electric eld
quickly enough that electrons produced at different times hit a photo plate at different positions.
The electric eld ramp has to be fast enough to allow electrons produced at temporal delays
of only picoseconds by the light pulse to be clearly separable at the photo plate. This way,
a temporal structure in the ultrashort light pulse can be recorded on the photo plate with a
resolution of a few picoseconds. It is the temporal image of a pulse that resembles a streak
on the photo plate which gives its name to the device.
Streak cameras can be operated either in the single-shot or the multi-shot mode, depending
on whether a single x-ray pulse already produces enough electrons for an image or whether
averaging over many shots is necessary to obtain enough signal. In both cases, it is necessary
to use a trigger in order to start the transverse electric eld ramp. In the multi-shot mode, it is
also necessary to maintain a good shot-to-shot stability of the timing between the trigger and
the peak of the x-ray pulse, which is highly nontrivial. The need for this stability is relaxed
in the single-shot operation mode, since in this case a (slight) offset of trigger timing would
merely lead to a shift of the streaked temporal pulse image along the time axis.
When using streak cameras, particular care has to be taken in order not to lose temporal
resolution due to a variety of effects. If the x-ray intensity is too high, space charge effects
at the photocathode will lead to a temporal broadening of the electron pulse. Another severe
limitation is the production of secondary electrons that are produced when the inner-shell
474 T Pfeifer et al
vacancies, produced by the impinging x-ray pulse, rell by releasing Auger electrons. Electron
beam focusing on the streaking plane is also crucial to obtain a high temporal resolution. To
date, these limitations in streak cameras allow for a temporal resolution of 600 fs [108, 109].
For measuring shorter pulses, we have to resort to a different approach, which will be outlined
in the next section.
Combining a streak camera with an x-ray spectrometer can provide valuable spectroscopic
information about time-dependent processes [110]. Since for a standard streak camera, only
one dimension of the two-dimensional photo plate is identied as the time axis, we have a
left-over degree of freedom along the second dimension to encode another information, for
instance, the spectrum of the incoming x-ray light. As the streak camera image is typically
independent on the energy of the incident photons, we have to use a spectrometer before
photoelectron production to spatially separate the x-ray spectrum on the photocathode of the
streak camera in a direction perpendicular to both, the photoelectron beam and the electric
eld of the ramp. Using electron optics, the photocathode can be imaged onto the photo plate,
maintaining the spectral resolution in the streaked image.
3.2. Correlation techniques
In the optical frequency domain, the use of correlation techniques has allowed for the
complete characterization of the electric eld of ultrashort laser pulses down to the few cycles
limit [111113]. Correlation refers to scanning the temporal shape of the ultrashort pulse to be
characterized by another (if possible even shorter) reference pulse. In the ideal case, we could
use an innitesimally short pulse to directly sample the electric eld of the ultrashort pulse
to be characterized. This, in fact, is the way how femtosecond laser pulses in the terahertz
frequency range are characterized [114, 115]. However, most of the times there is no shorter
reference pulse available. In this case, the ultrashort pulse can be split into two copies where
one copy serves as the reference. This most simple correlation technique is therefore termed
autocorrelation. Whenever a different pulse is used as the reference, the term crosscorrelation
is used.
How does the sampling work? If we know the exact temporal pulse shape of the reference
pulse, it is enough to have both reference and unknown pulse co-propagating with a small
temporal offset. In analogy to the spatial double slit interference pattern in the k-vector
domain, the temporal interference of two pulses will create a fringed interference pattern in the
frequency domain (spectrum) where the two pulses overlap. The frequency dependent fringe
spacing can be used in order to completely characterize the unknown pulse fromthe information
we have about the reference pulse. This technique is known as spectral interferometry [116].
If the reference pulse is unknown, this method cannot be directly employed. The crucial
information determining the pulse shape is the spectral phase of the pulse. Together with
the spectral amplitude, which is simply the spectrum of the laser pulse, the exact laser pulse
shape in frequency and thus also time (which is the Fourier transform) can be retrieved. Since
in linear interferometry only the interference of identical frequencies can be monitored with
a slow detector (where slow refers to the response time being much longer than the pulse
duration), we cannot straightforwardly record the relative phase between different spectral
components constituting the pulse. To do that, we have to articially shift the spectral regions in
frequency, whichmeans we are boundtouse a nonlinear (sum-frequency, difference-frequency)
mixing technique. Autocorrelation for example works by creating two copies of the same
unknown laser pulse and recording the light produced by second-harmonic generation (SHG)
in a nonlinear optical crystal versus the time delay. The signal will be maximum where the
pulses overlap completely in time and the dependence on time delay carries information about
Femtosecond x-ray science 475
the temporal structure. While the complete information about the laser pulse shape cannot
be retrieved from this simple autocorrelation measurement, an extension of the technique,
where the spectrum of the SHG light is recorded for each time delay, is able to do so. The
latter approach is termed frequency-resolved optical gating (FROG) and has been pioneered by
Trebino et al [117, 118]. Another way of retrieving the spectral phase from an unknown pulse
without the help of an external known reference is called spectral interferometry for direct
electric-eld reconstruction (SPIDER) and was developed by Iaconis and Walmsley [119].
Here we will present the rst steps in the eld of ultrafast x-ray science to implement
these methods of pulse characterization that were originally developed for the optical frequency
region. However, a number of newcharacterization techniques employing the cross-correlation
idea have been developed specically for x-ray applications and will also be presented in the
following.
3.2.1. Autocorrelation. An important issue in the transfer of autocorrelation to the x-ray
spectral region lies in the fact that nonlinear susceptibilities and in general the interaction
cross section of two-photon processes are rapidly decreasing with increasing photon energy
used for excitation. In order to observe the nonlinear process that is vital for the performance
of autocorrelation, very high intensities of x-ray light have to be used. So far, it has been
possible to create these highly intensive x-ray pulses by only two methods: free-electron lasers
(section 2.4.3) and high-harmonic generation (section 2.1). The rst autocorrelation in the soft
x-ray range was reported by Kobayashi et al [120] and later by Tzallas et al [121, 122] and
we will review the latter in the following paragraph. In these works, two-photon ionization of
He has been employed as the nonlinear detection signal.
The fundamental laser pulse used for high-harmonic generation was an 790 nm, 130 fs
duration, 10 mJ pulse energy laser pulse produced by a Ti:sapphire laser operating at 10 Hz.
The (odd) harmonic orders 7 through 15 were selected by an 0.2 m thickness indium lter.
From earlier theoretical studies it is known that the two-photon response to the selected
harmonic orders is rather at, which is important for second-order autocorrelation. It was
also veried that the amount of He
+
produced scaled quadratically with the harmonic intensity.
The intensity of the harmonic pulse at the focus was 10
11
Wcm
2
and roughly 90 ions per
laser pulse were produced in agreement with the known two-photon absorption cross-section
of He. The harmonic pulse to be measured was focused and split with a special mirror, where
half of the mirror surface could be moved with respect to the other half by means of a piezo
element. Controlling the offset of the mirror halves allowed us to control the temporal delay
between the two XUV pulses. After accounting for the effect that the two identical pulses
do not co-propagate it has been possible to extract the autocorrelation trace as the amount
of He
+
ions created versus time delay. From the measurement, a pulse duration for each of
the attosecond pulses constituting the XUV attosecond pulse train could be determined to be
800 as.
Usinga different regime of high-harmonic generationandthe secondharmonic (400 nm) of
a Ti:sapphire laser as the fundamental radiation, an autocorrelation in a higher photon energy
regime could recently be carried out [123] (section 4.6). In this case, a single attosecond
pulse could be observed as a result of the autocorrelation measurement (gure 17), which was
conducted by looking at the rst ATI-photoelectron peak (two-photon absorption) produced
by the XUV light in He gas.
3.2.2. X-FROGmeasurements. While it is possible to obtain the duration or even the temporal
intensity of the x-ray pulses by performing a cross-correlation measurement, it is surely better
to record the shape of the pulse in a more general way, extracting its temporal intensity and
phase. If this is done, it is possible to reconstruct the electric eld of the pulse, except for a phase
476 T Pfeifer et al
(a)
(b)
(c)
Figure 17. Experimental measurement of an XUV autocorrelation of XUV pulses generated with
high-harmonic generation for two different durations of the driving pulse: (a) 8 fs and (b) 12 fs. The
lighter grey lines are gaussian ts to the data, which indicate pulse durations of (a) 950 90 as and
(b) 1.30.1 fs after deconvolution. The darker greylines in(a) and(b) are calculatedautocorrelation
traces that correspond to the Fourier transformof the corresponding spectra shown in (c), assuming
a at spectral phase. Lighter and darker grey lines in (c) indicate EUV spectrum obtained with
the 8 fs and 12 fs driving pulse, respectively. (Reprinted by permission from Macmillan Publishers
Ltd; Nature [123]. Copyright (2004).)
constant. The measurement method that can accomplish this has actually been developed for
laser pulses in the NIRto visible domain and is called FROG[117,118]. The only difference of
the method to a cross-correlation measurement, where the nonlinear-optical cross-correlation
signal is measured depending on time delay between known and unknown pulse, is that the
cross-correlation signal is nowalso spectrally dispersed and recorded again separately for each
time-delay. This creates a two-dimensional set of data, the so-called FROG trace. Initially
based on the autocorrelation technique (where the same laser pulse is used as the unknown
and the reference pulse), it was later extended to characterize an unknown pulse using a
known reference pulse in a cross-correlation measurement, referred to as X-FROG[124]. This
technique is particularly benecial if the unknown pulse is very weak. Using this technique,
it has so far been possible to measure the electric eld of a VUV pulse produced in the high-
harmonic generation process [125, 126].
For this measurement technique, it is particularly important to keep a very small timing
jitter between the known reference and the unknown x-ray pulse. If this is not the case the
FROG-trace will be signicantly distorted and reconstruction of the x-ray pulse shape cannot
be done.
3.2.3. Crosscorrelation. A way to circumvent the low nonlinear response in the x-ray range
is to perform crosscorrelation measurements. In this case, the x-ray pulse is interacting with
an intense pulse of visible radiation [127, 128]. Typically, the interaction is chosen to take
place in a medium where the ionization potential I
p
is lower than the photon energy h
x-ray
of the x-ray pulse but much larger than the laser photon energy h
laser
to keep ionization
by the laser pulse alone at low levels. In the absence of the laser pulse, photoelectrons at
Femtosecond x-ray science 477
Figure 18. Measuring the chirp of high-order harmonic pulses by XUV-IR cross-correlation.
Interaction of intense 800 nm light with photoelectrons produced by the high-harmonic XUV
radiation in argon produces sidebands in the photoelectron spectrum that are in between the main
peaks at the odd harmonic orders (which are the only ones present in the XUVpulse). The tilt of the
photoelectron sidebands with respect to the time-axis contains information about the instantaneous
frequency of the XUV pulse. (Reprinted gure with permission from [129]. Copyright (2004) by
the American Physical Society.)
energy E = h
x-ray
I
p
are produced. In the presence of the intense laser light we obtain
photoelectrons with energies at E
n
= h
x-ray
+ nh
laser
, where n = . . . , 2, 1, 0, 1, 2, . . ..
This can be understood in the framework of the laser-assisted photoelectric effect, described
by a Volkov-state formalism [128]. The additional peaks in the photoelectron spectrum are
called sidebands of the nth order. The difference in electron energy between the highest order
sideband and the 0th order scales linearly with the ponderomotive potential U
p
. Furthermore,
the high-intensity laser pulse will create an ac-Stark shift (also called ponderomotive shift)
of the ionization potential, which is on the order of the ponderomotive potential U
p
of the
laser eld. This effect shifts the electron energy of the directly photoionized (n = 0) peaks.
Both these effects can be employed to measure the temporal duration of an x-ray pulse [128].
The temporal delay between x-ray and laser pulse is varied while the ponderomotive shift
or the sideband intensity is recorded. By deconvolution, the duration of the x-ray pulse can
be reconstructed. This method is particularly valuable for characterization of x-ray pulses
directly produced by femtosecond lasers, since an optical femtosecond pulse to be used for
correlation is available. The problemthat arises for measuring x-ray pulses that are not directly
produced by a femtosecond laser, such as free-electron laser pulses (section 2.4.3), is the lack
of synchronization with a femtosecond laser pulse. The temporal resolution in this case is
mainly determined by the temporal jitter, except for single shot experiments.
The spectral position of the sidebands (gure 18) can be used in order to acquire
information about the chirp-rate of the x-ray pulse, which has recently been demonstrated
by Mauritsson et al [129].
In particular, if a comb of high-harmonic (section 2.1) frequencies is produced and
overlapped with the fundamental laser radiation in a collinear propagation geometry, they
can be used as a quite sensitive probe with much more temporal resolution than the intensity
crosscorrelation. Since in this specic case, the x-ray pulse and the laser pulse are coherent to
each other (i.e. they exhibit a xed phase relation), each +1st order sideband of the (2n 1)th
harmonic order interferes with the 1st order sideband of the (2n + 1)th harmonic order.
Monitoring the constructive and destructive interference of these sidebands as a function of
delaybetweenharmonic andfundamental pulse inthe crosscorrelationregionreveals a subcycle
modulation of the sideband intensity. By comparing the phases of subsequent sidebands, it is
possible to retrieve the relative phase of the high-harmonic orders, enabling the reconstruction
of the temporal shape of the attosecond pulse train produced in the process of high-harmonic
478 T Pfeifer et al
generation [130, 131]. The sideband width increases for shorter and shorter x-ray pulse trains,
eventually leading into the structure observed for the ponderomotive streaking method, which
is presented next.
3.2.4. Ponderomotive streaking. The shortest measurable pulses of light that are currently
available have durations on the order of a few hundreds of attoseconds (1 as = 10
18
s). These
pulses are produced by a high-harmonic generation of laser light (section 2.1) and suitable
ltering of the created broadband soft x-ray spectrum[132,133]. Since these pulses are derived
from the fundamental laser pulse in a coherent fashion, they are perfectly locked to the laser
pulse. This means that the timingjitter betweenthe attosecondpulses andthe laser pulse (phase)
is negligibly small. Ponderomotive streaking with the laser electric eld can thus be used for
measuring the pulse duration of these sub-femtosecond x-ray pulses. Asub-femtosecond x-ray
pulse photoionizes an atom. It creates a burst of photoelectrons with duration equal to the one
of the x-ray pulse. If this free-electron burst is created under the inuence of a high-intensity
laser eld, it will be signicantly accelerated or decelerated depending on the phase of the
electric eld of the light. By scanning the delay between the laser and the sub-cycle x-ray
pulse, the photoelectrons emitted into a small solid angle along the laser polarization (detected
for instance by photoelectron time-of-ight detection) exhibit a characteristic modulation of
the photoelectron energy. Comparing these transient photoelectron energy distributions with
a simulation, the free parameter of x-ray pulse duration can be extracted [133]. Since the
measurement in this case is based on the effect of the laser eld instead of its intensity, no
deconvolution of the cross-correlation result with the duration of the streaking laser pulse is
necessary. It is enough to knowthe frequency of the streaking eld, which is much more easily
accessible. This technique can be regarded as the optical analogue of a streak camera, where
the deecting electric eld is ramped up in a very short time since it is produced by an intense
optical laser eld.
3.2.5. Laser-induced absorption gating. An ultrafast change in the absorption properties of
a material can also be used to perform a measurement of the duration of a femtosecond x-ray
pulse. An example where a structural modication of a solid on a femtosecond time-scale leads
to an abrupt change in the transmission characteristic has been observed by Decamp et al [86].
When a crystal is irradiated by x-ray light, the light is partially absorbed along its way through
the crystal. However, when the x-ray wavelength and (grazing) incidence angle full the
Bragg-condition = 2d sin for a given lattice plane distance d in the crystal, it has been
previously observed that x-ray light can be transmitted through the crystal even for thicknesses
of many absorption lengths [134, 135]. This is because internal multiple Bragg-reection
along the propagation of the x-rays in the crystal results in the creation of two stationary
modes (standing wave patterns). One of these standing waves (-solution) has zero intensity
at the positions of the atoms (lattice planes) while the other one (-solution) has maximum
intensity at these positions. Since absorption is higher in a vicinity close to the atom, the
-solution will be attenuated very efciently while the -solution can propagate much longer
and also has a different phase velocity than its companion solution. At the exit of the crystal,
both - and (if present) -solution are projected onto the free-space solutions, which are the
Bragg-reected and the forward-directed (unscattered) x-ray beam. This happens in a coherent
manner, meaning that the relative phase between - and -solution (which is changing with
propagation distance) decides about the relative intensity between forward and Bragg-scattered
x-ray beam. An ultrashort (femtosecond) laser pulse can be used close to the exit surface of the
crystal to repopulate the extinguished -solution from the surviving -solution. Depending
Femtosecond x-ray science 479
on the time delay between an x-ray pulse travelling through the crystal and a laser pulse
(launching the repopulating crystal lattice dislocation upon incidence on the crystal, which
then moves from the surface into the crystal at approximately the speed of sound) the relative
phase of - versus -solution can be controlled, leading to a controlling of the intensity in the
forward and Bragg-scattered x-ray light. It has been shown that the change of x-ray intensity
proceeds on a time-scale of 100 ps, limited by the pulse duration of the x-ray pulse used in the
study. This method thus provides a fast switch of x-ray intensity. Therefore, it can be used
to temporally scan across an x-ray pulse to measure its duration. While Decamp et al used
acoustic phonons, it could be possible to use optical phonons [136] that have higher phonon
frequencies. The temporal resolution of this technique is then expected to give access to the
femtosecond time-scale [136]. Another application is an ultrafast shutter in order to produce
(cut out) sub-picosecond pulses of x-rays from a longer pulse.
Another crosscorrelation method is based on laser-gated absorption in the gas phase. In an
experiment conducted by Oguri et al [137], a particular ionization state of Kr (singly ionized)
exhibits a resonance with the soft x-ray light to be characterized. If an intense femtosecond
laser pulse is used to irradiate the gaseous sample of Kr atoms, different ionization states are
sequentially produced with the lower ones being created in the leading edge of the laser pulse
and the higher ones shortly after the centre of the pulse. The abundance of Kr
+
thus initially
increases before falling off towards later times, as ionization continues to produce Kr
2+
and
higher charge states. Efcient absorption of the x-ray light thus occurs only during a short
window of time, which can be used to sample the x-ray pulse shape again by varying the
time-delay between x-ray and optical pulse.
3.3. Control of pulse parameters
Controlling the parameters of x-ray pulses is critical for optimizing with respect to particular
experimental requirements. Alongthese lines, enhancement of single laser-generatedharmonic
ux has been carried out by adaptive laser-pulse shaping [138, 139]. In addition, the spectral
position of a comb of harmonics could be controlled by similar means [140]. For the case
of laser-generated plasma x-rays, an optimization of the yield has also been carried out using
automated algorithms [141]. While these studies focused on single parameter optimization of
the soft-x-ray light, our group has recently demonstrated that it is possible to control the overall
spectral shape of coherent XUV/soft-x-ray radiation created by the process of high-harmonic
generation. Control can be achieved by adaptive shaping of the spectral phase of the driving
femtosecond laser pulse. Under standard experimental conditions the XUV high-harmonic
spectrum exhibits a characteristic property: starting at lowest orders and moving to higher
ones, the high-harmonic intensity drops for a fewharmonic orders, then stays roughly constant
in the so-called plateau region (up to orders 300 [54]) and nally vanishes abruptly at a
particular photon energy (cutoff). This large bandwidth of coherent radiation can be used to
generate XUV light pulses of a few attoseconds duration. These, in turn, provide a tool to
study the electronic dynamics of all kinds of atomic [5] and molecular systems.
To the same extent as this large bandwidth might be benecial to attosecond pulse
generation and observation of electronic dynamics, it is unfavourable for monitoring and
controlling molecular dynamics. For this purpose, a small-bandwidth XUV source would
clearly be advantageous. In particular for applications in ultrafast photoelectron spectroscopy
a high degree of monochromaticity of the XUV light is a very desirable property. Beyond
spectroscopy, zone-plate nanoscale microscopy using high-harmonic radiation could also be
improved by exploiting the same property [142]. Thus it seems reasonable to investigate
whether it might be possible to control the process of HHG in such a manner that just one
480 T Pfeifer et al
particular harmonic order is generated while all the others are suppressed. Even beyond this,
there is the question whether it is feasible to control the overall shape of the harmonic spectrum.
This could open the door to a newera of quantumcontrol where we could coherently manipulate
electrons on their natural time scale.
We have been able to show that it is indeed possible to not only enhance specic harmonic
orders while suppressing neighbouring ones but also to engineer the harmonic XUV spectral
shape in more general ways [143]. Selective generation and suppression of single and groups
of harmonics is feasible. Our work can be regarded as the rst signicant step towards quantum
control in the XUV spectral region. Additionally, our experiments represent a direct example
for coherent control of electronic motion, since we are able to control the harmonic spectrum
emitted by the extended and nonlinear motion of the electron in the vicinity of its parent ion.
In our setup [144], we use a deformable membrane mirror to shape our laser pulses. This
mirror replaces the commonlyusedretroreectingmirror inthe prismcompressor installedafter
a hollow-bre compression stage. By applying control voltages to each of its 19 electrodes, we
are able to create different surface shapes that translate into different spectral phase functions of
the laser pulse. A hollow bre setup [39] is used for HHG. Aluminium lters are employed to
separate the harmonic light fromthe intense fundamental laser pulse. For spectral analysis, we
use a grazing incidence spectrometer equipped with a backside-illuminated charge-coupled-
device camera. A computer is used to read out the spectrometer and to control the deformable
mirror. An evolutionary algorithm[145] iteratively optimizes the harmonic spectrumtowards a
given shape criterion. This is done by maximizing the tness. It increases when the spectrum
comes closer to the desired shape. Results of a typical shape optimization experiment are
presented in gures 19(a)(c). Our goal was to selectively enhance the 25th harmonic order
(integrated spectral signal in region Aindicated in gure 19) while at the same time suppressing
the generation of neighbouring orders (region B). The intensity of the 25th harmonic increased
by a factor of four while neighbouring orders are a factor of three less intense. It is thus possible
to generate single harmonic orders with high brightness (gure 19(b)). The setup was manually
optimized for maximum harmonic output (spectrum shown in gure 19(a)) for the reference
mirror shape prior to running the optimization algorithm. However, a much larger degree of
control is possible. Not only can we enhance a particular harmonic, but we have also already
demonstrated isolated harmonic emission at different orders as well as selective generation of
groups of harmonics. Another example is the selective minimization of a harmonic, which is
depicted in gure 19(d).
We performed simulations based on the solution of the Schr odinger equation for a one-
dimensional model atom to elucidate structural changes of the produced XUV pulses in the
time domain. The same evolutionary optimization strategy as employed in the experiments was
used for the simulation both for enhancement (gure 20(a)) and suppression (gure 20(b)) of
a single harmonic order. The corresponding trains of attosecond pulses are shown in gure 20
besides the optimized spectra. Although the changes in the infrared driver pulses are miniscule,
the temporal XUV structure changes dramatically, demonstrating comprehensive control over
XUV-pulse shape. Remarkably, the ratio of suppressed harmonic orders versus enhanced ones
is much less than in the experiments. This agrees with the common theoretical knowledge about
the single-atom response and proves that controlling spatial propagation effects is important
to generate arbitrarily shaped XUV-spectra and attosecond pulses. By applying spatial pulse
shaping methods we could recently provide conclusive evidence of this result [146]. These
ndings about control over coherent XUV spectra represent a key to opening up a new eld
of quantum control. The controllable frequency ranges and bandwidths match typical values
associated with electronic inner shell processes and electron dynamics in general. Not only can
basic molecular vibration and rotation be addressed using femtosecond infrared laser pulses
Femtosecond x-ray science 481
f
i
t
n
e
s
s
(c)
30 35 40
0
5
10
15
20
25
30
35
40
I
n
t
e
n
s
i
t
y

