Paper C2: Laser Science and Quantum Information Processing -

Laser Physics
Prof. Simon Hooker
Michaelmas Term 2012
ii
Contents
1 Basic Laser Physics 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Recommended texts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Lineshapes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3.1 Homogeneous broadening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.2 Inhomogeneous broadening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3.3 Comparison of homogeneous and inhomogeneous transitions . . . . . . . . . . . . . . . . . 4
1.4 The optical gain cross-section . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4.1 Laser rate equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Gain saturation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5.1 homogeneous broadening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5.2 Inhomogeneously broadened transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Threshold behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6.1 Homogeneously broadened laser transitions . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6.2 Inhomogeneously broadened systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2 Solid State Laser Materials 15
2.1 General considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.1.1 Radiative transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.2 Non-radiative transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.3 Line broadening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Trivalent rare earths, 4f
n
4f
n
transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.1 Energy level structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.2 Transition linewidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.3 Nd:YAG laser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.4 Other crystalline hosts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.5 Nd:Glass laser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.6 Erbium lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3 Trivalent iron group, 3d
n
3d
n
transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.1 Energy level structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.2 The ruby laser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3.3 Alexandrite laser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.3.4 Ti:sapphire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3.5 Host materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
iii
iv CONTENTS
3 Q-Switching 43
3.1 Laser Spiking and relaxation oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.1.1 Rate equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.1.2 Cavity lifetime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.1.3 Numerical analysis of laser spiking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.1.4 Analysis of relaxation oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.2 Q-switching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2.1 Techniques for Q-switching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2.2 Analysis of Q-switching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4 Modelocking 57
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 General ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2.1 Simple analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.2.2 Further general comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3 Active modelocking techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3.1 AM modelocking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3.2 Synchronous pumping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.3.3 Pulse duration of actively modelocked, homogeneously broadened lasers . . . . . . . . . . 66
4.4 Passive modelocking techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.4.1 Saturable absorbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.4.2 Kerr lens modelocking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.5 Examples of modelocked lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.5.1 CW modelocking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.5.2 Pulsed modelocking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5 Ultrafast Lasers 71
5.1 Propagation in dispersive media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.1.1 Propagation through an arbitrary system . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.1.2 Propagation of Gaussian pulses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.1.3 Nonlinear eects: self-phase-modulation and the B-integral . . . . . . . . . . . . . . . . . 76
5.2 Dispersion control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2.1 The grating pair . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2.2 The prism pair . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.2.3 Introduction of positive GDD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2.4 Chirped mirrors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.3 TW and PW laser systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.3.1 Regenerative and multipass ampliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.3.2 Multipass ampliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.3.3 Low-energy, TW laser systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.3.4 High-energy, PW laser system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.4 Applications of ultrafast lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6 Semiconductor & Fibre lasers 89
6.1 Review of semiconductor physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.1.1 Distribution over energy levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
6.1.2 Doped semiconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
6.2 Radiative transitions in semiconductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2.1 Joint density of states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
CONTENTS v
6.2.2 Density of possible transitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.3 The gain coecient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.2.4 Condition for gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6.2.5 Generating gain: the injection density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.3 Homojunction diode lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.3.1 Structure of homojunction diode lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.4 Double heterostructure lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.4.1 Forming heterostructures: Material properties . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.5 Geometries of heterostructure diode lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.6 Output characteristics of diode lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.7 Fibre lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.7.1 Pumping congurations and bre materials . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.7.2 An example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.7.3 Erbium-doped bre lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7 Precision frequency control 109
7.1 Single-mode operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.1.1 Short cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.1.2 Intra-cavity etalon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.1.3 Ring resonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.2 Output linewidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.2.1 Theoretical limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.2.2 Practical limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
7.3 Frequency locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.3.1 Locking to atomic or molecular transitions . . . . . . . . . . . . . . . . . . . . . . . . . . 116
7.3.2 Locking to an external cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
vi CONTENTS
Lecture 1
Basic Laser Physics
1.1 Introduction
In these lectures we build on the basic laser physics covered last year in Paper BIII. All of the material covered
in those lectures will be required for this course, and hence you should revise all of that work.
I would be very grateful if you were to bring to my attention any errors or unclear passages
(email: [email protected]).
1.2 Recommended texts
The lectures and notes should give you a good base from which to start your study of the subject. However,
you will need to do some further reading. The following books are at about the right level, and contain sections
on almost everything that we will cover:
1. Laser Physics, Simon Hooker & Colin Webb, Oxford University Press.
2. Principles of Lasers, Orazio Svelto, fourth edition, Plenum Press.
3. Lasers and Electro-Optics: Fundamentals and Engineering, Christopher Davies Cambridge University
Press.
4. Laser Fundamentals, William Silfvast, Cambridge University Press.
You may also want to consult:
5. Lasers, Anthony Siegman, University Science Books. This book approaches laser physics from more of
an engineering viewpoint, but contains a wealth of information. It is particularly strong on cavity modes
and Gaussian beams.
1.3 Lineshapes
As we saw on the BIII course, a variety of mechanisms can (and will) lead to broadening of the frequency spec-
trum of an optical transition. These mechanisms can be classied as being homogeneous or inhomogeneous.
The key property of homogeneous broadening mechanisms is that the broadening mechanism aects all atoms
in the sample in the same way. As a consequence all atoms in the same energy level interact with radiation of
frequency with the same strength. In contrast, inhomogeneous broadening arises from processes which alter
1
2 LECTURE 1. BASIC LASER PHYSICS
the frequency of an atomic transition by an amount that depends on some property of the atom, such as its
velocity or position. In this case, the atoms may be considered to belong to dierent classes, depending on
the frequency the frequency of the transition. The strength with which an atom will interact with radiation of
frequency then depends on the class to which it belongs.
1.3.1 Homogeneous broadening
Homogeneous line broadening was discussed in detail as part of the BIII course. Examples include: lifetime,
or natural, broadening; pressure broadening; and phonon broadening. In each case it was found that the
spontaneous emission from a sample of atoms in some energy level E
2
on a transition to a lower level E
1
gave
rise to a spectrum proportional to g
H
(
0
), where
0
is the central frequency of the transition,
0
= E
2
E
1
.
The function g
H
(
0
) is normalized
1
such that
_

0
g
H
(
0
)d = 1. For example, in the case of lifetime
broadening we nd that g
H
(
0
) is a Lorentzian function:
g
H
(
0
) =
1

/2
(
0
)
2
+ (/2)
2
, (1.1)
where is the full-width at half maximum of the lineshape. This is equal to the sum of the contributions from
the upper and lower levels:
=
1

rad
2
+
1

rad
1
(1.2)
where
rad
i
is the radiative lifetime of level i.
1.3.2 Inhomogeneous broadening
Inhomogeneous broadening mechanisms cause the transition frequency of dierent atoms to be shifted by dif-
ferent amounts. Such broadening mechanisms are said to be inhomogeneous.
Inhomogeneous broadening will always be associated with some homogeneous broadening since natural
broadening is always present, and the atoms may also be subject to other homogeneous broadening mechanisms
such as phonon broadening. The transition lineshape observed from a sample subjected to both homogeneous
and inhomogeneous broadening processes is given by the convolution of the homogeneous and inhomogeneous
lineshapes. It should be clear, however, that when the frequency width of the inhomogeneous distribution is
much greater than that of the homogeneous distribution the observed transition lineshape will simply be the
inhomogeneous lineshape.
Doppler broadening
For gas lasers the most important line broadening mechanism is usually Doppler broadening, which arises
from a combination of the Doppler eect and the thermal motion of the atoms: the observed frequency
emitted by atoms moving towards the observer with velocity v
z
will be
0
= (v
z
/c)
0
, and hence any
distribution of velocities will give rise to a distribution of observed frequencies. If the distribution is Maxwellian
then the proportion P(v
z
)dv
z
of atoms with velocities in the range v
z
to v
z
+dv
z
is given by :
P(v
z
)dv
z
=
_
M
2k
B
T
exp
_

Mv
2
z
2k
B
T
_
dv
z
, (1.3)
1
Actually the function g
H
(x) is normalized such that

g
H
(x)dx = 1. In eqn (1.1) the argument of g
H
(x) extends only to

0
. However, since the linewidth is always very small compared to
0
, extending the lower limit to will have negligible eect
on the integral.
1.3. LINESHAPES 3

0

g
D
(

0
)

D
(b)
v
v
z
spectrometer
0 v
z
P(v
z
)
(a)
z
Figure 1.1: The formation of an inhomogeneously broadened spectral line by Doppler broadened gas atoms.
Radiation emitted by the atoms in the z-direction is detected by the spectrometer. The distribution of atomic
velocities in the z-direction (a) leads, via the Doppler eect, to a distribution of frequencies measured by the
spectrometer, and hence broadening of the spectral line (b).
where M is the mass of each atom, and k
B
the Boltzmann constant. The observed lineshape g
D
(
0
) is then
found by making the substitution v
z
/c = (
0
)/
0
, to yield
g
D
(
0
)d =
_
M
2k
B
T
exp
_
Mc
2
2k
B
T
_

0

0
_
2
_
c

0
d. (1.4)
This is a Gaussian distribution with a full-width at half maximum, in this context known as the Doppler
width
D
, of

D
=

8 ln 2
_
k
B
T
Mc
2

0
. (1.5)
It is often convenient to re-write the lineshape in terms of
D
, i.e.
g
D
(
0
) =
2

D
_
ln 2

exp
_

_

0
(
D
/2)
_
2
ln 2
_
. (1.6)
This lineshape is normalized so that
_

g
D
(x)dx = 1.
4 LECTURE 1. BASIC LASER PHYSICS
Broadening in amorphous solids
In a number of important laser systems the lasing species are ions doped into a solid. If the solid is a good quality
crystal, the laser transition will be broadened homogeneously by natural and phonon broadening. However if
the solid is non-uniform, ions in dierent regions will experience dierent local environments. Of particular
importance is the local value of the strain of the crystal lattice since this aects the electric eld experienced
by the active ion, which in turn aects the energy levels of the ion through the Stark eect. Other aspects of
the local environment which can aect the transition frequencies of an active ion are the presence of impurity
ions, or variations in the orientation of the crystal lattice. Since such eects change the centre frequency of
the active ions according to their location in the medium, this contribution to the broadening of the frequency
response of a macroscopic sample is inhomogeneous.
These eects are particularly important for ions doped in glasses, such as the Nd
3+
ions in the Nd:glass
laser, since in a glassy material the local environment varies signicantly with position, leading to substantial
inhomogeneous broadening. Very often the distribution of centre frequencies is found to follow a Normal, i.e.
Gaussian, distribution, in which case the transition lineshape is also Gaussian.
Treatment of inhomogeneous broadening
We can develop a mathematical treatment of inhomogeneous broadening by considering that atoms with dierent
centre frequencies behave as if they were dierent types of atom. The corollary of this is that atoms with
similar centre frequencies are of the same type, or class. The frequency range over which each class can
interact appreciably is determined by the (smaller) degree of homogeneous broadening experienced by the class.
For example, the strength with which the class of atoms with centre frequency
c
interacts with radiation of
frequency is determined by the homogeneous optical cross-section
21
(
c
). Note that the strength with
which each class of atoms interacts with radiation of frequency depends on the detuning of from
c
, not
on the detuning from
0
, the average centre frequency of the entire inhomogeneous lineshape.
In treating the interaction of inhomogeneously broadened atoms with radiation we can treat each class
separately. We can dene the number density of atoms in the upper and lower levels with centre frequencies
between
c
and
c
+
c
as N
2
(
c
)
c
and N
1
(
c
)
c
respectively. For example, suppose inhomogeneously
broadened atoms interact with radiation of total intensity I in a narrow band near
L
. The rate equation for
atoms in the upper level would then be of the form:
dN
2
(
c
)
dt
= R
2
(
c
) N

(
c
)
21
(
L

c
)
I

N
2
(
c
)

2
+. . . , (1.7)
where R
2
(
c
) is the pump rate of atoms of class
c
into the upper level, N

(
c
) = N
2
(
c
) (g
2
/g
1
)N
1
(
c
)
is the population inversion density for atoms with a centre frequency
c
, and
2
is the uorescence lifetime of
the upper level (assumed to be the same for all classes). In principle the rate equations for each class can then
be solved separately, and the results integrated over
c
to give the total population densities in the upper and
lower levels.
1.3.3 Comparison of homogeneously and inhomogeneously broadened laser tran-
sitions
Imagine passing a probe beam of narrow-band, tunable radiation through a sample of atoms, as shown schemat-
ically in Figure 1.2. If we tune the frequency of the probe beam in the region around one of the absorption
transitions of the atoms, we can record the uorescence spectrum that is, the spectrum of radiation emitted
spontaneously from the excited level
2
.
2
Note that it is important to ensure that we record the spectrum of spontaneous emission rather than scattered radiation. In
principle this can be achieved by using a pulsed probe beam and turning the detector on gating it a short time after the
1.3. LINESHAPES 5
probe
laser
spectrometer

0
f
l
u
o
r
e
s
c
e
n
c
e

s
i
g
n
a
l
homogeneous
lineshape
probe frequency

0
f
l
u
o
r
e
s
c
e
n
c
e

s
i
g
n
a
l
inhomogeneous
lineshape
homogeneous
lineshape
probe frequency
Homogeneous broadening Inhomogeneous broadening
Figure 1.2: Fluorescence spectra recorded as a function of the frequency of a probe laser for the cases of
homogeneous and inhomogeneous broadening.
The behaviour of homogeneously and inhomogeneously broadened transitions is compared in the Table
below:
passage of the probe pulse. Scattered radiation, which is prompt and (in the absence of nonlinear eects) has the same frequency
as the incident radiation, is not then detected.
6 LECTURE 1. BASIC LASER PHYSICS
Homogeneous transition Inhomogeneous transition
For a given frequency of the probe laser, all the
atoms in the sample will interact with the probe
beam with the same strength
For a given frequency of the probe beam, only
those atoms with centre frequencies
c
close to
will be excited.
The strength of the interaction increases as is tuned
close to the centre frequency
0
of the transition since
the homogeneous cross-section increases
The strength of the interaction increases as is
tuned close to
0
since more atoms have this
centre frequency
Atoms excited to the upper level will emit radiation
with the same spectrum described by g
H
(
0
)
Atoms excited to the upper level will emit a
homogeneously broadened spectrum centred at

c
=
The shape and centre frequency of the uorescence
spectrum will be unchanged as the frequency of the
probe laser is tuned across the transition
The shape of the uoresence spectrum is
unchanged, but the centre frequency tunes with .
The intensity of the uorescence changes as the probe
frequency is tuned; the uorescence will be brightest
when is tuned close to
0
, and be negligibly weak
when tuned many homogeneous linewidths away from

0
The intensity of the uorescence changes as the
probe frequency is tuned; the intensity of the
spectrum will be greatest for probe frequencies
close to
0
since more atoms can interact with the
probe beam.
1.4 The optical gain cross-section
In the BIII course it was shown that the growth of a beam of radiation of spectral intensity I(, z) interacting
with atoms via a homogeneously broadened transition is given by,
I
z
= N

21
(
0
)I(, z), (1.8)
where the population inversion density is dened as,
N

= N
2

g
2
g
1
N
1
, (1.9)
and the optical gain cross-section is given by,

21
(
0
) =

0
c
B
21
g
H
(
0
) =

2
c
2

2
0
A
21
g
H
(
0
). (1.10)
Note that it is often useful to dene the gain coecient
21
(
0
):

21
(
0
) =
1
I
I
z
= N

21
(
0
). (1.11)
1.5. GAIN SATURATION 7
1.4.1 Laser rate equations in terms of the optical cross-section
We will often need to write rate equations to describe the transfer of population between levels of an atom. In
terms of the Einstein coecients, the rate equation for (say) the upper laser level can be written:
dN
2
dt
= R
2
(N
2
B
21
N
1
B
12
)
_

0
g
H
(
0
)()d +. . . (1.12)
= R
2
N

_

0

21
(
0
)
I()

d +. . . (1.13)
where R
2
is the rate of pumping of the upper laser level, () is the spectral energy density of the beam, and
the +. . . indicate that in general there may be other processes to consider. The second line follows from the
relations between the Einstein B-coecients and the denition of the optical gain cross-section.
For narrow-band radiation the gain cross-section varies slowly over the spectral width of the radiation and
so I() acts like a Dirac delta function: I() = I
T
(
L
) where I
T
is the total intensity and
L
the centre
frequency of the beam. We then have:
dN
2
dt
= R
2
N

21
(
L

0
)
I
T

L
+. . . , (1.14)
In a similar way, we can integrate eqn (1.8) over the bandwidth of the beam to nd the growth equation for
the total intensity.
dI
T
dz
= N

21
(
L

0
)I
T
=
21
(
L

0
)I
T
. (1.15)
Equations (1.14) and (1.15) will be used frequently in our analysis of laser systems.
1.5 Gain saturation
Gain saturation refers to the phenomenon whereby the optical gain on a laser transition depends on the in-
tensity of the radiation interacting with it. Figure 1.3 shows schematically the processes which aect the level
populations of a laser operating under steady-state conditions in the presence of an intense beam of radiation.
In the absence of radiation on the laser transition, pumping of the upper and lower levels at rates R
2
and R
1
,
respectively, will form a population inversion N

(0). However the populations in the upper and lower laser levels
will be altered by the presence of radiation on the laser transition owing to stimulated transitions from 2 1
and absorption from 2 1. Hence the population inversion in general will be a function of the intensity
3
I of
the (assumed narrow-band) radiation and the frequency detuning of this radiation from the centre frequency of
the transition.
1.5.1 homogeneous broadening
For a homogeneously broadened laser transition the rate equations for the laser levels may be written as,
dN
2
dt
= R
2
N

21
(
L

0
)
I

N
2

2
(1.16)
dN
1
dt
= R
1
+N

21
(
L

0
)
I

L
+N
2
A
21

N
1

1
. (1.17)
3
We will usually drop the subscript T, and distinguish spectral intensity by writing it in the form I().
8 LECTURE 1. BASIC LASER PHYSICS
(
g
2
/
g
1
)
N
1

2
1
I
/

L
N
2

2
1
I
/

L
N
2
A
2
1

1
R
2
R
1
2
1
Figure 1.3: Processes aecting the upper and lower laser levels in a laser operating under steady-state conditions
in the presence of an intense radiation beam.
In last years lectures we showed how these may be solved straightforwardly to give the steady-state popu-
lation inversion:
N

(I) =
N

(0)
1 +I/I
s
(
L

0
)
. (1.18)
where I
s
is the saturation intensity (Units Wm
2
), and is given by
I
s
(
L

0
) =

L

21
(
L

0
)
R
, Saturation intensity (1.19)
where we have emphasized that the saturation intensity depends (through the optical cross-section) on the
detuning of the intense radiation from the centre-frequency of the transition
4
. The recovery time
R
is given by

R
=
2
+
g
2
g
1

1
[1 A
21

2
]. (1.20)
The recovery time depends on the lifetimes of both the upper and lower laser levels. It can usefully be
thought of as giving an approximate measure of the time taken for the population inversion to re-establish itself
after a perturbation, such as the passage of an intense laser pulse.
4
The saturation intensity is not a spectral quantity; it has units of intensity! However, just like the optical gain cross-section
itself, it does depend on the frequency detuning. In order to avoid clutter, we will often drop the explicit frequency dependence of
the saturation intensity and just write I
s
.
1.5. GAIN SATURATION 9

0
)
Figure 1.4: The gain coecient measured for a homogeneously broadened transition by: a weak probe beam
(solid line); a weak probe beam in the presence of an intense beam of intensity I
s
(
L

0
) at frequency
L
(long-dashed line); an intense probe beam of intensity I
s
(0) (short-dashed line).
The saturated gain coecient can be found directly from the saturated population inversion,

I
(
0
) = N

(I)
21
(
0
) =

0
(
0
)
1 +I/Is
. (1.21)
Note that we can calculate the saturation intensity for absorption in a similar way. The result is of the same
form as eqn (1.19), but with a dierent expression for the recovery time
R
.
To illustrate how the gain coecient saturates in a homogeneously broadened laser medium, we imagine
measuring the gain coecient under the following conditions:
1. Using a weak, tunable probe beam in the absence of any other radiation;
2. Using a weak, tunable probe beam in the presence of a second (intense) beam of radiation at a frequency

L
of intensity I
s
(
L

0
).
3. An intense, tunable probe beam of constant intensity I
s
(0).
Clearly in the rst case the gain coecient measured as a function of the frequency of the probe beam
will simply be the small-signal gain coecient
0
(
0
). In the second case, however, the presence of the
intense beam will decrease the population inversion density. Since the intense beam has an intensity equal to
the saturation intensity (by denition, whatever the frequency
L
of the intense beam), the population inversion
will burnt down to half that generated by the pumping in the absence of the intense beam. As the frequency of
the probe beam is tuned across the laser line, the gain measured will therefore always be half of the small-signal
gain. Finally, for the case of the intense probe beam, the degree of saturation of the population inversion
depends on the detuning of the probe from the centre frequency of the transition. In this case the intensity of
the probe is equal to the saturation intensity when tuned to the line-centre. Hence when =
0
the measured
gain coecient will be half the small-signal gain coecient. At large detuning, however, the intensity of the
10 LECTURE 1. BASIC LASER PHYSICS

0

L
Centre frequency
c
P
o
p
u
l
a
t
i
o
n

i
n
v
e
r
s
i
o
n

N
*
(

c

-

0
)
spectral hole
(a)

0

L
Frequency
G
a
i
n

c
o
e
f
f
i
c
i
e
n
t

0
)
spectral hole
(b)
Figure 1.5: Saturation eects on inhomogeneously broadened transitions. (a) The eect of a beam of intense,
narrow-bandwidth radiation of total intensity I with frequencies close to
L
on the distribution of the population
inversion over centre frequencies. The dotted curve shows the unperturbed distribution for I = 0, and the solid
curve the distribution for I I
s
(0). For this plot
H
/
D
= 0.1. (b) Spectral hole burning in the gain
prole of an inhomogeneously broadened laser transition by a narrow-bandwidth beam of intensity I and with
frequencies close to
L
. The gain prole measured by a weak probe beam is shown for I = 0 (dotted), I = I
s
(0)
(dashed), I = 10I
s
(0) (solid). For this plot
H
/
D
= 0.03.
probe beam will be small compared to the saturation intensity; here the beam interacts only weakly with the
atoms and hence perturbs their populations only slightly. The measured gain coecient must, then, tend to
the small-signal value at large detuning. The three measured gain proles are shown in Figure 1.4.
1.5.2 Inhomogeneously broadened transitions
The phenomenon of gain saturation is considerably more complicated when the laser transition is inhomoge-
neously broadened since we must now consider separately the extent to which the population inversion of each
frequency class is saturated, and the contribution by each such class to the overall gain.
We will restrict ourselves to a qualitative discussion.
5
Consider an inhomogeneously broadened gain medium
subjected to a beam of intense, narrow-band radiation with frequencies close to
L
. The extent to which each
class of atoms is saturated that is, the extent to which, for each class of atoms, the population density of the
upper level is reduced depends on the detuning of the intense beam from the centre frequency of the class. It
should be clear that signicant saturation will only occur for those classes of atoms with centre frequencies such
that [
c

L
[
H
; for these atoms the population inversion will be strongly burnt down. However, those
classes with centre frequencies signicantly dierent from that of the radiation (on the scale of
H
) will not
be saturated to any appreciable extent. As such, a spectral hole will be burnt into the population inversion,
as illustrated in Fig. 1.5(a). The width of this spectral hole will be approximately
H
.
The gain coecient measured by a weak probe is given by integrating the contribution to the gain from
each class. Hence we expect the measured gain coecient in this case to exhibit a spectral hole, reecting that
in the population inversion. This phenomenon is shown in Figure 1.5(b).
6
5
A mathematical treatment may be found in Chapter 5 of H&W.
6
It can be shown that in the limit of extreme inhomogeneous broadening i.e. the inhomogeneous linewidth is much greater
than the homogeneous linewidth and if the homogeneous lineshape is Lorentzian, the spectral hole burnt in the gain prole is
also Lorentzian, with a full-width at half maximum of
hole
=
H

1 +

1 + I/I
s
(0)

. For this case we see that if the intensity

of the saturating beam is small compared to the saturation intensity
hole
2
H
. This result is consistent with our expectation
1.6. THRESHOLD BEHAVIOUR 11
Intense probe The saturation of the gain as measured by an intense probe (with no other beam of radiation
present) is very dierent from that shown in Figure 1.5(b) since as the probe is tuned across the transition the
class of atoms which provide gain are also the class which experiences saturation. Hence in this situation we do
not expect a spectral hole to be burnt into the gain prole.
It can be shown that for the case of extreme inhomogeneous broadening (i.e.
D

H
), the gain
coecient measured by an intense probe is equal to a constant fraction of that measured by a weak probe.
Further, if the homogeneous lineshape is Lorentzian, it may be shown that the saturated gain coecient is,

D
I
(
0
) =

D
0
(
0
)
_
1 +I/I
s
(0)
. (Special case) (1.22)
Notice that this looks rather similar to eqn (1.21), which was obtained for homogeneous broadening. However,
this similarity is very mis-leading since eqn (1.21) refers to the gain measured for a homogeneously- broadened
transition by a weak probe in the presence of a saturating beam at =
L
; eqn (1.22) refers to the gain
measured for an extreme inhomogeneously broadened transition by an intense probe in the absence of any other
saturating radiation. The situations couldnt be more dierent!
1.5.3 Summary
To summarize: for homogeneously broadened gain media an intense beam of radiation at =
L
causes the
entire population inversion to be burnt down by a factor [1+I/I
s
(
L

0
)] such that the gain prole measured
by a weak probe beam has the same shape as that measured in the absence of the saturating beam, but with
the gain being reduced by the factor [1 +I/I
s
(
L

0
)]; for inhomogeneously broadened gain media a spectral
hole is burnt into the population inversion, and the gain measured by a weak probe beam is only reduced for
frequencies close to
L
. For homogeneously broadened systems the gain measured by an intense probe beam
is reduced by a greater extent at frequencies close to the line centre, leading to a attening and broadening of
the measured gain prole; for strongly inhomogeneously broadened media the gain is everywhere reduced by
the same factor
_
1 +I/I
s
(0), and the measured gain prole has the same shape as measured by a weak probe
beam.
1.6 Threshold behaviour
We now examine how lasers operating on homogeneously and inhomogeneously broadended transitions behave
as they are brought above threshold.
Let us suppose that the gain medium is inserted into a optical cavity of length L. In general the modes of
an optical cavity are described by the set of integers (l, m, p). The integers l and m determine the transverse
prole of the right- and left-going beams in the cavity, and are said to determine the transverse modes of the
cavity; the integer p determines the number of half-wave loops of the radiation eld along the axis of the cavity,
and so describes the longitudinal modes. The frequencies of the cavity modes are determined primarily by
p; for example, for the case of a plane-plane cavity the longitudinal modes are separated in frequency by an
amount,

p,p1
=
c
L
. (1.23)
As discussed in the BIII course, in order for laser oscillation to occur on a cavity mode the round-trip gain
must equal the round-trip loss.
that the spectral hole burnt in the population inversion is approximately
H
wide, and the range of frequency classes with which
the probe interacts is also approximately equal to
H
.
12 LECTURE 1. BASIC LASER PHYSICS
Threshold
longitudinal
modes
i
n
c
r
e
a
s
i
n
g

p
u
m
p

p
o
w
e
r
(a) (b)

I
(

0
)

o
u
p
u
t

p
o
w
e
r

o
u
p
u
t

p
o
w
e
r

o
u
p
u
t

p
o
w
e
r

Threshold
longitudinal
modes

I
(

0
)

Threshold
longitudinal
modes

I
(

0
)

single-mode
oscillation
(a) A
Threshold
longitudinal
modes
i
n
c
r
e
a
s
i
n
g

p
u
m
p

p
o
w
e
r

I

(

0
)

o
u
p
u
t

p
o
w
e
r

o
u
p
u
t

p
o
w
e
r

o
u
p
u
t

p
o
w
e
r
multimode
oscillation
spectral
holes
D

I

(

0
)
D

I

(

0
)
D
(a) (b)
(b) B
Figure 1.6: Behaviour of the small-signal gain (a) and output power (b) as the pumping power is increased
above the threshold value for lasers operating on : (A) homogeneously; and (B) inhomogeneously broadened
transitions
1.6.1 Homogeneously broadened laser transitions
As the pumping is increased from zero the population inversion N

, and hence the small-signal gain coecient

0
(), will increase. For low levels of pumping, spontaneous emission will increase the energy in all of the cavity
modes with frequencies lying within the linewidth of the laser transition. The intensity will be low, and the
number of photons per mode will typically be much less than unity.
With further increases in pumping the small-signal gain coecient will increase until the threshold condition
is met for the cavity mode closest to the line centre, causing the onset of oscillation on that cavity mode. The
onset of oscillation will be accompanied by a very large (by a factor of order 10
15
) increase in the energy density
and number of photons in the oscillating mode.
Further increases in the pumping do not increase either the gain coecient or the population inversion.
In fact this must be the case since in the steady-state the round-trip gain must always be balanced by the
round-trip loss. Instead, as the pumping is increased the intensity of the oscillating mode increases this,
after all, is what we are trying to achieve! The increased intensity of radiation circulating within the cavity
reduces, or burns down, the population inversion to the threshold value. In other words, the laser transition is
saturated. For a given level of pumping the intensity of the oscillating mode is determined by the condition that
the saturated round-trip gain is equal to the round-trip loss, or, equivalently, that the population inversion is
burnt down from the value, N