[
a
r
b
.

u
.
]
wavelength [nm]
27
25
23
(b)
25 30 35 40
0
5
10
15
I
n
t
e
n
s
i
t
y

[
a
r
b
.

u
.
]

wavelength [nm]
(a)
B A B
25 25 30 35 40
0.0
0.2
0.4
0.6
0.8
1.0
1.2
(d)
23
25
27
29
21
wavelength [nm]
I
n
t
e
n
s
i
t
y

[
a
r
b
.

u
.
]
5 10 15 20
-5
0
5
10
15
20
25
30
fittest
average
reference
generation #
Figure 19. Engineering of coherent XUV spectra by controlling electron motion. Optimization
experiments were carried out to maximize (a)(c) and minimize (d) single selected harmonics. The
tness of the ttest laser pulse shape increases during the run of the evolutionary algorithm. The
reference pulse shape stays at a roughly constant value, indicating stable experimental conditions.
but we can also nowdirectly engineer electronic states for further applications. In section 4.1.3
we will highlight the potential of the presented coherent soft x-ray spectral shaping method
for optimal control experiments in the future.
In addition, the technique of high-gain harmonic generation has been developed by the
group of Yu at Brookhaven National Lab [94, 95], where an intense ultrashort-laser modulated
electron beam is used to efciently produce higher harmonics of the modulating laser pulse. If
shaped soft x-ray pulses can be used for modulation, this technique allows us to create shaped
pulses of yet higher photon energy and large numbers of photons per pulse up to the hard x-ray
region.
4. Experiments
We will now describe a few benchmark experiments that employed the presented sources of
ultrafast x-rays for various scientic applications. As mentioned in the beginning, due to
their short wavelength and short temporal duration, ultrafast x-rays are the tool of choice for
monitoring atomic-scale microscopic changes on their natural (femtosecond to picosecond)
time scale. In principle, there are two ways of extracting time dependent information: either
use both ultrashort pump and probe pulses with a slow detector or use a fast detector (for
instance a streak camera) in which case only the pump pulse would need to be ultrashort
[147]. Due to the limited temporal-resolution capability of fast detectors in the femtosecond
regime, most experiments are performed in the ultrafast-pumpultrafast-probe scheme. A
number of methods have been developed, ranging from x-ray photoelectron spectroscopy and
ponderomotive streaking to ultrafast x-ray diffraction. Each of these methods is well suited
to a particular set of applications and many of the methods are mutually complementary
482 T Pfeifer et al
-8 -6 -4 -2 0 2 4 6 8
0
10
20
30
40
-0.10
-0.05
0.00
0.05
0.10