0
, that would be achieved at that level of pumping, to the threshold value N

thresh
.
1.6. THRESHOLD BEHAVIOUR 13
The behaviour of the saturated gain coecient
I
(
0
) and the output power of the laser as the level of
pumping is increased is illustrated schematically in Figure 1.6(a).
Spatial hole burning
In principle, for a homogeneously broadened laser system only the cavity mode with the largest gain can ever
oscillate; since the population inversion is clamped to the threshold value, and all atoms interact with radiation
of a given frequency with the same strength, the other modes will always be below the threshold for oscillation.
However, in practice in some circumstances more than one mode can oscillate. This behaviour arises from
the fact that the spatial distribution of the intensity of the oscillating mode is not uniform. In particular the
oscillating mode will form a standing wave within the cavity. Near the anti-nodes of the standing wave the
population inversion will be burnt down to threshold value; near the nodes the intensity will be low, and the
population inversion essentially unsaturated. The non-uniform burning down of the population inversion in
a homogeneously broadened laser system is known as spatial hole-burning, by analogy with the spectral
hole-burning discussed in the next section. Spatial hole-burning allows other modes, with slightly dierent
frequencies, to feed o regions of unsaturated population inversion and reach the threshold for oscillation.
In such circumstances two or more cavity modes can oscillate, leading to multimode oscillation at several
frequencies within the linewidth of the transition.
1.6.2 Inhomogeneously broadened systems
The above-threshold behaviour of inhomogeneously broadened lasers is quite dierent since dierent classes of
atoms interact with dierent frequencies, and hence dierent cavity modes. Once again, as the pumping is
increased to the threshold value the mode with the largest gain will start to oscillate. Now, however, as the
pumping is increased the population inversion is only clamped to the threshold value for those classes of atoms
which can interact with the frequency of the oscillating mode; other classes will be essentially unaected, and
for these classes the population inversion will continue to increase as the pumping increases. As a consequence
the gain coecient
D
I
(
0
) develops a spectral hole at the frequency of the oscillating mode.
Further increases in the pumping level will allow other modes, of dierent frequencies, to reach the threshold
for oscillation. Thus, in general as the pumping level is increased in an inhomogeneously broadened laser
oscillator, the number of oscillating modes increases as well as the output power of each oscillating mode. The
process is illustrated schematically in Figure 1.6(b).
14 LECTURE 1. BASIC LASER PHYSICS
Lecture 2
Solid State Laser Materials
A large number of technologically and scientically important lasers operate by optically pumping transitions
within ions doped as an impurity species into a variety of crystalline or glass hosts. The term solid state
lasers is usually reserved for systems of this type, lasers operating between levels of the electron band structure
found in semiconductors being referred to as semiconductor or simply diode lasers.
Solid state laser materials oer advantages over liquid and gas lasers in that they are robust, chemically
inert, require no special handling, and do not degrade or become contaminated with use.
The spectroscopy of ions doped into solid hosts is a complex subject. Fortunately for our purposes we need
only understand a few key concepts in order to appreciate how dierent laser systems work, and why they
behave as they do. In particular we would like to know the energy level structure of the ions, the lifetimes of
the levels, and the linewidth of the transitions between them.
2.1 General considerations
Ions doped into a solid host dier from free ions in that they are subjected to the electric eld the crystal
eld arising from the ions of the solid host. The interaction with the crystal eld leads to an additional
term in the Hamiltonian of the form,
H
c
= eV
c
, (2.1)
where V
c
is an electrostatic potential describing the crystal eld. In general the eect of this interaction is to
split and shift the energy levels from their positions in the absence of the crystal eld, an eect known as the
Stark eect
1
.
The crystal eld has a symmetry reecting that of the crystal lattice. Consequently a proper treatment of
the eect of the crystal eld involves detailed consideration of this symmetry; indeed in some circumstances
these symmetry properties are used to label the energy levels arising from the interaction with the crystal eld.
The nature of the inuence of the crystal eld on the energy level structure depends critically on its strength
relative to the other terms in the Hamiltonian. Three limiting cases may be identied:
Weak eld: [H
c
[ [H
so
[ [H
re
[. In this case the interaction of the crystal eld is weak compared to
the spin-orbit and residual electrostatic interactions (as well the interaction with the central eld) so that
the energy levels are changed only slightly from those in the free ion. The eect of the crystal eld in this
case is to cause a small splitting and shifting of the levels of the free ion to form a so-called manifold of
1
Named after the German physicist Johannes Stark (1874-1957).
15
16 LECTURE 2. SOLID STATE LASER MATERIALS
closely-spaced levels. Since the energy shifts are small, the labelling of the manifolds is simply that of the
energy levels of the free ion. Further, in the weak eld case the energy level structure of an ion will be
almost independent of the crystal host in which it is embedded. A good example of the weak eld case
are the energy levels formed by the 4f
n
congurations in trivalent rare earth ions discussed in Section 2.2.
Intermediate eld: [H
so
[ [H
c
[ [H
re
[. In this situation the crystal eld interaction must be
considered as a perturbation acting on the terms formed by the residual electrostatic interaction before
the spin-orbit interaction is taken into account; the crystal eld acts on the terms formed by the residual
electrostatic interaction and splits and shifts them. The spin-orbit interaction leads to further splitting
of the energy levels. For intermediate crystal elds it is no longer possible to label the energy levels with
the quantum numbers L, S, and J. The rst-row transition metal ions, such as Cr
3+
frequently exhibit
crystal elds of intermediate strength. It is more dicult to calculate the energy levels for intermediate
crystal elds than for weak- or strong-elds. Frequently, therefore, the energy levels of ions experiencing
intermediate crystal elds are labelled with either a weak- or strong-eld notation.
Strong eld: [H
so
[ , [H
re
[ [H
c
[. Here the spin-orbit or residual electrostatic interactions may be
ignored (in the rst approximation) and the crystal eld acts on the single-electron orbitals of the central
potential to give single-electron crystal-eld orbitals which reect the symmetry of the crystal eld.
As discussed below, the energies of the orbitals depends on the spatial distribution of the orbital wave
function with respect to the crystal lattice. The energy levels are labelled by the symmetry properties of
the electron wave function using a notation derived from group theory. Strong crystal elds can arise in
the second- and third-row transition-metal ions.
For each electron the relative strengths of the residual electrostatic, spin-orbit, and crystal eld interactions
depends very strongly on the radial distance r
i
of the electron from its nucleus, and the distance d from the
active ion to the neighbouring ions. For example, for neighbouring ions distributed with octahedral symmetry
the energy of interaction is proportional to,

r
4
i
_
d
5
where

r
4
i
_
is the expectation value of r
i
. In contrast, the energy shift arising from the spin-orbit interaction is
proportional to,
_
1
r
3
i
_
Hence the ratio of the interaction with the crystal eld to the spin-orbit interaction varies approximately as,
E
c
E
so

r
7
i
_
d
5
. (2.2)
As a consequence very small changes in the relative size of the mean electron radius or the nearest-neighbour
distance can change the relative strength of the interaction with the crystal eld enormously.
2
2
Frequently it is said that the relative strength of an electrons interaction with the crystal eld depends on the extent to which
it is shielded by other electrons in the ion. For example, the 4f electrons in trivalent rare earth ions are often said to be shielded by
the outer 5s and 5p electrons. Shielding eects of this type do occur, but calculation shows them to be approximately an order of
magnitude smaller than the eects of changing mean electron radius and nearest-neighbour distance. For a discussion of this point
see G. Burns, Solid State Physics, Academic Press.
2.1. GENERAL CONSIDERATIONS 17
2.1.1 Radiative transitions
For all the solid-state laser transitions discussed in this lecture, the laser transition occurs between two levels
within the same electron conguration. The parity of the electron wave function does not change in such
transitions, and hence they are forbidden by the selection rules of electric dipole radiation. Whilst this would
indeed strictly be the case for transitions between the equivalent levels in a free ion, for ions doped into a
solid, these electric dipole transitions may occur for two reasons. The rst is simply that the presence of the
crystal eld means that there is no longer inversion symmetry at the site of the active ion: in other words the
electron wave function no longer has a denite parity and under the transformation r
i
r
i
the electron
wave function can change magnitude (as well as possibly change sign). A pure electron conguration must have
a well-dened parity since it is a solution of the Scrodinger equation in a spherically-symmetric central eld.
However, the crystal eld interaction is not spherically symmetric and therefore introduces small admixtures
of congurations of opposite parity to form a wave function which is no longer spherically symmetric, and no
longer has a well-dened parity. These admixtures of congurations of opposite parity can allow transitions to
occur between two levels which nominally have the same electron conguration.
A second mechanism by which electric dipole transitions may occur between two levels of the same cong-
uration is a dynamically-induced transition strength through vibrations of the crystal lattice which destroy the
inversion symmetry at the site of the active ion. This second mechanism becomes important when the active
ion is located at a crystal site exhibiting a high degree of symmetry. As might be expected, in such cases the
radiative lifetime of the level depend strongly on the temperature of the crystal.
Notwithstanding the above, the dipole-forbidden nature of transitions within a single conguration means
that the transition rates are signicantly slower than fully dipole-allowed transitions. The radiative lifetimes of
the upper levels of visible transitions within ions doped into a solid are 3 to 6 orders of magnitude longer than
fully dipole-allowed transitions in free atoms and ions.
Finally we note that the strength of a radiative transition can depend on the orientation of the polarization
of the radiation with respect to the axes of the crystal. This is true for both absorption and emission. In
such cases the orientation of the crystal axes with respect to the axis of the laser cavity and the direction and
polarization of the pump radiation can be important, and the laser output can be partially or totally polarized.
2.1.2 Non-radiative transitions
Non-radiative transitions play a very important role in the operation of solid-state lasers: rapid non-radiative
decay can provide an ecient route for feeding population from one or more excited levels into the upper laser
level, and similarly can help to keep the population of the lower laser level low.
Phonon de-excitation
The energy levels of an active ion are coupled to the vibrations of the crystal lattice via the crystal eld
interaction. The lattice vibrations have a spectrum of normal modes,
3
and their excitation is quantized; the
unit of excitation being known as a phonon. Transitions between energy levels may therefore occur not by
emission of radiation, but by the emission or absorption of one or more lattice phonons. The rate of such
phonon de-excitation is found to decrease very rapidly with the number of phonons involved. For example,
for the case of weak coupling it may be shown that the rate of de-excitation on a transition of energy E
21
by phonons of energy E
p
varies as exp[(E
21
/E
p
)], where is a positive constant which depends on the
transition. In practice the de-excitation rate is found to be small if the number of phonons required is greater
than approximately 5, and consequently the dominant contribution to the decay rate is from the phonons with
the largest possible energy.
3
There are two classes of these modes, known as acoustic and optical modes. A discussion of normal modes of crystals may be
found in textbooks on solid-state physics.
18 LECTURE 2. SOLID STATE LASER MATERIALS
1
2
virtual
level
Figure 2.1: Schematic diagram of the two-phonon Raman process leading to line broadening. In this process
phonons of two dierent energies transfer ions from level 1 to level 2 via a virtual level, followed by the reverse
process which returns the ion to level 1. The net result is no change in either the level occupied by the ion or
the phonon spectrum but a broadening of the energy level by an amount proportional to the rate at which this
process occurs.
The rate of phonon de-excitation depends strongly on the strength of coupling between the electrons of the
ions and the crystal eld, and on the phonon spectrum of the lattice and in particular on the maximum
phonon energy. The de-excitation rates therefore vary widely for dierent combinations of impurity ions and
hosts; the rate of phonon de-excitation also increases rapidly with temperature which reects the increasing
density of phonons in a given mode with temperature (determined by the Bose-Einstein distribution). For small
energy dierences the rate of non-radiative transitions can be extremely high: at room temperature the rate is
of order 10
11
- 10
12
s
1
.
The strong dependence of the rate of phonon de-excitation on the energy gap between the initial and nal
levels of the ion means that if a level is to be suitable as an upper laser level it should lie signicantly above
the nearest lower-lying level. This is necessary if the level is to have a long uorescence lifetime, which allows
a large population to build up in the level; and ensures that it decays predominantly by emission of radiation,
which enables the population in the upper laser level to be used eciently.
2.1.3 Line broadening
The natural linewidths of transitions between the levels of an impurity ion are typically very small, owing to
the long radiative lifetimes of the levels. In addition a temperature-dependent lifetime broadening arises from
the increased rate of decay of the levels caused by non-radiative transitions. In practice, however, the measured
linewidths of transitions are signicantly broader than can be accounted for by the measured lifetimes of the
upper and lower levels, and hence other processes must be responsible for the additional broadening.
A variety of mechanisms can cause this additional broadening. One of these, two-phonon Raman scattering
is illustrated in Fig. 2.1. In this process two phonons of dierent energy are successively absorbed and emitted
so as to remove, and then return an ion from its energy level. Since the ion is returned to its initial level
there is no change in the lifetime of the level, but a broadening occurs proportional to the rate of scattering.
As might be expected the rate of scattering, and hence its contribution to the linewidth, depends strongly on
temperature.
Phonons may also play a more direct role in determining the linewidth of a radiative transition. For
some ion-host combinations radiative transitions can be accompanied by the emission or absorption of one or
more phonons. These so-called vibronic transitions (discussed in Section 2.3.1) have very broad, temperature-
dependent linewidths, as discussed below.
2.2. TRIVALENT RARE EARTHS, 4F
N
4F
N
TRANSITIONS 19
2.2 Trivalent rare earths, 4f
n
4f
n
transitions
2.2.1 Energy level structure
A large number of solid-state lasers operate on 4f
n
4f
n
transitions in trivalent rare earth ions, and in particular
the lanthanide series the actinides being radioactive, and hence dicult to work with. Important trivalent
rare-earth laser ions include: Ce
3+
, Nd
3+
, Ho
3+
, and Er
3+
.
Neutral atoms of these rare earths have a ground state electronic conguration of the form [Xe] 4f
n+1
6s
2
,
or [Xe] 4f
n
5d 6s
2
where n 0 and [Xe] represents the ground state conguration of xenon:
1s
2
2s
2
2p
6
3s
2
3p
6
3d
10
4s
2
4p
6
4d
10
5s
2
5p
6
.
In forming the trivalent ion, the three loosest-bound electrons are lost:
4
the two 6s electrons and either one of
the 4f electrons or the 5d electron, to give a ground-state conguration which is always of the form [Xe] 4f
n
. The
absorption spectrum of the ions corresponds to transitions within the 4f conguration (4f
n
4f
n
transitions),
or 4f
n1
5d 4f
n
transitions to the lowest-lying empty orbital, 5d.
5
Of key importance to the spectroscopy of the trivalent rare earth ions in crystalline media is the fact that the
wave function of the 4f orbital is more compact than those of the 5s and 5p orbitals. As such the mean radius
of the 4f electrons are relatively small compared to the size of the ions, and hence to the nearest-neighbour
distance. The interaction with the crystal eld is therefore weak, producing a series of manifolds which are only
slightly perturbed from those of the levels of the free ion. As a consequence the energy level structure of the 4f
congurations in the trivalent rare earths is approximately independent of the crystal host, although the optical
cross-sections of transitions may vary signicantly.
The manifolds are labelled by the levels of the free ion from which they arise, i.e.
2S+1
L
J
. If the total spin
quantum number S is an integer, the maximum number of levels within the manifold is given by (2J + 1); if
S is half-odd-integer the levels of the crystal eld are all doubly degenerate,
6
and consequently the maximum
number of levels within the manifold is reduced to (2J + 1)/2. It should be emphasized that these are the
maximum number of non-degenerate levels; the actual number formed depends on the symmetry of the crystal
eld.
2.2.2 Transition linewidth
The natural linewidth of the 4f
n
4f
n
transitions is very narrow owing to the long radiative lifetimes of the levels.
This linewidth is increased substantially by phonon collisions to give a homogeneously-broadened transition with
a linewidth which depends on temperature. At room temperature the linewidth is typically in the range 0.1 -
3 THz (10 - 100 cm
1
).
Inhomogeneous broadening can arise if the crystal eld varies with position in the crystal. For example, the
presence of crystal defects, strain, or impurity ions can cause variation in the crystal eld and hence local shifts
of the ion energy levels. For good quality laser crystals these eects should be small, in which case the line
broadening will be predominantly homogeneous. In contrast, ions doped into glasses experience a very wide
range of local environments. The energies, broadening, and even the number of levels varies signicantly from
site to site and consequently all transitions will be strongly inhomogeneously broadened. The inhomogeneous
linewidth of the 4f
n
4f
n
transitions in ions doped into a glass host are of order 30 THz (1000 cm
1
).
2.2.3 Nd:YAG laser
Perhaps the most important example of a laser based on a 4f
n
4f
n
transition in a trivalent rare earth ion is the
Nd:YAG laser. Neodymium-based lasers are very widely used in science and industry. They are frequently used
4
In the free ion the electrons are removed from the atom, in a solid they form bonds with the atoms of the host lattice.
5
The 5s and 5p orbitals already being occupied.
6
This is known as Kramers degeneracy.
20 LECTURE 2. SOLID STATE LASER MATERIALS
as pump lasers for dye lasers, or other solid-state lasers such as Ti:sapphire. In medicine they nd applications
in removing secondary cataracts or tissue removal. High-power Nd:YAG lasers can be used in laser drilling and
welding.
Crystal properties
In Nd:YAG the Nd
3+
ion replaces the Y
3+
ion. Since the size of the Nd ion, which has a radius of 98 pm,
is greater than that of the Y ion, of radius 90 pm, it is not possible to introduce Nd ions with an atomic
concentration much above 1.5% without straining the crystal lattice unduly. It is worth noting that an atomic
concentration of Nd ions equal to 1% corresponds to a density of Nd ions of 1.386 10
20
cm
3
.
Doped YAG crystals are grown using the Czochralski method. For Nd:YAG the growth rate must be rather
slow in order to avoid inhomogeneities within the nished rods, a typical pulling rate is 0.5 mmhr
1
i.e. it
takes several weeks to grow one boule.
Energy levels
Figure 2.2 shows the energy levels of the Nd
3+
ion and how these are split into manifolds by the crystal eld of
YAG. Broad pump bands are provided by several closely-spaced manifolds with energies between approximately
12 000 and 33 000cm
1
(300 - 800 nm). Particularly strong pumping occurs at 810 nm (12 300 cm
1
) and 750 nm
(13 300 cm
1
), which can be accessed by diode laser pumping.
The manifolds excited by optical pumping on these bands are relatively closely spaced compared to the
maximum phonon energy in YAG of 850 cm
1
. Consequently excited ions cascade down through these manifolds
by very rapid non-radiative transitions until the two levels (denoted R
1
and R
2
) of the
4
F
3/2
manifold is reached.
The levels of the
4
F
3/2
manifold are metastable because the nearest lower-lying level the top of the
4
I
15/2
manifold is separated by some 4 698 cm
1
, corresponding to more than 5 phonons. The rate of non-radiative
decay of the
4
F
3/2
manifold is therefore slow, and it decays instead almost entirely by radiative transitions to
levels of the
4
I manifolds, with a uorescence lifetime of 230 s. In contrast, the levels of the
4
I manifolds, are
separated by less than 1 500 cm
1
, and consequently decay non-radiatively with lifetimes of order 100 ps.
It can be seen, therefore, that the energy levels of Nd:YAG are very well suited to achieving laser oscillation:
the levels of the
4
F
3/2
manifold may be populated eciently by cascade from broad pump bands, and they are
metastable ideal properties of an upper laser level. Suitable lower laser levels exist within the
4
I manifolds;
these lie signicantly above the ground state and decay very rapidly to it. The system is therefore an almost
ideal example of a four-level laser system. Figure 2.3 shows a simplied energy level scheme for the strongest
Nd:YAG laser transition; transitions to the levels of other
4
I manifolds are analogous, with the exception of
those to the lowest-lying manifold, I
9/2
. Lasers operating on transitions to this manifold behave quite dierently
than those operating on transitions to the higher-lying manifolds, since the levels of the I
9/2
manifold can have
a signicant thermal population. As such, these lasers behave more like three-level systems.
Lasing has been achieved from the
4
F
3/2
manifold on more than 20 transitions to the I
13/2
, I
11/2
, and I
9/2
manifolds, with wavelengths near 1319 nm, 1064 nm, and 946 nm respectively. Of these, the strongest and
certainly the most commonly used transition is the
4
F
3/2

4
I
11/2
transition.
7
The R
1
and R
2
levels of the
4
F
3/2
manifold are separated by 84 cm
1
. Rapid non-radiative transitions
ensure that the populations of these two levels is maintained in thermal equilibrium and, importantly, the rate
at which the population in these two levels is mixed is suciently fast that this remains the case even during
laser oscillation. Hence at room temperature the relative populations of these levels are approximately 60%
and 40% respectively. Of the possible
4
F
3/2

4
I
11/2
transitions, the strongest is the
2
transition between
the R
2
and Y
3
levels with a vacuum wavelength of 1064.15 nm. At room temperature this transition dominates
the laser output, since the population of the R
2
levels is suciently large that the gain on this transition is
7
Indeed, in order to obtain lasing to the other manifolds it is necessary to employ frequency-selective cavities which suppress
lasing to the
4
I
11/2
manifold and enhance the feedback on the desired transition.
2.2. TRIVALENT RARE EARTHS, 4F
N
4F
N
TRANSITIONS 21
0
19100
17000
15900
14600
13500
12500
11520
6330
4210
2240
4
I
9/2
0 Z
1
132 Z
2
200 Z
3
311 Z
4
852 Z
5
2028 Y
2
2002 Y
1
2110 Y
3
2146 Y
4
2481 Y
5
2514 Y
6
11423 R
1
11507 R
2
3933
3922
4034
4055
4440
4500
5765
6725
4
G
9/2
4
G
5/2
+
2
G
7/2
2
H
11/2
4
F
9/2
4
S
3/2
+
4
F
7/2
4
H
9/2
+
4
F
5/2
4
F
3/2
4
I
15/2
4
I
13/2
4
I
11/2
4
I
9/2
4
F
3/2
4
I
15/2
4
I
13/2
4
I
11/2
0
5000
10000
15000
20000
Energy
(cm
-1
)
(a) (b)
Figure 2.2: Energy level diagrams for the Nd
3+
ion in YAG. (a) shows the lowest-lying levels of the free ion
formed from the [Xe] 4f
3
conguration; (b) illustrates how these levels are split by the crystal eld in YAG to
form manifolds of closely-spaced levels. Note that in (b) the levels of each manifold are shown on an expanded
energy scale, and the spacing between manifolds is not drawn to scale. The energies of the levels is given in
cm
1
.
22 LECTURE 2. SOLID STATE LASER MATERIALS
4
F
3/2
4
I
11/2
4
I
9/2
1064 nm
rapid
relaxation
rapid non-
radiative decay
optical
pumping
2
0
1
3
{
Figure 2.3: Simplied energy level diagram of the
4
F
3/2

4
I
11/2
laser transition in Nd:YAG showing the
four-level nature of the laser scheme.
2.2. TRIVALENT RARE EARTHS, 4F
N
4F
N
TRANSITIONS 23
largest. In contrast, if the crystal is cooled to low temperatures almost all the population of the
4
F
3/2
manifold
resides in the R
1
level and lasing occurs from this level on the
1
transition from R
1
to the Y
2
level of
4
I
11/2
at
a vacuum wavelength of 1064.40 nm.
Broadening
In Nd:YAG the
4
F
3/2

4
I transitions are homogeneously broadened by phonon collisions, with a full-width at
half-maximum of approximately 190 GHz (6.5 cm
1
). Note that this is much larger than the natural broadening
dominated by the short lifetime of the lower laser level which is of order 1 GHz.
Eective gain cross-section
Calculation of the optical gain is complicated by the fact that both the
1
and
2
transitions in Nd:YAG can
contribute to the gain. In such cases it is useful to use a single eective cross-section which takes into account
the contribution to the gain from all the transitions involved. The eective cross-section is dened in such a
way that the gain coecient is given by multiplying it by the total population of the manifold of levels, rather
than the population of any single level.
The peak optical gain cross-sections of the
1
and
2
transitions are 1.9 10
19
cm
2
and 7.1 10
19
cm
2
respectively. Hence the dominant contribution comes from the
2
transition; if we neglect the contribution
of the gain from the
1
transition, the peak value of the eective gain cross-section is given by multiplying
the peak cross-section of the
2
transition by the proportion of population in the R
2
level (39%):
e
(0)
f
R
2

R
2
(0) = 2.8 10
19
cm
2
. In fact, contribution from the spectral wings of the
1
transition increases the
eective cross-section to 3.7 10
19
cm
2
.
Practical implementation
Nd:YAG lasers are one of the most common types of laser in use, and can be operated in a wide variety of
dierent congurations: they can run continuously or in a pulsed mode; they can be pumped by ashlamps or
laser diodes; and they can provide mean output powers from a few milliwatts up to several kilowatts. Very often
Nd:YAG lasers are frequency doubled, tripled, or quadrupled to generate radiation of wavelength 532, 355, or
266 nm respectively.
Flashlamp-pumped systems employ pump chambers with single or multiple ashlamps. In such systems the
laser rod is typically of about 5 mm diameter and 30 - 150 mm long, and the optical cavity is of order 0.5 m in
length. With pulsed pumping the laser is often Q-switched at a pulse repetition of 10 - 50 Hz to yield output
pulses of 5 - 10 ns duration and energies of a few tens to several hundred millijoules. The addition of one or
more Nd:YAG amplier stages can increase the output pulse energy to several Joules.
Flashlamp-pumped systems can generate pulsed output with a much higher pulse repetition frequency by
employing continuous pumping and employing acousto-optic modulators to Q-switch the cavity at a repetition
rate of 10 - 20 kHz. The mean output power of this type of system can be as high as 15 W.
ND:YAG lasers may be pumped by GaAs diode lasers operating at 808 nm which closely matches the peak
absorption corresponding to excitation of levels of the
2
H
9/2
and
4
F
5/2
manifolds. The excited levels lie only
900 cm
1
above the upper laser levels, and consequently much less energy is deposited in the crystal in the form
of heat (phonons) than is the case for ashlamp pumping which excites all manifolds up to some 10 000 cm
1
above the upper laser levels.
Diode-pumped Nd:YAG lasers can generate continuous-wave output with powers of: 10 W or more with end-
pumping; above 100 W with side-pumping; and of order 1 kW with slab-pumping. Q-switching and modelocking
is also possible. The slope eciencies of diode-pumped Nd:YAG can be as high as 50 - 60% for end-pumped
congurations, and 25 - 40% for side-pumping. These gures are much higher than for ashlamp-pumped lasers,
for which the slope eciency is typically 3%.
24 LECTURE 2. SOLID STATE LASER MATERIALS
Table 2.1: Important parameters of Nd:YAG, Nd:YLF, Nd:GdVO
4
, and Nd:Glass (Hoya LHG-5 phosphate)
lasers.
Nd:YAG Nd:YLF Nd:GdVO
4
Nd:Glass
(nm) 1064 1053 1063 1054

2
(s) 230 450 90 290

1
(ps) 100s 100s 100s 100s
(GHz) 160 380 180 5 000

21
(10
20
cm
2
) 37 18.7 76 4.1
(Wm
1
K
1
) 13.0 6.0 12.3 1.19
2.2.4 Other crystalline hosts
Neodymium ions have exhibited laser oscillation in several crystalline hosts including Lithium yttrium uoride
(LiYF
4
usually written as YLF) several vanadates such as YVO
4
and GdVO
4
. Vanadates are particularly
useful in conjunction with diode-pumping since in these hosts the pump bands are some 80% broader, thereby
reducing the tolerance required on the wavelength of the diode laser.
2.2.5 Nd:Glass laser
A wide variety of oxide-, uoride-, and sulphide-based glasses have been developed as hosts for Nd ions. The
wavelength of the
4
F
3/2

4
I
11/2
transition varies from approximately 1054 to 1062 nm depending on the glass
host. The lifetime of the upper laser level is similar to that in YAG.
In glass hosts the laser transition is strongly inhomogeneously broadened, the linewidth increasing by a
factor of approximately 50 to typically 6.5 THz ( 25 nm) compared to that found in Nd:YAG. Largely as
a result of the greater linewidth, the gain cross-section of Nd:Glass is an order of magnitude smaller than that
in Nd:YAG.
The small optical gain cross-section of Nd:Glass allows a large population inversion density to be generated
without the onset of amplied spontaneous emission. The material is therefore frequently employed to amplify
laser pulses to large energies. The low thermal conductivity of Nd:Glass restricts the pulse repetition rate of
ampliers based on this material to no more than a few pulses per second; and for very high energy ampliers,
the maximum pulse repetition rate is much lower. Finally we note that the large linewidth supports amplication
of pulses as short as 100 fs.
2.2.6 Erbium lasers
Laser action has been achieved in erbium ions doped into a variety of garnet and uoride crystalline hosts, as
well as in several types of glass. There are two transitions of interest, with wavelengths of approximately 2.9 m
and 1.5 m. Radiation at 2.9 m is absorbed very strongly by water, and consequently erbium lasers operating
on this transition have found applications in medicine. The 1.5 m transition matches the third transparency
window of optical bres, and consequently is important for optical communications, particulary in the form of
the erbium-doped bre amplier (EDFA), the construction of which is discussed in Lecture ??. Further,
this wavelength falls into the so-called eye-safe window,
8
and consequently erbium lasers may also be used in
telemetry and laser-ranging applications.
8
Lasers operating between 1.45 and 1.70 m are sometimes known as eye-safe since radiation in this region is strongly absorbed
by the cornea of the eye, and so cannot reach the retina. The term, however, is something of a misnomer since a suciently high-
power laser operating in the eye-safe region could still cause damage to the cornea or lens of the eye.
2.2. TRIVALENT RARE EARTHS, 4F
N
4F
N
TRANSITIONS 25
4
I
15/2
4
I
13/2
1
5
2
0

-

1
5
6
0

n
m
rapid non-
radiative decay
o
p
t
i
c
a
l

p
u
m
p
i
n
g
Er
3+
rapid
relaxation
1
4
8
0

n
m
9
8
0

n
m
4
I
11/2
Figure 2.4: Energy level structure of the Er
3+
ion in a glass host showing some possible laser transitions. Inset:
simplied energy level diagram for the important transition at 1550 nm used in the EDFA.
26 LECTURE 2. SOLID STATE LASER MATERIALS
G
a
i
n

c
o
e
f
f
i
c
i
e
n
t

(
a
r
b
.