19 21 23 25 27
0
1
2
s
p
e
c
t
r
a
l

i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
.
]

-8 -6 -4 -2 0 2 4 6 8
0
5
10
15
20
-0.10
-0.05
0.00
0.05
0.10
i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
.
]

19 21 23 25 27
0
1
2
harmonic order
e
l
e
c
t
r
i
c

f
i
e
l
d

[
a
.
u
.
]
time [fs]
a)
b)
Figure 20. Simulation results of coherent XUV engineering. Selective optimization of the 23rd
harmonic order with respect to its neighbour harmonics; (a) maximization and (b) minimization.
The corresponding XUV pulse shapes are plotted to the right. The temporal structure of the XUV
pulses changes dramatically even though the infrared driving laser pulses are only slightly modied.
and serve different purposes. Although this experimental section will be devoted primarily to
experiments carriedout usingfemtosecondpulses, we shall mentionthe experiments performed
with x-ray light produced at synchrotron facilities, where tremendous progress in the analysis
of time-resolved dynamics in biology and chemistry has been made recently. Since in the
case of standard synchrotrons the pulse duration is always on the order of 100 ps or higher
(limited by the revolving-electron-bunch length), the temporal resolution is limited but enough
to detect important and scientically interesting dynamical processes. For instance, 100 ps
x-ray pulses were used to monitor the structural dynamics of an iodine sample in solution after
photoexcitation or -dissociation by an ultrafast laser pulse [148, 149]. X-ray diffraction with
100 ps synchrotron pulses was also used in powder-diffraction studies to analyse the geometry
of short-lived transiently excited electronic states in organic solids [150]. The monitoring
of structural phase transitions was another target of studies [151], where a paraelectric to
ferroelectric transition was observed after photoexcitation with an ultrashort laser pulse. Metal-
insulator transitions were followed in real time by core-level photoelectron spectroscopy [152].
Structural changes in melting silicon could be temporally resolved in another experiment [153].
The thermal expansion and melting properties of metal nanoparticles heated by femtosecond
laser pulses were studied using 100 ps synchrotron x-rays [154, 155]. Moreover, the structural
intramolecular dynamics of a large protein (myoglobin mutant) following the excitation by a
picosecond laser pulse could be monitored on the 100 ps time scale [156]. The diffusion
of a CO molecule within the protein could be tracked in time and space with angstrom
resolution. Apart from x-ray diffraction, x-ray uorescence measurements have been shown
Femtosecond x-ray science 483
to be a versatile tool for coreshell absorption spectroscopy in GaAs [157]. On the 10 ns time
scale, synchrotron x-ray pulses have been employed for the structural analysis and life-time
measurements of transient intermediate states created by photoexcitation of large molecular
systems in solution [158]. In this case, the technique of x-ray absorption ne structure (XAFS)
was used. Absorption techniques are also useful for studying the change of oxidation state in
chemical complexes [159]. In particular, the technique of x-ray absorption near-edge structure
(XANES) in conjunction with short x-ray pulses is an efcient tool to monitor transient
chemical shifts. A recent review [6] contains more details about ultrafast x-ray absorption
techniques with particular emphasis on synchrotron experiments.
We also mention the progress in the eld of time-resolved ultrafast electron
diffraction [160166], where not electromagnetic but matter (electron) waves are used to
provide structural information about the illuminated samples. Also here, an ultrafast laser
pulse is used to excite a sample system and a time-delayed high-energy electron pulse diffracts
off the sample. Due to the smaller penetration depths of electrons as compared with x-rays, this
method is of particular importance for surface analysis. Another recently introduced method
pioneered in the group of Corkum [167171] allows for the generation of sub-optical cycle
durationelectronpulses. These pulses are generatedbytunnel-ionizationof atoms or molecules
and can be used to measure sub-femtosecond dynamics. Molecular nuclear wavepackets and
electronic structure has been the subject of study in these works.
This section is structured as follows: we will describe experiments carried out in different
aggregation states and phases of matter. Section 4.1 will contain experiments conducted in
the gas phase. Afterwards, we will discuss clusters in section 4.2. Then we will treat surface
science in section 4.3 before we turn to the report of the experiments conducted in the solid
phase in section 4.4. Recent results in the study of dense-plasma dynamics will be summarized
in section 4.5. Finally, section 4.6 will be devoted to the latest developments in high-intensity
x-ray science, where it now becomes possible to drive nonlinear lightmatter interactions with
ultrashort intense x-ray pulses.
4.1. Gas-phase
If we want tostudythe dynamics of single particles, we have touse densities inthe probe volume
that are low enough to effectively rule out interactions between the particles. In particular,
when we are working with ultrafast pulses, we only have to ensure that inter-particle collisions
or other interactions do not occur on a time-scale that is smaller than the dynamical time-scale
that we are interested in monitoring. This low-density environment is provided in the gas-
phase. We will therefore start out by describing experiments performed in the gas-phase with
the goal of monitoring these single-particle dynamics. Molecular dynamics will be the subject
of section 4.1.1, before we address the time-domain study of electron dynamics during Auger
decay in section 4.1.2. The latter highlights an important aspect of ultrafast soft-x-ray science:
due to the smaller wavelength and the shorter optical cycle in the x-ray frequency region, we
can eventually leap from the femtosecond study of molecular dynamics to the observation
of electron dynamics on the attosecond time scale. Section 4.1.3 will describe an approach
that could lead to controlling electron dynamics by applying taylored soft x-ray light elds in
the future.
4.1.1. Probing of molecular dynamics. In molecular systems, a lot of information about a
specic molecule can be extracted from knowing the potential energy curves (for diatomic
molecules) or potential energy surfaces (for molecules containing more than two atoms).
According to the BornOppenheimer approximation, different electronic states, populated by
the excitation of a ground-state molecule, correspond to different potential energy curves
484 T Pfeifer et al
or surfaces the shapes of which are given by the binding energies of the electron for a
given conguration of bond lengths and angles between atoms in the molecule. Using IR-
spectroscopy, one is able to obtain information about the shape of the molecular potential
energy surfaces by extracting the vibrational level spacing close to the vicinity of a local energy
minimum of the surface. It would be interesting of course to completely map out the energy
surface landscape, not only in the region close to bound states but also for a dissociative
system. In addition, since the BornOppenheimer approximation breaks down as soon as
rapid atomic motion (vibration, dissociation) occurs, we have to nd a way of how to measure
the progression of a quantum-mechanical molecular wave-packet [172] as it evolves along a
multiplicity of energy surfaces in an excited molecule. All this can be done by employing
ultrafast soft x-ray pulses.
Whereas a wealth of information about molecular dynamical processes can be extracted
by the use of ultrafast (femtosecond) laser pulses [173, 174], the exact shape of the molecular
potential explored by the molecular wavepacket often remains obscure. This is due to the fact
that in most cases multiphoton excitation processes are at work for either pump or probe pulse
(or even both) to extract a measurable signal, such as an ionization yield or a photoelectron
energy distribution. In addition, one is limited to the part of the potential energy surface,
where the difference potential between ionized and bound state is smaller than the photon
energy of the near infrared or visible laser pulse. All these restrictions do not apply for
the case of the femtosecond soft-x-ray pulses used as a probe. With the energy of a single
photon, they are able to map virtually any bound state molecular wavepacket far into the
ionization continuum. In particular, this works even if the bound state has a high ionization
potential, which would thus be not directly accessible by a visible or infrared laser pulse.
This makes it possible to measure for instance the dynamically changing ionization potential
encountered during dissociation of a molecule into separate atoms in real time, as was shown by
Nugent-Glandorf et al [175, 176].
In this particular work, a Br
2
molecule was the system under study. Using a 400 nm 80 fs
excitation (pump) pulse of 140 J pulse energy, the excited repulsive C state of the molecule
(gure 21(a)) was populated from the initial molecular ground state. The probe pulse was a
200 fs soft x-ray pulse, employing the 17th and the 19th harmonic of an 800 nm fundamental
laser pulse containing 3 10
6
photons, created in the interaction of 1.7 mJ 70 fs laser pulses
with an argon gas. The laser repetition rate was 1 kHz and the Br
2
density in the interaction
region was on the order of 7 10
3
Pa. The selection of a single harmonic order out of the
comb of plateau harmonics was accomplished using a double grating monochromator setup that
facilitated to keep the pulse duration of the selected harmonic soft-x-ray pulse on the order of
a few100 fs [176] (other methods to select a certain photon energy out of harmonic spectra use
a single grating [177, 178] or a monochromator based on XUV-multilayer mirrors [179, 180]).
The 400 nm pump-pulse created a molecular wavepacket in Br
2
that moves towards larger
internuclear separations. The soft x-ray pulse was used to ionize the system at different times,
and photoelectron energy spectra were acquired. Figures 21(b)(d) show the photoelectron
energy measured depending on the pumpprobe delay. The photoelectron energy peaks that
are associated with the ionization of atomic Br increase for larger and larger delay times (see
gure 21(c)). A small feature at lower binding energy (gure 21(b)) can be associated with
the wavepacket propagating on the excited dissociative molecular state. Plotting the integrated
signal of the atomic peak versus time (gure 21(d)) and comparing it with the instantaneous
cross-correlation signal between the pump and the probe pulse, it can be seen that single atoms
can be detected with a time delay of 40 fs after the excitation to the dissociative state. This is
in accord with theory, where after 40 fs the Br
2
internuclear distance reaches 3 (compared
with 2.23 equilibrium bond length) where atomic behaviour starts to become apparent.
Femtosecond x-ray science 485
X
C
X
A
Br
2
Br
2
+
4
0
0