u
n
i
t
s
)
Wavelength (m)
i
n
c
r
e
a
s
i
n
g
p
u
m
p
Figure 2.5: Measured gain coecient as a function of wavelength for the
4
I
13/2

4
I
15/2
Er:Glass laser transition
for dierent pump laser intensities.
Figure 2.4 shows the energy levels of Er
3+
ions in a glass host, and the important
4
I
13/2

4
I
15/2
transition
at 1.54 m. The upper and lower manifolds of levels are split by the crystal eld into 7 and 8 degenerate levels
respectively. In a glass host the 56 possible transitions between these levels are broadened homogeneously by
lifetime and phonon broadening, and inhomogeneously by spatial variations in the local environment. The
broadening characteristics of the laser transition are therefore complex, and to some extent the relative impor-
tance of homogeneous and inhomogeneous broadening depends on the properties of the glass host employed.
However, the linewidth of the laser transition is approximately an order of magnitude greater than in a crystalline
environment and hence the transition may be considered to be predominantly inhomogeneously broadened.
The most important pump band occurs at 980 nm (10 200 cm
1
) to the
4
I
11/2
manifold since this can be
accessed by InGaAs/GaAs diode lasers. In-band pumping can also occur at 1480 nm (6 800 cm
1
) to the higher
levels of the
4
I
13/2
manifold. For either pump band, excitation is followed by rapid non-radiative decay to the
lowest level of the
4
I
13/2
manifold, which acts as the upper laser level.
As for the case of Er:YAG, since the
4
I
13/2
level lies well above the next lower level, those of the ground
manifold, the rate of multiphonon decay of this level is low and the level decays predominantly radiatively with
a uorescence lifetime of approximately 8 ms. Lasing occurs from the bottom of the
4
I
13/2
manifold to the
levels of the ground-state manifold,
4
I
15/2
. Note that for the
4
I
13/2

4
I
15/2
transition up-conversion
9
depletes
the upper laser level, and consequently the concentration of Er
3+
ions must be kept low, generally to below
1 10
20
cm
3
.
The Er:Glass laser is interesting in that it exhibits both three- and four-level characteristics. Figure 2.5
shows the measured gain coecient as a function of wavelength for dierent pump intensities. It is seen that at
long wavelengths positive gain is achieved at low levels of pumping, i.e. four-level laser behaviour. In contrast,
at shorter wavelengths gain only occurs for high pump intensities, which is characteristic of a three-level laser.
This complex behaviour reects the fact that the lower manifold of states extends over approximately 850 cm
1
,
compared to k
B
T 210 cm
1
. Lasing at long wavelengths occurs to the top of the
4
I
15/2
manifold which has
only a low thermal population and which undergoes rapid relaxation. However, lasing at shorter wavelengths
9
Up-conversion proceeds through the same ion-ion interactions that cause concentration quenching. In up-conversion ion A
decays from an intermediate level to the ground state, causing ion B to be excited from an intermediate level to a higher-lying
level. If the two ions are of the same species this process is known as up-conversion, and can proceed very rapidly if the energies
of the two transitions are closely-matched. As for concentration quenching, up-conversion transitions are non-radiative.
2.3. TRIVALENT IRON GROUP, 3D
N
3D
N
TRANSITIONS 27
Table 2.2: Important parameters of Er:YAG and Er:Glass lasers.
Er:YAG Er:Glass
4
I
11/2

4
I
13/2
4
I
13/2

4
I
15/2
4
I
11/2

4
I
13/2
4
I
13/2

4
I
15/2
(nm) 2940 1646 1540

2
(s) 100 7700 100 8000

1
(s) 7700 8000
(GHz) 3500

21
(10
20
cm
2
) 2.6 0.5 0.7
(Wm
1
K
1
) 13 13 1 1
occurs to lower-lying Stark levels which have a large thermal population; this population must be signicantly
reduced by the pumping before a population inversion is achieved.
The properties of some Er lasers are listed in Table 2.2.
Er:Glass lasers have been operated in a variety of formats and in both continuous and pulsed mode. Q-
switched operation is able to generate pulses of nanosecond duration, which is of interest in eye-safe laser-
ranging. Alternatively, modelocked microlasers can generate pulses of approximately 20 ps duration with a
pulse repetition rate as high as several GHz.
2.3 Trivalent iron group, 3d
n
3d
n
transitions
2.3.1 Energy level structure
The ground-state conguration of neutral atoms in the iron group of the transition metals is of the form
[Ar] 3d
n+1
4s
2
or [Ar] 3d
n+2
4s, and hence that of the trivalent ions is [Ar] 3d
n
. The spectroscopy of these ions
is very dierent from that of the 4f
n
conguration in the trivalent rare earths owing to the fact that the size of
the 3d orbitals are larger than those of the 4f orbitals in the trivalent rare earths, and the ionic size (and hence
d) is smaller for the iron group than the rare earths. As a consequence the interaction with the crystal eld is
more than an order of magnitude bigger for the 3d electrons in the iron group than the 4f electrons in the rare
earths, to the extent that the interaction with the crystal eld is stronger than the spin-orbit interaction; the
LS coupling scheme no longer applies.
It is highly instructive to consider in outline the form of the wave functions and energy levels in the strong
crystal eld limit. In this case the eect of the crystal eld must be considered before the residual electrostatic
and spin-orbit interactions, and consequently we determine the eect of the crystal eld acting on the single-
electron orbitals formed in the central potential of the active ion.
The crystal eld removes the spherical symmetry of the free ion, and hence is no longer a good quantum
number. Instead the modied electron states are labelled by its symmetry properties, and by the electron spin,
which is unchanged by the interaction with the crystal eld.
As an example, we consider the case of a d-electron moving in a crystal eld with octahedral symmetry.
10
Octahedral symmetry of the crystal eld arises when the active ion is located at the centre of a cube, and
identical neighbouring ions are located on each face of the cube, as illustrated in Fig. 2.6. In the absence of
the crystal eld the d-electron is ve-fold degenerate, corresponding to m = 2, 1, 0, 1, 2. A crystal eld with
octahedral symmetry partially lifts this degeneracy to form a triply-degenerate crystal eld orbital labelled t
2
,
10
We note that other, generally lower, symmetries of the crystal eld are often treated by considering the dierence between
the true symmetry and octahedral symmetry as a perturbation which acts on the energy levels found for the case of octahedral
symmetry.
28 LECTURE 2. SOLID STATE LASER MATERIALS
active
ion
neighbouring
ion
e
t
2
Figure 2.6: Crystal eld orbitals of a d-electron in a eld with octahedral symmetry. The top row shows the
relative positions of the active ion and the neighbouring ions, and a surface of constant probability density for
the two degenerate wave functions corresponding to the e
g
orbitals. The bottom row shows equivalent plots for
the three degenerate wave functions corresponding to the t
2
orbital. Notice that the wave functions of the e
orbitals have lobes pointing towards the neighbouring ions, and hence have higher energy than the t
2
orbitals
for which the lobes in the wave function point between the neighbouring ions.
2.3. TRIVALENT IRON GROUP, 3D
N
3D
N
TRANSITIONS 29
d
4
e
t
2
d
6
e
t
2
d
7
e
t
2
d
8
e
t
2
d
9
e
t
2
d
10
e
t
2
d
e
t
2
d
2
e
t
2
d
3
e
t
2
Figure 2.7: Schematic diagram showing the ground-state congurations arising from splitting of a d
n
congu-
ration a strong crystal eld with octahedral symmetry. The arrows indicate electron spin.
and a higher-lying doubly-degenerate orbital e.
11
The spatial distributions of the t
2
and e orbitals are very
dierent, as illustrated schematically in Fig. 2.6. The e wave functions have regions of high probability density
centred on the axes connecting the active ion with the neighbouring ions, whereas for the t
2
orbitals these
regions are centred between the axes. Consequently electrons in the e orbitals are on average closer to the
neighbouring ions than those in the t
2
orbitals and therefore their interaction with the neighbouring ions
that is, with the crystal eld causes their energy to be increased from the unperturbed energy by a greater
amount than electrons in the t
2
orbitals.
We may now consider the congurations formed in the crystal eld as the number of d-electrons is increased.
The ground state conguration of a single d-electron is simply t
2
. Adding a second and third d-electron to
the system gives rise to ground-state congurations comprised of two or three t
2
electrons respectively, since
the t
2
orbital is three-fold degenerate. Since the spin-orbit interaction is minimized if the number of aligned
spins is maximized,
12
the spins of the electrons will all point in the same direction.
13
A fourth d-electron added
to the system may also occupy the t
2
orbital, but its spin must point in the opposite direction to satisfy the
Pauli exclusion principle.
14
Considerations of this kind lead straightforwardly to the ground-state congurations
illustrated schematically in Fig. 2.7.
The possible congurations of excited levels may be identied in a similar way. For example, for the case
of three d-electrons two dierent excited congurations may be formed: one of the spins of the t
2
electrons
may be reversed to give a conguration lying above the ground-state owing to the larger spin-orbit interaction;
alternatively, one of the electrons can be promoted to the e orbital. The two types of excited congurations
give rise to two types of transition, as illustrated in Fig. 2.8. In a spin-ip transition all the electrons remain
in the same orbital, but the spin of one of the electrons is reversed, as illustrated in 2.8(a); in a conguration
transition one electron makes a transition from an e orbital to a t
2
orbital.
15
As we shall explore further below, these two types of transition have very dierent characteristics. In a
11
Just as in molecular spectroscopy, a subscript g (standing for gerade) may also be included in the label to indicate that the
orbitals have even parity, i.e. e
g
and t
2g
. Orbitals with odd parity can be labelled with a subscript u, standing for ungerade.
12
This is Hunds rule.
13
Hence the ground-state congurations of t
2
2
will have S = 1 (a triplet), and that of t
3
2
will have S = 3/2 (a quartet).
14
We note that if the crystal eld were weak, the ground-state conguration would be t
3
2
e, with all spins parallel since this
minimizes the spin-orbit interaction; in other words the increase in energy associated with promoting an electron to the e orbital
would be smaller than the increase in energy caused by reversing the spin of the electron.
15
The spin of the electron may, or may not, also change.
30 LECTURE 2. SOLID STATE LASER MATERIALS
d
3
e
t
2
d
3
e
t
2
spin-flip
transition
d
3
e
t
2
d
3
e
t
2
configuration
transition
Figure 2.8: Spin-ip and conguration transitions of a d-electron in a crystal eld with octahedral symmetry.
spin-ip transition the spatial distribution of the total electronic wave function is unchanged. As a consequence
the strength of the interaction of the active ion with the neighbouring ion is unchanged and, as we will see, this
results in narrow-band transitions in which there is no change in the vibrational energy of the crystal lattice.
Transitions of this type are known as zero-phonon transitions.
16
In contrast, in a conguration transition the spatial distribution of the total electron wave function changes
signicantly. Since the interaction of the active ion with the neighbouring ions is very dierent for the upper
and lower levels, the neighbouring ions move during the transition. Conguration transitions are therefore asso-
ciated with changes in the vibrational motion of the lattice, and are characterized by broad-band, temperature-
dependent lineshapes the broad bandwidth arising from the emission or absorption of varying numbers and
energies of lattice phonons. Transitions of this type are therefore known as phonon-assisted or vibronic
transitions and are the basis of an important class of widely-tunable solid-state laser.
Conguration co-ordinate diagrams
Further insight into the properties of energy levels in the strong-coupling limit is provided by the conguration
co-ordinate diagram. In this model the active ion and the surrounding ions are treated as a single system.
To simplify matters this multi-dimensional problem
17
is reduced to one-dimension, and the distances between
the active ion and the surrounding ions is parameterized by a conguration co-ordinate Q. We can think of Q
as representing some sort of average value of the distance between active and neighbouring ions.
18
Within the conguration co-ordinate picture the Schrodinger equation for the active ion may be written in
the form,
[T
n
+T
e
+V (Q, r)] (Q, r) = E(Q, r), (2.3)
where: r represents the co-ordinates of the electrons; T
n
is the kinetic energy operator of the nuclei of the active
and surrounding ions; T
e
is the sum of the kinetic energy operators of the electrons in the active ion
19
; and
V (Q, r) is the electrostatic potential energy, comprised of terms arising from the attraction of the electrons
to the nuclei of the active and neighbouring ions, mutual repulsion between the electrons of the active and
16
The 4f
n
4f
n
transitions considered above are zero-phonon transitions since the interaction with the crystal eld is weak for
both the upper and lower levels.
17
If there are N neighbouring ions there are in principle N dierent values of the spacing between the active ion and the nearest
neighbours. Whether all N values are dierent depends on the nature of the lattice vibrations, as discussed in footnote 20.
18
Although see Note 20.
19
We will ignore the electrons in the neighbouring ions since they will not be involved in transitions.
2.3. TRIVALENT IRON GROUP, 3D
N
3D
N
TRANSITIONS 31
spectrum Q
A
b
s
o
r
p
t
i
o
n
E
m
i
s
s
i
o
n S
t
o
k
e
s

s
h
i
f
t
(a)
z
e
r
o
-
p
h
o
n
o
n
l
i
n
e
spectrum Q
(b)
Figure 2.9: Transitions between electronic levels: (a) when the equilibrium conguration co-ordinate is essen-
tially equal for the upper and lower electronic levels; (b) when the equilibrium conguration coordinates are
dierent for the two levels, causing the potential curves to be displaced. Vibrational wave functions are illus-
trated schematically for the lowest vibrational level and an excited vibrational level. In (b) the position of the
zero-phonon line is shown by the dotted curve.
neighbouring ions, and interactions between the nucleus of the active ion and the electrons and nuclei of the
neighbouring ions.
Just as for the case of diatomic molecules discussed last year, we may make the Born-Oppenheimer
approximation and assume that the total wave function may be written as a product of an electronic wave
function
e
(r) and a nuclear (or vibrational) wave function
n
(Q). We assume also that the electronic wave
function depends only weakly on Q. It is straightforward to show that the Schrodinger equation may then be
separated:
[T
e
+V (Q, r)]
e
(r) = E(Q)
e
(r) (2.4)
[T
n
+E(Q)]
n
(Q) = E
n
(Q). (2.5)
In principle, these equations may be solved as follows. The Schrodinger equation for the electronic wave function
can be solved for xed Q to yield an energy eigenvalue, E(Q), that is a function of the conguration co-ordinate.
We see from eqn (2.5) that this energy eigenvalue then acts as a potential well within which the nuclei of the
ions move. The energy of the nucleus moving in this potential is the total energy of the system (i.e. the energy
of the active ion and the vibrational motion of the lattice).
The potential well in which the ions move resembles that in which the nuclei of a diatomic molecule move,
and hence in a similar way the ion will exhibit a ladder of vibrational energy levels as shown.
The strength of an optical transition between two electronic levels is proportional to the square modulus of
the dipole (or higher order) matrix element. Since the dipole operator does not operate on the vibrational wave
function
n
(Q), the strength of optical transitions between two electronic levels is proportional to the square
32 LECTURE 2. SOLID STATE LASER MATERIALS
of the overlap integral of the vibrational wave functions of the upper and lower level just as in the case of
transitions in a diatomic molecule.
An important parameter of the potential well is the value of the conguration co-ordinate, Q
0
, for which the
potential is a minimum; this is the equilibrium value of the conguration co-ordinate.
20
If the spatial distribution
of the electronic wave functions is similar for the upper and lower electronic levels then the interaction of the
ion with the neighbouring ions will be similar for the two levels, and hence the upper and lower potential wells
will have similar values of the equilibrium values of the conguration co-ordinate and so lie vertically above one
another, as shown in 2.9(a). In contrast, if the spatial distributions of the wave functions for the upper and
lower levels is very dierent, the interaction of the electrons of the active ion will be dierent, and consequently
Q
0
will be dierent for the two levels. This situation is shown in 2.9(b).
We may now use the conguration co-ordinate picture to deduce the form of the absorption and emission
spectra of active ions for the two types of transition, as illustrated in Fig. 2.9. Suppose that the lower electronic
level corresponds to the electronic ground-state, and let us consider optical excitation to the upper electronic
level. The vibrational wave functions for the lowest vibrational level generally resemble a Gaussian curve, whilst
those of high-lying levels are strongly peaked near the classical turning points.
21
Absorption from the lower
electronic level will be dominated by transitions from the lowest vibrational level, since this will have by far
the largest thermal population. The strongest absorption will therefore by from here to the lowest vibrational
level of the upper electronic level, as shown in 2.9(a), since the overlap integral for these two vibrational wave
functions will be large. The overlap integral will be comparatively small for transitions to high-lying vibrational
levels since the vibrational wave function of such levels show rapid oscillations and are only large near the
classical turning points, where the amplitude of the wave function of the lowest vibrational level of the ground
state is small. In any case, excitation to excited vibrational levels of the upper electronic level will be followed
by extremely rapid de-excitation to the lowest vibrational level by phonon de-excitation. The emission spectrum
will therefore be dominated by emission from the lowest vibrational level of the upper electronic level. Just
as for the absorption spectrum, and for the same reason, the emission will be dominated by transitions to the
lowest vibrational level of the electronic ground-state. In other words, when the equilibrium co-ordinates of
the upper and lower electronic levels are similar or, equivalently, when the spatial distribution of the wave
functions of the upper and lower levels are similar, such as in spin-ip transitions the absorption and emission
spectra are dominated by a single, zero-phonon transition as we would expect from our earlier considerations.
In contrast, consider the case shown in 2.9(b) in which the spatial distribution of the electronic wave functions
are dierent for the upper and lower levels, and consequently the equilibrium conguration co-ordinates are
dierent. Now absorption from the lowest vibrational level of the ground electronic state is distributed over
many vibrational levels of the upper electronic level, leading to absorption extending to higher frequencies
than that of the zero-phonon line. Absorption at frequencies above that of the zero-phonon line corresponds
to electronic excitation accompanied by excitation of lattice vibrations. Following absorption, ions in excited
vibrational levels of the upper electronic level will undergo rapid phonon de-excitation to the lowest-lying
vibrational level. Just as for the absorption, emission will then occur to a wide range of vibrational levels of
the ground electronic state, producing a broad emission spectrum shifted to frequencies below that of the zero-
phonon line. The absorption and emission spectra are widely separated, the separation in frequency between
the peaks of the two spectra being known as the Stokes shift. Hence, as expected from our earlier discussion,
20
We see now that Q need not be thought of as the average value of the distance between active and neighbouring ions. Rather,
when Q = Q
0
all the neighbouring ions are in their equilibrium positions; when Q = Q
0
+Q all the neighbouring ions have increased
their separation from the active ion by a distance Q. The conguration co-ordinate model is therefore eectively considers the
lattice vibrations to be such that all the neighbouring ions change their separation from the active ion by the same amount, and
consequently can be represented by a single value of Q. Vibrational modes of this type are known as breathing modes.
21
For a given vibrational energy level the classical turning points are the values of Q for which the potential curve has an energy
equal to the total vibrational energy. In classical mechanics the kinetic energy of a particle is zero at such points, and hence they
represent the extrema of its oscillatory motion. Classically, the probability of nding a particle is greatest at the turning points,
since there it is stationary. The Correspondence Principle therefore tells us to expect the wave functions of high-lying (i.e. with
large energy) vibrational levels to be strongly peaked at these points.
2.3. TRIVALENT IRON GROUP, 3D
N
3D
N
TRANSITIONS 33
Cr
3+
:Al
2
O
3
(ruby) Cr
3+
:BeAl
2
O
4
(alexandrite)
2
E
4
T
2
4
T
1
4
A
2
2
E
4
T
2
4
T
1
4
A
2
o
p
t
i
c
a
l

p
u
m
p
i
n
g
l
a
s
e
r

t
r
a
n
s
i
t
i
o
n
o
p
t
i
c
a
l

p
u
m
p
i
n
g
z
e
r
o
-
p
h
o
n
o
n
v
i
b
r
o
n
i
c
4
A
2
4
T
2
4
T
1
2
E
2
T
1
2
T
2
2
A
1
4
A
2
4
T
2
4
T
1
2
E
2
T
1
2
T
2
2
A
1
(a) (b) (c) (d)
t
2
3
t
2
2
e t
2
3
t
2
2
e
Figure 2.10: The crystal eld congurations and energy levels (a and c) and simplied conguration co-ordinate
diagrams (b and d) for ruby and alexandrite. Note that the diagrams are not to scale, and for clarity not all
the energy levels shown in (a) and (c) are shown in the corresponding conguration co-ordinate diagram.
transitions between electronic levels with dierent spatial distributions (and hence dierent values of Q
0
), such
as conguration transitions, are broad-band, vibronic transitions. Note that the width of the absorption and
emission bands, and the magnitude of the Stokes shift, both increase with the magnitude of the dierence in
the equilibrium conguration co-ordinates of the two levels.
2.3.2 The ruby laser
The ruby laser is of enormous signicance since it was the rst optical maser. It is much less commonly used
today than it once was, but it still nds applications in holography (where the relatively short wavelength for
a solid-state laser is better suited to the response of photographic emulsions), pulsed interferometry, and some
medical applications. It is also the best known example of a three-level laser.
Crystal properties
Ruby is formed by doping Cr
3+
ions into a sapphire (Al
2
O
3
) host with a doping level
22
of approximately
0.05 wt%, corresponding to an active ion density of 1.58 10
19
cm
3
. The chromium ions, of radius 62 pm,
substitute for the aluminium ions, of radius 53 pm, with little distortion of the host lattice. Ruby may be grown
by the Czochralski method, and has excellent mechanical, thermal, and chemical properties. Ruby laser rods
typically have a diameter of between 3 and 25 mm and a lengths as long as 200 mm.
Energy levels
Figure 2.10 shows the energy levels of Cr
3+
ions doped in Al
2
O
3
together with a simplied conguration co-
ordinate diagram. In ruby the crystal eld is strong, and hence, as discussed in Section 2.3.1 the crystal eld
22
In ruby gemstones the doping level of chromium ions is much higher, of order 1%, which gives them a rich red colour. Ruby
laser rods typically appear pink.
34 LECTURE 2. SOLID STATE LASER MATERIALS
conguration of the ground state is t
3
2
. This conguration forms crystal eld terms labelled
23
as
4
A
2
(the ground-
state),
2
E,
2
T
1
, and
2
T
2
where the superscripts are equal to 2S + 1. The lowest-lying excited conguration is
t
2
2
e, which forms the levels
4
T
2
,
4
T
1
and
2
A
1
. Of particular importance is the fact that since the crystal eld
is strong the
4
T
2
and
4
T
1
terms (which arise from the t
2
2
e conguration) lie above the
2
E term (which arises
from the t
3
2
conguration.
24
Following the argument above, the wave functions of the levels of the t
2
2
e conguration have a very dierent
spatial distribution than those of the t
3
2
conguration, and consequently, as shown in Figure 2.10 the equilibrium
conguration co-ordinate of the
4
T
2
and
4
T
1
levels is quite dierent from that of the
4
A
2
and
2
E levels.
The absorption spectrum of ruby is dominated by the
4
T
2

4
A
2
and
4
T
1

4
A
2
transitions since these
are fully spin-allowed. Since these are conguration transitions, as discussed in Section 2.3.1, they form broad,
vibronic pump bands some 50 nm wide and centred at approximately 554 nm and 404 nm respectively.
25
These
may be pumped conveniently by ashlamp radiation. Absorption to the doublet levels are much weaker since
they are spin-forbidden.
Absorption on the
4
T
2,1

4
A
2
pump bands is followed by rapid non-radiative relaxation to the lowest-lying
excited level: the
2
E level. This level is metastable since it can only decay on the spin-forbidden
2
E
4
A
2
transition; in fact it has one of the longest uorescence lifetimes of any level in the solid state, approximately
3 ms. Since the
2
E
4
A
2
transition involves no change in the crystal eld orbitals, it is a narrow-band,
zero-phonon transition.
The
2
E level is actually split
26
into two levels separated by only 29 cm
1
. Population is rapidly transferred
between these two levels by phonon collisions, so that the levels may be considered to be in thermal equilibrium,
with a slightly larger population in the lower of the two
2
E levels. The transitions from the two levels of
2
E to
4
A
2
are denoted R
1
and R
2
.
Laser parameters
The R
1
and R
2
transitions are homogeneously broadened by phonon interactions to give a temperature-
dependent linewidth, which at room temperature is approximately equal to 450 GHz. The optical gain cross-
section of the R
1
transition in ruby is approximately 2.5 10
20
cm
2
.
The energy levels of ruby therefore have many desirable features for a laser transition: broad pump bands
allow for ecient absorption of broad-band pump radiation; almost all the ions excited on these transitions
are transferred by rapid non-radiative transitions to the upper laser level; the upper laser level has a long
uorescence lifetime, and therefore acts as a storage level for excitation. The main diculty, of course, is
that the laser transition occurs to the ground-state and consequently the laser is a three-level laser system as
illustrated schematically in Fig. 2.11.
In order to achieve a population inversion the optical pumping must be suciently strong to deplete the
population of the ground-state by approximately one-half, as discussed in the course last year. The splitting
of the
2
E level causes the lasing transition to be split into two, closely-spaced lines denoted R
1
and R
2
with
wavelengths of 694.3 nm and 692.8 nm respectively. Of these two transitions, the R
1
line at 694.3 nm usually
dominates owing to the fact that it arises from the lower of the
2
E levels, and consequently has a higher thermal
population.
23
Just as for levels in a free atom or ion, single-electron orbitals are labelled with lower-case letters, crystal eld terms are labelled
with upper-case letters.
24
The
4
T
2
and
4
T
1
levels may also be denoted in an alternative notation as
4
F
2
and
4
F
1
respectively.
25
It is these strong absorption bands in the green and violet part of the spectrum which are responsible for the striking colour of
ruby gemstones.
26
The splitting arises from a combination of the spin-orbit interaction and lowering of the symmetry of the crystal eld from
octagonal symmetry. The
4
A
2
ground state is also split by the same interactions, but its splitting is even smaller. The split levels
of the
2
E level are denoted
2
E
3/2
and
2
E
1/2
, the latter being the lower. In the alternative notation mentioned in Note 2.3.2 these
levels are denoted

E and 2

A.
2.3. TRIVALENT IRON GROUP, 3D
N
3D
N
TRANSITIONS 35
4
A
2
2
E
694 nm
rapid non-
radiative decay
optical
pumping
2
1
3
{
4
T
2
4
T
1
29 cm
-1
R
2

=

6
9
2
.
8

n
m
R
1

=

6
9
4
.
3

n
m
2
E
Figure 2.11: Simplied energy level scheme for the ruby laser. Inset: Splitting of the upper laser level.
Practical implementation
Since ruby is a three-level laser, it is almost invariably operated in a pulsed mode. Optical pumping is usually
provided by a medium-pressure xenon ashlamp.
The long lifetime of the upper laser level means that it is straightforward to Q-switch the laser, and indeed
this is necessary if strong laser spiking is to be avoided. Q-switched ruby lasers typically generate pulses with
an energy of several Joules, and a duration between 10 and 50 ns. Ruby lasers may also be actively or passively
modelocked to give output pulses of 5 to 10 ps duration.
Table 2.3: Important parameters of the ruby, alexandrite, and Ti:sapphire laser transitions.
Ruby Alexandrite Ti:sapphire
(nm) 694.3 680.4, 700 - 820 660 - 11 80

2
(s) 3000 260 (vibronic) 3.8

1
100 ps (vibronic) 100 ps

21
(10
20
cm
2
) 2.5 1 0.34
(GHz) 330 330,26 000 100 000
Doping level (%) 0.05 0.1 0.1
Pump bands (nm) 375 - 425, 525 - 575 380 - 630 380 - 620
36 LECTURE 2. SOLID STATE LASER MATERIALS
2.3.3 Alexandrite laser
Alexandrite, or chromium-doped chrysoberyl (Cr
3+
:BeAl
2
O
4
), was the rst commercially important tunable
solid-state laser. Alexandrite lasers have proved useful in applications where their wide tunability is useful, such
as detection of pollutant species in the atmosphere and in several medical applications.
Crystal properties
The Aluminium ions of the host material occupy two types of sites within the lattice: one in which the crystal
eld has a mirror symmetry, and one for which the eld exhibits inversion symmetry. Typical doping levels are
typically 0.04 - 0.1 at.%, the chromium ions substituting for aluminium ions, with a signicant majority (78%)
going into the sites with mirror symmetry.
The host is optically biaxial, leading to a strong polarization dependence on its optical properties which
dominates thermal or stress-induced birefringence. The host has a very high thermal conductivity, a high
optical damage threshold, is chemically stable, and has good mechanical strength and hardness. Laser rods of
high optical quality may be produced with diameters of about 5 mm diameter and lengths up to 100 mm.
Energy levels
Figure 2.10(b) and (d) shows the energy levels of alexandrite. In this material the crystal eld is of intermediate
strength and consequently the
4
T
1
and
4
T
2
levels are closer to the
2
E level than is the case for Cr
3+
ions in
Al
2
O
3
(i.e ruby). For example, in alexandrite the
4
T
2
level lies only 800 cm
1
above the
2
E level, compared to
approximately 1500 cm
1
in ruby.
As in the case of ruby, broad pump bands are provided by the vibronic
4
T
2

4
A
2
and
4
T
1

4
A
2
transitions
lying at 550 - 650 nm and 350 - 450 nm respectively. These may be pumped by xenon or mercury ashlamps,
or by argon and krypton ion lasers.
Absorption on the
4
T
2

4
A
2
pump bands is followed by rapid non-radiative relaxation to the lowest
vibrational level of the
4
T
2
level and to the
2
E level. Since the
4
T
2
level lies only 800 cm
1
above the
2
E level,
population is transferred between these levels by rapid non-radiative transitions to form a thermal distribution
over the two levels. The radiative lifetime of the
2
E level is similar to that of the same level in ruby: 2.3 ms;
that of the
4
T
2
level is somewhat shorter, 7.1 s, since it may decay to the ground state via a spin-allowed
transition. Since the population in these levels is shared, the eective lifetime of ions in either level depends on
the relative population in each level, and is therefore strongly temperature dependent. At room temperature
the eective lifetime of either level is approximately 300 s.
Laser parameters
Lasing can occur in alexandrite on the analogous transition to the ruby laser transition: the zero-phonon
2
E
4
A
2
transition at 680.4 nm, which operates as a three-level scheme. However, of more importance is the
4
T
2

4
A
2
since this is a vibronic transition and hence is widely tunable between approximately 700 nm and
818 nm.
The vibronic laser transition operates as a four-level laser: Pumping on the broad-band
4
T
2,1

4
A
2
transitions is followed by rapid non-radiative relaxation to the bottom of the
4
T
2
level and to the
2
E level; the
2
E level acts as a population storage level, increasing the eective lifetime of the upper laser level considerably;
lasing can then occur to excited vibrational levels of the
4
A
2
ground electronic level, followed by rapid non-
radiative decay to the lowest vibrational level of the ground electronic state. A simplied energy level diagram
of the operation of the vibronic laser transition in alexandrite is shown in Fig. 2.12.
Since the populations of the
4
T
2
and
2
E levels are closely-coupled, it is appropriate to use an eective
optical gain cross-section together with the total population in the two levels, as discussed on page 23 for the
2.3. TRIVALENT IRON GROUP, 3D
N
3D
N
TRANSITIONS 37
2
E
4
T
2
4
T
1
4
A
2 o
p
t
i
c
a
l

p
u
m
p
i
n
g
700 - 818 nm
rapid
relaxation
rapid non-
radiative decay
optical
pumping
2
0
1
3
{
4
T
2
*
4
T
1
4
T
2
4
A
2
*
4
A
2
(a) (b)
energy
storage
Figure 2.12: Simplied energy level scheme for the vibronic laser transition in alexandrite. (a) Simplied
congurational co-ordinate diagram illustrating how the lowest vibrational level of the
2
E level acts as a storage
level and is in thermal equilibrium with the lowest vibrational level of
4
T
2
; (b) Energy level scheme showing
the four-level nature of the laser transition. Note in (b) an asterisk indicates that the level is in an excited
vibrational level.
case of the Nd:YAG laser. The eective cross-section for the vibronic transition in alexandrite is approximately
1 10
20
cm
2
, although this is strongly dependent on temperature.
The main parameters of the laser transitions in ruby and alexandrite are given in Table 2.3
Practical implementation
Alexandrite lasers operate in a similar conguration to Nd:YAG lasers. The main dierence is that the laser
rod may be heated to between 50 and 70

C in order to increase the relative population in the

4
T
2
level.
27
Alexandrite lasers are usually ashlamp-pumped. Although they may in principle be operated continuously,
the relatively low gain cross-section means that the threshold population inversion density is high, and so pulsed
operation is more usual. Pulse repetition rates between 10 and 100 Hz are typical, with output pulse energies
of up to about 1 J.
Alexandrite lasers naturally produce output pulses with a duration close to the eective lifetime of the upper
laser level, i.e. of 200 - 300 s. Q-switching reduces the duration of the output pulse to typically 50 ns.
2.3.4 Ti:sapphire
Crystal properties
In titanium-doped sapphire, Ti:Al
2
O
3
, titanium ions (ionic radius 74.5 pm) substitute for aluminium ions (ionic
radius 53 pm) with a typical concentration of 0.1 wt%. High-quality crystals may be grown by the Czochralski
technique.
27
Heating to higher temperatures would increase the population of the
4
T
2
level still further. However, raising the temperature
of the laser rod also increases the thermal population of the lower laser level, and hence the optimum temperature is below that
which would maximize the thermal population of the upper laser level.
38 LECTURE 2. SOLID STATE LASER MATERIALS
2
E
2
T
2 o
p
t
i
c
a
l

p
u
m
p
i
n
g
2
1
3
rapid
relaxation
rapid
relaxation
vibronic laser transition
660 1180 nm
0
Figure 2.13: Simplied energy level structure and laser transitions of the Ti
3+
ion in Al
2
O
3
(Ti:sapphire).
Energy levels
The Ti
3+
ion has only a single d-electron, and consequently its energy level structure is rather simple: the
ground crystal eld conguration is t
2
, which forms a
2
T
2
term; the only other possible crystal eld orbital for
a d-electron is e, which gives rise to a
2
E term, as shown in Figure 2.13.
Radiative transitions between these two electronic levels obviously involve a change in the crystal eld
conguration, and are therefore broad, vibronic transitions. It is worth noting that the sparse energy level
structure of Ti:sapphire means that losses arising from excited-state absorption are negligible.
Figure 2.14 shows the absorption and emission spectra of Ti:sapphire. The very broad bands, and their wide
separation is a consequence of the large dierence in equilibrium conguration co-ordinates for the upper and
lower levels. Absorption from the ground electronic level occurs on the strong vibronic
2
E
2
T
2
transitions
to the excited vibrational levels of the
2
E level, with wavelengths between 630 nm and 400 nm. Ions in excited
vibrational levels decay rapidly by non-radiative transitions to the lowest vibrational level of the
2
E level, which
forms the upper level of the laser transition. At room temperature ions in this level decay predominantly
radiatively to the vibrational levels of the
2
T
2
level, with a uorescence lifetime of approximately 3.2 s, the
relatively short lifetime reecting the fact that the transition to the ground state is spin-allowed.
Laser parameters
The Ti:sapphire laser is a four-level laser: lasing may occur from the lowest vibrational level of the
2
E level
to excited vibrational levels of the
2
T
2
level, at wavelengths tunable between 660 nm and 1180 nm. Indeed
Ti:sapphire has the largest bandwidth of any laser.
Although the bandwidth of the laser transition is very large, the large radiative decay rate of the laser
transition ensures that the optical gain cross-section remains relatively high for a solid-state laser: approximately
3 10
19
cm
2
, which is comparable to that in Nd:YAG
2.3. TRIVALENT IRON GROUP, 3D
N
3D
N
TRANSITIONS 39
Wavelength (nm)
A
b
s
o
r
p
t
i
o
n

/

e
m
i
s
s
i
o
n

(
a
r
b
.

u
n
i
t
s
)
Figure 2.14: Absorption and emission bands of the Ti:sapphire laser.
Practical implementation
A major dierence between Ti:sapphire and all the lasers discussed so far in this chapter is the short uorescence
lifetime of the upper laser level. Flashlamp pumping is therefore not usually very ecient owing to the much
longer duration of the ashlamp pulse. Instead, laser pumping is usually employed, the most commonly employed
pump lasers being frequency-doubled Nd lasers (YAG, YLF or vanadate).
Figure 2.15 shows the construction of two common types of Ti:sapphire laser oscillator. In both designs the
Ti:sapphire laser rod is 2 - 10 mm long, with end faces cut at Brewsters angle so as to eliminate reection
losses for one polarization. The z-folded cavity compensates for astigmatism. This arises because it is necessary
to focus the pump radiation to a small cross-sectional area in order to achieve sucient population inversion.
The design of the optical cavity design must therefore be such that the transverse modes are also brought to a
focus at the Ti:sapphire crystal. However, the Brewster-cut faces of the crystal introduce astigmatism to the
intra-cavity beam i.e. the divergence of the beam becomes dierent in the horizontal and vertical planes. In
the cavity shown in Fig. 2.15 this is corrected by the two concave mirrors used away from normal incidence,
which introduce astigmatism of the opposite sign to that caused by refraction at the laser rod.
Continuously-tunable c.w. Ti:sapphire lasers may be tuned between approximately
28
700 nm and 1000 nm,
with output powers of 1 - 2 Watts. Wavelength tuning may be accomplished by a variety of techniques. In the
design illustrated in Figure 2.15(a), tuning is achieved by adjusting a birefringent lter.
29
Other techniques,
28
The upper and lower limits to the tuning curve may be extended to approximately 100 nm and 675 nm respectively by employing
cavity mirrors optimized for these wavelengths. However, it is not possible continuously to tune from 675 nm to 1100 nm without
changing the cavity mirrors.
29
At its simplest a birefringent lter is simply a plate of birefringent material with its optical axis in the plane of the plate. The
retardance introduced by the plate depends on the angle between the optic axis and the electric eld of the incident light, and on
the wavelength of light. As the plate is rotated the wavelength(s) for which the retardance is equal to an integer times 2 varies.
40 LECTURE 2. SOLID STATE LASER MATERIALS
Ti:sapphire
rod
Pump laser
output
coupler
turning
mirror
high
reflector
dispersion compensating
prisms
(a)
(b)
Ti:sapphire
rod
Pump laser
output
coupler
turning
mirror
high
reflector
birefringent
plate
Figure 2.15: Two types of Ti:sapphire laser oscillator: (a) a wavelength-tunable oscillator; (b) a modelocked
oscillator incorporating intra-cavity dispersion control.
such as prisms or gratings, may be employed for selecting the output wavelength. For the oscillator shown in
Fig. 2.15(a) the linewidth of the laser output is typically about 40 GHz; this may be reduced to better than
1 GHz by employing an intra-cavity etalon. Further reduction of the linewidth to around 10 MHz is possible
by employing a ring cavity, and by locking the operating frequency to an external cavity the linewidth can be
reduced to as little as 50 kHz.
Ti:sapphire lasers are widely used to generate very short optical pulses by the Kerr modelocking technique
discussed in Lecture 4. The cavity design shown
30
in 2.15(b) is typical of modelocked Ti:sapphire lasers, and
incorporates a pair of prisms for dispersion control as discussed in Lecture 5. The train of modelocked laser
pulses from such a system typically has a pulse repetition rate of 80 MHz, a mean power of approximately
0.5 W (i.e. 5 nJ per pulse), with pulse durations of 35 - 50 fs. Modelocked oscillators of this type can also be
wavelength tuned, although the tuning range is more restricted: 780 - 820 nm is common.
The polarization of these wavelength(s) is unchanged, unlike other wavelengths which in general become elliptically polarized, and
consequently experience lower losses in passing through a linear polarizer (or any other polarization-sensitive element, such as the
laser rod). In order to reduce losses, the lter can be mounted at Brewsters angle, and tuning achieved by rotating the lter about
an axis parallel to its normal.
30
In designing suitable cavities it is desirable to maximize the ratio of the round-trip losses under c.w. operation to those when
Kerr lens modelocking occurs.
2.3. TRIVALENT IRON GROUP, 3D
N
3D
N
TRANSITIONS 41
2.3.5 Host materials
The host into which the active ion of a solid-state laser is doped plays an important role in the operation of
the laser through its physical properties and its inuence on the energy level structure of the ion. Desirable
physical properties include: hardness, so that the material surface may be polished to an optical-quality nish;
high thermal conductivity, so that waste heat may be conducted away rapidly; and ease of growth, so that the
material may be grown to the required size whilst maintaining high optical quality throughout the sample. It
is also desirable that the crystal exhibits low thermal lensing
31
and low thermally-induced birefringence.
32
At
the microscopic level, the host material must be capable of accepting the active ion into its structure.
Solid state hosts fall into three classes: crystals, glasses, and ceramics, as discussed below.
Crystalline hosts
Crystalline hosts have thermal conductivities which are typically an order of magnitude greater than in glass
hosts, and can therefore sustain higher mean pumping powers before the thermal lensing of the laser rod
becomes problematic. Laser transitions in crystal hosts are usually homogeneously broadened, which makes
single mode operation easier to achieve, and gives narrower linewidths for non-vibronic transitions (see Section
2.3.1). Disadvantages of crystal hosts are that they are often birefringent, making them polarization sensitive,
and it can be dicult to grow large crystals of high optical quality.
Glass hosts
Glass hosts have several advantages over crystals: glass can be cast into many forms from optical bres to
metre-sized discs with good homogeneity and high optical quality; glass is inexpensive; and it has low, or
zero, birefringence. Further, to some extent, the physical properties of the glass and the properties of the laser
transition can be controlled by adjusting the composition of the glass. The biggest drawback of glass is the
relatively low thermal conductivity, which makes thermal management more dicult. Optical transitions within
ions doped into a glass host are always inhomogeneously broadened, with linewidths several orders of magnitude
larger than non-vibronic transitions in crystalline hosts. This oers advantages in providing broad pump bands,
but means that the laser transition will also be broad. As a consequence, the optical gain cross-section will
be low, making it more dicult to achieve laser oscillation in an oscillator owing to the lower round-trip gain
(although this restriction does not apply to bre lasers). The low gain cross-section has signicant advantages,
though, in laser ampliers. In an amplier an upper limit on the single-pass gain is set by the onset of parasitic
laser oscillation caused by small Fresnel reections from the ends or edges of the laser rod, or depletion of the
gain by strong amplied spontaneous emission. The low gain cross-section obtained with glass laser materials
allows a larger population inversion to be generated before this is a problem, and this allows more energy to be
extracted by an injected laser pulse.
To summarize: the higher gain and good thermal conductivity of crystalline hosts means that they are often
employed in laser oscillators and lower-power ampliers; glass hosts are frequently used in high-energy pulse
ampliers (with low mean power) owing to the low optical gain cross-section and ease with which large rods
and discs may be fabricated.
31
Thermal lensing refers to a spatial variation of the refractive index of the laser gain medium arising from spatial variations
in temperature and thermal stresses within the material. Frequently the induced variation in refractive index and, for a solid
material, possible changes in the surface gure of the laser rod or disk causes the gain medium to act as a thin lens of focal length
f. For solid-state lasers f may be positive or negative, with a magnitude that can be as small as a few 10s of centimetres. Further
details may be found in Koechner and Bass Solid-State Lasers: A Graduate Text, Springer (2003).
32
The change in refractive index resulting from thermally-induced strain within the crystal can be dierent for dierent polariza-
tions. For example, in a laser rod the change in refractive index will in general be dierent for the radial and azimuthal polarization
components.
42 LECTURE 2. SOLID STATE LASER MATERIALS
Lecture 3
Q-Switching
3.1 Laser Spiking and relaxation oscillations
Two types of transient behaviour are frequently observed in the output of laser systems:
1. Laser spiking: This can occur when a laser is rst turned on, and comprises large amplitude, somewhat
irregularly-spaced, pulses of laser radiation.
2. Relaxation oscillations: These are more regular oscillations in the output power of the laser, typically
with an exponentially-decaying amplitude. Relaxation oscillations can occur in the tail of a series of laser
spikes, or can follow a small perturbation to the laser system such as disturbance of a cavity mirror, or a
change in the pumping power.
The two phenomena are related; both arise from the fact that the population inversion density and the
intra-cavity laser intensity are coupled together via stimulated emission and feedback from the cavity mirrors,
and that the inversion density and intra-cavity intensity respond to changes in the operating conditions of the
laser on dierent time scales.
Consider rst the case of laser spiking. Immediately after the pumping is switched on, the population in the
upper laser level will grow from zero in a time of order the upper level lifetime
2
. At this point the density of
photons in the cavity will be very low since the photons arise from spontaneous emission only. At some point
the population inversion density will reach the threshold for laser oscillation, N

th
, but laser oscillation will not
commence since the density of photons in the relevant cavity mode will still be low. As pumping continues, the
population inversion grows beyond the threshold value, and the density of photons in the laser mode will build
up by stimulated emission. Eventually the density of photons in the laser mode will be suciently high to burn
the population inversion down towards N

th
, further increasing as it does the rate of stimulated emission and
the photon density. Inevitably, the population inversion will be driven well below the threshold value, leading
to reduction in the photon density and the cessation of lasing. The dramatic growth and rapid decline of the
intra-cavity photon density correspond to a spike in the output power of the laser. As pumping continues, the
population inversion will start to grow again and the cycle can repeat. Laser spiking of this type will not usually
continue indenitely, even in continuously-pumped lasers, owing to the fact that the population inversion and
photon densities do not decay to zero after the spike. Consequently, after each spike these quantities tend to end
up closer to their steady-state values, in which case the spiking will decay. Figure 3.1 shows a typical example
of laser spiking from the rst ever ruby laser.
It should be noted that laser spiking is often associated with rapid jumps in the longitudinal and/or transverse
mode of the laser. This occurs for essentially the same reasons as spatial hole-burning in lasers operating on
homogeneously broadened transitions: immediately after a laser spike, signicant population inversion may
43
44 LECTURE 3. Q-SWITCHING
0 1 t/ms
flashlamp
intensity
laser
output
Figure 3.1: Laser spiking in the rst ruby laser. The top trace shows the ashlamp pulse, the lower the output
from the ruby laser as recorded by a photodiode. Note that the sharp feature at t = 0 is caused by ashlamp
radiation reaching the photodiode.
remain in regions near nodes of the longitudinal nodes, or in regions where the transverse prole of the mode of
the previous pulse was low in intensity. This gain can be exploited by other modes which, as pumping continues,
may therefore reach threshold sooner than the mode (or modes) of the previous spike.
Laser spiking tends to occur when the time scale on which the population inversion responds to the pumping,
typically of order
2
, is slow compared to that in which the photon density changes in the cavity, which is of
order the cavity lifetime
c
. Hence we expect laser spiking to occur if,

2

c
Approximate condition for laser spiking. (3.1)
3.1.1 Rate equations
For a homogeneously broadened laser the rate equations for the upper and lower laser level may be written
simply as
1
,
1
In writing the rate equations in this form we have made several simplifying assumptions: we assume that the laser oscillates
on a single longitudinal and transverse mode; and we ignore transverse variations in the intra-cavity laser intensity and in the
population inversion.
3.1. LASER SPIKING AND RELAXATION OSCILLATIONS 45
dN
2
dt
= R
2
(t) N

(t)
21
I(t)

N
2
(t)

2
(3.2)
dN
1
dt
= R
1
(t) +N

(t)
21
I(t)

N
1
(t)

1
+N
2
(t)A
21
, (3.3)
where R
i
is the pump rate of level i, and
L
and I(t) are the angular frequency and total intensity of the
oscillating laser mode.
Notice that in the absence of stimulated emission (I(t) = 0) the population of the upper laser level grows in
a time of order
2
.
The rate per unit volume at which photons are stimulated into the oscillating mode is simply N

21
I/
L
.
Hence if we suppose that the fraction of the volume of the oscillating mode lying within the gain medium is
equal to f
c
, the rate equation for the density of photons is simply,
dn
dt
= f
c
N

(t)
21
I(t)

n(t)

c
. (3.4)
The last term gives, in terms of the cavity lifetime
c
,the rate of loss of photon density by coupling through
the output mirror (plus any other losses). We derive an expression for
c
in Section 3.1.2.
This equation may be simplied by noting that I(t) = n(t)
L
c. Then,
dn
dt
=
_
N

th
1
_
n

c
, (3.5)
where,
N

th
=
1
f
c

21
c
c
(3.6)
and we have dropped the time dependence of N

(t) and n(t) to avoid clutter. We see that N

th
is the threshold
population inversion density since we must have N

> N

th
for the photon density to grow (by stimulated
emission). Equation (3.5) also shows that the time scale in which the photon density changes is
c
divided by
a factor which depends on the extent to which the population inversion exceeds N

th
. As such we expect the
photon density to change in a time of order
c
, as stated earlier.
3.1.2 Cavity lifetime
To calculate the cavity lifetime we consider some photons circulating within a laser cavity. In the absence of
further photons being supplied by any source (an injected beam, or by spontaneous or stimulated emission),
the number of photons within the cavity will decrease with time. Suppose that at some moment the cavity
contains n photons per unit volume. The ux i.e. the number of photons per unit area, per unit time of
photons striking each mirror is
1
2
nc, the factor of one-half arising since half the photons travel to the left, and
half to the right. For each mirror i the proportion of these photons which are transmitted is equal to T
i
. Hence,
ignoring any other losses, the density of photons within the cavity will decay according to
dn
dt
=
1
2
ncA
mode
(T
1
+T
2
)
V
mode
,
46 LECTURE 3. Q-SWITCHING
50 0 100 150 200 250 300
1
0
2
3
4
1
0
2
3
4
5
t / c
50 0 100 150 200 250 300
t / c
5 0 10 15 20 25 30
t / c
5 0 10 15 20 25 30
t / c
5 0 10 15 20 25 30
t / c
N
*
(
t
)

/

N
*
t
h
1
0
2
3
4
N
*
(
t
)

/

N
*
t
h
1
0
2
3
4
N
*
(
t
)

/

N
*
t
h
n
(
t
)

/

n
t
h
1
0
5 0 10 15 20 25 30
t / c
2
3
4
5
n
(
t
)

/

n
t
h
1
0
2
3
4
5
n
(
t
)

/

n
t
h
(a)
(b)
(c)
Figure 3.2: Numerical analysis of laser spiking in an ideal four-level laser with a lling factor f
c
of unity pumped
with an over-pumping ratio r = 4 and: (a)
2
/
c
= 100; (b)
2
/
c
= 10; (c)
2
/
c
= 1. In each case the plots
show the temporal evolution of the population inversion density and the photon density normalized to the
steady-state value, n
th
.
where A
mode
and V
mode
are the area and volume of the oscillating mode. Hence we may write,
dn
dt
=
n

c
,
where the cavity lifetime is given by

c
=
2V
mode
A
mode
c
1
T
1
+T
2
.
Now V
mode
A
mode
L
c
, and hence

c
=
2L
c
c
1
T
1
+T
2

L
c
c
1
1 R
Cavity lifetime
where R =

R
1
R
2
and the approximation holds if R
1
R
2
1.
3.1.3 Numerical analysis of laser spiking
Fig. 3.2 shows numerical solutions of these rate equations for a system with a lling factor f
c
of unity and
constant pumping at a rate corresponding to an over-pumping ratio (see Section 3.1.4) r = 4. In each of these
3.1. LASER SPIKING AND RELAXATION OSCILLATIONS 47
simulations the pumping would, in the absence of stimulated emission, produce a population inversion density
which was a factor of r greater than the threshold value. However, once the population inversion rises above
N
th
the rate of stimulated emission exceeds the rate of loss of photons from the cavity, increasing the density
of photons within the cavity, and causing in turn the population inversion to burn down towards the threshold
value. In all the cases shown the population inversion and photon densities tend to their steady-state values
N
th
and n
th
respectively at long times after the pumping is switched on at t = 0. Substantial laser spiking is
observed when
2
/
c
1, consistent with our discussion above, and the number and amplitude of the spikes
in the laser output is seen to decrease as the ratio
2
/
c
is reduced. In particular, for the case
2
/
c
= 1 there
is essentially no spike in the photon density. Instead the population inversion density photon density reaches
rN
th
before the smoothly-rising photon density burns the population inversion down to the steady-state value.
3.1.4 Analysis of relaxation oscillations
Relaxation oscillations, being somewhat gentler than laser spiking, are amenable to algebraic analysis. We con-
sider the simple example of an ideal homogeneously broadened, four-level laser system in which the population
of the lower laser level is essentially zero at all times. In this case eqn (3.2) is also the rate equation for the
population inversion density:
dN

dt
=
dN
2
dt
= R
2
N

21
I

N
2

2
(3.7)
= R
2

1
f
c
N

th
n

2
(3.8)
Notice that in the absence of any stimulated emission (I = 0) the population inversion would reach a steady-state
value of R
2

2
.
Under steady-state conditions N

(t) = N

th
. Substituting this into eqn (3.8) then yields an expression for
the steady-state photon density,
n
th
=
_
R
2

th

2
_
f
c

c
(3.9)
= (r 1)
N

th
f
c

2
, (3.10)
where the parameter r = R
2

2
/N

th
is the ratio of the population inversion that would be produced in the
absence of any stimulated emission to N

th
. It is also equal to the over-pumping ratio that will appear in our
discussion of Q-switching.
We now suppose that having established steady-state conditions, the laser is subject to a small perturbation
that shifts the population inversion and photon densities to:
N

(t) = N

th
+ N

(t) (3.11)
n(t) = n
th
+ n(t) (3.12)
We may then linearize the dierential equations for N

(t) and n(t) by substituting eqns (3.11) and (3.12)

into (3.5) and (3.8), and ignoring the small product terms in N

(t)n(t). We nd:
dN

dt
= r
N

1
f
c
n

c
(3.13)
dn
dt
= (r 1)f
c
N

2
(3.14)
48 LECTURE 3. Q-SWITCHING
We now assume solutions of the form,
N

(t) = a exp(mt) (3.15)

n(t) = b exp(mt). (3.16)
Substitution into eqns (3.13) and (3.14) then gives,
_
m+
r

2
_
a +
1
f
c

c
b = 0 (3.17)
(r 1)
f
c

2
a mb = 0. (3.18)
These two simultaneous equations can be solved by the usual matrix method: non-trivial solutions require the
determinant to be zero, yielding,
m =
r
2
2

_
r
2
2
_
2

r 1

c
. (3.19)
Clearly, relaxation oscillations will occur if m is complex, i.e. if
_
r
2
2
_
2
<
r 1

c
(3.20)

2
>
r
2
r 1

c
4
. (3.21)
In other words if the upper level lifetime is suciently short, relaxation oscillations do not occur and perturba-
tions to N

(t) and n(t) decay exponentially back to their steady-state solutions. Note that for r 1 eqn (3.21)
is in agreement with our earlier estimate (3.1).
If the above condition is not met relaxation oscillations will occur with an angular frequency

ro
=

r 1

_
r
2
2
_
2

_
r 1

c
, (3.22)
where the approximation holds if
2

c
. Notice that very often r 1 1, in which case
ro
is close to the
geometric mean of the upper level and cavity decay rates.
The condition (3.21) explains why relaxation oscillations (and also laser spiking) are unusual in gas lasers.
Such systems typically operate on allowed transitions with upper level lifetimes of 10s to 100s of nanoseconds
2
.
This is reasonably short compared to a typical cavity lifetime of 10 ns. In contrast, the upper level lifetime
for many solid state laser transitions is of the order of microseconds or even milliseconds. In such cases laser
spiking and relaxation oscillations are to be expected.
2
A notable exception is the CO
2
laser which has an upper level lifetime of 4 ms, and hence can exhibit relaxation oscillations.
3.2. Q-SWITCHING 49
3.2 Q-switching
Q-switching is a technique that generates pulses of laser light with a peak power very much greater than the
mean power that can be produced in steady-state operation. The high peak power is particularly useful in
applications that require a high peak intensity, such as: laser cutting or drilling; or in pumping other laser
systems such as Ti:sapphire ampliers.
In a Q-switched laser the cavity is deliberately spoiled imagine, for the moment, that one of the cavity
mirrors is blocked whilst the pumping builds up the population inversion. Since the losses of the modied
cavity are so high, the population inversion can build up to much greater levels than the threshold inversion for
lasing in the normal cavity without any lasing occurring. Once the population inversion has grown to a large
value the mirror is unblocked, thereby switching the Q of the cavity from a low value to a high value. High
rates of stimulated emission then cause the density of photons in the cavity to grow very rapidly, burning the
population inversion density well below the threshold value. The output from the laser takes the form of a giant
pulse of radiation essentially a single laser spike of the type discussed above.
Note that the pumping in Q-switched lasers may be pulsed or continuous. With pulsed pumping the ratio
of the peak population inversion to the threshold value can be larger than can be achieved with continuous
pumping, leading to a higher peak output power (although the mean output power may be similar).
3.2.1 Techniques for Q-switching
A wide variety of techniques have been used to implement Q-switching.
Rotating mirror
Historically one of the earliest techniques was to rotate one of the cavity mirrors with a motor. The cavity would
then be completed briey whenever the mirror was rotated into alignment. It is best to use a roof prism as
shown schematically in Figure 3.3, rather than a plane mirror, since any ray incident in a plane perpendicular to
the axis of the roof is reected in the anti-parallel direction to the incident ray. Hence for the system illustrated
alignment in the vertical plane is automatic this is useful given the inevitable vibration in a mechanical
system. Alignment in the horizontal plane is achieved once per rotation.
The disadvantages of this approach are obvious. The switching is rather slow and the timing of the output
pulse has a relatively high uncertainty. In order to estimate the switching time, we suppose that the mirror
has to be aligned to an angular tolerance about equal to the divergence of the oscillating cavity mode. This
divergence will typically be of the order of 1 mrad. Hence, for a motor running at 24 000 r.p.m. the switching
time is about 400 ns. The disadvantages are such that this technique is now used only very rarely.
Electro-optic switching
A more modern approach is the use of an electro-optic switch, as illustrated schematically in Figure 3.4. The
refractive index of an electro-optic material changes when it is subjected to an external electric eld. In the
Pockels eect
3
, observed in non-centro-symmetric materials, the induced refractive index varies linearly with
the applied electric eld. Suitable materials are KD*P or lithium niobate for visible to near infrared wavelengths,
or cadmium telluride for the mid-infrared. In the Kerr eect
4
the change in refractive index is proportional
to the square of the electric eld. We note that all materials exhibit the Kerr eect to some extent. Since
the Kerr eect arises from higher-order processes than in the Pockels eect (via the third-order susceptibility
rather than the second-order), larger electric elds are usually required. Since these must be switched quickly
for Q-switching to be eective, it is more usual to use the Pockels eect in this application.
3
Named after the German physicist F. C. A. Pockels (1865 - 1913).
4
Named after the Scottish physicist John Kerr (1824 - 1907).
50 LECTURE 3. Q-SWITCHING
laser medium
fixed output
coupler
turntable
(a)
(b)
Figure 3.3: Q-switching with a rotating mirror: (a) schematic diagram of the layout of the optical cavity; (b)
use of a roof prism as the rotating element. A roof prism has the advantage of automatically ensuring alignment
of the reected beam in one plane (in this case the vertical plane).
We can treat electro-optic devices of this type as wave plates with a birefringence which depends on the
strength of the applied electric eld. In the arrangement shown in Figure 3.4 a Pockels cell is oriented with
its axis at 45
o
to the axis of a polarizer. The cavity is held in a low-Q state by applying a certain voltage
(the quarter-wave voltage) to the Pockels cell, which then behaves as a quarter-wave plate. Consequently
vertically-polarized light transmitted by the polarizer becomes circularly polarized after it transmission through
the Pockels cell. Upon reection from the end mirror of the cavity the handedness of the circular polarization
changes, and hence on passing through the Pockels cell again it is converted to horizontally-polarized light and
so reected out of the cavity by the polarizer. Hence, with the quarter-wave voltage applied to the Pockels cell
the end mirror of the cavity is eectively blocked.
The mirror can be unblocked by switching the voltage across the Pockels cell to zero, which reduces the
birefringence of the Pockels cell to zero
5
. Vertically-polarized light can now pass through the polarizer and
Pockels cell with essentially zero loss, and the Q-switched pulse can build up. A Kerr cell would be employed
in essentially the same way.
Electro-optic switching has the advantages of fast switching times (of the order of 10 ns) and a high hold-o
ratio, which allows the population inversion to be built up to many times the threshold value for the high-Q
cavity. The disadvantages are that the technique requires several optics to be placed in the cavity (each with
some insertion loss and subject to possible damage), and the devices and associated electronics are relatively
expensive. However, the advantages are such that the technique is very widely used.
5
It would also be possible to devise an arrangement in which the low-loss condition was reached by switching on the Pockels
cell. However, it is usually technically easier to switch o voltages rapidly a technique known as crowbaring than to switch
them on quickly.
3.2. Q-SWITCHING 51
Pockels cell
Polarizer
Rejected beam
Gain medium
vertical polarization
horizontal polarization
circular polarization
Figure 3.4: Electro-optic Q-switching. The system is shown with the Pockels cell voltage set so as to reject the
beam reected from the right-hand mirror.
Acousto-optic switching
If an acoustic wave is launched through an acousto-optic crystal the refractive index is changed slightly at the
peaks and troughs as a result of the local expansion or contraction of the crystal lattice. The resulting periodic
variation in the refractive index will cause incident radiation to be diracted. We can think of this as Bragg
reection from the high- and low-index planes established in the crystal.
A wide variety of acousto-optic materials are available, including: fused silica, tellurium dioxide and lithium
niobate for visible wavelengths; gallium arsenide for use in the infrared; and silicon for radiation in the 10 m
region.
Figure 3.5 shows an acousto-optic switch employed in a Q-switched laser. With the radio-frequency (RF)
voltage applied to the crystal an acoustic wave propagates along the length of the crystal and is damped at the
opposite end, the damping often being achieved by cutting the end of the crystal to form a wedge. Some fraction
of the laser radiation incident on the crystal will be Bragg reected away from the cavity axis and hence lost
from the cavity mode. With the RF voltage applied, then, the acousto-optic crystal increases the losses of the
laser cavity. Turning o the RF voltage removes this additional loss, and sets the cavity to a high-Q condition.
The advantages of acousto-optic switching are that it is relatively cheap, and insertion losses may be made
low by orienting the crystal at Brewsters angle. However the hold-o ratio is low, and hence the technique
is usually employed in continuously-pumped Q-switched lasers for which the over-pumping ratio is relatively
modest. The switching times achieved are also relatively slow, the switching time corresponding to the time
taken for the acoustic wave to propagate out of the region of the cavity mode. Taking the speed of sound in
the crystal to be 5 kms
1
, we can estimate the switching time to be 200 ns for a 1 mm diameter beam. The
switching time can be decreased by focusing the cavity mode through the acousto-optic crystal.
Saturable absorbers
A very simple, and quite widely used, method for achieving Q-switching is to insert a saturable absorber into
the laser cavity. During the pump pulse the absorber will cause the laser mode to experience a high loss and
lasing is prevented. However once the population inversion reaches a high value the absorber will be subjected
52 LECTURE 3. Q-SWITCHING
Acousto-optic modulator
Rejected beam
Gain medium