n
m
1
7
t
h
1
7
t
h
1
7
t
h
1
7
t
h
(a) (b)
(d)
(c)
Figure 21. Measuring dissociative molecular dynamics in Br
2
with ultrafast XUV-radiation.
(a) Energy level diagram: a 400 nm femtosecond pulse excites a dissociative wavepacket in Br
2
that is probed at different time delays with the 17th harmonic generated with a fundamental 800 nm
femtosecond laser pulse. (b) photoelectron energy spectrum for different time delays in the region
of the two-photon absorption fromthe neutral Br
2
ground state to the Br
+
2
Xstate. (c) photoelectron
energy spectrum in the region of the atomic binding energy. From the process of above-threshold
ionization by the combined action of XUV and visible laser elds (most prominent peak in (b)),
the temporal resolution can be obtained (cross-correlation, see (d) solid line). The onset of the
atomic line is slightly displaced with respect to the cross-correlation signal (d) ( ) from which
the timescale of dissociation can be extracted. (Reprinted gure with permission from [175].
Copyright (2001) by the American Physical Society.)
This study opens the door to a variety of applications of the presented technique to the
monitoring of ultrafast molecular wavepacket dynamics. This is of particular interest when
it will be applied to the observation of nonadiabatic processes or the passage of wavepackets
through conical intersections.
4.1.2. Auger lifetime. Recently, the generation of high-order harmonic radiation resulted in
the production of the rst pulses with durations smaller than the optical cycle of the visible
laser light. These pulses of a few hundreds of attosecond in duration [133, 181] are produced
at photon energies that are in the soft x-ray region. Due to these properties, they are ideally
suited to the study of inner-shell electron dynamics. It is well known from spectroscopic
analysis of Auger line widths that the lifetimes of inner-shell vacancies are on the order of
a few femtoseconds, becoming even smaller for more tightly bound electronic states. It is
possible to use the ponderomotive streaking technique described in section 3.2.4 to measure
the temporal response of the Auger-electron release in a core-excited atomic (or molecular)
system (gure 22(a)).
In the experiment conducted by Drescher et al [5], the attosecond soft x-ray pulse was
used as a pumppulse in order to create the coreshell vacancy (core hole) in a Kr atom. The
attosecond pulse itself was produced by employing a few-cycle 800 nm laser pulse focused to
high intensities into a Ne gas cell for high-harmonic generation. A collinearly propagating
486 T Pfeifer et al
(a) (b)
Figure 22. Direct time-domainmeasurement of Auger lifetime inKr usinga sub-femtosecondXUV
pulse. The XUV pump pulse created a coreshell vacancy in Krypton, the temporal evolution of
which was monitored (probed) by the same few-cycle 800 nm pulse that was initially used to create
the XUV pulse by high-harmonic generation. The pulse shape of the 800 nm pulse was obtained
by a separate measurement and was used to deconvolute the pumpprobe transient to extract the
temporal distribution of the Auger electron emission. (Reprinted by permission from Macmillan
Publishers Ltd; Nature [5]. Copyright (2002).)
portion of the fundamental pulse with variable time-delay was used as the probe pulse. The
probe process was based on the ponderomotive acceleration of the Auger electron. The nal
velocity (energy) of the Auger electron depends on the phase of the probe laser electric eld
at which it detaches from the atom. From the crosscorrelation data acquired by the transient
Auger-electron-energy spectra it has been possible for the rst time to measure an Auger
decay process of an atom directly in the time-domain (gure 22(b)). The same technique
can be expected to produce more knowledge about inner-shell electron processes in atoms,
molecules, clusters and solids. For a recent reviewon the eld of attosecond physics see [182].
While the presented experiment is a demonstration of the versatility of ultrafast x-rays for time-
resolved coreshell electron spectroscopy in the gas phase, section 4.4.3 contains an example
of monitoring the corehole decay in solid state materials.
4.1.3. Adaptive control with spectrally shaped soft x-rays. The control of chemical
reactions is a central topic of optimal quantum control. It has always been a dream of
chemists and physicists to use laser light in order to steer chemical reactions towards desired
targets. In particular the idea of closed loop quantum control [103] has been a major
breakthrough [104, 105] in the eld. For these early experiments, shaped 800 nm laser pulses
were used. The transfer of pulse shaping techniques into the visible [183], mid-infrared
(MIR) [184] and UV [185] (down to 200 nm) wavelength range has already been achieved in
the past and is available for control experiments.
With the technique of coherent soft x-ray engineering introduced in section 3.3 it becomes
possible for the rst time to conduct closed loop learning control experiments in the soft x-ray
Femtosecond x-ray science 487
spectral range. As we show here [186], the branching ratios of a photodissociation reaction
of sulfur hexauoride (SF
6
) can be controlled by adaptive soft x-ray spectral shaping. This
study can be understood as a demonstration of the principal technical versatility of the adaptive
coherent soft x-ray source for this type of experiments, regardless of the physical process at
work. Many more applications that directly observe the electronic dynamics are expected
to follow up on the herein developed technological advances in the future. Soft x-rays are
particularly useful to obtain selectivity in chemical reactions as they provide a means for
site-specic excitation in molecules [187].
The earliest experiments to demonstrate selectivity in chemical reactions by applying
shaped optical laser light elds focused on photofragmentation [105,188,189]. This is because
of the experimental simplicity to prepare the system and to detect the products. In gas-phase
photochemistry, the detection of ionic fragments is particularly simple. A time-of-ight mass
spectrometer is ideally suited to serve as a detector for the product ions.
SF
6
is an interesting, well-known system for the study of photofragmentation. When
this molecule interacts with light in the soft x-ray spectral region (ionization potential
I
p
= 15.32 eV), different positively charged ionic fragments are produced. This can be
explained by dissociative photoionization [190193], which is the dominant fragmentation
process in the soft x-ray range for many molecules. It means that electrons are transferred to
high-lying excited ionic states by the high-energy photons. The potential energy curves for
these states can be repulsive for a particular bond, which leads to dissociation. One or more
uorine atoms can be detached.
The presented results concentrate on the two most dominant fragmentation channels, SF
+
5
and SF
+
3
. As an optimization criterion (tness function) for the evolutionary algorithmwe used
the relative yield of ions (product branching ratio).
The branching ratio, i.e. the value of the tness function, was recorded by the computer.
The evolutionary algorithm was programmed to perform a maximization and afterwards a
minimization of the product ratio. The results of these experiments are summarized in
gure 23(a). The relative product yields could be maximized and minimized by 25% with
respect to the reference yield obtained for the unmodulated spectrum prior to optimization.
The soft x-ray spectra recorded for the case of maximization and minimization of the ion
yield, along with the spectrum before optimization are presented in gure 23(b). Since the
aluminium lter was used the recorded spectra contain the full spectral intensity information
about the light used in the experiment.
It is important to note that the short- and long-term stability of the shaped coherent soft-
x-ray light fromthe high-harmonic generation process is high enough to allowthe evolutionary
algorithmto nd an optimal set of control parameters even though the branching ratio can only
be slightly changed. This demonstrates the technical versatility of the presented method to
perform, in principle, any optimal control experiment in the soft-x-ray spectral region. This
experiment is a rst demonstration of the applicability of high-harmonic spectral engineering
in optimal control of quantum dynamics with soft-x-ray light. In the future, quantum control
will not only concentrate on the control of molecular motion and vibrational dynamics but
will also include the control of electronic dynamics in tightly bound core levels of atoms and
extended systems.
4.2. Clusters
Clusters are at the interface between the areas of gas and condensed phase. Therefore, they
represent an interesting medium to study the transition from a single-particle to a many-
particle system. Regarded in the light of optics, clusters are an interesting candidate species
488 T Pfeifer et al
i
o
n

f
r
a
g
m
e
n
t

r
a
t
i
o
SF
5
/ SF
3
+ +
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8

b
e
f
o
r
e

o
p
t
i
m
i
z
a
t
i
o
n

m
i
n
i
m
i
z
a
t
i
o
n

m
a
x
i
m
i
z
a
t
i
o
n
25 30 35 40
0.0
0.2
0.3
0.5
0.7
0.8
1.0
s
p
e
c
t
r
a
l

i
n
t
e
n
s
i
t
y

[
a
r
b
.

u
.
]
wavelength [nm]
before optimization
minimization
maximization
SF
5
/ SF
3
+ +
(b) (a)
Figure 23. Control of the branching ratio of SF
6
photofragmentation by shaped soft x-ray
light. The two most dominant fragmentation channels SF
+
5
and SF
+
3
were optimized. (a) Both
minimization and maximization of product ratios as compared with an unmodulated reference
harmonic emission were achieved. (b) Soft x-ray spectra of shaped harmonic emission to optimize
photofragmentation of SF
6
. Optimal harmonic spectra are shown that maximize ( ) or
minimize (- - - -) the branching ratio of SF
+
5
versus SF
+
3
. The unmodulated reference harmonic
spectrum before optimization is also given for comparison.
for nonlinear lightmatter interactions, since their (larger than atomic) size typically comes
with an increased interaction cross-section. This can be made use of for the study of nonlinear
optics in the x-ray regime.
One of the expectations associated with the development and construction of x-ray FEL
is the enormous photon ux per pulse that is achievable. Although in general nonlinear
interaction cross sections become lower and lower in the XUV to x-ray spectral region the
intensities of these lasers will be and are high enough to drive nonlinear optical processes.
In the rst study on optical ionization of cluster targets, Wabnitz et al [97] used ultrashort
100 fs pulses of VUV radiation (98 nm wavelength, 12.7 eV photon energy) produced by
a FEL. The ionization potential of Xe atoms is 12.1 eV and thus smaller than the photon
energy used. When a gas composed of single atoms was irradiated by the VUV pulses,
only singly ionized Xe
+
atoms could be detected. However, when the single atoms were
replaced by clusters and the experiment repeated, highly charged Xe ions could be produced,
ranging up to Xe
8+
at a high pulse intensity (2 10
13
Wcm
2
) and cluster size (30 000
atoms per cluster). Along with the highly charged ion states produced, each of the ions
was recorded with high kinetic energies that could be attributed to a Coulomb explosion
behaviour. For clusters of 1500 atoms interacting with VUV pulses of 7 10
13
Wcm
2
,
the average kinetic energy per atom was analysed to be 400 eV or 30 photons. The
question to be answered was what is the mechanism driving this very efcient absorption of
VUV light?
For the interaction of high-intensity visible near infrared femtosecond laser pulses with
clusters, Coulomb explosion is a common phenomenon. Complete disintegration of clusters
occurs for ultrashort pulses of infrared light of intensity 10
16
Wcm
2
[194, 195] and higher.
There the process leading to the high ionization states is very well understood in the framework
of ponderomotive acceleration of electrons to kinetic energies (kinetic energy is on the order
of the ponderomotive potential of the laser) above the ionization potential of the cluster atoms
in a single laser cycle. This enables the high energy electrons (initially ionized by tunnel
ionization) to create more and more free electrons in an avalanche type process. However,
Femtosecond x-ray science 489
since the ponderomotive potential U
p
decreases rapidly with photon energy (U
p
1/( h)
2
),
this process cannot be the one at work in the VUV case. Here, the ponderomotive potential
for the given experiment was smaller than 150 meV, which is much smaller than the ionization
potential of the Xe atoms and thus not sufcient to efciently drive ionization in the same way
as with near-infrared lasers.
In fact, the explanation [98] can be found by considering the fact that in a VUV laser
eld of intensity as high as the one used in the experiment, each atom in the Xenon cluster
is ionized after only 10 fs of the 100 fs laser pulse. A high-density (metal-like) electron gas
is thus created. Due to the presence of the atomic ion cores, the free electrons will scatter
multiple times during their interaction with the laser eld and are thus able to acquire signicant
amounts of energy (known as inverse-bremsstrahlung heating). In this way, the electron gas
can effectively extract energy from the VUV-pulse. As the plasma temperature increases,
electronion collisions can result in the ionization of more tightly bound electrons, leading
to the highly charged Xe atoms observed in the experiment. Using a rate model to describe
the dynamics of the process, the experimental chargestate distributions could be reproduced
numerically with high accuracy [98], conrming the validity of the theoretical approach and
thus our understanding of the process.
4.3. Surfaces
When an extended piece of matter interacts with another particle or multiple particles, it
is naturally always the surface through which the interaction is mediated. This explains the
particular interest of physicists and chemists in the nature and the dynamics of the surface. Our
understanding about surface electronic structure and reactions is important froma fundamental
but also an applied point of view, considering the engineering of catalysts for the chemical and
pharmaceutical industries.
Since x-rays are ideally suited to resolve structures on the atomic scale, they are a versatile
tool for surface analysis. Moreover, x-ray photons can be used for photoelectron spectroscopy
to map electronic structure into the ionization continuum without having electron transitions
to proceed via different intermediate (and possibly only roughly known) bands. Combined
with temporal resolution when using ultrashort pulses of x-rays, even the transiently excited
states and the intermediate local structural changes on the surface during a chemical reaction
can be monitored on the natural time scale of these systems.
Section 4.3.1 presents an experiment in which ultrafast x-rays are used to elucidate the
temporal dynamics of electronic structure on the surface of Ge crystals. The chemically
important dynamics of molecular bonding to a surface will be the subject of section 4.3.2.
4.3.1. Surface electron dynamics. Ultrafast x-ray photoelectron spectroscopy (with an
x-ray probe pulse) can be used to obtain time-resolved information about the occupied and
unoccupied electronic states at the surface of some material. The main advantage of using
a high-photon energy for the probe process is that it is possible to map the local density of
states directly into the continuum. This resolves the issue of usually just being sensitive to
the joint density of states of initial and nal electronic bands. If an ultrashort laser pulse
is used for excitation of the surface, the transiently occupied electronic states can also be
monitored by the x-ray probe pulse. Pioneering work in this area was carried out by Haight
and co-workers [196198].
In one of the early experiments performed by employing high-energy photons as the probe
pulse, the nature and dynamics of the antibonding state of As adsorbed to a Ge(111) surface has
been illuminated [197]. XUV-light at frequencies of 10, 14, 18, and 22 eV was produced by
490 T Pfeifer et al
high-harmonic generation (section 2.1) with a 0.5 mJ, 700 fs, 610 nm fundamental laser pulse
at a 540 Hz repetition rate, stemming from an amplied dye-laser system. Laser pulses of
0.1 mJ derived from the same laser system served as the pump pulses to populate the normally
empty As derived surface band. After excitation, the population of this state was monitored by
scanningthe time-delaybetweenthe pumpandthe probe pulses. The fact that highharmonics at
different photon energies are available facilitated the discrimination between bulk and surface
states in the photoelectron spectrum, since peaks corresponding to surface states usually stay at
a constant ionization potential while peaks associated with bulk states shift as a consequence
of intraband electronic dispersion. For very small time delays, the As surface state peak
at 0.43 eV has a pronounced shoulder, which disappears in the rst few picoseconds after
excitation. Furthermore, the peak experiences an overall shift towards lower electron energies.
This can be explained as follows: at the relevant L-point in k space, the surface state is located
0.2 eV below the conduction band of the bulk (photoemission from which accounts for the
shoulder on the high-energy side). Laser excitation injects electrons into the bulk conduction
band from where they scatter into the surface state. Thermalization quickly redistributes
conduction band electrons into the surface state, accounting for the fast disappearance of the
shoulder. The subsequent shifting of the peak can be explained by the cooling of the hot electron
distribution in the surface state, initially obtained right after the conduction-bandsurface-state
scattering. Monitoring the surface state peak versus time reveals a rapid population (on the
order of a few picoseconds) and a lifetime (decay) of several hundreds of picoseconds, which
is in agreement with expectations if only radiative recombination is considered.
In another experiment it has been possible to measure the recombination dynamics
of the transiently photoexcited surface electronic states in tris (8-hydroxy quinoline)
aluminium [199]. The authors extracted a surface recombination velocity from the results.
Ultrafast tracking of the chargecarrier dynamics on a GaAs surface has been the subject of
the study carried out by Siffalovic et al [200]. In this experiment, core-level photoelectrons
released by femtosecond high-harmonic soft x-ray pulses have been used to measure the
temporal evolution of the electric eld in the surface layer of the semiconductor material after
excitation with visible laser pulses.
4.3.2. Surface photochemistry. It is of fundamental importance in physics and chemistry
to analyse and understand the dynamics of molecules adsorbed to a surface. This interaction
between adsorbate and substrate is at the heart of the catalytic properties of surfaces for
mediating and accelerating particular chemical reactions. It is crucial to understand the
dynamics of excited states of surface adsorbed molecules. From the analysis of this dynamics
we can infer important information about the properties of the adsorbatesubstrate chemical
bond. It is well known that the techniques of valence- and core-level spectroscopy (UPS/XPS)
can be employed to acquire detailed information about the electronic structure and thus the
chemical state of surfaceadsorbate complexes [201].
In an experiment performed by Bauer et al [202,203] molecular oxygen was adsorbed on a
Pt(111) surface. The amplied Ti:sapphire laser systemwas operated at a 1 kHz repetition rate.
A55 fs duration laser pulse (800 nmcentral wavelength) of on-target uence of 10 J cm
2
was
used to excite the adsorbed oxygen. The probe pulse was generated through capillary phase-
matched high-harmonic generation (see section 2.1) at a duration of less than 25 fs. Because
of the extended phase-matched conversion length for the capillary-generated harmonics, the
spectrum consists of only 35 harmonics (23rd31st) at photon energies from 34.5 to 46.5 eV.
Filteringusinga 200 nmthickness Al-lm(for removingthe fundamental light) anda multilayer
mirror for beamsteering and focusing resulted in a further decrease of EUV-bandwidth. Half of
the energy of the EUVpulse arriving at the target was contained in the 29th harmonic (43.5 eV)
Femtosecond x-ray science 491
(a) (b)
Figure 24. Observing transient changes of surface photochemistry. Oxygen adsorbed on Pt can
undergo a charge transfer between different oxidation states of the molecular surfaceadsorbate
complex or photodissociate to yield highly excited oxygen atoms. Both of these processes lead to
a transient peak observed in the photoelectron spectrum (a) after irradiation with an 800 nm pump
and a 43.5 eV XUV harmonic probe pulse. As a function of the time-delay, the amplitude of the
peak is changing indicating an ultrafast surface reaction on a time-scale of 550140 fs. (Reprinted
gure with permission from [202]. Copyright (2001) by the American Physical Society.)
with low contributions of the neighbouring orders. A time-of-ight electron spectrometer was
used to measure the photoelectron energies.
The platinum surface was kept at a temperature of 195