B
2
B

a
RF
R
F
Bragg condition: m = 2
a
sin
B
Absorber
Figure 3.5: Acousto-optic Q-switching. Note that the Bragg angle
B
has been greatly exaggerated. A value
closer to 1
o
would be more typical. The insertion loss of the acousto-optic crystal can be eliminated (for one
polarization) by rotating the crystal out of the plane of the paper through Brewsters angle.
to an ever increasing ux of spontaneous and weak stimulated emission. As population is transferred to the
upper level of the absorbing species the absorption starts to saturate, increasing the feedback from the cavity
mirror and so increasing the rate of stimulated emission. This positive feedback leads to rapid saturation (or
bleaching) of the absorber until, at full saturation, it represents essentially zero loss. The cavity is now in a
high-Q state, and a giant pulse develops. It is clear that the absorber must have a strong absorption transition
at the laser wavelength and also that the lifetime of the upper level must be reasonably long so that the bleached
state is maintained for the duration of the Q-switched pulse. On the other hand, the upper level lifetime should
be short compared to the desired time between Q-switched pulses. Many types of saturable absorbers have been
used, such as solutions of dyes, and doped crystals such as Cr
4+
:YAG.
Saturable absorbers are a simple (there is no external drive circuit) and cheap way of achieving Q-switching.
The relatively slow turn on corresponds to many cavity round-trips, which allows good mode discrimination.
As a consequence it is quite straightforward to achieve Q-switched lasing on a single transverse mode of the
cavity. The disadvantages are that the shot-to-shot uctuation in the output energy and in the timing of the
output pulse can be relatively high, and that the absorber can degrade over time.
3.2.2 Analysis of Q-switching
An analysis of Q-switching is relatively straightforward, providing we make a few simplifying assumptions: we
neglect pumping during the development of the Q-switched pulse, and ignore spontaneous transitions. With
these assumptions, the rate equations (3.2) and (3.3) become:
3.2. Q-SWITCHING 53
dN
2
dt
= N

(t)
21
I

L
(3.23)
dN
1
dt
= +N

(t)
21
I

L
. (3.24)
The equations above correspond to the case of severe bottlenecking in which the rate of spontaneous decay
of the lower laser level is negligible. As such, stimulated emission causes the population to build up in the lower
laser level. The rate equation for the evolution of the population inversion during the Q-switched pulse can be
found simply by subtraction:
dN

dt
=
dN
2
dt

g
2
g
1
dN
1
dt
(3.25)
= N

(t)
21
I(t)

L
, (3.26)
where = 1 +g
2
/g
1
.
A second case is also of interest, however: that of an ideal four-level laser. In these systems the lower level
decays so quickly that throughout the Q-switched pulse N
1
(t) 0. In this case N

(t) N
2
(t) and hence the
rate equation for the population inversion is identical to that for N
2
(t).
It is convenient, therefore, to combine these two cases into one set of equations for the population inversion
and the photon density:
dN

dt
=

f
c
N

th
n

c
(3.27)
dn
dt
=
_
N

th
1
_
n

c
, (3.28)
where again we have used I(t) = n(t)c
L
and eqn (3.6), and where
=
_
1 ideal 4-level
1 +
g
2
g
1
severe bottlenecking
(3.29)
The general form of the solution to these equations is shown in Fig. 3.6. Notice that the population inversion
density is exactly equal to the threshold value, N

th
, when the photon density reaches its peak. Since stimulated
emission will continue to occur whilst the photon density falls, the population inversion can be burnt down to
a level well below the threshold value N

th
.
If we integrate eqn (3.28) over the complete duration of the pulse we nd,
_
n(t=)
n(t=)
dn =
_

th
n

c
dt
_

c
dt = 0 (3.30)
where we have used the fact that the rst integral is zero since the initial and nal densities of photons in the
cavity are both zero.
Now, the rate at which photons are coupled out of the laser is equal to nV
c
/
c
, where V
c
is the volume taken
up by the oscillating mode in the cavity. Hence the power output from the laser is,
54 LECTURE 3. Q-SWITCHING
2 4 6 8 10 0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
N
i
*
N
f
*
n(t)
N (t)
*
N (t) = N
th
* *
t/
c
N

(
t
)
/
N
t
h
*
*
Figure 3.6: General temporal evolution of the population inversion (dashed) and photon (solid) densities during
a Q-switched laser pulse for the case N

i
= 4N

th
. The Q-switch is opened at t = 0. Notice that it takes several
cavity lifetimes for the Q-switch pulse to develop.
P =
L
V
c
n

c
. (3.31)
Applying this result to eqn (3.30), the energy of the output pulse is given by,:
E =
L
V
c
_

c
dt (3.32)
=
L
V
c
_

th
n

c
dt (3.33)
=
L
V
c
f
c

_
N

f
N

i
dN

(from eqn (3.27)) (3.34)

=

L

V
g
(N

i
N

f
), (3.35)
where N

i
and N

f
are the initial and nal population inversion densities, and V
g
= f
c
V
c
is the volume of the
oscillating mode in the gain medium. We therefore can write,
E = N

i
V
g

L
, (3.36)
where the energy utilization factor is dened as,
=
N

i
N

f
N

i
. (3.37)
3.2. Q-SWITCHING 55
The interpretation of this result is straightforward: The initial number of inverted atoms is equal to N

i
V
g
; for
an ideal four-level laser transition the number of photons emitted in the pulse is simply (N

i
N

f
)V
g
; whereas in
the case of severe bottlenecking this number is reduced by a factor of since for these systems each stimulated
emission event reduces the population inversion by . Eqns (3.36) and (3.37) then follow from the fact that
each photon carries an energy
L
.
We may gain further information by dividing eqn (3.28) by (3.27) to give,
dn
dN

=
f
c

_
N

th
N

1
_
(3.38)

_
t

dn =
f
c

_
t

_
N

th
N

1
_
dN

, (3.39)
which gives a useful relation between the photon and inversion densities at all times during the Q-switched
pulse:
n(t) =
f
c

_
N

th
ln
_
N

(t)
N

i
_
+N

i
N

(t)
_
. (3.40)
After the pulse the photon density is zero, and hence setting t in the above yields,
N

i
N

f
= N

th
ln
_
N

f
N

i
_
(3.41)

i
N

f
N

i
=
N

th
N

i
ln
_
N

f
N

i
_
. (3.42)
We dene the over-pumping ratio as the ratio of the initial population inversion density to the threshold
value:
r =
N

i
N

th
. (3.43)
Equation (3.42) may then be rearranged to give a relation between the over-pumping ratio and the energy
utilization factor:
r =
ln(1 )

. (3.44)
This tells us, for example, that to extract 80% of the energy stored in the initial population inversion in an
ideal four-level laser requires pumping the initial population inversion to twice the threshold value. Figure 3.7
plots as a function of the over-pumping ratio. We see that once r reaches approximately 5, virtually all the
available energy is extracted from the population inversion.
56 LECTURE 3. Q-SWITCHING
2 3
4
5
0. 2
0. 4
0. 6
0. 8
1.0
1
overpumping ratio r
E
n
e
r
g
y

u
t
i
l
i
z
a
t
i
o
n

Figure 3.7: The product of the energy utilization factor and plotted as a function of the over-pumping ratio
r.
Peak output power and pulse duration
The peak density of photons in the cavity occurs when dn/dt = 0, or, N

= N

th
. Equation (3.40) then gives
the peak photon density as,
n
p
=
f
c

th
(r 1 ln r) . (3.45)
From (3.31) the peak output power is then,
P
p
=
n
p
V
c

c
=
N
th

L
V
g

(r 1 ln r) . (3.46)
We may then estimate the duration
Qs
of the Q-switched pulse by dividing the pulse energy E by the peak
power, to give:

Qs
=
r
r 1 ln r

c
. (3.47)
We note that as r , 1 and hence
Qs

c
. In this limit the Q-switched pulse rises very rapidly owing
to the sudden onset of a high rate of stimulated the population inversion is extracted almost instantaneously
followed by an exponential decrease with a time constant of
c
as the photons leak out of the cavity.
Lecture 4
Modelocking
4.1 Introduction
Modelocking is an extremely important technique that is applied to a wide variety of laser systems. Its value
lies in the ability to generate pulses of laser radiation which have a short duration typically picoseconds
(1 ps 10
12
s) to femtoseconds (1 fs 10
15
s) and a peak power many orders of magnitude greater
than the mean power that can be extracted from the laser medium. Pulses from modelocked lasers have found
applications in a wide variety of scientic and technical applications including: imaging techniques such as
two-photon laser-scanning uorescence imaging, ballistic imaging, and optical coherence tomography; ultrafast
chemistry, including coherent control of chemical reactions; pump-probe measurements of solid-state materials;
and short-pulse laser machining and processing of materials.
4.2 General ideas
We recall that the longitudinal modes of a laser cavity of length L are spaced in angular frequency by an amount,

p,p1
= 2
c
2L
=
2
T
c
, (4.1)
where T
c
is the time taken for light to travel once round the cavity. In order to avoid clutter, it will be convenient
to let
p,p1
throughout this lecture.
The right- and left-going waves of a cavity mode may be written in the simplied form,
E(z, t) = a
p
exp (i[k
p
z
p
t +
p
]) , (4.2)
where a
p
is the amplitude of the wave,
p
is the phase of the wave, the upper (lower) sign refers to waves
propagating towards positive (negative) z, and we have ignored the details of the transverse prole of the
waves.
Suppose that the laser oscillates simultaneously on N such modes. For simplicity, we will take there to be an
odd number of oscillating modes labelled by p = P/2 . . . P/2. The amplitude of the laser radiation emerging
from the output coupler would have the form,
E(z, t) =
p=p
0
+P/2

p=p
0
P/2
a
p
exp (i[k
p
z
p
t +
p
]) , (4.3)
57
58 LECTURE 4. MODELOCKING
-1 -0.5 0.5 1
-1 -0.5 0.5 1
t / T
c
t / T
c
0.0
0.0

p
= 2/
osc

p
= 2/
osc
T
c
T
c
(a)
(b)
Figure 4.1: Calculated intensity as a function of time emerging from a laser with N = 61 modes oscillating
simultaneously with equal amplitudes and (a) random (but xed) phases; (b) identical phases. Notice that
for both plots the pattern repeats every cavity round-trip time T
c
, and that the sharpest features have the
same width. It should also be emphasized that the vertical scales of the two plots are quite dierent: the peak
intensity of the modelocked pulses are approximately N times greater than the mean intensity in (a).
where p
0
is the number of the central longitudinal mode.
Figure 4.1(a) shows a calculation of the output intensity of a laser oscillating simultaneously on 61 modes
with equal amplitudes, but with random though constant relative phases
p
. The output has two key
features,
1. During an interval equal to T
c
the output contains sharp uctuations in intensity which are randomly
distributed;
2. The sharpest of these peaks has a duration of order 2/
osc
, where
osc
= (N1) is total bandwidth
occupied by the oscillating modes;
3. The pattern is periodic with a period equal to the cavity round-trip time T
c
.
In contrast, Figure 4.1(b) shows the intensity calculated for the case when the modes all have the same
phase (
p
= 0). Now the output consists of a train of short pulses separated by the cavity round-trip time.
This situation is known as modelocking.
4.2. GENERAL IDEAS 59
4.2.1 Simple analysis
To gain further insight, let us calculate the output for the case where all the modes have the same amplitude,
E
0
. The frequencies of the oscillating modes may be written,

p
=
p
0
+p, (4.4)
where
p
0
is the frequency of the centre mode. Similarly, we consider the case in which the phases of adjacent
modes dier by a constant amount, , so that we have,

p
=
p
0
+p. (4.5)
From eqn (4.3) the amplitude of the radiation emitted by the laser will then be:
E(r, t) = E
0
p=+P/2

p=P/2
exp [i(k
p
z
p
0
t pt +
p
0
+p)] . (4.6)
For convenience we consider the amplitude at z = 0 and also take the special case
1
in which = 0. Then,
E(0, t) = E
0
exp [i(
p
0
t
p
0
)]
p=+P/2

p=P/2
exp (ip[t]) . (4.7)
The sum appearing in the above is a geometric progression, which may be summed in the usual way to give,
E(0, t) = E
0
exp [i(
p
0
t
p
0
)]
sin
_
N
2
t
_
sin
_
1
2
t
_ . (4.8)
Hence the output takes the form of a (travelling) wave of mean frequency
p
0
multiplied by an envelope function
sin(
N
2
t)
sin(
1
2
t)
.
Taking the intensity to be given by [E(0, t)[
2
(i.e. ignoring constants of proportionality):
I(t) = E
2
0
sin
2
_
N
2
t
_
sin
2
_
1
2
t
_ . (4.9)
This is just the function plotted in Fig. 4.1(b).
The temporal behaviour calculated above corresponds to a train of intense modelocked pulses. Notice that
the peak intensity of each pulse is equal to N
2
E
2
0
, and hence the peak intensity is a factor of N greater than
the mean intensity of N modes oscillating with random phases
2
.
Pulses occur at times given by t
m
= 2m, and hence the separation of pulses is simply 2/ = T
c
.
The width,
p
, of the pulses may be taken to be the interval from the peak of a modelocked pulse to the rst
zero. It is straightforward to show that this is equal to:
1
It can be shown quite simply that if = 0 the peaks of all the modelocked pulses shift in time by a constant value of
t = /, but that otherwise the train is unchanged.
2
Note that the mean intensity is the same whether or not the N modes are modelocked or not. Modelocking is an interference
phenomenon, and interference neither destroys or creates energy!
60 LECTURE 4. MODELOCKING

p
=
2
N
=
T
c
N

2

osc
, (4.10)
where
osc
= (N 1) is the (angular) frequency bandwidth covered by the oscillating modes.
It should be clear from the above that if we were able to lock the phase of N longitudinal modes of a laser
we could generate a stream of pulses of high peak intensity and short duration. In practice the number of modes
that can be locked ranges from several hundred to several thousand, leading to a huge increase in peak power.
The duration of the pulses is determined by the frequency width of the laser transition; transitions with broad
linewidths can (in principle) generate shorter pulses. In practice pulses with durations from a few picoseconds
to a few femtoseconds can be generated; a very active area of current research is the development of techniques
related to modelocking to generate pulses in the attosecond range (1 as 10
18
s).
4.2.2 Further general comments
Modelocking techniques can be divided into two classes: active and passive, as described below.
In the general discussion above we did not consider in any detail the properties of the laser gain medium. As
we might expect, the modelocking behaviour of a laser depends on whether the laser transition is homogeneously
or inhomogeneously broadened. We have already seen that an inhomogeneously broadened laser transition can
oscillate simultaneously on all longitudinal modes for which the round-trip gain exceeds the round-trip loss.
The discussion above applies quite well to such lasers, and in such cases the modelocking technique only
needs to lock the phases of modes that are already oscillating simultaneously. As a rst approximation, the
duration of modelocked pulses output from a laser operating on an inhomogeneously broadened transition will
be
p
2/
D
, where
D
is the inhomogeneous linewidth of the transition.
Homogeneously broadened laser transitions, however, usually only oscillate on a single longitudinal mode
(unless spatial holeburning occurs). The modelocking must therefore play a stronger role, and encourage laser
oscillation on many modes. As such, the dependence of the pulse duration on the strength of the modelocking
will be dierent from the case of inhomogeneous broadening, and the duration of the modelocked pulses will
generally be longer than 2/
H
, where
H
is the homogeneous linewidth.
4.3 Active modelocking techniques
Active modelocking techniques fall into two classes: amplitude modulation (AM) and frequency modu-
lation (FM). We note that we usually require a single modelocked pulse to circulate within the laser cavity.
For a linear laser cavity this is achieved by locating the active component near one of the end mirrors, and
driving the component at a frequency equal to the longitudinal mode spacing . However, it is possible to
generate N such modelocked pulses within the laser cavity by locating the modulator suitably and driving it
at N. Harmonic modelocking of this type is occasionally used to increase the pulse repetition frequency.
In principle it can also be used to generate shorter pulses in homogeneously broadened lasers (see eqn (4.19)).
4.3.1 AM modelocking
Once modelocking is achieved, the intra-cavity radiation takes the form of a short pulse which circulates round
the laser cavity. An obvious way, therefore, to achieve modelocking is to introduce some sort of shutter into the
cavity which lets through short pulses, but which blocks or increases the loss of longer pulses. Radiation
which happens to reach the shutter when it is open will pass through with low loss, will experience net round-trip
gain, and hence will grow. In contrast, the radiation arriving at the shutter outside this interval will experience
a round-trip loss which is greater than the round-trip gain, and so will be absorbed.
4.3. ACTIVE MODELOCKING TECHNIQUES 61
0.5 1 1.5 2 2.5 3
t / T
c
T
c
Round-trip
loss
t / T
c
Round-trip
loss
(a)
(b)
Figure 4.2: Schematic diagram of AM modelocking. In (a) the pulses arrive perfectly synchronized with the
shutter and hence experiences minimum loss. The pulse shown in (b) arrives slightly too early, and hence
the leading edge is reduced in intensity. The eect is to move the peak of the pulse to later times, closer to
synchronization with the shutter.
After many round-trips a pulse will form which propagates round the cavity in synchronization with the
shutter, as illustrated schematically in Figure 4.2. The duration of the modelocked pulse so formed is determined
by an equilibrium between two competing eects. The shutter tends to decrease the duration of the pulse (and
hence increase the bandwidth) since the losses of the leading and trailing edges of the pulse are greater than
those experienced by the centre of the pulse. However, the gain medium can only amplify a nite frequency
bandwidth which limits the minimum possible duration of the modelocked pulse.
The above description is said to be in the time domain. It is also possible to consider the problem from
the perspective of the frequency domain. To do this we suppose that the amplitude transmission of the shutter
varies harmonically as,
T(t) = T
0
+

2
cos t. (4.11)
Suppose that a wave of angular frequency
p
is incident on the shutter, such that we may write its amplitude
as the real part of,
E(t) = E
0
exp(i
p
t). (4.12)
62 LECTURE 4. MODELOCKING
The amplitude of the transmitted wave will be:
E(t) = E
0
exp(i
p
t)
_
T
0
+

4
exp(it) +

4
exp(it
_
(4.13)
= E
0
T
0
exp(i
p
t)
+
E
0

4
exp(i[
p
]t) +
E
0

4
exp(i[
p
+ ]t. (4.14)
In other words, the amplitude modulation adds frequency sidebands at
p
. Since the modulation is
driven at the frequency dierence between longitudinal modes, the frequencies of the sidebands of a longitudinal
mode p lie exactly at the frequencies of the adjacent longitudinal modes p 1.
This frequency domain picture provides further insight into how it is possible to achieve modelocking in lasers
operating on homogeneously broadened transitions. For each longitudinal mode, the amplitude modulation
couples energy into the adjacent modes; and it does so with a denite phase. As such, we can consider the
sidebands as injection seeding the adjacent longitudinal modes. The system is complex, since each of many
modes is coupled to the adjacent modes, but it is clear that the amplitude modulation causes energy to ow
from the modes closest to the line centre out to the wings. The amplitude modulation therefore tends to force
the laser out of single-mode oscillation, and broaden the frequency spectrum. A steady state is reached when
this broadening is balanced by the frequency narrowing arising from the nite frequency response of the gain
medium.
We note that for a homogeneously broadened laser transition the round-trip gain of the modes near the line
centre is actually greater than the round-trip losses; in the wings of the modelocked pulse the round-trip gain is
less than the round-trip loss, as shown schematically in Fig. 4.3. In both cases the dierence in energy is made
up by energy transfer to or from adjacent modes so that a steady-state is maintained.
Finally we note that for both homogeneously and inhomogeneously laser transitions the system adopts the
lowest loss conguration, which corresponds to simultaneous oscillation on a large number of longitudinal modes,
phase-locked so as to form a short pulse circulating round the cavity.
Experimental implementation
Amplitude modulation may be achieved by using a Pockels cell in much the same conguration as is used
to generate Q-switched pulses. The Pockels cell is driven with a sinusoidal voltage of frequency . Since
the amplitude modulation does not need to be 100%, the driving voltage need only be some fraction of the
quarter-wave voltage.
It is also possible to use an acousto-optic modulator to implement AM modelocking, using a similar congu-
ration to that used for Q-switching, as shown in Fig. 4.4. However, for modelocking the faces of the acousto-optic
block are cut parallel, and the length of the block is cut to be an integer number of half (acoustic) wavelengths
long. A standing wave can then be driven in the acousto-optic block, rather than the travelling acoustic wave
used to Q-switch. The amplitude of the standing wave passes through zero twice per period, corresponding to
windows of minimum loss. Hence, if the block is driven with an angular frequency
a
, AM modelocking will be
achieved if 2
a
= .
FM modelocking
Frequency modulation modelocking is achieved by modulating the refractive index of an intra-cavity component
at the mode spacing frequency , as illustrated schematically in Fig. 4.5. As we will see, this aects the
frequency spectrum of an incident wave and hence the active device is known as a frequency modulator.
We consider rst the eect on a wave of angular frequency of a modulator of length L
mod
and time-
dependent refractive index n(t). The incident wave will emerge with an amplitude equal to the real part of,
4.3. ACTIVE MODELOCKING TECHNIQUES 63
Frequency
gain > loss
energy flow
loss
gain
energy flow
longitudinal
modes (b)
(a)
Frequency
(c)
}
}}
}}
}}
}}
Figure 4.3: Energy ow and coupling of modes in an AM modelocked laser operating on a homogeneously
broadened transition. The frequency dependence of the optical gain and loss in the absence of the amplitude
modulation is shown in (a). As illustrated in (b), the amplitude modulation causes energy to ow from those
modes for which the round-trip gain exceeds the round-trip loss to modes in the wings of the gain prole
for which the gain is lower than the loss. This energy transfer allows steady-state laser oscillation to be
maintained simultaneously on many modes. The amplitude modulation causes each mode to develop sidebands
at frequencies equal to those of the adjacent longitudinal modes, and hence couples adjacent modes as illustrated
schematically in (c).
E(L
mod
, t) = E
0
exp
_
i
_
n(t)
L
mod
c
t
__
. (4.15)
Let us now suppose that the refractive index is modulated harmonically, such that,
n(t) = n
0
+

2
sin t.
The transmitted wave will be
E(L
mod
, t) = E
0
exp (it) exp(i
0
) exp(i sin t), (4.16)
where
0
= n
0
L
mod
/c is the (constant) phase shift experienced by light of angular frequency , and =
L
mod
/2c is the amplitude of the modulation in phase introduced by the modulator.
64 LECTURE 4. MODELOCKING
acousto-optic
modulator
rejected
beam
gain medium

B
2
B

a
RF
R
F
m = 2
a
sin
B
reflection
RF
(a) (b)
(c)
standing
wave
Figure 4.4: Amplitude modulation modelocking with an acousto-optic modulator. In (a) the Bragg condition
for reection of a beam of radiation by incident at an angle
B
on an acoustic wave of wavelength
a
. In contrast
to the use of acousto-optic crystals in Q-switching, discussed in Lecture 3, in order to achieve modelocking the
cavity losses must be modulated at . This is achieved by cutting the acousto-optic crystal so that its two ends
are parallel allowing reection of the acoustic wave and separated by an integer of half wavelengths when
the modulator is driven at /2. This conguration allows a standing wave to be set-up with an amplitude
that passes through zero with a frequency . A typical layout of the optical cavity is shown in (c). Note that
the Bragg angle
B
has been greatly exaggerated. Note also that the insertion loss of the acousto-optic crystal
can be eliminated (for one polarization) by rotating the crystal out of the plane of the paper through Brewsters
angle.
Now we may use the identity,
3
exp(i sin t)

m=
J
m
() exp(imt) (4.17)
where J
m
(x) is the Bessel function of the rst kind of order m.
Thus we may write the amplitude of the transmitted wave as,
E(L
mod
, t) = E
0
exp (it) exp(i
0
)

m=
J
m
() exp(imt), (4.18)
and we see that the modulator introduces a series of sidebands at frequencies of m, where m is an integer.
Applying these ideas to an intra-cavity modulator driven at the mode spacing frequency , we see that for
each mode of frequency
p
, the modulator generates frequency sidebands at frequencies of
p
m thereby
3
Consult any text on mathematical methods, for example Riley, Hobson & Bence Mathematical methods for physics and
engineering: a comprehensive guide.
4.3. ACTIVE MODELOCKING TECHNIQUES 65
n(t)
L
mod
incident wave
modulated wave
modulator drive
Figure 4.5: Schematic diagram of FM modelocking.
coupling the longitudinal modes in a similar way to amplitude modulation. The width of the pulse generated
in this way is determined by the balance between the bandwidth broadening caused by the modulator and the
frequency narrowing produced by amplication in the gain medium.
As shown schematically in Fig 4.6, the modelocked pulses are found to be located at the maxima and minima
of n(t), when the modulator produces no shift in frequency. This may be understood as follows. The eect of
the phase modulation introduced by the frequency modulator is equivalent to modulating the cavity length by
driving one of the cavity mirrors sinusoidally. In general a pulse reected from the moving cavity mirror will
be Doppler-shifted in frequency and, since the mirror is driven with a frequency of , when it returns one
round-trip later it will be further shifted in frequency by the same amount. Eventually the pulse will be shifted
outside the bandwidth of the amplifying transition and will no longer be amplied. However, this progressive
frequency shift is minimized for radiation which strikes the mirror at one of the extrema of its motion or,
equivalently, if it passes through the modulator near the maxima or minima in n(t).
Notice that there are two stable positions for the modelocked pulse. An undesirable consequence of this is
that the inevitable disturbances present in a real laser can cause the modelocking to ip between these two
states, or even that two modelocked pulses circulate within the cavity.
Experimental implementation
Frequency modulation may be achieved by orienting a Pockels cell with its optical axis parallel with the polar-
ization of the intra-cavity laser radiation. Applying a voltage to the Pockels cell will then change the refractive
index experienced by the radiation, but will not alter its polarization state.
4.3.2 Synchronous pumping
Modulating the gain at the longitudinal mode spacing frequency can result in modelocking in some laser
systems by processes analagous to the AM modelocking discussed above. It should be clear that in order for
the gain of the laser transition to respond to the modulation, the population inversion must decay in a time
which is short compared to the cavity round-trip time.
For example, dye lasers can be modelocked by optically pumping the dye with pulses from a laser which
66 LECTURE 4. MODELOCKING
0.5 1 1.5 2 2.5 3
t / T
c
T
c
n(t)
Figure 4.6: Location of the modelocked pulses with FM modelocking.
is itself modelocked
4
. Synchronous pumping can also be achieved in semiconductor lasers by modulating the
driving current.
4.3.3 Pulse duration of actively modelocked, homogeneously broadened lasers
For completeness, we note that it can be shown
5
that the duration of the pulses generated by AM modelocking
a CW laser operating on a homogeneously broadened laser transition is given by,

p
0.45
_
2
I
L
g

_
1/4
2

mod

H
, (4.19)
where
I
is the (saturated) gain coecient, L
g
the length of the gain medium,
mod
is the driving frequency
of the modulator (usually set equal to , but it may be a harmonic of this), and
H
is the homogeneous,
linewidth. Notice that the pulse duration decreases with increasing modulation, , but only rather slowly. In
fact, for most systems the rst bracketed term will be close to unity, and hence,

p
0.45
2

mod

H
. (4.20)
A similar result can be obtained for FM modelocking. Notice that since
H
the pulses output by an
actively modelocked laser operating on a homogeneously broadened laser will be signicantly longer than an
inhomogeneously broadened transition of the same linewidth.
4.4 Passive modelocking techniques
Modelocking can also be achieved by inserting into the laser cavity an element with transmission properties
which depend on the intensity of the incident radiation. Passive modelocking avoids the need to drive modulators
4
Obviously care must be taken to ensure that the repetition frequency of the modelocked pump pulses corresponds to the
longitudinal mode spacing of the dye laser: in other words the cavity round-trip times must be the same for the two lasers
5
See, for example, Siegman.
4.4. PASSIVE MODELOCKING TECHNIQUES 67
at high frequencies, and is therefore relatively simple to implement practically. Passive modelocking techniques
are perhaps of more practical importance than active techniques, and are able to generate the shortest possible
laser pulses.
From a theoretical point of view, however, it is more dicult to calculate the pulse duration of the modelocked
pulse than in active modelocking, and treatments tend to be specic to the particular case under consideration.
In any case, the lower limit of the duration will always be given by approximately by 2/
H,D
, where
H,D
is the linewidth of the laser transition.
4.4.1 Saturable absorbers
Just as there is for a transition exhibiting optical gain, there are several key parameters for an absorption
transition. In particular, the saturation intensity, I
s
, and the saturation uence (units of Wm
2
),
s
:
I
s
=

21

R
(4.21)

s
=

21
, (4.22)
where
R
is the recovery time of the transition. We note that the time taken for the absorption to return to its
high, unsaturated value once an intense, saturating beam of incident radiation is switched o is of the order of
the recovery time.
The saturation intensity is relevant to steady-state conditions, or situations in which the duration, , of an
incident pulse is much longer than the recovery time. In this case, I
s
divides two regions: low intensities for
which the incident radiation experiences high, unsaturated absorption; high intensities, in which the absorption
is strongly reduced by saturation of the transition.
The situation is somewhat dierent in pulsed conditions. If the duration of an incident laser pulses is short
compared to the recovery time the absorption experienced at any time t during the pulse depends only on the
intensity integrated over time to that point, i.e. on
_
t

I(t

)dt

. The reason for this is simply that atoms

transferred to the upper level of the absorption transition will not decay during the incident laser pulse, and
hence the density of atoms in the upper level will depend only on the number of incident photons per unit area,
not on the details of the pulse shape. As such, the absorption will start to saturate after a certain number of
incident photons per unit area, or, equivalently after a certain incident uence (energy per unit area). In this
case, then, the behaviour is determined by the uence of the incident pulse relative to the saturation uence.
Fast saturable absorbers
In the context of modelocking, a fast saturable absorber is a medium for which the recovery time is very
much shorter than the cavity round-trip time, and in fact is short compared to the typical spacing of the noise
spikes observed in the non-modelocked, multimode laser (See Fig. 4.3(b)). Typically, then, the recovery time of
a fast saturable absorber will be a few tens of picoseconds, or less.
If a fast saturable absorber is placed within a laser cavity, near one of the laser mirrors, the round-trip loss
will be lower for intense noise spikes than for low-intensity spikes. Consequently the most intense spike will tend
to grow at the expense of the others, and will start to saturate the available gain. The result is a single, intense
pulse propagating round the laser cavity: modelocking. We can think of the saturable absorber as acting as a
shutter which is activated by the pulse itself, as illustrated in Figure 4.7.
Slow saturable absorbers
Under some circumstances it is also possible to achieve modelocking using a slow saturable absorber, for
which the recovery time is comparable to although it must still be less than the cavity round-trip time.
68 LECTURE 4. MODELOCKING
T
c
gain
loss
window of
net gain
modelocked pulses
Figure 4.7: Passive modelocking with a fast saturable absorber.
gain
loss
window of
net gain
modelocked pulse
slow recovery of saturated
gain and loss
Figure 4.8: Schematic diagram of modelocking with a slow absorber. The interplay of saturation of the gain
and loss ensures that net gain only exists in a short window. The saturated gain and loss must recover within
a cavity round-trip time T
c
.
Figure 4.8 illustrates schematically how modelocking is achieved with a slow absorber. For simplicity we
imagine the saturable absorber and the gain medium lie in the same plane. The leading edge of the modelocked
pulse will experience a net round-trip loss, owing to the fact that the (unsaturated) absorption is higher than
the available gain. As such the front of the pulse will be eroded. As the pulse passes through the absorber, the
absorption will gradually decrease as population is transferred to the upper level of the absorption transition.
At some point the round-trip loss introduced by the saturable absorber will fall below the round-trip gain, and
the pulse will experience amplication. In order for the generated pulse to be short, however, it is essential that
shortly after this moment the growing pulse saturates the round-trip gain below the (saturated) round-trip loss
so that the trailing edge of the pulse is also attenuated. Further, in order for the system to settle down to a
steady-state in which a single, modelocked pulse circulates within the cavity both the gain and the absorption
must recover to their initial values in less than a cavity round-trip time.
It should be appreciated, therefore, that modelocking with a slow saturable absorber requires a delicate
balance between the energy of the modelocked pulse, and the saturation uences and recovery times of both
the absorber and the gain medium. The requirement that the gain can be returned to its initial value by
the pumping mechanism in less than one round-trip time means that the technique is limited to lasers with a
relatively short upper level lifetime, such as dye lasers or semiconductor lasers.
4.5. EXAMPLES OF MODELOCKED LASERS 69
intensity
r
a
d
i
a
l

p
o
s
i
t
i
o
n
optical Kerr
medium
intensity
r
a
d
i
a
l

p
o
s
i
t
i
o
n
Figure 4.9: Schematic diagram of self-focusing arising from the optical Kerr eect.
4.4.2 Kerr lens modelocking
Kerr lens modelocking (KLM) is one of the most important types of modelocking used today, and is able to
generate the shortest optical pulses obtainable directly from a laser oscillator.
As we have already seen, all materials exhibit the Kerr eect, in which the refractive index depends on the
square of any applied electric eld. If an intense optical pulse is incident on a medium, the optical Kerr eect
will cause the refractive index to be modied by the electric eld of the pulse itself. Since the square of the
electric eld is proportional to the intensity, I, of the pulse, we may write the refractive index of a material as,
n(I) = n
0
+n
2
I. (4.23)
It is found that n
2
is positive for nearly all optical materials, with values typically of order 10
16
cm
2
W
1
.
The optical Kerr eect is responsible for a variety of phenomena, such as self-focusing and self-phase-
modulation. The former lies at the heart of Kerr lens modelocking. Figure 4.9 shows the propagation of a beam
with a Gaussian transverse prole through an optical Kerr medium. The intensity of the beam will be greater
on the propagation axis than in the transverse wings, and hence the refractive index of the medium in the axial
region will be greater than in the wings. The optical Kerr medium therefore looks like a thin lens which focuses
the beam slightly as it propagates through it.
Figure 4.10 shows schematically how the optical Kerr eect can cause modelocking. The optical Kerr
medium is usually the (solid) laser gain medium itself. The laser cavity is designed so that in the absence of
any self-focusing the transverse dimensions of the lowest-order cavity mode are comparable to, or larger than,
the size of an aperture placed within the cavity. As such, low-intensity radiation experiences a relatively high
round-trip loss. Higher intensity radiation, however, will be self-focused by the Kerr eect and so experience
lower round-trip losses. Modelocking then arises by essentially the same processes as with a fast saturable
absorber: the optical Kerr medium acts as a fast, self-activated shutter.
In order for KLM to be eective, the laser cavity must be operated close to being unstable in order that
the weak Kerr lens causes a large change in the transverse extent of the mode.
4.5 Examples of modelocked lasers
4.5.1 CW modelocking
In CW lasers, active modelocking is usually implemented with acousto-optic modulators owing to their lower
insertion loss compared to a Pockels cell. The modulator should be placed near one of the end mirrors of the
cavity if only a single pulse circulating within the cavity is to be generated.
Some examples are given in Table 4.1.
70 LECTURE 4. MODELOCKING
optical Kerr
medium
aperture
low power
beam high power
beam
Figure 4.10: Schematic diagram of Kerr lens modelocking.
4.5.2 Pulsed modelocking
The behaviour of pulsed modelocked lasers is determined by the dynamics of the population inversion during
the pump pulse. If the lifetime of the upper laser level is short, the duration of the burst of modelocked pulses
is determined by the duration of the pumping. However if the lifetime of the upper laser level is long, pulsed
modelocking tends to be accompanied by Q-switching. In this case the duration of the burst of modelocked
pulses will be approximately equal to the duration of the Q-switched pulse
QS
.
The higher gain found in pulsed lasers means that it is possible to use a Pockels cell to achieve active
modelocking, in either AM or FM mode. Passive modelocking of pulsed lasers is achieved with fast saturable
absorbers.
Some examples are given in Table 4.1.
Table 4.1: Parameters of some modelocked lasers
Laser pulsed / c.w. Method Pulse duration
He-Ne c.w. acousto-optic 1 ns
CO
2
c.w. acousto-optic 10-20 ns
Ar
+
c.w. acousto-optic 150 ps
Nd:YAG c.w. acousto-optic 100 ps
He-Ne c.w. slow sat. abs. 300 ps
GaAs c.w. slow sat. abs. 5 ps
Dye c.w. slow sat. abs. 25 fs
Dye c.w. synch. pumping 1 - 30 ps
Ti:sapphire c.w. Kerr lens 10 - 30 fs
Nd:YAG pulsed Pockels cell 40 ps
CO
2
pulsed fast sat. abs. 1 ns
Dye pulsed fast sat. abs. 1 ps
Nd:YAG pulsed fast sat. abs. 40 ps
Nd:Glass pulsed fast sat. abs. 5 ps
Ruby pulsed fast sat. abs. 10 ps
Lecture 5
Ultrafast Lasers
5.1 Propagation of short laser pulses in dispersive media
It may be shown that the root-mean-square (RMS) angular frequency bandwidth,
RMS
, of a pulse is related
to the RMS duration of the pulse, t
RMS
, by,

RMS
t
RMS
. (5.1)
In general the value of the time-bandwidth product
RMS
t
RMS
depends on:
the exact shape of the pulse envelope (i.e. Gaussian, square, sech
2
etc.);
the amount of frequency chirp or phase structure within the pulse.
In addition, of course, dierent denitions of and t such as the full-width at half maximum will
yield dierent time-bandwidth products, although any reasonable denition of these quantities should give a
result similar to eqn (5.1).
Short laser pulses, therefore, are associated with a large bandwidth. For propagation in vacuo, this presents
no diculties, and the pulse will propagate over arbitrary distances without distortion. However, optical
materials are generally dispersive that is their refractive index varies with frequency which can lead to
distortion of the pulse as it propagates. It is no use generating a short laser pulse if it is stretched and distorted
beyond recognition when it reaches the interaction region in an experiment. An understanding of the eect of
dispersion, and how it may be controlled, is crucial for both applications and generation of short laser pulses.
In general we may represent an optical pulse by a superposition of harmonic waves. For example, a pulse
propagating through a medium is given by the real part of,
E(z, t) =
_

0
a() exp i[k()z t] d, (5.2)
where we take the axis of propagation to be the z axis, and a() is the amplitude of each angular frequency .
The wave vector k() will in general be a function of frequency of the wave, as determined by the dispersion
relation for the medium. For example, in terms of the refractive index n() the wave vector is given by,
k() =

c
n(). (5.3)
71
72 LECTURE 5. ULTRAFAST LASERS
In general we can expand the wave vector about its value at some representative central frequency
0
:
k() = k(
0
) +
k

0
(
0
) +
1
2

2
k

0
(
0
)
2
+. . . (5.4)
If we retain just the rst two terms in this expansion, substitution into eqn (5.2) yields,
E(z, t) = exp i[k(
0
)z
0
t]
_

0
a() exp
_
i()
_
z
vg
t
__
d(), (5.5)
where =
0
, and the group velocity of the pulse is given by,
v
g
=

k

0
. (5.6)
It is easy to see that in this approximation the pulse comprises a carrier wave of frequency
0
, the wavefronts
of which propagate at the phase velocity
v
p
=

0
k(
0
)
, (5.7)
multiplied by a pulse envelope determined by the integral. Within the approximation we have made here, the
pulse envelope has a constant shape and moves at the group velocity. If we had included higher-order terms in
the expansion of k(), we would nd that the envelope of the pulse would change with propagation distance.
5.1.1 Propagation through an arbitrary system
In determining the behaviour of a short pulse as it propagates between two planes z = z
1
and z = z
2
, it is useful
to calculate the total phase accumulated:
() =
_
z
2
z
1
k()dz (5.8)
= (
0
) +

0
(
0
) +
1
2

0
(
0
)
2
+. . . (5.9)
For a single medium of length L, this expansion is simply equal to eqn (5.4) multiplied by L.
We consider each of the terms above:
(
0
). This is the total phase accumulated at
0
. Together with the group delay, it determines where
the peaks of the carrier wave are located with respect to the envelope of the pulse. As such it becomes
important for pulses which comprising only a few optical cycles; techniques for measuring and controlling
(
0
) are now an active area of research.

0

(1)
(
0
). This term is known as the group delay. To see this, we note that for propagation
through a single medium of length L,
(1)
(
0
) = L/v
g
, and gives therefore the time taken for the envelope
to propagate between the two planes.
5.1. PROPAGATION IN DISPERSIVE MEDIA 73
Dispersive
medium
Figure 5.1: Distortion of an initially un-chirped optical pulse by propagation through a dispersive medium with
positive GDD. The transmitted pulse is stretched in time, and develops a positive frequency chirp.

0

(2)
(
0
). This term is known as the group delay dispersion (GDD) or the quadratic phase.
It is the lowest term in the expansion responsible for distortion of the pulse envelope with propagation.
Positive dispersion is dened as
(2)
(
0
) > 0. The GDD is generally expressed in units of fs
2
.

0

(n)
(
0
). There are of course higher terms in the expansion: the cubic, quartic, quintic... phase.
For the production and propagation of suciently short pulses, these will also be important.
Group delay dispersion (GDD)
Let us consider rst the frequency dependence of the group delay. Denoting the transit time through the system
for quasi-monochromatic waves with frequencies near as (), we have from our discussion above,
() =

. (5.10)
Expanding about the mean frequency yields,
() = (
0
) +

0
(
0
) +. . . (5.11)
=
(1)
(
0
) +
(2)
(
0
)(
0
) +. . . . (5.12)
We see that the term in
(2)
(
0
) causes the transit time to increase linearly with frequency. As such, for
positive values of the GDD, the high frequency components of the pulse will move to the rear of the pulse; the
low frequency components will lead. This will cause the pulse to develop a positive frequency chirp, that is
for a stationary observer the frequency of the radiation increases as the pulse passes. Further, the dierence in
group delays will cause the pulse to increase in duration by an amount approximately equal to

GDD
=
(2)
(
0
), (5.13)
where is the bandwidth of the pulse. These processes are illustrated schematically in Fig. 5.1
5.1.2 Propagation of Gaussian pulses
Considerable insight may be obtained by applying the general methods discussed above to the special case of
an optical pulse with a Gaussian temporal prole. The general conclusions we draw will apply to pulses with
arbitrary temporal prole.
74 LECTURE 5. ULTRAFAST LASERS
Properties of a Gaussian pulse
We consider a generalized Gaussian pulse of the following form,
1
E
in
(t) = e
t
2
e
i
0
t
(5.14)
where the complex Gaussian parameter describing the pulse is given by,
a + ib. (5.15)
Let us rst establish the main properties of this pulse. The full-width at half maximum duration of the
temporal prole of the pulse intensity is easily shown to be,

p
=
_
2 ln 2
a
=

2 ln 2
1()
, (5.16)
from which we note that the duration of the pulse depends only on 1().
If b ,= 0 the instantaneous frequency changes within the pulse. To see this we note that the phase of the
electric eld is given by (t) =
0
t bt
2
, and the instantaneous frequency given by,
2
(t)
d(t)
dt
=
0
+ 2bt. (5.17)
Thus in general the frequency of the pulse changes linearly with time about the mean value of
0
; this frequency
chirp will be positive if b is positive.
To investigate how such a pulse propagates through an optical system it is necessary to nd the frequency
spectrum of the pulse by taking the inverse Fourier transform:
a() =
1

2
_

E
in
(t)e
it
d (5.18)
a() =
1

2
e
(
0
)
2
/4
(5.19)
=
1

2
exp
_
1
_
1
4
_
(
0
)
2
_
exp
_
i
_
1
4
_
(
0
)
2
_
. (5.20)
It is clear that in general the phase () of the amplitudes of the frequency components varies quadratically
with (
0
) with a curvature determined by (1/). The power spectrum, which is proportional to [a()[
2
,
has a full-width at half maximum of,

p
= 2

2 ln 2
1(1/)
= 2

2 ln 2
_
a[1 + (b/a)
2
]. (5.21)
1
Note that Siegman writes the pulse in the form exp(t
2
) exp(j
0
t) with a jb. Consequently equations in Siegmans
book may be converted to our notation by making the transformation j i, and vice versa.
2
Equation (5.14) may be written as E
in
(t) = exp(at
2
) exp [i(
0
+ bt)]. However, it is not correct to deduce from this that
the eective frequency is
0
+ bt.
5.1. PROPAGATION IN DISPERSIVE MEDIA 75
-2 -1 0 1 2
E(t)
t

(
t
)

-

0
t /
p
I(t)
t /
p
-2 -1 0 1 2

p
slope = 2bt
|a()|
2
(
0
) /
p
-2 -1 0 1 2
()
(
0
) /
p
-2 -1 0 1 2

p
Figure 5.2: Properties of a Gaussian optical pulse with a pulse parameter = a +ib for the case b > 0. Shown
are the electric eld E(t), the instantaneous frequency (t), the temporal prole of the intensity I(t) the power
spectrum [a()[
2
and phase ().
In terms of the full-widths at half maxima, the time-bandwidth product of the Gaussian pulse is found from
eqns (5.16) and (5.21) to be

p
= 4 ln 2
_
1 + (b/a)
2
. (5.22)
The time-bandwidth product clearly takes the smallest possible value if b = 0, i.e. if the pulse has no frequency
chirp or, equivalently, if the phases of the frequency components are all the same.
3
Figure 5.2 summarizes the
properties of Gaussian optical pulses of this type.
Propagation of a Gaussian pulse
We now consider the propagation of a Gaussian pulse through an optical system for which the linear and
quadratic dispersion are nite, but the higher-order dispersion may be ignored:
() =
(0)
+
(1)
(
0
) +
1
2

(2)
(
0
)
2
. (5.23)
To calculate the form of the output pulse we multiply the frequency spectrum of the input pulse by exp[i()]
and take the inverse Fourier transform. As explored in the problems, this turns out to be relatively straightfor-
ward, and we nd
E
out
(t) =
_

out

in
exp[i(
(0)

0
t)] exp
_

out
_
t
(1)
_
2
_
, (5.24)
3
If the widths in eqn (5.22) are replaced with root-mean-square values the factor of 4 ln 2 is replaced by . We then see that the
case b = 0 reduces to the minimum possible value according to eqn (5.1).
76 LECTURE 5. ULTRAFAST LASERS
where
1

out
=
1

in
2i
(2)
. (5.25)
We see that the transmitted pulse still has a Gaussian temporal prole, but that the duration of the pulse
envelope and the frequency chirp will in general be dierent from the input pulse since the parameter will be
changed. There will also be a relative shift between the peak of the Gaussian pulse envelope which occurs at
t =
(1)
, i.e. it is shifted in time by the group delay and the carrier wave which is shifted in phase by
(0)
.
Since in general
(0)
/
0
,=
(1)
the carrier-envelope oset of the pulse will be altered by the optical system.
Note that there are strong parallels between eqn (5.24) describing the propagation of a Gaussian optical
pulse and the equations describing the propagation of a lowest-order Gaussian beam. Comparison of these two
situations reveals that the parameter 1/ plays an equivalent role to that of the complex radius of curvature q,
and the dimension of time (shifted by
(1)
) is equivalent to the transverse co-ordinate r. Thus compression
(stretching) of a Gaussian pulse are analogous to focusing (defocusing) of a Gaussian beam. Notice too that
compression of a chirped Gaussian pulse requires the introduction of a medium with quadratic phase variation
in frequency; whereas as focusing a Gaussian beam requires a lens which imposes a phase shift which varies
quadratically with the transverse co-ordinates of the beam.
These points are made more obvious by considering the case of a uniform medium. Equation (5.24) may
then be written as
E
out
(t) = exp[i
0
(t z/v
p
)]
. .
carrier wave
pulse envelope
..
_

out

in
exp[
out
(t z/v
g
)
2
] (5.26)
since
(0)
= k(
0
)z and the group delay
(1)
= z/v
g
.
In general the pulse duration and frequency chirp of the optical pulse are altered by transmission through
the optical system since the Gaussian parameter is changed. Note, however, that the frequency width of the
pulse is not altered by the propagation, as must be the case for a linear system.
4
Whether the optical pulse is
compressed or stretched in time, and whether the pulse becomes more or less chirped, depends on the relative
sign of the initial frequency chirp and the GDD of the optical system.
5.1.3 Nonlinear eects: self-phase-modulation and the B-integral
As we saw in Lecture 4, at high intensities the refractive index of a material becomes nonlinear with the intensity:
n = n
0
+n
2
I. (5.27)
In propagating a distance L a wave of constant intensity I therefore accumulates a nonlinear phase

NL
B =
2

0
_
L
0
n
2
I(z)dz (5.28)
in addition to that which would be accumulated by a beam of low intensity. Here
0
is the vacuum wavelength
of the radiation. The integral above is universally known as the B-integral for the propagation.
4
This may be seen from eqn (5.21); the width of the spectrum is determined by (1/) and from eqn (5.25) (1/
out
) = (1/
in
).
5.2. DISPERSION CONTROL 77
t
t
I(t)

0
blue-shifted
trailing edge
red-shifted
leading edge
Figure 5.3: Generation of a frequency chirp by self-phase-modulation of an intense optical pulse.
The intensity of a beam of radiation will generally vary with transverse position within the beam, and, as
we have seen, this usually leads to self-focusing of the beam. For a laser pulse the nonlinear phase will also
vary with time during the pulse. Ignoring dispersion, the total accumulated phase of a wave after propagating
a distance L into a medium will be given by,
=
2

0
[n
0
+n
2
I(t)] L t. (5.29)
At any point in the pulse, the eective frequency,

, is given by the (negative) total rate of change of the phase

with time:

=
d
dt
=
2

0
n
2
L
I
t
. (5.30)
The nonlinear refractive index therefore causes the phase of the carrier wave in the pulse to vary nonlinearly
with time, corresponding to a variation of frequency within the pulse. This process is known as self-phase-
modulation.
Figure 5.3 shows this process schematically. The leading edge a pulse must have I/t > 0 corresponding
to a decrease in frequency, or red shift, for materials with n
2
> 0. Similarly, the trailing edge of the pulse will
be blue-shifted, and as a consequence the pulse will develop a positive chirp. The rate of change of frequency
with time is approximately linear over the central part of the pulse, as shown in Fig 5.3.
Self-phase-modulation and self-focusing are nearly always undesirable in a laser system, in which case they
must be kept to a minimum. A rule of thumb often employed by system designers is that the total B-integral
should be kept below 1 radian.
5.2 Dispersion control
The linear and nonlinear dispersion discussed above causes distortion of short laser pulses as they propagate
within a system. In order to overcome this or at least alleviate it it is necessary to employ techniques
in which additional, controllable dispersion can be introduced. A simple way of achieving this is to pass the
pulse through a series of optical components for which the physical (and hence optical) path is dierent for
dierent wavelengths. In this way controllable, wavelength-dependent phases can be introduced. The dispersion
introduced in this way is known as geometric dispersion.
Dispersion control is particularly important for modelocked lasers producing very short pulses. In order to
generate the shortest possible pulses by modelocking it is necessary for the longitudinal modes must be evenly
78 LECTURE 5. ULTRAFAST LASERS

A
B
C
E
F
D
1 2
Figure 5.4: Schematic diagram of a general optical system exhibiting geometric dispersion.
spaced in frequency, but the presence of dispersion in the cavity will prevent this. Clearly the problem will be-
come more acute as the duration of the modelocked pulses is reduced, and the associated bandwidth is increased.
In practice it is found that dispersion control is necessary to generate pulses shorter than approximately 150 fs.
Most materials exhibit positive dispersion, and hence geometric dispersion introduced into a laser cavity needs
to be negative (and controllable).
Figure 5.4 shows a general system exhibiting geometric dispersion. Upon reaching the plane 1, light is
refracted or diracted through a wavelength-dependent angle . After propagating some distance it is re-
fracted/diracted again so as to render the direction of propagation parallel to the initial direction. Since
depends on the wavelength, the optical path, and hence the group delay, will depend on wavelength; the system
therefore introduces dispersion.
In calculating the phase accumulated by a given wavelength, we need to dene two planes (an input plane
and an output plane) through which all waves pass. There is no diculty in identifying the input plane: it is
the plane passing through A, perpendicular to the incident ray. However, the output plane needs to be dened
more carefully. For example, the plane BC is not a suitable candidate since this plane is dierent for each
wavelength. There is still a lot of exibility: the output plane could be any xed plane perpendicular to the
rays leaving the system. One candidate would be such a plane passing through E.
However, it is more convenient to move the reference output plane back along the undeected ray to A
i.e. the same plane as the input plane! Using these reference planes the additional optical path travelled by the
deected ray is then equal to the distance AF AE.
5
Now,
5
The calculated accumulated phase depends on where the input and output planes are dened to be. However, the calculated
GDD (and higher orders) do not depend on these locations; all that will change is the total accumulated phase and the group delay
for propagating between the two reference planes, just as we would expect.
5.2. DISPERSION CONTROL 79
AE =
D
cos
(5.31)
AF = AE cos( ). (5.32)
Hence the optical path dierence is given by,
OP =
D
cos
[1 cos( )] , (5.33)
and hence the phase introduced is simply,
() =

c
D
cos
[1 cos( )] . (5.34)
It is a simple matter (in principle) to calculate the GDD etc. by dierentiating eqn (5.34), providing the relation
between and is known.
5.2.1 The grating pair
Figure 5.5 shows a ray propagating via a pair of parallel reection grating. Light of wavelength will be
diracted at an angle from the rst grating according to,
m = d[sin sin ] m = 0, 1, 2, 3, . . . (5.35)
where is the incident angle and d the grating spacing. At the second grating the incident angle will now be
and consequently the second grating will diract the light at an angle if the order of diraction is m for the
rst grating and m for the second. The diracted ray therefore leaves parallel to the incident ray.
The optical phase introduced by the grating pair is given by eqn (5.34), and this, together with eqn (5.35)
allows one to calculate the GDD and other phase terms. For this simple case, however, it is easy to deduce the
sign of the GDD. Consider the time taken for light to propagate from A to (say) the position of the mirror or
roof prism, which we may take as a convenient reference plane. We know that the time taken for the undeected
light (dashed line) to reach this plane is a minimum, since the path it takes is the same as if the gratings were
replaced by mirrors. All other paths must take a longer time, and since the diraction angle increases with
wavelength, we conclude that the group delay is smaller for shorter wavelengths (higher frequencies). Hence the
group delay decreases with frequency, and the GDD must be negative
6
. It should also be clear that the GDD
will increase linearly with the grating separation D, and consequently the grating pair can provide controllable
negative GDD. Clearly the dierent wavelengths leave the grating pair at dierent heights above the incident
ray. Consequently in the exiting beam there is a correlation between position and wavelength, a situation known
as spatial chirp. In general this is undesirable, and the spatial chirp is undone by reecting the beam back
through the grating pair to form a beam counter-propagating along the axis of the incident. This counter-
propagating beam may be selected by allowing the incident beam to enter the system at a small angle to the
6
Note that the group delay is NOT equal to the optical path dierence given by eqn (5.33) divided by the speed of light. To use
the optical path we must nd the phase introduced by the system and then see how it varies with frequency. Unfortunately this
is not as straightforward as the method used in the text to deduce the sign of the GDD: as increases the optical path dierence
decreases, but () is equal to this path dierence multiplied by /c and hence it is not obvious whether overall () increases or
decreases with . The method above works because we can use our insight to deduce that the group delay,
(1)
(), decreases with
frequency.
80 LECTURE 5. ULTRAFAST LASERS

F
E
C
B
A
g
r
a
t
i
n
g

1
g
r
a
t
i
n
g

2
D
mirror or
roof prism

Figure 5.5: A grating pair used to introduce negative GDD.

plane of the page; the exiting beam therefore leaves above the incident beam and may be picked o by a small
mirror. Alternatively, and somewhat better, the beam may be returned by reection from a roof mirror which
returns the light parallel to the incident rays, but at a dierent height. Obviously, double-passing the beam
through the grating pair will introduce twice the GDD generated from a single pass.
5.2.2 The prism pair
Dispersion control may also be achieved with the arrangement of prisms shown in Figure 5.6
7
. It is clear that the
optical paths are dierent for dierent wavelengths, and consequently the arrangement will introduce geometric
dispersion. Note that the plane AB is a plane of symmetry and consequently the total GDD is simply twice
that introduced by the rst prism pair. The advantage, however, of the arrangement of four prisms is that the
emerging beam propagates along the same axis as the incident beam. This means that the optical system (in
particular a laser cavity) can be aligned without the prisms in place, and once this is done the prisms can be
introduced without changing the alignment of other components.
It may be shown that for the rst two prisms, the optical path and phase dierence between a ray refracted
at an angle and the ray which passes between the apexes of the prisms are also given by eqns (5.33) and
(5.34). There is no simple argument for deducing the sign of the GDD in this case: indeed it may be positive
or negative. To evaluate the GDD, and other terms, one must use Snells Law to calculate as a function of
frequency. This procedure is not dicult, but it is somewhat messy and gives us no particular insight. In fact
it is found
8
that the GDD introduced by a prism pair is also negative, provided that l is suciently large. The
GDD is proportional to l = D/ cos , and hence may be controlled by adjusting l. In practice, however, it is
usual to introduce precise control of the GDD by moving prisms II and III perpendicular to their bases, as
indicated in Fig. 5.6. This reduces l and introduces more prism material into the path, and therefore increases
7
Note that the diagram given in Svelto appears to be incorrect in that the higher frequency light is refracted through a smaller
angle.
8
This was rst shown by R. L. Fork et al., Optics Letters vol. 9, p. 150 (1984).
5.2. DISPERSION CONTROL 81
A
B
blue
red
I
II III
IV

D
l
Figure 5.6: A prism pair used to introduce negative GDD. Note that the plane AB is a plane of symmetry; it
would be possible to place a mirror here, reecting rays back through the rst two prisms to yield a counter-
propagating beam with the same GDD as produced by the four prisms shown.
the GDD.
The prisms are usually oriented at Brewsters angle so as to minimize the insertion loss; and the apex angles
of the prisms are chosen so that they operate at minimum deviation (i.e. the incident and exit angles are equal).
The GDD introduced by prism pairs is typically much smaller than that provided by grating pairs, but have the
advantage that the insertion loss can be much smaller. As such prism pairs are often used to provide dispersion
control in modelocked laser oscillators.
5.2.3 Introduction of positive GDD
The two systems above both introduce negative GDD. Clearly we would also like to be able to introduce positive
GDD (especially for CPA - see later), but this would appear to be dicult owing to the fact that for both these
systems the GDD is negative for a positive separation D.
A clever solution to this problem is to introduce an imaging system between the two elements which can
project the rst element in front, behind, or at the same position as the second. In this way the eective
separation of the elements may be made to be positive, zero, or negative which allows the GDD to be negative,
zero, or positive.
In practice this system is usually implemented with diraction gratings, as shown in Figure 5.7. The two
lenses form an image, with unit magnication, of grating G1. By adjusting the position of G1, its image, G1

,
may be positioned either side of the second grating G2. Notice that the longitudinal magnication is equal to
1, and hence the imaging system changes the orientation of G1

with respect to G1.