C. In this case, it is known


from the literature (see in [202]) that chemisorbing molecular oxygen prefers two different
congurations on the surface, peroxo (O
2
2
) or superoxo (O

2
). The likelihood for an adsorbed
molecule to be found in either of these congurations depends on the coverage, with the
superoxo state being dominant at high coverage and the peroxo state at low coverage. In the
experiment, a distinct peak at 6 eV below the Fermi level appeared for time delays between
0 and 5 ps (see gure 24(a)). A rapid buildup of this peak on a timescale of a few hundred
femtoseconds could be observed (gure 24(b)). According to earlier UPS studies a peak at
6 eV indicates the presence of either atomic oxygen or peroxo conguration molecular oxygen
adsorbed to the Pt surface while it is absent for the superoxo state. In the study of Bauer et al,
this peak is only transiently visible, disappearing for pumpprobe time delays longer than 5 ps.
So it was possible to monitor the structural change of an adsorbed molecular species after laser
excitation directly in the time domain. By monitoring the photoelectron distribution at the
Fermi level it was shown [203] that the change in molecular bonding structure is preceded by
the thermalization of an initially created hot-electron distribution. It is the time scale of the
energy transfer to the lattice that matches the dynamics of the structural rearrangement of the
oxygen molecules on the surface.
4.4. Solid state
Many materials in the solid phase exhibit an important property: the atoms are arranged in a
regular manner. The unit cell is the smallest part of the solid consisting of one or more atoms
which is repeated in three dimensions according to the lattice structure of the crystal. This
lattice structure gives rise to the characteristic diffraction pattern (narrow reection peaks) that
arises when x-rays are scattered off a crystal. This diffraction pattern not only reveals the
lattice structure but also contains information about the unit cell. One of the main benets of
x-rays is their capability of resolving the structure of solid material. For that reason it is not
astonishing that many experiments using ultrafast x-rays focused on the study of time-resolved
structural dynamics in solids.
492 T Pfeifer et al
Beyond the fact that x-rays are well suited to study the solid phase, this particular state
of matter has intrinsically interesting physical and chemical properties. Since many atoms are
packed into a small space, interactions between particles govern the electronic and structural
behaviour of solids. Processes can be explained by employing statistical methods, since
the typical number of equal particles in a crystal is very large. These statistical processes
give rise to collective phenomena such as ferromagnetism and superconductivity, where
electron correlations play a dominant role. In general, the solid state provides for a large
variety of structural and electronic phase transitions. X-rays, due to their short wavelength
and short optical cycle, are ideally suited to resolve small-scale spatial structures and to
eventually resolve correlated electron dynamics on the natural time-scale of electronic motion.
Extended reviews on the eld of time-resolved spectroscopy in solids can also be found in the
literature [204, 205].
In section 4.4.1 we will describe the experiments that illuminate solid-state dynamics with
atomic spatial and temporal resolution. It will be followed by section 4.4.2 where the same
methods will be used to study an ultrafast structural phase-transition. Section 4.4.3 describes
an experiment aiming at the analysis of Auger decay in solid materials. Finally, section 4.4.4
shows that ultrafast x-rays can be used to learn about the interaction between the electronic
and the lattice structure of solids.
4.4.1. Lattice dynamics
Propagating coherent acoustic phonons. Hard x-rays generated with femtosecond laser-
induced plasma-based sources (section 2.2) allow for monitoring of structural changes on the
natural time scale of nuclear vibrations. In fact, these sources currently represent the easiest
to operate x-ray source that provides both, hard x-ray (1 wavelength) radiation paired
with ultrashort femtosecond pulse duration. The oscillation of atoms (phonons) in any solid
material takes place on the femtosecond to picosecond time-scale, depending on the phonon
momentum. For that reason and since the x-rays have a wavelength that is short enough to
resolve the lattice structure of crystalline solids, they are an exquisite tool for the real-time
observation of solid state lattice dynamics.
It has been possible to monitor spatially-resolved lattice dynamics in crystals for the rst
time in an experiment performed by Rose-Petruck et al [206]. They used a 10 mJ, 30 fs,
800 nm laser pump pulse (20 Hz repetition rate) to heat a solid sample of crystalline GaAs
(4 mm
2
area illuminated by at-top beam prole), which was replaced (moved) after each
laser shot. A part (50 mJ) of the primary 200 mJ laser pulse was focused onto a moving Cu
wire (spot size/source size 25 m) to create the femtosecond probe pulse of hard x-rays, mainly
concentrated within the K

lines (K
1
= 1.5407 , K
2
= 1.5439 ). The produced x-ray ux
could be estimated as 5 10
10
photons/(4sr s).
The primary heating mechanism initiated by the femtosecond 800 nm pump pulse in this
experiment was the excitation of electrons from the valence band to the conduction band of
the GaAs material and the subsequent single and multiphoton absorption of bound and free
carriers. While varying the time delay between pump and x-ray probe pulse, the Bragg peaks
corresponding to (111)-GaAs scattering of the K

photons were modied on a picosecond time-


scale (gure 25). For small delay times, the Bragg peaks become broader while after a fewtens
of picoseconds each of the Bragg peaks develops a secondary peak at an angle of 150 arcs
less than the original peak. For larger and larger delays, the separation between the secondary
and primary Bragg peaks decreases with a characteristic time of 100 ps. These observations
can be interpreted considering the optical excitation of a coherent acoustic phonon wavepacket,
Femtosecond x-ray science 493
250
200
150
100
50
0
P
u
m
p
-
p
r
o
b
e

t
i
m
e

d
e
l
a
y

(
p
s
)

B
(arcsec)
200 100 0 100 200 200 100 0 100 200 200 100 0 100 200
Figure 25. Measuring strain wave propagation in solids with ultrafast hard x-ray pulses from
a laser-induced plasma source. An intense femtosecond infrared pulse creates a localized strain
which propagates into the bulk of the GaAs sample. It is probed by Bragg reection of the x-ray
pulse, which arrives at variable times with respect to the infrared pulse. The positions of the Bragg
peaks shift as a function of time in the measurement (a), which agrees well with a simulation (b)
and an iterative inversion (c) to nd the strain distribution as a function of time from the measured
data. (Reprinted by permission from Macmillan Publishers Ltd; Nature [206]. Copyright (1999).)
launched by the femtosecond laser pump pulse. Anumerical simulation was carried out which
used the following three parameters to t to the data of the experiment: peak strain (which
can be directly extracted from the data), the acoustic velocity and the x-ray absorption length
that was estimated to be 300 nm. Using this model, a 100 ps acoustic pulse moving from the
surface into the crystal with a peak bipolar strain of 0.12% or 3 m could be extracted
from the experimental result. In a following study, the anharmonic damping of coherent
lattice oscillations in a thin lm of germanium could be measured and understood in terms of
four-body elastic dephasing processes [207].
Several other studies on coherent phonon creation by laser excitation and monitoring by
femtosecond x-ray diffraction have been carried out [208210]. The separate contributions of
surface and the bulk of InSb to the temporally diffracted x-ray signal after impulsive phonon
excitation could be observed in the study of Reis et al [208]. Another experiment on InSb
was performed by Synnergren et al [209]. A detail worth mentioning: the setup in their work
used a grating in order to create a spatial chirp of the pump laser pulse on the sample, allowing
the recording of x-ray transients of 200 ps in a single shot (in contrast to scanning the time
delay sequentially). Another very recent work used a kilohertz-repetition-rate femtosecond
laser-plasma x-ray source to study laser-induced oscillatory changes in the superlattice of a
nanostructure [210].
Unit-cell level structural changes. It has been possible to follow the oscillatory behaviour of
lattice vibrations in a laser-excited crystal. The study carried out by Sokolowski-Tinten et al
shed light onto a structural phase transition in bismuth [211]. This work shows that one can
use time-resolved ultrafast x-ray spectroscopy to map out the dynamic atomic displacement
within one unit cell of the crystal. The bismuth structure in the most common state (-arsenic)
is governed by the PeierlsJones mechanism (for references see [211]). Two bismuth atoms
494 T Pfeifer et al
(a) (b)
Figure 26. Observing coherent lattice vibrations in bismuth with femtosecond x-ray pulses. (a)
Geometrical structure factor for x-ray diffraction as a function of the distance d of the two basis
atoms in the unit cell. Vibrational motion of the two atoms with respect to each other was initiated
by a femtosecond 800 nm pump pulse, where the structural change was monitored by a time-
delayed femtosecond x-ray pulse. The transient intensity (b) of the corresponding (222) Bragg
diffraction peak was observed to exhibit long term changes (several picoseconds) as well as short
term oscillations (467 fs). (Reprinted by permission from Macmillan Publishers Ltd; Nature [211].
Copyright (2003).)
are present in the unit cell, one at the atomic sites of a slightly distorted face-centred cubic
lattice and another close to but not at the centre along the body diagonal of the unit cell. It is
known that this equilibrium structure can be modied not only by external parameters such
as pressure or temperature but also by electronic excitation. When the distance between the
two unit-cell atoms is changed by, for instance, laser-induced electronic excitation, the atoms
perform a damped oscillation around the new equilibrium position of the excited structure
(gure 26).
In the experiment [211], femtosecond laser-induced plasma generation was employed to
create ultrashort 4.51 keV x-ray pulses for recording the intensity of the Bragg-peaks both for
the (111) and the (222) reection. The pump pulse was an 800 nm laser pulse with 120 fs
pulse duration, focused to a spot size of 80 m on the 50 nm thick bismuth lm surface.
The authors measured the diffraction efciency of the (111) and the (222) reection as a
function of time delay between the pump and the probe pulse. For a pump pulse uence of
6 mJ cm
2
, a rise of the (222) reection signal is observed in less than a picosecond while the
(111) reection intensity drops on the same timescale. Afterwards, an oscillatory behaviour
of the reection efciency is recorded for (222) which is in antiphase to an oscillation for
the (111) reection. Upon an increase in pump laser uence to 20 mJ cm
2
, the oscillatory
behaviour in the (222) reection pumpprobe trace disappears and only the initial sub-
picosecond increase and a slow decay afterwards (on a timescale of several picoseconds)
remains. The antiphase behaviour of the (111) and (222) signals can be explained by
considering the geometrical structure factors of the (111) and the (222) lattice planes (see
gure 26(a)). When the distance between the two unit-cell bismuth atoms is changed, the
structure factors of (111) and (222) decrease and increase, respectively. Therefore, it can
be stated that an oscillation in the relative bismuth position could be monitored at small
laser uences, representing the rst measurement of intra-unit-cell atomic dynamics by x-
ray diffraction. The oscillation period was 467 fs (as seen in gure 26(b)), corresponding to
a frequency of 2.12 THz. This frequency is signicantly downshifted as compared with the
relevant optical phonon frequency of 2.92 Thz. This downshift can be explained in terms of
Femtosecond x-ray science 495
a change in the crystal potential due to large oscillation amplitudes. It could be calculated that
the nearest-neighbour distance (d
nn
= 0.31 nm) of the Bi atoms is changed by 58%, which
is close to the Lindemann stability limit. When the mean displacement of the atoms reaches
1020%of the lattice constant, a solid-to-liquid phase transition occurs. This could explain the
behaviour at higher laser uences, where the oscillations disappear, possibly due to the local
creation of a liquid phase. This would be further supported by the slow (several picoseconds)
decrease in diffraction intensity that could be indicative of melting.
Besides the experimental efforts, theoretical studies and simulations of dynamical
diffraction off crystals for femtosecond x-ray pulses are carried out [212214] in order to
better understand the nature of the probing process. One has to take into account, for instance,
that the pulse duration of the scattered x-ray pulse can increase signicantly when a large
number of lattice planes contribute to scattering.
4.4.2. Nonthermal melting. When a highly intense laser pulse interacts with a surface, it is the
electrons, not the matrix of atoms, to which energy is primarily transferred in the rst step. This
is because the electron mass mis much lighter than the nuclear mass and can therefore be easily
accelerated in the laser electric eld (average kinetic energy E
kin
U
p
1/mponderomotive
potential). Ahot electron plasma is thus created. If the pulse duration of a heating and a probing
pulse is longer than the time it takes the electrons to transfer their energy to the ion cores (to
reach equilibrium between electronic temperature and lattice temperature) the probing pulse
will merely record the commonly known thermal melting of the material. However, if the pulse
duration of these pulses is short enough to deposit a high amount of energy into the electron
gas before it is able to equilibrate with the atomic matrix, this nonequilibrium situation gives
rise to a newphenomenon called nonthermal melting. In this regime, the matrix behaves much
like a cluster irradiated with a high-intensity laser pulse (see section 4.2) undergoing Coulomb
explosion, since the cohesive action of the electrons is lost at a signicant amount of ionization.
While in thermal melting the local structure of the solid starts changing to liquid around certain
nucleation centres close to the surface (creating a melt front that propagates into the solid with
roughly the speed of sound), the nonthermal case creates a situation in which the electronic
structure of many atoms (even several tens of nanometres belowthe surface) is modied within
a time shorter than a picosecond. This can be explained by fast electrons penetrating deep into
the material where they nally release their energy by ionizing other bound electrons. Details
about the mechanisms for ultrafast laser pulses interacting with solids (including nonthermal
melting) can be found in the literature [215].
The rst direct measurement of nonthermal melting was performed by Siders et al [216].
An intense 100 fs laser pulse (uence 0.5 J cm
2
) of 800 nm at 20 Hz repetition rate was
used as a pump pulse in order to drive the nonthermal melting process on a 160 nm thick
Ge surface, which was moved in order to provide a fresh sample to every impinging laser
pulse. A split-off part of the same laser pulse that served as the pump was used to create
femtosecond pulses of Cu-K