Many ingenious improvements have been made to this simple idea. More advanced systems replace the
lenses with curved mirrors, avoiding problems caused by chromatic aberration. It is also possible to use a folded,
single-grating design which automatically ensures alignment of the two gratings as the other components are
adjusted.
5.2.4 Chirped mirrors
A recently developed way of introducing controlled GDD is reection from a chirped mirror. A standard
dielectric mirror comprises a stack of alternating layers of high and low refractive index material, the thickness
of each mirror being equal to a quarter of the wavelength in the medium. In a chirped mirror the thickness of
the layers is varied within the stack so that dierent wavelengths are reected most strongly at dierent depths
in the stack, as illustrated schematically in Figure 5.8. With careful design and manufacture, it is possible to
82 LECTURE 5. ULTRAFAST LASERS
G1 G2, G1'

(2)
= 0
D = 0
(b)
G1 G2

(2)
< 0
D > 0
(a)
G1 G2

(2)
> 0
D < 0
(c)
G1'
G1'
Figure 5.7: Grating congurations giving (a) negative, (b) zero, and (c) positive GDD.
produce any desired GDD as well as controlled higher-order phase terms. If the GDD produced by one mirror
proves insucient, it may be increased by multiple reections.
Chirped mirrors have the advantage that they are simple to align and set up, and introduce a minimum of
optical elements into the system. A disadvantage is that they must be designed specically for each application,
and are therefore less exible.
5.3 TW and PW laser systems
Perhaps the supreme example of the application of dispersion control is the generation of laser pulses with peak
powers in the terawatt (1 TW 10
12
W) or even the petawatt (1 PW 10
15
W) range.
9
Imagine trying to generate laser pulses with a peak power of 1 TW. If the pulse had a duration of 1 s, our
1 TW laser pulse would require an energy of 1 MJ! The rst trick, then, in generating high peak powers is to
keep the duration of the pulses short; modelocked pulses in the femtosecond to picosecond ranges are used for
ultra-high power laser systems.
The laser pulses generated by modelocked lasers typically have very low energies a few nanojoules is quite
common. In order to reach the TW or PW level the pulse energy must be amplied by factors of 10
6
to 10
10
. If
we were to attempt to do this in a simple chain of ampliers, we would soon reach peak intensities that would
cause strong self-focusing, leading to damage of the optical components in the amplier chain. In practice the
intensity to which the pulses can be amplied is often even lower than the limit set by optical damage; at lower
intensities self-phase-modulation leads to the development of a complicated phase structure to the pulse which
cannot be compensated easily by controlled dispersion. An obvious way to keep the intensity reasonable is to
9
An excellent review of TW and PW laser systems may be found in S. Backus, Review of Scientic Instruments vol. 69 p. 1207
(1998). Further information on PW lasers systems can be found at the NIF website: www.llnl.gov/nif/.
5.3. TW AND PW LASER SYSTEMS 83
Figure 5.8: Schematic diagram showing: (a) the principle of operation of a chirped mirror; (b) the calculated
standing wave pattern in a chirped mirror as a function of wavelength. Note how for this mirror the penetration
depth increases with wavelength, giving negative GDD.
increase the diameter of the laser beam. However, this approach quickly requires large diameter laser rods and
optics, which are both expensive (if they are available) and unwieldy.
The solution to this problem is chirped pulse amplication (CPA). In this approach the low energy
pulses to be amplied to high energy, high peak power pulses in three stages:
1. The pulses are stretched in time by passing them through a pulse stretcher with a large GDD. Stretch
factors of order 10
3
to 10
5
are common.
2. The stretched pulses are then amplied in a chain of one or more ampliers. Since the duration of the
stretched pulses is many orders of magnitude longer than the original pulses, they may be amplied to
higher energies than in the absence of stretching by a factor approximately equal to the stretch factor.
3. The amplied pulse is then re-compressed to (approximately) the duration of the original pulse.
The CPA process is illustrated schematically in Figure 5.9
The pulse stretcher usually comprises the grating pair illustrated in Figure 5.7(c), set to produce positive
dispersion. One reason for this choice is that the material of the following amplier chain will tend to introduce
more positive dispersion, and therefore there is no danger of partially re-compressing the laser pulses as they
are amplied, which could lead to higher than expected peak pulse intensities in the laser ampliers.
5.3.1 Regenerative and multipass ampliers
In a CPA laser there will typically be several stages of amplication. Immediately after the pulse has been
stretched, there will be one or more pre-ampliers in which most of the gain occurs (perhaps a factor of 10
7
).
After the pre-ampliers the energy of the stretched pulses will be suciently high that they can extract energy
eciently from one or more power ampliers. Most of the energy in the nal pulse is added by the power
ampliers.
There are two main classes of amplier used, as illustrated schematically in Fig. 5.10.
84 LECTURE 5. ULTRAFAST LASERS
Compressor
amplifiers
stretcher
short
input pulse
stretched
and chirped pulse
amplified
chirped pulse
amplified
short pulse
Figure 5.9: Schematic diagram of chirped pulse amplication.
Faraday rotator
Pockels cell TFP
TFP
input beam
half-wave plate
gain medium
cavity
mirror
cavity
mirror
input beam
gain medium
amplified
output
beam
pump
beam
pump
beam
(b)
(a)
Figure 5.10: Schematic diagram of (a) a regenerative and (b) a multipass amplier.
5.3. TW AND PW LASER SYSTEMS 85
Regenerative ampliers
In a regenerative amplier the amplication takes place in an optical cavity. Laser pulses are injected into the
cavity by applying a voltage pulse to a Pockels cell as shown schematically in Fig. 5.10(a). In the absence of any
voltage on the Pockels cell, the incident laser pulses will, after reection from the cavity end mirror be reected
out of the cavity by the polarizer. However, if a quarter-wave voltage is applied to the Pockels cell, the pulses
will return to the intra-cavity polarizer with the orthogonal polarization and hence will be trapped in the cavity
and be amplied. After amplication the pulses are switched out by applying a second quarter-wave pulse to
the Pockels cell
10
.
Since the gain medium is located within an optical cavity, the pumping must be at a level which prevents the
single-pass gain from becoming too high, or lasing or at least strong amplication of spontaneous emission
will occur even in the absence of an injected pulse. As such, in order to extract energy from the gain medium
it is necessary for the pulse to undergo many cavity round trips, perhaps of order 20.
The advantages of regenerative ampliers are that the optical cavity imposes a good transverse mode struc-
ture on the amplied pulses. As such, the beam quality should be good. There are several disadvantages,
however. Leakage through the polarizers and Pockels cell can give rise to a series of lower-energy pulses propa-
gating ahead of the main pulse. Prepulses of this type can be problematic in many applications. Further, the
many round-trips required to extract the energy stored in the population inversion introduces a long length of
material into the path of the pulse (100s of millimetres). The consequent linear and non-linear phase structure
introduced into the pulse can be dicult to compensate for. Notwithstanding these diculties, regenerative
ampliers have been used to generate TW pulses with pulse durations as short as 30 fs.
5.3.2 Multipass ampliers
In a multipass amplier the pulse is passed through the gain medium several times by reection from a series
of mirrors sometimes referred to as angular multiplexing. Since the gain medium is not located within a
cavity the single-pass gain may be much higher without causing too much amplied spontaneous emission
(ASE). Multipass ampliers therefore have the advantage of introducing less material into the path of the pulse,
as well as not suering from prepulses. The disadvantages are that the overlap between the pump laser beam
and the amplied laser pulse is not as good as in a regenerative amplier, and consequently the eciency will
not be as high as in a regenerative amplier. A typical eciency for a multipass pre-amplier is 15%; although
multipass power ampliers may reach 30%, more or less equal to that achieved in a regenerative amplier.
5.3.3 Low-energy, TW laser systems
A good example of a low-energy TW laser is the system used in the Clarendon Laboratory, which has the
following key components:
1. A Ti:sapphire oscillator pumped by a frequency-doubled CW Nd:YLF laser. The oscillator generates a
stream of modelocked pulses with a pulse repetition rate of 86 MHz. The duration and energy of each
pulse are approximately 35 fs and 10 nJ respectively.
10
The Faraday rotator is included in the arrangement in order to prevent pulses being reected back into earlier parts of the laser
chain (such as the laser oscillator), since such reections can cause damage or instabilities. Its inclusion complicates a description
of the operation of the regenerative amplier somewhat, although the essential features are included in the discussion above. Let
us suppose that the input pulses are vertically polarized. The half-wave plate is oriented at 45

to the vertical so as to rotate the

plane of polarization to the horizontal plane. On on passing through the Faraday rotator the plane of polarization is rotated by
(say) +45

. In the absence of any voltage on the Pockels cell, the pulses will be rejected by the cavity and their polarization further
rotated by the Faraday rotator in the same sense, i.e. another +45

, to give vertically-polarized light. The half-wave plate then

ips the polarization to the horizontal plane, and consequently the pulses are reected by the external polarizer. Trapping laser
pulses within the cavity operates as discussed above, i.e. applying a voltage to the Pockels cell rotates the polarization of the laser
pulses so that they are transmitted by the intra-cavity polarizer. The pulses are rejected from the cavity by applying a second
pulse to the Pockels cell, whereupon they are reected by the external polarizer in the manner discussed in this note.
86 LECTURE 5. ULTRAFAST LASERS
2. A pulse stretcher which stretches the pulses to a duration of approximately 300 ps.
3. A subset of the stretched pulses are selected by a Pockels cell and injected into a regenerative amplier
operated at a repetition rate of 1 kHz. The regenerative amplier comprises a 5 mm diameter Ti:sapphire
rod pumped by pulses from a diode-pumped, frequency-doubled Q-switched Nd:YAG laser operating at a
pulse repetition rate of 1 kHz The regenerative amplier increases the energy of the pulses to 0.7 mJ.
4. A subset of the pulses from the regenerative amplier are injected into a multipass amplier comprising a
20 mm diameter Ti:sapphire rod pumped by a ashlamp-pumped, frequency-doubled, Q-switched Nd:YAG
laser which delivers approximately 1 J of green pump light per pulse. This amplier increases the energy
of the pulses to 200 mJ.
5. Finally the amplied pulses are compressed by a double-passed grating pair to a duration of approximately
40 fs. Losses in the compressor are dominated by the four reections from the diraction gratings, and
are such that the compressed laser pulses have an energy of approximately 120 mJ.
6. The compressed pulses therefore have a peak power of 3 TW and a duration quite close to those from the
oscillator.
5.3.4 High-energy, PW laser system
In order to reach very high pulse energies (100s of Joules) it is necessary to increase the beam diameter to
100 mm or more. At present it is not possible to grow large-diameter crystals of Ti:sapphire with sucient
optical quality for this purpose. As a consequence, high-energy systems use as the active medium Nd:Glass
since the technology for producing optical-quality rods and disks of this material is well developed.
The laser transition in Nd:Glass is inhomogeneously broadened with a large linewidth of approximately
50 THz, and hence in principle can produce modelocked laser pulses with durations as short as approximately
100 fs. In practice, pulses of duration 0.5 - 1 ps can be generated. Modern high-energy glass laser systems
amplify pulses from a Ti:sapphire oscillator and pre-amplier system, the Ti:sapphire being tuned in wavelength
to match the peak of the gain of the Nd:Glass laser. Typically a series of multipass rod ampliers is used for
amplication to pulse energies of several tens of Joules. Amplication to higher energies requires the use of
larger beam diameters, and hence large-area discs of Nd:Glass are used.
A good example of a PW class laser (there are not very many!) is the Vulcan laser at the Rutherford
Appleton Laboratory, just down the road near Didcot. This was the rst laser in Europe to reach the PW level.
The laser has the following features:
Pulses with an energy of 5 nJ, a duration of 120 fs, and a wavelength of 1055 nm are generated by a
modelocked Ti:Sapphire oscillator.
These pulses are stretched by a factor of 4 10
4
to a duration of 5 ns, and then amplied to an energy
of order 10 mJ using a new technique known as optical parametric chirped pulse amplication (OPCPA).
In this approach the gain medium is replaced by a non-linear, optical parametric amplier and the pulses
are amplied by conversion of energy from the pump laser beam to the wavelength of the pulses. This
method avoids the diculties of gain narrowing and reduces the pedestal level normally associated with
regenerative ampliers.
The pre-amplied pulses are then introduced into a chain of ashlamp-pumped rod and disk Nd:Glass
ampliers, which amplies the pulse energy to 670 J.
Two gold-coated holographic diraction gratings (940 mm diameter!) with a groove spacing of 480 lines mm
1
re-compress the pulse to 650 fs; somewhat longer than the original pulses, but not bad considering that
they have been amplied by a factor of 10
11
!
5.4. APPLICATIONS OF ULTRAFAST LASERS 87
After losses caused by reection at the compressor gratings and beam-steering optics, the energy of the
pulses available to the target chamber is approximately 500 J, i.e. a peak power of 0.8 PW. This can
be focused to a spot size of order 7 m diameter, corresponding to a peak focused intensity of order
10
21
Wcm
2
. Note that at this intensity the radiation pressure is 11 orders of magnitude greater than
atmospheric pressure!
5.4 Applications of ultrafast lasers
At low pulse energies (fraction of a millijoule), ultrafast lasers nd a myriad of applications in pump-probe
experiments in which an event, such as a chemical reaction, is triggered by a short laser pulse, and the subsequent
evolution is studied using a second, short probe laser pulse. The ability to control the phase structure of short
laser pulses is opening up the eld of coherent control in which, for example, the products of photo-initiated
chemical reactions are selected by controlling the precise variation of the electric eld within the laser pulse.
At somewhat higher pulse energies (several millijoules), focusing the laser pulse to a beam diameter of a few
tens of microns results in a peak laser intensity of 10
15
Wcm
2
and higher. At such intensities the electric eld
of the electromagnetic wave is comparable, or stronger, than the Coulomb eld binding the valence electrons in
an atom. A number of applications arise:
High-harmonic generation. Generation of very high-order (of order 100) harmonics of the driving laser
wavelength to produce beams of coherent radiation with wavelengths in the vacuum ultraviolet to soft x-ray
spectral region. High-harmonic generation requires relatively low driving intensities of order 10
15
Wcm
2
.
Driving table-top soft x-ray lasers. Typical intensities required for this are 10
16
- 10
17
Wcm
2
.
Plasma accelerators. At intensities of order 10
18
Wcm
2
, a laser pulse propagating through a plasma will
drive a longitudinal plasma wave that trails the laser pulse in much the same way that a wake follows a
boat travelling across a lake. The longitudinal electric elds within the plasma wave can be at least three
orders of magnitude greater than those used in conventional particle accelerators, such as those at CERN.
For example, recently experiments by groups in Oxford and Lawrence Berkeley National Laboratory
demonstrated acceleration of electrons to 1 GeV energy (c.f. the electron rest mass of 0.5 MeV) in an
laser-driven plasma accelerator only 33 mm long.
At somewhat higher pulse energies (several Joules), a variety of exotic phenomena become possible:
Laser-driven fusion. Of particular interest is so-called fast-ignitor inertial connement fusion, in which a
pre-heated core of material is rapidly heated by an intense ultrashort laser pulse to initiate fusion.
Generation of ultra-intense magnetic elds (up to 10
5
T) during the interaction of intense laser pulses with
a solid target.
Generation of intense, short-pulse proton or heavy-ion beams.
Laboratory Astrophysics. The extreme temperatures, pressures, and magnetic elds that can be created
when a ultra-high intensity laser interacts with matter allows us to recreate (briey) the conditions found
in several types of astrophysical objects.
88 LECTURE 5. ULTRAFAST LASERS
Lecture 6
Semiconductor & Fibre lasers
6.1 Review of semiconductor physics
As the constituent atoms of a semiconductor crystal are brought together from innity, the wave functions of
the valence electrons begin to overlap. At inter-atomic separations characteristic of the solid state, the wave
functions are suciently overlapped that the valence electrons are delocalized and can no longer be associated
with a particular atom.
The electron wave functions can then be written as Bloch wave functions:
(r) = u
k
(r) exp(ik r), (6.1)
where u
k
(r) has the periodicity of the crystal lattice.
The energy of the electron is a function of the wave vector k. For a free electron the energy and momentum
of the electron would be given by,
p = k (6.2)
E =

2
k
2
2m
e
, (6.3)
where m
e
is the rest mass of a free electron. However, for electrons moving in the periodic potential of the
crystal lattice this relation no longer applies. Deviation from the parabolic dependence of energy on k becomes
particularly strong when the wave vector k approaches one of the Brillouin zones
1
, which is to be expected
since this is just the condition for strong Bragg reection from the crystal planes. The distortion of the electron
energy from the free space value leads to some energies being forbidden and the formation of allowed energy
bands with energies E
n
(k) which are functions of k and an index n = 1, 2, 3, . . .. The formation of the band
structure is illustrated schematically in Figure 6.1.
We note that the Bloch wave functions are not momentum eigenstatates
2
, and hence k is not the momentum
of the electron. However, the quantity k does play a role akin to momentum in conservation laws describing
the interaction of electrons with elds or phonons. For this reason it is known as the crystal momentum.
The number of electron states per band is equal to twice the number of unit cells in the crystal. A semi-
conductor contains an even number of electrons per unit cell, and consequently at absolute zero all bands are
1
The rst Brillouin zone is dened as the Wigner-Seitz primitive cell of the reciprocal lattice; the Wigner-Seitz primitive cell
about any lattice point is the region of space which is closer to that lattice point than to any other.
2
Fundamentally this relates to the fact that the Hamiltonian is not invariant under arbitrary translations.
89
90 LECTURE 6. SEMICONDUCTOR & FIBRE LASERS
E(k)
k
/a /a 2/a 2/a 3/a 3/a 0
E(k)
k
/a /a 0
(a) (b)
Figure 6.1: Schematic diagram of the formation of band structure in a one-dimensional crystal lattice of lattice
spacing a. In (a) is shown the parabolic dependence of the energy E(k) on the wave vector k for a free electron
(dashed curve). This dependence is strongly distorted for values of k close to a Brillouin zone, leading to the
formation of bands of allowed energies separated by energy gaps. In (b) the bands are shown in the reduced
zone scheme in which reciprocal lattice vectors are added to (or subtracted from) k so as to bring everything
into the rst Brillouin zone. This last picture is usually the one used when discussing semiconductors.
either completely full or completely empty. As such, at absolute zero a semiconductor has zero conductivity.
What dierentiates a semiconductor from an insulator is that the energy gap between the top of the highest full
energy band, the valence band, and the next highest (empty) band, the conduction band, is not too large
compared to k
B
T. Typical values of this band gap energy are 0.5 - 3.0 eV, compared to k
B
T 0.025 eV. Ther-
mal excitation of electrons from the valence band to the conduction band leads to a nite electrical conductivity
which is between that of insulators and conductors and increases strongly with temperature.
Eective mass
It may be shown that under a force F an electron in one of the energy bands moves according to,
F =
2
_

2
E
n
k
2
_
1
dv
g
dt
, (6.4)
where v
g
is the group velocity of the electron. As such, it is useful to dene the eective mass:
6.1. REVIEW OF SEMICONDUCTOR PHYSICS 91
m

=
2
_

2
E
n
k
2
_
1
. (6.5)
Hence the eective mass of the electron depends on the curvature of the band energy E(k). Note that for a free
electron, the denition above gives m

= m
e
.
Holes
Figure 6.2 shows a semiconductor in which some electrons have been excited from the valence band to the
conduction band to give two partially-lled bands. In describing the partially-lled valence band, it is often
easier to focus attention on the relatively few empty electron states than on the many lled states. Empty
electron states are known as holes, and have the following properties:
k
h
= k
e
, where k
h
is the wave vector of the hole, and k
e
that of the missing electron.
E
h
= E
e
, where E
h
is the energy of the hole, and E
e
is the energy that an electron would have if it
occupied the empty state.
v
h
= v
e
m

h
= m

e
. This follows from the rst and second points above.
The hole moves under the inuence of external electromagnetic elds as if it had a charge of +e.
Hence the holes behave as particles of positive charge, with an energy and mass equal to minus that of the
missing electron. Notice that for the partially-lled valence band, the curvature of the electron energy in the
region near the top of the band is negative, and hence their eective mass would be negative. In contrast, the
holes appear to have a positive eective mass which is another reason for dealing with the hole states rather
than the electron states.
6.1.1 Distribution over energy levels
The probability of a level with an energy between E and E +dE being occupied by an electron is given by the
Fermi-Dirac distribution:
f(E)dE =
1
exp([E E
F
]/k
B
T) + 1
, (6.6)
where E
F
is the Fermi energy. From eqn (6.6) we see that the Fermi energy is that energy for which a level
has a probability of 1/2 of being occupied. Alternatively, as shown in Figure 6.3(a), the Fermi energy gives the
energy to which the band is lled at T = 0.
Imagine that we transfer N
e
electrons per unit volume from the valence band into the conduction band. After
a few picoseconds the electrons will equilibrate within the conduction band to form a local thermal equilibrium
at temperature described by T and a quasi-Fermi level E
Fc
. Similarly the electrons in the valence band will
equilibrate to a thermal distribution described by a quasi-Fermi level E
Fv
. Hence the probabilities of a level of
energy E being occupied are :
f
c
(E) =
1
exp([E E
Fc
]/k
B
T) + 1
(6.7)
f
v
(E) =
1
exp([E E
Fv
]/k
B
T) + 1
, (6.8)
for the conduction and valence band respectively.
92 LECTURE 6. SEMICONDUCTOR & FIBRE LASERS
conduction band
valence band
electrons
electrons
holes
0
E
electron
E
hole
E
g
Figure 6.2: Partially-lled valence bands following promotion of electrons from the conduction to the valence
band.
6.1.2 Doped semiconductors
In an ultra-pure sample of a semiconductor compound such as GaAs the number of electrons excited across the
energy gap up into the conduction band at room temperature is very small. As indicated in Fig. 6.4(a) for
such an ultra-pure semiconductor the Fermi energy would lie halfway up the energy gap between the top of the
valence band and the bottom of the conduction band, corresponding to the energy for which the probability of
occupation is
1
2
.
Higher concentrations of electrons in the conduction band are required for most practical applications. This
is achieved by adding carefully controlled concentrations of particular doping species to the bulk semiconductor
compound. For example, for the case of GaAs, adding small proportions of Si atoms or Te atoms
3
we introduce
a new energy level corresponding to the impurity species which lies just below the conduction band of the host
material, as indicated schematically in Figure 6.4. Atoms of these particular added species (Si or Te) have
one more electrons in their outer shells than the atoms they replace and can therefore act as donor species
by providing a source of electrons which go into the conduction band, because the gap between the impurity
level and the conduction band of the host material is small and the impurity atoms are easily ionized at room
temperature. This type of material, with an excess of electrons in the conduction band is called n-type, and at
light doping levels provides a number density of electrons in the conduction band, which although much larger
than in the case of the intrinsic undoped material, nevertheless depends strongly on temperature and tends to
zero at low temperatures. Under conditions of light doping the Fermi energy lies between the impurity level
and the bottom of the conduction band.
3
Si belongs to group IV and substitutes for atoms originally on Ga sites. Te belongs to group VI and substitutes for atoms
originally on As sites. In the case of n-type GaAs, it is Si which is used most commonly.
6.2. RADIATIVE TRANSITIONS IN SEMICONDUCTORS 93
-20 -15 -10 -5 5 10
0.2
0.4
0.6
0.8
1
0
(E - E
F
) / k
B
T
f(E)
T > 0
T = 0
Figure 6.3: The Fermi-Dirac distribution for T = 0 (dashed line) and T > 0 (solid line).
In contrast if the same n-type material is heavily doped the impurity level merges into the conduction band,
and the Fermi level moves up to sit close to the top of the occupied states in the conduction band as indicated
in Figure 6.4(c).
In an analogous way, by doping GaAs with electron acceptor atoms such as Be, Zn, or C, p-type materials
with excess holes in the valence band can be produced and at heavy doping levels the Fermi level of such
materials moves down into the valence band
4
.
Heavily doped semiconductor materials are known as degenerate semiconductors. Because they have
partially lled bands (like metals) they remain electrically conducting even at very low temperatures.
6.2 Radiative transitions in semiconductors
Suppose that an incident harmonic wave of angular frequency causes an electron to make a transition from
an upper level of energy E
2
in the conduction band to a level in the valence band of lower energy E
1
. If we
write the electric eld of the incident wave in the form,
E(r, t) = E
0
exp [i(k
opt
r t)] , (6.9)
the perturbation takes the form,
H

= eE r. (6.10)
The rate of transitions induced by the perturbation is given by Fermis golden rule:
W
21
=

2

2
[H
21
[
2
(
21
), (6.11)
4
Be belongs to group IIA and Zn to group IIB, and these atoms substitute for Ga to give rise to excess holes. C belongs to
group IVB and substitutes for As.
94 LECTURE 6. SEMICONDUCTOR & FIBRE LASERS
conduction band
valence band
conduction band
valence band
impurity
levels
conduction band
valence band
Fermi
level
impurity
levels
Fermi
level
Fermi
level
(a) (c) (b)
Figure 6.4: Occupancy of conduction and valence bands and position of the Fermi level for: (a) ultra-pure
semiconductor; (b) after the introduction of a small density of an electron donor impurity species; (c) after the
introduction of a large density of electron donor species.
where,

21
= E
2
E
1
, (6.12)
and,
H
21
=
_

2
(r) [er E
0
exp(ik
opt
r)]
1
(r)d (6.13)
=
_
u

2
(r) er E
0
exp [i(k
opt
+k
1
k
2
) r] u
1
(r)d. (6.14)
We see from eqn (6.14) that the integrand will oscillate rapidly during the integration, and hence be zero, unless
k
2
= k
1
+ k
opt
. (6.15)
This, then, is a selection rule for optical transitions between electron levels in a semiconductor; it resembles a
statement of conservation of momentum, except that the quantities k
2,1
are crystal momenta. In the optical
region [k
opt
[ [k
1,2
[ and hence the selection rule becomes,
k
2
k
1
, (6.16)
and hence k is unchanged in the transition the transitions are vertical in our diagrams.
We also note that since the perturbation does not operate on the electron spin s, we also have the selection
rule,
s = 0. (6.17)
6.2. RADIATIVE TRANSITIONS IN SEMICONDUCTORS 95
6.2.1 Joint density of states
We now consider a transition between two levels of energy E
2
= E
c
(k
2
) and E
1
= E
v
(k
1
) of a given value of
k = k
1
k
2
. In any particular band the density of states with wavevectors between k and k + dk is given by
the well known result:
g(k)dk =
k
2

2
dk. (6.18)
We wish to nd the density of transitions with a given energy dierence . To this end we note that within
the parabolic approximation we may write the energies of the upper and lower electron states as,
E
2
= E
g
+

2
k
2
2m

c
(6.19)
E
1
=

2
k
2
2m

v
, (6.20)
where m

v
is the mass of an electron in the valence band, i.e. m

v
= m

h
such that m

v
> 0.
Hence the energy of the transition is written in terms of k as,
= E
2
E
1
= E
g
+