photons by focusing it onto a solid Cu target (section 2.2). This


radiation was used for transient Bragg scattering analysis of the Ge structure. The angular
region around the diffraction angle corresponding to the Ge(111) lattice place was monitored
as the time delay between the pump and the probe pulse was varied (see gure 27). For
negative and very long positive time delays the Bragg peak at the illuminated position of
the Ge surface was identical to the neighbouring positions not hit by the intense pump laser
pulse. However, a few picoseconds after illumination with the pump the Bragg peak intensity
started to decrease. At a delay of 40 ps the decrease in Bragg diffraction peak intensity stops
and for larger delay times the peak signal recovers (gure 27(b)). Along with the recovery,
the Bragg peak is shifted towards smaller diffraction angles (gure 27(a)), indicating local
496 T Pfeifer et al
(a) (b)
Figure 27. Time-resolved measurement of nonthermal melting with ultrafast x-ray pulses. An
intense femtosecond 800 nm pump pulse is used as a driver of nonthermal melting on a Ge surface
while an x-ray pulse is used as a probe pulse at different time delays. (a) Bragg peak as a function
of position across sample and time delay. The position of the impinging pump pulse is marked by
two dashed horizontal lines. The dashed ellipses indicate regions, where a single laser pulse hit the
sample; recovery is almost complete. The transient Bragg peak intensity is shown in (b), where a
rapid drop occurs followed by a slower decline and subsequent recovery. The solid red line was
obtained for the centre of the pumped region, while the dashed black line denotes the intensity at a
point vertically displaced by 0.2 mmfromthe centre. (Reprinted gure with permission from[216].
Copyright (1999) by AAAS.)
expansion of the (111) lattice planes. If one assumes a thermal melting process, the melt
front can be calculated to move at speeds 10
4
ms
1
, which are in excess of the expected
interface velocity by an order of magnitude and surpass the highest possible value by a factor
of 3. Since thermal melting can thus be ruled out this measurement constitutes the rst
observation of nonthermal melting of a solid. The shifting of the Bragg peak to smaller
angles occurs on a time scale that is compatible with a thermal expansion process of the
lattice.
Other experiments on nonthermal melting have also been carried out in the past [217219].
In the work of Sokolowski-Tinten et al [218], a very fast (few hundreds of femtoseconds)
nonthermal melting process was observed. Rischel et al [217] have been concentrating
on organic thin lms in their study. The study of Rousse et al [219] shed light on the
femtosecond nonthermal melting dynamics in semiconductors. Very recent experiments
conducted with 100 fs hard-x-ray pulses from a linac (SPPSSLAC) [89] observed anisotropic
inertial dynamics in the nonthermal melting of silicon [90, 220].
4.4.3. Solid-state inner-shell electron dynamics. One of the rst examples of time-resolved
spectroscopy of inner-shell solid-state electron dynamics has been performed by Shimizu et al
[221]. The objective was to measure the Auger decay rate as it depends on the temperature.
For temperatures up to 200 K, the decay times are larger than 100 ps and thus still accessible
by synchrotron radiation pulses. For higher temperatures, these decay times rapidly decrease
into the single picosecond regime, making them accessible only by using femtosecond pulses.
In the experiment of Shimizu et al , high-harmonics (harmonic orders 1123 (16.637.5 eV))
of a Ti:sapphire 1 kHz amplied laser were employed to create a coreshell vacancy in a
CsBr crystal. There are two competing relaxation mechanisms (gure 28(a)): (a) Auger
decay, where an electron from the Br

4p valence band lls up the Cs


+
5p inner-shell hole by
promoting an electron from the valence band to the conduction band (Cs
+
6s) and (b) Auger-
free luminescence, where a valence-band electron lls the inner-shell vacancy by emitting a
photon. The luminescence signal could be measured as a function of time and temperature
Femtosecond x-ray science 497
0.0
0.2
0.4
0.6
0.8
1.0

0.0
0.2
0.4
0.6
0.8
1.0
I
n
t
e
n
s
i
t
y

(
a
.
u
.
)
8 6 4 2 0
Time (ps)
b) 340K
a) 320K
(a) (b)
Figure 28. Measuring time-dependent Auger electron relaxation in solids (from[221]). (a) Energy
level diagram of CsBr. A femtosecond high-harmonic XUV pulse was used as the pump pulse, to
create an outermost-core-shell vacancy. Subsequent luminescence at two different photon energies
was recorded as a function of time (b) by a difference-frequency-mixing method. While the decay
of the luminescence always has the same time-constant of 5.0 0.5 ps in both cases, the rise
times are different: a pulse-width limited rise-time for the high-energy and a slower rise-time of
950100 fs for the low-energy luminescence signal. (Reprinted gure with permission from[221].
Copyright (2003) by the American Physical Society.)
by frequency downconversion with a femtosecond pulse. It is known that the radiative decay
rate does not depend on temperature and has been previously determined to be 1.1 ns from
synchrotron radiation measurements. Therefore, the temperature dependence of the extracted
transients (see gure 28(b)) was used to extract the temperature-dependent decay rate of the
Auger recombination. As the Auger process proceeds through the Urbach exciton tail states,
the Urbach tail model can be used to t the decay times. This resulted in the measurement
of an extrapolated value of the Auger-decay lifetime of 2.4(+3.8/1.5) fs, in good agreement
with theory.
4.4.4. Electronlattice interaction dynamics. It is the position of the atomic nuclei that make
up the lattice structure of a certain solid-state material. On the other hand, as their name
implies, it is the electrons that give rise to the electronic transport and absorption properties.
For most of the situations it is most likely that both electrons and atoms are not only playing
among themselves but most of the times they are strongly interacting with each other. To study
this interaction, one has to basically observe along two ways: (a) change the structure of the
lattice in different dened ways and look at the modications of the electronic properties and
(b) excite different dened electronic transitions and study the ensuing dynamics of the lattice.
The following question could also arise: if we photoexcite a certain material and we
record some change in the electronic structure, is this process mediated by a reconguration
of the lattice or is the lattice modied after the electronic structure has changed, where the
lattice modication does not signicantly inuence the electronic structure? To answer this
question for the specic case of crystalline VO
2
, in which an insulator to metal transition exists,
Cavalleri et al have conducted a series of measurements, employing both ultrafast optical and
x-ray techniques [222224].
By conducting a femtosecond pumpprobe experiment and measuring the reectivity of
VO
2
lms, they have been able to extract two timescales of reectivity change: a very fast
498 T Pfeifer et al
one (100 fs time constant) and a slower one which is only visible in higher-thickness lms on
the 3 ps time scale [224]. The rst one can be attributed to a very fast (electronic) structural
change whereas the latter is caused by an increase in temperature in the lower layers of the
photo-excited lm, which are slowly heated above the insulator-to-metal critical temperature
after the laser pulse has been absorbed by the sample. It was also observed that the time scale of
the transformation depends on the uence of the femtosecond laser pulse [223]. For excitation
with 50 fs laser pulses, this transformation time varied between several tens of picoseconds
for uences between 5 and 10 mJ cm
2
and 100 fs for uences around 25 mJ cm
2
. For the
very fast transformation times it would be interesting to know whether the structural change
of the lattice is on the same time scale. The authors show from x-ray scattering data (acquired
by employing ultrafast laser-induced plasma x-rays) that this is indeed the case [223] down to
the sub-picosecond pulse durations. The question that remained to be answered, and which is
of course crucial to our understanding of ultrafast phase transitions is, is it only the optically
excited electronic structure that gives rise to the photo-induced phase transition (with the atomic
positions merely responding to it) or is the optical electronic excitation rst of all giving rise
to a lattice structural reorganization that then results in a change of the electronic properties
(metallic instead of insulating)? This question was only recently answered [225] by employing
very short laser pulses. The phase transition was observed to have proceeded via structural
rearrangement.
4.5. Dense plasmas
Dense plasmas are extreme states of matter which are of fundamental importance in particular
for applications such as inertial connement fusion or laboratory astrophysics. Plasmas
efciently reect photons of frequency lower than the plasma frequency. The higher the
density of the plasma the higher the plasma frequency (see equation (31)). For that reason,
optical wavelengths cannot be used for the characterization of very dense plasmas. Since
the laser-induced generation of a plasma can be a very fast process (depending on the pulse
duration of the laser pulse), femtosecond x-ray pulses can be used for resolving the dynamics
of a dense plasma [226, 227]. Very recently it was possible to followthe creation and evolution
of a femtosecond-laser produced plasma on a time-scale of 100 fs with a femtosecond soft
x-ray pulse [227]. The plasma was generated by the action of an intense (3 10
17
Wcm
2
)
60 fs laser pulse (pump) of high temporal contrast (i.e. no prepulse) impinging on a 100 nm
thick foil of polypropylene. The probe pulse was obtained from high-harmonic generation in
argon gas. The transmission of harmonic orders 19 and 21 was recorded as a function of time
delay between the plasma-generating pump pulse and the high-harmonic pulse. Fromthe data,
an initial plasma density of 2.93.510
23
cm
3
was noted since only the 21st but not the 19th
harmonic was transmitted. After 200 fs it dropped to a density below 2.9 10
23
cm
3
, visible
from the onset of transmission for the 19th harmonic. Using a hydrodynamics simulation,
the initial plasma temperature of 220 eV could be extracted from the measurement with an
uncertainty of 15%.
4.6. Nonlinear x-ray optics
The benet of ultrashort pulses is twofold: on the one hand, they come with a duration that
is matched to or faster than molecular dynamics, allowing for time-resolved measurements
of microscopic structural changes of matter. On the other hand, the concentration of a
certain amount of (pulse) energy into a very short time interval creates a high intensity of
electromagnetic radiation. With the invention of the laser it has been possible for the rst
Femtosecond x-ray science 499
time to access intensities of light that are high enough to perform nonlinear mixing of optical
frequencies, for example creating the second harmonic of light in a dielectric sample [228].
It has been shown that nonlinear susceptibilities (responsible for the creation of nonlinear
polarizationof the conversionmedium) are quicklydecreasingwithdecreasingelectromagnetic
wavelength, making observation of nonlinear effects in the x-ray region a difcult endeavour.
Using suitable focusing, it has recently been possible to generate high intensity soft-x-ray
radiation of up to 10
14
Wcm
2
[229] at a spot size of 1 m. A few experiments have been
reported to date, where it has indeed been possible to use these high intensity light elds to
observe a two-photon transition in the XUV/soft-x-ray spectral region [123, 230, 231]. In all
these works, high-harmonic generation (see section 2.1) of laser light was employed to produce
high-intensity femtosecond soft-x-rays.
In the work reported by Papadogiannis et al [230], the nonlinear interaction of XUV-light
with He atoms has been shown to produce a He
+
ion yield that scales quadratically with the
intensity of the XUV pulse. This behaviour is indicative of a two-photon transition. This
nonlinear interaction of XUV light with matter has been employed for the rst autocorrelation
measurement of XUV pulses [120] (section 3.2.1).
The most recent study on nonlinear interactions of (ultrashort) intense soft-x-ray radiation
with matter observed double ionization of He atoms. Nabekawa et al [231] used a 10 Hz
repetition rate, 800 nmwavelength Ti:sapphire amplied laser systemto create high-harmonics
with a 23 fs driver pulse of 70 mJ pulse energy in argon gas. The 27th harmonic (42 eV) is
selected and focused by a multilayer mirror made of siliconcarbide and magnesium. The
intensity of this soft-x-ray pulse was estimated to be 1.7 10
13
Wcm
2
. A time-of-ight
mass spectrometer was used to detect the amount of He
2+
produced as a function of the soft-
x-ray intensity. A slope of 2.00 was found for the double-logarithmic plot of He double-
ionization versus He single-ionization signal. The latter can be considered a meter of the
intensity for a given xed beam radius and He density.
In another very recent work of Sekikawa et al [123], the 9th harmonic of an intense
(3.9 10
14
Wcm
2
) 400 nm femtosecond laser pulse (8.3 or 12 fs, 1 kHz repetition rate)
was used to induce ATI in He atoms. The photon energy of the 9th harmonic is 27.9 eV.
A characteristic photoelectron energy peak at 31.2 eV was observed as a function of time-
delay between two ultrashort soft-x-ray pulses that were produced by two copies of the same
femtosecond laser pulse. By monitoring the total amount of ATI electrons versus time-delay
it was possible to measure the autocorrelation of the single (harmonic) XUV pulse to be on
the order of 1 fs.
5. Conclusion
With the parallel development of several experimental x-ray light sources and measurement
techniques, the eld of ultrafast x-ray science grows and advances at an ultrafast pace.
In this paper, we have reviewed some of the most important sources of ultrafast
(femtosecond) x-ray pulses and some of their most important applications by describing
and referring to past and recent works in the eld. The strength of this research area is the
widespread approach towards covering the extremely large x-ray spectral region from several
tens of electronvolts (soft x-rays, XUV) ranging up to several tens of kiloelectronvolts (hard
x-rays). Truly, it is not possible to nd a particular x-ray source that is the optimal source
for any scientic application in the x-ray region. Each of the presented sources has unique
capabilities in terms of the important parameters such as pulse duration, photon number,
average or peak brightness, spatial and temporal coherence properties, repetition rate and
synchronizability to external pulsed light sources. Each of the given sources, by virtue of its
500 T Pfeifer et al
characteristic properties, is best suited to particular types of experiments while it cannot be
employed for a class of different scientic cases. In the design of a new experiment, it is
of great importance to choose the suitable light source in order to provide the parameters of
interest.
A very important parameter is the duration of the x-ray pulses produced, which we thus
treated with particular emphasis. As for the light sources, there is a multiplicity of techniques
to perform the measurement of x-ray pulse duration. The choice of the pulse characterization
method is based on the photon energy range and pulse duration range. In almost any case, the
availability of an optical (ultrafast) trigger/crosscorrelation pulse is necessary, except when
the x-ray pulse energy is high enough to efciently drive nonlinear lightmatter interaction
by itself. In the latter case, an autocorrelation measurement, employing x-ray light only,
is feasible.
We reviewed a number of important scientic studies that made use of ultrafast x-rays.
As can be seen by the wide variety of applications, it can be judged that ultrafast x-rays are
quickly nding their way into virtually any eld of chemistry and the physical sciences. We
have discussed studies performed on molecules and atoms in the gas phase, cluster dynamics,
surface science, solid state structural and electronic dynamics and nally nonlinear interactions
of x-rays with matter. Although this already represents an impressive list of scientic cases,
manyof the presentedexperiments canbe consideredproof-of-principle-type demonstrations,
with many further applications to follow. The herein described experiments were meant to
highlight the potential of ultrafast x-ray techniques which range from analysis of structural
dynamics on atomic length and time-scales via mapping out of molecular potential energy
curves and fragmentation dynamics to monitoring of electronic dynamics on surfaces and in
the bulk. Our understanding of these and many more fundamental physical processes can be
expected to increase signicantly in the future by making extended use of these and newly
developed techniques. We also showed a way to control and tailor the spectral shape of coherent
soft x-rays. With this at hand, we are about to step out into a new eld devoted to the quantum
control of (core) electron dynamics.
A very exciting future perspective opens up by considering the development of very-
high-brightness sources such as free-electron lasers or possibly high-harmonic amplication
schemes which would deliver an amount of photons on time scales that are small to
the molecular motion. These pulses could be used to acquire the complete structural
information about a certain single molecule with only one x-ray pulse, eliminating the need
for crystallization techniques. Both small and large molecules (such as proteins or other large
biologically relevant molecules) can then be analysed. Due to the shortness of the pulses, both
the ground state (unperturbed) structure and the excited state dynamics can be observed in real
time and with sub-angstrom resolution.
With all this progress, x-rays can, are already and nally will be used as the ultimate probe
for microscopy at atomic resolution in four-dimensional space-time.
Acknowledgments
We would like to thank a number of persons who contributed to the work done in W urzburg
that was presented in this review: F Dimler, A Flettner, M Hafner, R Kemmer, J K onig,
M B Mason, A Paulus, T Sokollik, R Spitzenpfeil, D Walter, U Weichmann, C Winterfeldt
and S Zipfel. Particular thanks go to Professor A M Sergeev, who provided the initial code
for the simulation and Professor R D uren for a substantial part of the vacuum apparatus that
was used. We acknowledge nancial support of the DFG (grant SP 687/1-2), the Austrian
Science Fund (grant F016 (ADLIS)), the European Coherent Control Network (COCOMO),
Femtosecond x-ray science 501
the German-Israeli Cooperation in Ultrafast Laser Technologies (GILCULT) and the Fonds
der chemischen Industrie. TP acknowledges the support of a Feodor Lynen Fellowship of the
Alexander von Humboldt-Foundation.
References
[1] Roentgen W C 1896 Nature 53 274
[2] Schenkel B et al 2003 Opt. Lett. 28 1987
[3] Yamane K et al 2003 Opt. Lett. 28 2258
[4] Sekikawa T, Yamazaki T, Nabekawa Y and Watanabe S 2002 J. Opt. Soc. Am. B 19 1941
[5] Drescher M et al 2002 Nature 419 803
[6] Bressler C and Chergui M 2004 Chem. Rev. 104 1781
[7] McPherson A et al 1987 J. Opt. Soc. Am. B 4 595
[8] Ferray M et al 1988 J. Phys. B 21 L31
[9] Corkum P B 1993 Phys. Rev. Lett. 71 1994
[10] Kulander KC, Schafer KJ and Krause J L1993 Proc. Work-shop on Super-Intense Laser AtomPhysics (SILAP)
III ed B Piraux (New York: Plenum) pp 95110
[11] Lewenstein M et al 1994 Phys. Rev. A 49 2117
[12] Antoine P et al 1996 Phys. Rev. A 53 1725
[13] Weichmann U 2001 Dissertation Universit at W urzburg
[14] Protopapas M, Keitel C H and Knight P L 1997 Rep. Prog. Phys. 60 389
[15] Brabec T and Krausz F 2000 Rev. Mod. Phys. 72 545
[16] Keldysh L V 1965 Sov. Phys.JETP 20 1307
[17] Ammosov M V, Delone N B and Krainov V P 1986 Sov. Phys.JETP 64 1191
[18] Augst S, Meyerhofer D D, Strickland D and Chin S L 1991 J. Opt. Soc. Am. B 8 858
[19] Krause J L, Schafer K J and Kulander K C 1992 Phys. Rev. Lett. 68 3535
[20] Chen S Y, Maksimchuk A and Umstadter D 1998 Nature 396 653
[21] Fittinghoff D N, Bolton P R, Chang B and Kulander K C 1992 Phys. Rev. Lett. 69 2642
[22] Walker B et al 1994 Phys. Rev. Lett. 73 1227
[23] Agostini P et al 1979 Phys. Rev. Lett. 42 1127
[24] Paulus G G et al 1994 Phys. Rev. Lett. 72 2851
[25] Schafer K J, Yang B, DiMauro L F and Kulander K C 1993 Phys. Rev. Lett. 70 1599
[26] Sali` eres P et al 2001 Science 292 902
[27] Bartels R et al 2000 Nature 406 164
[28] Christov I P, Bartels R, Kapteyn H C and Murnane M M 2001 Phys. Rev. Lett. 86 5458
[29] Gaarde M B et al 1999 Phys. Rev. A 59 1367
[30] Balcou P, Dederichs A S, Gaarde M B and LHuillier A 1999 J. Phys. B 32 2973
[31] Lewenstein M, Sali` eres P and LHuillier A 1995 Phys. Rev. A 52 4747
[32] Bellini M et al 1998 Phys. Rev. Lett. 81 297
[33] Lee D G, Kim J H, Hong K H and Nam C H 2001 Phys. Rev. Lett. 8724 243902
[34] Kim H T et al 2003 Phys. Rev. A 67 051801
[35] Wollenhaupt M et al 2002 Phys. Rev. Lett. 89 173001
[36] Tewari S P and Agarwal G S 1986 Phys. Rev. Lett. 56 1811
[37] Boller K J, Imamoglu A and Harris S E 1991 Phys. Rev. Lett. 66 2593
[38] Siegman A E 1986 Lasers 1st edn (Sausalito: University Science Books)
[39] Rundquist A et al 1998 Science 280 1412
[40] Schn urer M et al 1998 Appl. Phys. B 67 263
[41] Durfee C G III et al 1999 Phys. Rev. Lett. 83 2187
[42] Tamaki Y, Nagata Y, Obara M and Midorikawa K 1999 Phys. Rev. A 59 4041
[43] Marcatili E A J and Schmeltzer R A 1964 Bell Syst. Tech. J. 43 1783
[44] Sali` eres P, LHuillier A and Lewenstein M 1995 Phys. Rev. Lett. 74 3776
[45] Nisoli M et al 2002 Phys. Rev. Lett. 88 033902
[46] Lawrence Berkeley Laboratory 2001 Atomic Scattering Factors http://www-cxro.lbl.gov/opticalconstants
[47] Bartels R A et al 2002 Science 297 376
[48] Scrinzi A, Geissler M and Brabec T 1999 Phys. Rev. Lett. 83 706
[49] Geissler M, Tempea G and Brabec T 2000 Phys. Rev. A 62
[50] Seres E, Seres J, Krausz F and Spielmann C 2004 Phys. Rev. Lett. 92 163002
502 T Pfeifer et al
[51] Seres J et al 2005 Nature 433 596
[52] Zhou J et al 1996 Phys. Rev. Lett. 76 752
[53] Christov I P et al 1996 Phys. Rev. Lett. 77 1743
[54] Spielmann C et al 1997 Science 278 661
[55] Kan C, Burnett N H, Capjack C E and Rankin R 1997 Phys. Rev. Lett. 79 2971
[56] Kakehata M et al 2000 Appl. Phys. B 70 219
[57] Nisoli M et al 2000 Appl. Phys. B 70 215
[58] Chessa P et al 1999 Phys. Rev. Lett. 82 552
[59] Tamaki Y et al 1999 Phys. Rev. Lett. 82 1422
[60] Tosa V, Takahashi E, Nabekawa Y and Midorikawa K 2003 Phys. Rev. A 67 063817
[61] Constant E et al 1999 Phys. Rev. Lett. 82 1668
[62] Schn urer M et al 1999 Phys. Rev. Lett. 83 722
[63] Schn urer M et al 2000 Appl. Phys. B 70 S227
[64] Zeitoun P et al 2004 Nature 431 426
[65] Murnane M M, Kapteyn H C, Rosen M D and Falcone R W 1991 Science 251 531
[66] Rousse A et al 1994 Phys. Rev. E 50 2200
[67] Jiang Y, Lee T and Rose-Petruck C G 2003 J. Opt. Soc. Am. B 20 229
[68] Giulietti D and Gizzi L A 1998 Riv. Nuovo Cimento 21 1
[69] Schwoerer H 2004 Femtosecond Technol. Tech. Med. Appl. 96 235
[70] McNaught S J, Fan J, Parra E and Milchberg H M 2001 Appl. Phys. Lett. 79 4100
[71] D usterer S et al 2001 Appl. Phys. B 73 693
[72] Flettner A et al 2003 Appl. Phys. B 77 747
[73] Pfeifer T, Walter D, Gerber G, Emelin M Yu, Ryabikin M Yu, Chernobrovtseva M D and Sergeev A M 2004
Phys. Rev. A 70 013805
[74] Strelkov V V, Platonenko V T and Becker A 2005 Phys. Rev. A 71 053808
[75] Rousse A et al 2004 Phys. Rev. Lett. 93 135005
[76] van Wonthergem B M and Rentzepis P M 1990 Proc. SPIE 1204 784
[77] Tomov I V, Chen P and Rentzepis P M 1995 Rev. Sci. Instrum. 66 5214
[78] Chen P, Tomov I V and Rentzepis P M 1996 J. Chem. Phys. 104 10001
[79] Egbert A, Chichkov B N and Ostendorf A 2001 Europhys. Lett. 56 228
[80] Egbert A et al 2002 Appl. Phys. Lett. 81 2328
[81] Hinze U, Egbert A, Chichkov B and Eidmann K 2004 Opt. Lett. 29 2079
[82] Oulianov D A, Tornov I V, Dvornikov A S and Rentzepis P M 2002 Proc. National Acad. Sci. USA vol 99,
p 12556
[83] Tomov I V and Rentzepis P M 2004 Chem. Phys. 299 203
[84] Schoenlein R W et al 2000 Science 287 2237
[85] Chin A H et al 1999 Phys. Rev. Lett. 83 336
[86] Decamp M F et al 2001 Nature 413 825
[87] Schoenlein R W et al 1996 Science 274 236
[88] Treacy E B 1969 IEEE J. Quantum Electron. QE 5 454
[89] Bentson L et al 2003 Nucl. Instrum. Methods Phys. Res. A 507 205
[90] Lindenberg A M et al 2005 Science 308 392
[91] Cavalieri A L et al 2005 Phys. Rev. Lett. 94 114801
[92] Materlik G and Tschentscher T 2001 TESLA Technical Design Report: Part V, The X-Ray Free Electron Laser
(Hamburg: DESY)
[93] Andruszkow J et al 2000 Phys. Rev. Lett. 85 3825
[94] Yu L H et al 2000 Science 289 932
[95] Yu L H et al 2003 Phys. Rev. Lett. 91 074801
[96] Ayvazyan V et al 2002 Phys. Rev. Lett. 88 104802
[97] Wabnitz H et al 2002 Nature 420 482
[98] Santra R and Greene C H 2003 Phys. Rev. Lett. 91 233401
[99] Thurston R N, Heritage J P, Weiner A M and Tomlinson W J 1986 IEEE J. Quantum Electron. 22 682
[100] Weiner A M, Heritage J P and Kirschner E M 1988 J. Opt. Soc. Am. B 5 1563
[101] Brixner T and Gerber G 2001 Opt. Lett. 26 557
[102] Feurer T, Vaughan J C, Koehl R M and Nelson K A 2002 Opt. Lett. 27 652
[103] Judson R S and Rabitz H 1992 Phys. Rev. Lett. 68 1500
[104] Bardeen C J et al 1997 Chem. Phys. Lett. 280 151
[105] Assion A et al 1998 Science 282 919
Femtosecond x-ray science 503
[106] Schelev M Y, Richards M C and Alcock A J 1971 Appl. Phys. Lett. 18 354
[107] Bradley D J, Liddy B and Sleat W E 1971 Opt. Commun. 2 391
[108] Naylor G A et al 2001 Meas. Sci. Technol. 12 1858
[109] Liu J Y et al 2003 Appl. Phys. Lett. 82 3553
[110] Pelletier J F, Chaker M and Kieffer J C 1996 Opt. Lett. 21 1040
[111] Baltu ska A, Pshenichnikov M S and Wiersma D A 1998 Opt. Lett. 23 1474
[112] Gallmann L et al 1999 Opt. Lett. 24 1314
[113] Morita R et al 2002 Meas. Sci. Technol. 13 1710
[114] DeFonzo A P, Jarwala M and Lutz C 1987 Appl. Phys. Lett. 50 1155
[115] Wu Q and Zhang X C 1995 Appl. Phys. Lett. 67 3523
[116] Lepetit L, Cheriaux G and Joffre M J. Opt. Soc. Am. B 12 2467
[117] Trebino R and Kane D J 1993 J. Opt. Soc. Am. A 10 1101
[118] Kane D J and Trebino R 1993 Opt. Lett. 18 823
[119] Iaconis C and Walmsley I A 1998 Opt. Lett. 23 792
[120] Kobayashi Y, Sekikawa T, Nabekawa Y and Watanabe S 1998 Opt. Lett. 23 646
[121] Tzallas P et al 2003 Nature 426 267
[122] Nikolopoulos L A A et al 2005 Phys. Rev. Lett. 94 113905
[123] Sekikawa T, Kosuge A, Kanai T and Watanabe S 2004 Nature 432 605
[124] Linden S, Giessen H and Kuhl J 1998 Phys. Status Solidi B 206 119
[125] Sekikawa T, Katsura T, Miura S and Watanabe S 2002 Phys. Rev. Lett. 88 193902
[126] Sekikawa T, Kanai T and Watanabe S 2003 Phys. Rev. Lett. 91 103902
[127] Schins J M et al 1996 J. Opt. Soc. Am. B 13 197
[128] Glover T E, Schoenlein R W, Chin A H and Shank C V 1996 Phys. Rev. Lett. 76 2468
[129] Mauritsson J et al 2004 Phys. Rev. A 70 021801
[130] Paul P M et al 2001 Science 292 1689
[131] Mairesse Y et al 2003 Science 302 1540
[132] Drescher M et al 2001 Science 291 1923
[133] Hentschel M et al 2001 Nature 414 509
[134] Borrmann G and Hildebrandt G 1956 Z. Naturforsch. A 11 585
[135] Batterman W B and Cole H 1964 Rev. Mod. Phys. 36 681
[136] Nazarkin A et al 2004 Phys. Rev. A 69 061803
[137] Oguri K, Nishikawa T, Ozaki T and Nakano H 2004 Opt. Lett. 29 1279
[138] Bartels R et al 2000 Nature 406 164
[139] Bartels R et al 2001 Chem. Phys. 267 277
[140] Reitze D H et al 2004 Opt. Lett. 29 86
[141] Feurer T 1999 Appl. Phys. B 68 55
[142] Wieland M et al 2002 Appl. Phys. Lett. 81 2520
[143] Pfeifer T et al 2005 Appl. Phys. B 80 277
[144] Pfeifer T, Weichmann U, Zipfel S and Gerber G 2003 J. Mod. Opt. 50 705
[145] Baumert T et al 1997 Appl. Phys. B 65 779
[146] Pfeifer T et al 2005 Opt. Lett. 30 1497
[147] Larsson J et al 2002 Appl. Phys. A 75 467
[148] Neutze R et al 2001 Phys. Rev. Lett. 8719 195508
[149] Plech A et al 2004 Phys. Rev. Lett. 92 125505
[150] Techert S, Schotte F and Wulff M 2001 Phys. Rev. Lett. 86 2030
[151] Collet E et al 2003 Science 300 612
[152] Glover T E et al 2003 Phys. Rev. Lett. 90 236102
[153] Johnson S L et al 2003 Phys. Rev. Lett. 91 157403
[154] Plech A et al 2003 Europhys. Lett. 61 762
[155] Plech A et al 2004 Phys. Rev. B 70 195423
[156] Schotte F et al 2003 Science 300 1944
[157] Adams B W, Decamp M F, Dufresne E M and Reis D A 2002 Rev. Sci. Instrum. 73 4150
[158] Chen L X et al 2001 Science 292 262
[159] Saes M et al 2003 Phys. Rev. Lett. 90 047403
[160] Williamson J C et al 1997 Nature 386 159
[161] Ruan C Y et al 2001 Proc. National Acad. Sci. USA vol 98, p 7117
[162] Siwick B J, Dwyer J R, Jordan R E and Miller R J D 2002 J. Appl. Phys. 92 1643
[163] Siwick B J, Dwyer J R, Jordan R E and Miller R J D 2003 Science 302 1382
504 T Pfeifer et al
[164] Ruan C Y et al 2004 Science 304 80
[165] Ruan C Y et al 2004 Proc. National Acad. Sci. USA vol 101, p 1123
[166] Srinivasan R et al 2005 Science 307 558
[167] Niikura H et al 2002 Nature 417 917
[168] Niikura H et al 2003 Nature 421 826
[169] Itatani J et al 2004 Nature 432 867
[170] Niikura H, Villeneuve D M and Corkum P B 2005 Phys. Rev. Lett. 94 083003
[171] Niikura H, Legare F, Villeneuve D M and Corkum P B 2005 J. Mod. Opt. 52 453
[172] Tannor D J 2003 Introduction to Quantum Mechanics: A Time-Dependent Perspective (Sausalito: University
Science Books)
[173] Baumert T, Grosser M, Thalweiser R and Gerber G 1991 Phys. Rev. Lett. 67 3753
[174] Zewail A H 1994 Femtochemistry vol 1+2 (Singapore: World Scientic)
[175] Nugent-Glandorf L et al 2001 Phys. Rev. Lett. 87 193002
[176] Nugent-Glandorf L et al 2002 Rev. Sci. Instrum. 73 1875
[177] Tsilimis G, Kutzner J and Zacharias H 2003 Appl. Phys. A 76 743
[178] Tsilimis G, Benesch C, Kutzner J and Zacharias H 2003 J. Opt. Soc. Am. B 20 246
[179] Siffalovic P et al 2001 Rev. Sci. Instrum. 72 30
[180] Drescher M, Siffalovic P, Spieweck A and Heinzmann U 2002 J. Electron. Spectrosc. Relat. Phenom.
127 103
[181] Kienberger R et al 2004 Nature 427 817
[182] Agostini P and DiMauro L F 2004 Rep. Prog. Phys. 67 813
[183] Zeidler D, Hornung T, Proch D and Motzkus M 2000 Appl. Phys. B 70 S125
[184] Eickemeyer F et al 2000 Opt. Lett. 25 1472
[185] Hacker M et al 2003 Appl. Phys. B 76 711
[186] Pfeifer T 2004 Dissertation Universit at W urzburg
[187] Eberhardt W et al 1983 Phys. Rev. Lett. 50 1038
[188] Levis R J, Menkir G M and Rabitz H 2001 Science 292 709
[189] Vajda S et al 2001 Chem. Phys. 267 231
[190] Mochiji K et al 1994 J. Vac. Sci. Technol. A 12 216
[191] Peterka D S, Ahmed M, Ng C Y and Suits A G 1999 Chem. Phys. Lett. 312 108
[192] Ono M and Mitsuke K 2002 Chem. Phys. Lett. 366 595
[193] Ono M and Mitsuke K 2003 Chem. Phys. Lett. 379 248
[194] Ditmire T et al 1996 Phys. Rev. A 53 3379
[195] Ditmire T et al 1997 Nature 386 54
[196] Baeumler M and Haight R 1991 Phys. Rev. Lett. 67 1153
[197] Haight R and Peale D R 1993 Phys. Rev. Lett. 70 3979
[198] Haight R 1995 Surf. Sci. Rep. 21 277
[199] Probst M and Haight R 1997 Appl. Phys. Lett. 71 202
[200] Siffalovic P, Drescher M and Heinzmann U 2002 Europhys. Lett. 60 924
[201] Feuerbach B, Fitton B and Willis R F (ed) 1978 Photoemission and the Electronic Properties of Surfaces
(New York: Wiley)
[202] Bauer M et al 2001 Phys. Rev. Lett. 8702 025501
[203] Lei C et al 2002 Phys. Rev. B 66 245420
[204] Rousse A, Rischel C and Gauthier J C 2001 Rev. Mod. Phys. 73 17
[205] Decamp M F et al 2005 J. Synchrotron Radiat. 12 177
[206] Rose-Petruck C et al 1999 Nature 398 310
[207] Cavalleri A et al 2000 Phys. Rev. Lett. 85 586
[208] Reis D A et al 2001 Phys. Rev. Lett. 86 3072
[209] Synnergren O et al 2002 Appl. Phys. Lett. 80 3727
[210] Bargheer M et al 2004 Science 306 1771
[211] Sokolowski-Tinten K et al 2003 Nature 422 287
[212] Wark J S and He H 1994 Laser Part. Beams 12 507
[213] Tomov I V, Chen P and Rentzepis P M 1998 J. Appl. Phys. 83 5546
[214] Wark J S and Lee R W 1999 J. Appl. Crystallogr. 32 692
[215] Bennemann K H 2004 J. Phys.: Condens. Matter 16 R995
[216] Siders C W et al 1999 Science 286 1340
[217] Rischel C et al 1997 Nature 390 490
[218] Sokolowski-Tinten K et al 2001 Phys. Rev. Lett. 8722 225701
Femtosecond x-ray science 505
[219] Rousse A et al 2001 Nature 410 65
[220] Gaffney K J et al 2005 Phys. Rev. Lett. 95 125701
[221] Shimizu T et al 2003 Phys. Rev. Lett. 91 017401
[222] Cavalleri A et al 2001 Phys. Rev. Lett. 8723 237401
[223] Cavalleri A and Schoenlein R W 2004 Top. Appl. Phys. 92 309
[224] Cavalleri A et al 2004 Phys. Rev. B 69 153106
[225] Cavalleri A et al 2004 Phys. Rev. B 70 161102
[226] Theobald W, Hassner R, Wulker C and Sauerbrey R 1996 Phys. Rev. Lett. 77 298
[227] Dobosz S et al 2005 Phys. Rev. Lett. 95 025001
[228] Franken P A, Weinreich G, Peters C W and Hill A E 1961 Phys. Rev. Lett. 7 118
[229] Mashiko H, Suda A and Midorikawa K 2004 Opt. Lett. 29 1927
[230] Papadogiannis N A et al 2003 Phys. Rev. Lett. 90 133902
[231] Nabekawa Y, Hasegawa H, Takahashi E J and Midorikawa K 2005 Phys. Rev. Lett. 94 043001

Você também pode gostar