2
k
2
2
_
1
m

c
+
1
m

v
_
= E
g
+

2
k
2
2
, (6.21)
where the reduced mass is given by,
1

=
1
m

c
+
1
m

v
. (6.22)
Now that we have established a relationship between the energy of the transition and the wavevector of the
electrons involved in the transition we may now use eqn (6.21) to re-write the density of states in terms of the
angular frequency of the transition:
g(k)dk =
k
2

2
dk = g
j
()d (6.23)
g
j
()d =

2
2
_
2

2
_
3/2
( E
g
)
1/2
d, (6.24)
where g
j
() gives the density of transitions of angular frequency , and is known as the joint density of
states.
6.2.2 Density of possible transitions
The joint density of states gives the density of transitions of angular frequency , but not all of these transitions
can occur since that requires that the initial state is occupied by an electron and the nal state is empty.
For example, the density of possible stimulated emission transitions with frequencies in the interval to
+ d is given by,
g
j
()f
c
(E
2
)[1 f
v
(E
1
)]d,
96 LECTURE 6. SEMICONDUCTOR & FIBRE LASERS
i.e. the density of transitions in this frequency interval multiplied by the probability that the upper level
is occupied and the lower level is unoccupied. Similarly the density of possible absorption transitions with
frequencies in the interval to + d is given by,
g
j
()f
v
(E
1
)[1 f
c
(E
2
)]d.
The density of net stimulated transitions with frequencies in the interval to +d is given by the dierence
of these two expressions:
g()[f
c
(E
2
) f
v
(E
1
)]d. (6.25)
6.2.3 The gain coecient
We are now in a position to write down the net rate of stimulated emission per unit volume with frequencies
between and +d. From eqns (6.11) and (6.25) we have:
R
SE
= W
21
g
j
()d[f
c
(E
2
) f
v
(E
1
)]. (6.26)
Note that for atoms we would write the above in terms of the population inversion density N

and optical gain

cross-section
21
() as,
R
SE
= N

21
()
I

= ()
I

, (6.27)
where () is the gain coecient of the transition. Hence we deduce that the gain coecient is given by
() = R
SE
/I. From eqns (6.26) and (6.14) we nd, after tidying and eliminating the intensity using
I = (1/2)
0
cnE
2
0
, where n is the refractive index of the medium,
() =
[D
21
[
2
n
0
c
3
(2)
3/2
( E
g
)
1/2
[f
c
(E
2
) f
v
(E
1
)]. (6.28)
6.2.4 Condition for gain
Notice that the term [f
c
(E
2
) f
v
(E
1
)] plays the role of the population inversion density which appears in the
expression for the gain coecient in atomic or molecular laser transitions. Clearly, in order for there to be
positive gain we must have,
f
c
(E
2
) > f
v
(E
1
) (6.29)

1
exp([E
2
E
Fc
])/k
B
T + 1
>
1
exp([E
1
E
Fv
])/k
B
T + 1
(6.30)
E
2
E
1
< E
Fc
E
Fv
. (6.31)
The left-hand side is simply the photon energy. The photon energy must also exceed the gap energy, and hence
we can write the condition for gain as,
E
g
< < E
Fc
E
Fv
. (6.32)
6.2. RADIATIVE TRANSITIONS IN SEMICONDUCTORS 97
conduction band
valence band
E
Fc
E
Fv

=

E
g

=

E
F
c

-

E
F
v
Figure 6.5: Illustrating the condition for gain for the case when T = 0.
In order for this condition to be met at any frequency, we must therefore have,
E
Fc
E
Fv
> E
g
Condition for gain. (6.33)
This condition originally derived in 1961 by Bernard and Duraourg holds for all temperatures T, but
is particularly easy to understand for the case T = 0 as illustrated schematically in Figure 6.5. At absolute zero
the gain corresponds to the upper level being occupied whilst the lower level is empty. At nite temperatures the
condition for gain is that the probability of a level of energy E
2
being occupied is greater than the probability
of a level of energy E
1
being occupied.
6.2.5 Generating gain: the injection density
Gain on optical transitions in semiconductor lasers is generated by raising N electrons per unit volume from
the valence band to the conduction band. The density of electrons moved is known as the injection density.
Since the density of electrons in the conduction band is equal to the density of holes in the valence band we
may write,
N =
_

0
g
c
(E
c
)f
c
(E
c
)dE
c
=
_

0
g
v
(E
v
)[1 f
v
(E
v
)]dE
v
, (6.34)
where g
c
(E) and g
v
(E) are the density of states of energy E in the conduction band and valence band respec-
tively. The density of states may be written in terms of energy by using eqns (6.18), (6.19) and (6.20).
The relations above then determine the positions of the quasi Fermi levels in the conduction and valence
bands, and hence the frequency width over which gain can be achieved.
For a given density N, the transcendental eqns (6.34) may be solved to nd the quasi-Fermi levels. Figure 6.6
shows the result of one such calculation for GaAs. Notice that for low injection densities the quasi-Fermi levels
98 LECTURE 6. SEMICONDUCTOR & FIBRE LASERS
N
tr
threshold for gain
Figure 6.6: Calculation of the variation of the quasi-Fermi levels in the conduction and valence bands of GaAs
as a function of the injected electron density N. For the conduction band the energy of the quasi-Fermi level
above the bottom of the band is plotted; for the valence band the energy of the holes (which have minus the
energy of the corresponding electron state) is plotted. The transparency condition therefore corresponds to the
sum of these terms equalling zero, which occurs at an injection density N
tr
as shown.
lie outside their respective bands. As the injection density is increased the Fermi level in the conduction band
rises more quickly than the Fermi level in the valence band drops. Hence the gain becomes positive.
Once the quasi-Fermi levels have been calculated as a function of the injection density, the gain coecient
may be found from eqn (6.28). Figure 6.7 shows the results of such a calculation. Two points are worth noting.
The gain coecient is typically large, some tens of cm
1
and hence laser oscillation is possible with very short
gain lengths. Second, the bandwidth of the gain is broad and increases as the injection density of electrons is
increased. For the example shown, the gain bandwidth is of order 10 THz, corresponding to a 30 nm bandwidth
at the operating wavelength of 840 nm.
The gain rst becomes positive when eqn (6.33) is met, corresponding to the injection density exceeding a
threshold density, known as the transparency density, N
tr
. The condition for gain can therefore be written
as N > N
tr
, which is analagous to N

> 0 for gain on an atomic transition.

6.3 Homojunction diode lasers
In a homojunction laser, gain is achieved in an intrinsic region of semiconductor located between layers of
degenerate p-type and n-type semiconductor, a so-called p-i-n junction.
Figure 6.8(a) shows schematically the positions of the conduction bands and Fermi levels at the (rather
theoretical!) moment when the three layers are rst contacted. Electrons from the n-type material can drift
across the intrinsic layer and recombine with holes in the p-type layer; in a similar way, holes drifting across
the intrinsic layer can recombine with electrons in the n-type layer. This process sets up an electric eld from
the n-type to the p-type layer, and equilibrium is reached when the rate of drift of electrons across the intrinsic
layer is balanced by the rate at which they are returned by the electric eld. The region in which the electric
eld exists is virtually devoid of charge carriers since they are quickly removed by the eld; it is known as the
depletion layer.
The electric eld raises the potential energy of electrons in the p-type layer, and lowers it in the n-type layer,
6.3. HOMOJUNCTION DIODE LASERS 99
1.424
80
40
0
G
a
i
n

(
c
m

1
)
40
2
.
5

1
0
1
8

c
m

3
1
.
5

1
0
1
8

c
m

3
80
120
160
1.46

0
(eV)
1.50
Figure 6.7: Calculated gain coecient as a function of photon energy for various values of N (in units of
10
18
cm
3
) for GaAs at T = 300k.
to give the equilibrium distribution shown in 6.8(b). Notice that under these equilibrium conditions the Fermi
energy lies at the same level in all three layers.
If a forward-bias voltage is now applied to the structure, the electric eld across the device will be reduced,
leading to a distribution of energy levels as shown in 6.8(c). Now there exists a region in the intrinsic layer in
which there are electrons in the conduction band and holes in the valence band; gain is possible provided the
quasi-Fermi levels obey eqn (6.33), or, equivalently, the density of electrons in the conduction band exceeds the
transparency density.
To relate the density of electrons in the conduction band to the diode current, we note that most of the
current is carried by the electrons (owing to their greater mobility). If the time for electron-hole recombination
in the intrinsic layer is
rec
, then in the steady-state, and in the absence of stimulated emission, the rate of ow
of electrons into the intrinsic layer must be balanced by the rate of recombination. Hence,

i
I
eV
g
=
N

rec
, (6.35)
where V
g
is the volume of the active region, and the internal quantum eciency
i
gives the fraction of
carriers that recombine radiatively in the active layer other carriers undergo non-radiative recombination at
the junction boundary, or ow round the active region. Hence the threshold current may be written as,
I
thresh
=
1

N
tr
eV
g

rec
. (6.36)
100 LECTURE 6. SEMICONDUCTOR & FIBRE LASERS
electron
diffusion
hole
diffusion
p n i p n i
E
F
p n i
E
Fv
E
Fc
electric
field
(a) (c) (b)
a
c
t
i
v
e

r
e
g
i
o
n
Figure 6.8: Energy levels and Fermi levels in a pin diode: (a) at the (theoretical) moment when the three
layers are brought into contact; (b) after the depletion layer has been established; (c) under the inuence of a
forward-bias voltage.
We see that the threshold current increases with N
tr
and the volume of the active region.
6.3.1 Structure of homojunction diode lasers
Figure 6.9 shows the structure of a homojunction diode laser. Notice that the gain length is very short, only
200 - 500 m, and perhaps 100 - 200 m wide. Optical feedback is usually provided by the uncoated surfaces
of the cleaved crystal planes at each end of the device. The high refractive index of GaAs (n = 3.6) ensures a
Fresnel reectivity of approximately 32 % which is sucient for lasing to be achieved.
One of the limitations of homojunction lasers is that the potential barrier preventing electrons from entering
the p-type material is very low when the forward-bias voltage is applied. Consequently electrons tend to diuse
into the p-type material, increasing the thickness of the active region beyond that of the depletion layer. In
practice diusion increases the thickness to around 1 m, much thicker than the thickness of the depletion layer.
This increases the threshold current to tens of amps, corresponding to a current density of order 10
5
Acm
2
.
This high value means that homojunction diode lasers cannot be operated continuously without damaging the
structure unless the whole device is cooled to cryogenic temperatures. For this reason, homojunction diode
lasers are of historical interest only and all modern diode lasers use heterostructures.
6.4 Double heterostructure lasers
All diode lasers currently used employ a double heterostructure, rather than the simple p-i-n geometry
discussed above. This structure can reduce the threshold current density by two orders of magnitude (to
approximately 10
3
Acm
2
) owing to:
Photon connement: The active layer is surrounded by layers of a higher refractive index. As such the
structure acts as a waveguide, and prevents the laser radiation from extending into regions of high loss.
Carrier connement: The surrounding layers also have large band gaps which connes the current
carriers to the active region. The resulting increase in carrier density increases the gain.
6.4. DOUBLE HETEROSTRUCTURE LASERS 101
tb]
Figure 6.9: Physical structure of a homojunction diode laser.
Reduced absorption: The large energy gap of the surrounding layers ensures that any laser radiation
which does extend beyond the active region experiences much lower absorption than it would if the band
gap were the same as that in the active layer.
Figure 6.10 shows schematically the structure of heterostructure diode lasers made from (a) AlGaAs and (b)
InGaAsP. In each case the active layer ( 0.15 m) is surrounded by thicker layers of n- and p-doped material
with a higher band gap. Figure 6.11 shows the variation of the refractive index, laser intensity, and band gap
along a line perpendicular to the p-i-n junction under conditions of forward bias. It is clear that the higher
band gap of the surrounding material prevents current carriers from moving beyond the active layer. Further,
the diode laser radiation is surrounded by layers of higher refractive index and hence the mode is guided.
6.4.1 Forming heterostructures: Material properties
The selection of materials from which to make the cladding layers is a crucial one; not only must their band
gap be larger than that of the intrinsic layer, but they must also be capable of being deposited via epitaxial
growth techniques on the active region to form crystals free from defects without grain boundaries or regions
of dislocation. This means that the alloy used for cladding must have a lattice constant very closely matched
to that of the intrinsic material.
As shown in Figure 6.12, if we start with pure GaAs, by adding increasing atomic proportions x of aluminium
to form the ternary alloy Ga
1x
Al
x
As the lattice spacing is virtually the same whatever the proportion of added
Al, even to the extent of replacing all of the Ga by Al to form AlAs since the lattice spacing of GaAs and AlAs
are almost identical: 0.564 and 0.566 nm respectively.
This is however a happy circumstance unique to the GaAs system. In general quarternary compounds
must be employed. For example, in In
1x
Ga
x
As
y
P
1y
the lattice spacing can be matched to the III-V semi-
conductor InP if y = 2.2x. With this ratio xed, adjusting x then varies the band gap energy, allowing the
output of the diode laser to be adjusted between 1150 nm and 1670 nm. This range is particularly useful since
102 LECTURE 6. SEMICONDUCTOR & FIBRE LASERS
~1 m
~1 m
~ 0.15 m
p - Al
0.3
Ga
0.7
As
n - Al
0.3
Ga
0.7
As
n
+
- GaAs
(substrate)
GaAs (active layer)
p - InP
n - InP
n
+
- InP
(substrate)
In
1-x
Ga
x
As
y
P
1-y
(active layer)
~1 m
~1 m
~ 0.15 m
(a) (b)
Figure 6.10: Schematic diagram of double heterostructure diode lasers made from: (a) AlGaAs; (b) InGaAsP.
Active region Material of cladding layers Wavelength range Substrate material
GaAs (AlGa)As 800 - 900 nm GaAs
GaInP (AlGa)InP 630 - 650 nm GaAs
InGaAs GaAs 900 - 1000 nm GaAs
InGaAs InGaAsP 1550 nm InP
InGaAsP InGaAsP 1300 - 1550 nm InP
InAsP InGaAsP 1060 -1400 nm InP
InGaAs InP 1550 nm InP
InGaN AlGaN, GaN 370- 460nm Al
2
O
3
,GaN
Table 6.1: Diode laser materials and wavelengths of operation.
it covers the second (1300 nm) and third (1550 nm) transparency windows of silica bres. InGaAsP diode lasers
therefore nd widespread application in telecommunications.
Table 6.1 lists the properties of some common types of diode laser.
6.5 Geometries of heterostructure diode lasers
Figure 6.13 shows the design of two types of heterostructure diode lasers. In both designs an oxide layer is
added immediately under one of the electrode layers to restrict the extent of the current ow, and hence the
gain region, in the y-direction. In the x-direction the region of gain is limited to the thickness of the active layer.
By controlling the volume of the gain region in this way, the threshold current for achieving laser oscillation is
reduced; although, of course, the laser power that can be extracted is also reduced.
For the design shown in Figure 6.13(a), the radiation mode is conned in the x-direction by the lower
refractive index cladding layer. In the y-direction the extent of the beam is determined by the width of the
gain region, and hence is said to be gain-guided. However, the connement is weak in this direction, and
the oscillating mode extends into regions which are not pumped, and so experiences high loss. This problem is
overcome in the buried heterostructure shown in Figure 6.13(b). Here, the active region is also surrounded
in the y-direction by cladding, or burying, layers which conne the radiation in that direction as well.
6.6. OUTPUT CHARACTERISTICS OF DIODE LASERS 103
E
g2
E
g2
E
g2
E
Fn
E
Fp
active
region
confinement
of carriers
n
2
n
2 n
1
transverse
beam profile
x - direction
(a)
(b)
(c)
Figure 6.11: Schematic diagram of the variation of: (a) the refractive index; (b) the laser intensity; and (c) the
band gap along the x-axis of a double heterostructure diode laser.
6.6 Output characteristics of diode lasers
Diode lasers have very high slope eciencies dP
out
/dP
in
= dP
out
/(V dI), where V is the diode voltage. This
can reach approximately 50% may be obtained in commercial devices.
The output power of a single double heterostructure diode laser is limited to powers less than a few hundred
mW by over-heating of the diode structure. Much large powers can be generated by manufacturing linear arrays
of diode lasers, known as diode bars. In turn, diode bars can be stacked to form diode arrays. A wide variety
of geometries are available. Diode arrays can give output powers of 1 kW or even higher, and are increasingly
used as replacements in ashlamps in solid state lasers such as Nd:YAG, and as the pump laser in bre lasers.
The frequency spectrum of diode lasers often depends on the driving current. Near threshold the laser mode
spectrum has the wide bandwidth characteristic of an LED, but as the excitation current is increased a dominant
axial mode establishes itself. This competes more and more eectively for the available inversion as the current
increases so that the bandwidth spanned by the oscillating modes decreases as the current increases. Diode
lasers can be operated in a single longitudinal mode by providing frequency-selective coupling. For example, a
bre Bragg grating (FBG) can be xed to the laser output; in a FBG the refractive index of the bre core
varies periodically, and hence only those wavelengths satisfying the Bragg condition for reection are eciently
coupled back into the gain region. Alternatively, wavelength-selective coupling can be provided by modulating
104 LECTURE 6. SEMICONDUCTOR & FIBRE LASERS
Figure 6.12: The band gap energy versus lattice spacing for some semiconductor compounds.
the thickness of the layers surrounding the diode laser gain region with a washboard pattern; for the selected
wavelength this provides continuous coupling between the left- and right-going waves in the gain region.
The biggest disadvantage of diode lasers is that they have a high divergence, which reects the small
dimensions of the gain region (typically L
x
= 1 L
y
= 5 m). Further, the divergence is dierent in the two
orthogonal directions (
x
25
o

y
5
o
) which is often undesirable in applications. Techniques have been
developed for making the beam more nearly circular, such as the use of prisms to expand the beam in one
direction, or anamorphic lenses which have dierent curvatures in the x- and y-directions. Nevertheless the
output beam quality of diode lasers remains poor in comparison to that from other types of laser.
Figure 6.14 shows the typical variation of output power with the diode current. Notice the low threshold
current, a few milliamperes, and the sensitivity of the threshold current with the device temperature. This
sensitivity reects the strong dependence on the quasi-Fermi energies on temperature.
6.7 Fibre lasers
Fibre lasers are particularly ecient at converting the often poor beam quality obtained from a diode laser into
diraction-limited laser radiation. In a bre laser the active ion is doped into the central core of an optical
bre, the ions being pumped by radiation propagating along the bre axis.
Fibre lasers have several advantages over the traditional rod or disc geometries:
Low pump power threshold: Achievement of the threshold population inversion density requires the
pump laser to exceed a threshold intensity. Since the cross-sectional area of an optical bre is so low, the
power required to reach threshold is orders of magnitude smaller than in a rod or disc where the area of
the gain medium is at least an order of magnitude larger. In practice threshold pump powers signicantly
below 1 mW can be achieved. Using the bre laser geometry it is possible to achieve lasing in lasers with:
low gain coecients; with high threshold pump intensities; and even in three-level laser systems.
Guiding of pump beam: The pump laser radiation is naturally guided by the bre, allowing very long
gain regions to be employed.
High eciency: The good overlap between the pump and generated laser radiation along the whole
length allows ecient conversion of the pump laser energy into laser output.
6.7. FIBRE LASERS 105
s
p - Ga
1-x
Al
x
As
n - Ga
1-x
Al
x
As
n - GaAs (substrate)
metal contact (-)
metal contact (-)
oxide
GaAs
s
n - Ga
1-z
Al
z
As
(burying layer)
p - Ga
1-z
Al
z
As
(current blocking)
(a) (b)
y
x
Figure 6.13: Geometries of heterostructure diode lasers: (a) gain-guided laser; (b) buried heterostructure,
index-guided laser.
Good thermal management: For any solid state laser system it is important to conduct away pump
energy deposited into the gain medium in the form of heat, in order to avoid thermal lensing. The high
surface area to volume ratio of the bre geometry allows heat to be conducted away more eectively.
Compact gain medium: Optical bres may be coiled, keeping the volume of the laser system small.
Robustness: Optical bres are robust and much less liable to cracking or other physical damage.
The main disadvantages of bre lasers are that the volume of the gain region is small. Consequently the
energy stored in the gain region will also be small, and hence bre lasers are not useful for amplifying laser
pulses to high energy. Further, the long propagation length through the bre core can be associated with the
development of non-linear loss processes such as stimulated Raman scattering or stimulated Brillouin scattering.
6.7.1 Pumping congurations and bre materials
In early bre lasers both the pump radiation and the generated laser radiation were guided in the inner core of
an optical bre (Fig. 6.15(a)). As such the pump laser must be close to diraction-limited for ecient coupling
into the inner core of the bre, which restricts the use of diode lasers to single-stripe devices.
Figure 6.15(b) shows a solution to this problem: cladding pumping. For this geometry the core of refractive
index is surrounded by two further layers: an outer core of lower refractive index, and a cladding layer of still
lower refractive index. The cladding layer will usually be protected by a nal protective layer. The pump laser
radiation is therefore guided in a multi-mode fashion by reections at the boundary between the outer core and
the cladding layer; with that part of the pump radiation passing through the core contributing to the pumping
of the bre laser. The laser radiation generated in the core can be guided mono- or multi-mode.
The use of cladding pumping allows much greater pump powers to be employed, increasing the output of
the bre laser from less than 1 W to greater than 100 W in a monomode beam and up to 1 kW in multi-mode
bre lasers.
6.7.2 An example
Figure 6.16 shows schematically the design of a high-power Nd:Glass bre laser. For this system the doped
core has a diameter of 10 m, with a numerical aperture of 0.17. The outer core has a diameter of 400 m and
106 LECTURE 6. SEMICONDUCTOR & FIBRE LASERS
Input current (mA)
O
u
t
p
u
t

p
o
w
e
r

(
m
W
)
Figure 6.14: Output power of Rohm RLD-78MA double heterostructure diode laser versus current for various
operating temperatures. Inset: photo of 5.6 mm diameter packaged diode laser. (From www.rohm.com)
numerical aperture of 0.38. Pump radiation, provided by a diode array operating at 808 nm, is coupled into
one end of the bre through a mirror which has a high transmission at 808 nm, but is highly reective at the
Nd:Glass wavelength. The output coupling mirror is also highly reective at the pump laser wavelength, but has
a transmission of approximately 90% for the Nd:Glass laser radiation. With this design, output powers of 32 W
have been achieved for a pump power of 72 W, corresponding to a conversion eciency of 44%. The output is
in the form of a single transverse mode, since although the core supports several low-order modes, the lowest
mode has the lowest losses and hence grows at the expense of higher-order transverse modes. Laser oscillation
occurs at approximately 1065 nm, with a large spectral width of 15 nm reecting the large inhomogeneous
width introduced by the glass host.
6.7.3 Erbium-doped bre lasers
Silica optical bres have transmission windows at 1.3 m and 1.5 m. Single-mode bres used at 1.3 m have
low dispersion and so the distance that signals can propagate before they must be amplied or regenerated is
determined primarily by the bre loss of 0.4 dBkm
1
. Ampliers are not available at 1.3 m, and so signals
at this wavelength are regenerated in a regenerative repeater, which: (i) detects and interprets the signal; (ii)
corrects errors; (iii) regenerates and retransmits the bitstream. The optimum repeater separation is typically
40 km.
At a wavelength of 1.5 m conventional bre has a lower loss (approximately 0.3 dB km
1
), but higher
dispersion than at 1.3 m. However in dispersion-shifted bre, in which material and waveguide dispersion
are balanced, the dispersion at 1.5 m can be reduced to 0.1 ps km
1
nm
1
compared to 17 ps km
1
nm
1
for
conventional bre at 1.3 m.
5
Hence dispersion-shifted bre can transmit signals some two orders of magnitude
5
In any optical bre the dispersion experienced by each mode the intra-modal dispersion has two contributions: dispersion
arising from the dispersion of the materials employed (material dispersion); and dispersion arising from the frequency dependence
of the wave vector of each mode (waveguide dispersion), which depends on the geometry of the bre. In dispersion-shifted bres
the dimensions of the bre are adjusted so that the total dispersion is low at the desired operating wavelength. In addition, for
any bre dierent modes have dierent phase velocities leading to so-called inter-modal dispersion; this can be avoided if only one
mode is excited.
6.7. FIBRE LASERS 107
doped core
cladding
(a)
(b) doped core
cladding
outer core
Figure 6.15: Fibre laser pumping congurations: (a) pumping of a conventional bre; (b) cladding pumping.
mode mixer
Figure 6.16: Schematic diagram of a high-power Nd:Glass bre laser.
further corresponding to distances of tens of thousands of kilometres before regeneration of the signal is
required. Of course, amplication of the signal is required more often to overcome the bre losses; typically
amplication is required every few tens of kilometres.
Amplication can be provided by an erbium-doped bre amplier (EDFA). The EDFA operates on the
4
I
13/2

4
I
15/2
transition at approximately 1550 nm. As discussed previously, pump radiation can conveniently
be provided by diode lasers operating at 800, 980, or 1480 nm.
Figure 6.17 shows schematically the conguration of an EDFA. The pump laser radiation is coupled into the
amplier by splicing together a bre carrying the pump radiation and a bre coaxial with the amplier. The
amplifying section is a monomode silica bre doped with around 0.1% erbium, and with a length ranging from
metres to hundreds of metres. The single-pass gain is typically of order 10
3
. Notice that optical isolators
based on the Faraday eect are inserted around the EDFA to prevent parasitic laser oscillation arising from
the small Fresnel reections at junctions in the bre.
EDFAs have been used in the latest 5 Gbit s
1
transatlantic bres which employ ampliers every 35 km,
but only regenerate the signal at the two land stations!
108 LECTURE 6. SEMICONDUCTOR & FIBRE LASERS
amplified
signal signal
erbium-doped fibre
isolator
pump
laser
Figure 6.17: Schematic diagram of an EDFA.
Finally we note that the very wide bandwidth (some 30 nm or 3.74 THz) of the erbium transition allows
amplication of multiple channels each transmitted at a slightly dierent wavelength. This technique, wave-
length division multiplexing (WDM), could allow, for example, 375 channels of 2.5 Gbit s
1
and spaced
by 10 GHz to be transmitted within the bandwidth of an erbium amplier. The implementation of WDM,
however, requires the development of suitable laser sources and techniques for de-multiplexing the transmitted
signals; a great deal of work is being done in this area.
Lecture 7
Precision frequency control
Here we discuss some of the issues aecting the frequency linewidth of an oscillating laser. We will often give
results in terms of the frequency (units of Hertz) rather than the angular frequency (units of radians per
second) since this allows easier comparison with real laser systems.
7.1 Single-mode operation
In principle a laser operating on a homogeneously broadened transition should oscillate on only a single longi-
tudinal mode, but in practice spatial hole-burning means that oscillation can occur on more than one mode.
Lasers operating on homogeously broadened transitions will typically operate on many longitudinal modes. In
addition, laser oscillation can occur on higher-order transverse modes, which in general have slightly dierent
frequencies.
It is clear that for a laser to achieve the narrowest possible linewidth it should operate on a single longitudinal
and transverse mode. Ensuring the latter is straightforward: by introducing a suitable aperture it is possible
to ensure that higher-order transverse modes experience suciently high losses that they cannot reach the
threshold for oscillation.
1
Several techniques can be used to force a laser to operate on only a single longitudinal mode.
7.1.1 Short cavity
The simplest approach is to ensure that the longitudinal mode spacing
p,p1
c/2L
c
is larger than the range
of frequencies,
osc
, for which the laser gain exceeds the threshold value. Assuming that one of the longitudinal
modes coincides with the centre of the laser transition, it is easy to show that single-mode operation will occur
if,
L
c
<
c

osc
. (7.1)
7.1.2 Intra-cavity etalon
An intra-cavity etalon may be used to increase the losses experienced by longitudinal modes adjacent to the
mode nearest line centre. The etalon must satisfy two conditions:
1
See Fig. 8.7 of Modern Classical Optics by Geo Brooker, also reproduced as Fig. 6.12 in H&W.
109
110 LECTURE 7. PRECISION FREQUENCY CONTROL
threshold
selected mode

osc
u
n
s
a
t
u
r
a
t
e
d
g
a
i
n

0
selected mode

0

p,p-1

1/2
Fabry-Perot
transmission
longitudinal
modes
gain medium

intra-cavity
etalon
cavity mirror cavity mirror
(a)
(b) (c)

Figure 7.1: SLM operation by using an intra-cavity etalon. (b) Relation between the peaks of the etalon
transmission and the laser lineshape when Condition (2) is satised. (c) Relation between the FWHM of the
etalon and the longitudinal mode spacing when Condition (1) is satised.
1. The width of the etalon peaks must be suciently narrow to discriminate between adjacent longitudinal
modes.
2. The free-spectral range of the etalon must be suciently large that adjacent etalon peaks lie outside the
range of frequencies for which laser oscillation is possible.
Figure 7.1 shows an idealized arrangement in which the orientation of the etalon has been rotated so that
one of its transmission peaks coincides with the longitudinal mode nearest the centre of the laser line.
We may express conditions (1) and (2) as:

1/2
2
<
p,p1
(7.2)

FSR
>

osc
2
, (7.3)
where
1/2
=
FSR
/T is the FWHM of one of the etalon peaks. Using the standard relations for a Fabry-
Perot etalon it is easy to show that the thickness d of the etalon must satisfy,

osc
2
<
c
2nd
< T
c
L
c
, (7.4)
7.2. OUTPUT LINEWIDTH 111
pump laser
Ti:sapphire
Faraday
isolator
birefringent
filter
laser output
piezo-mounted
mirror
Figure 7.2: Single longitudinal mode tunable Ti:sapphire laser employing a uni-directional ring cavity. Wave-
length tuning is achieved by rotating the birefringent lter. One of the cavity mirrors is mounted on a piezo-
electric crystal so that the length of the cavity can be adjusted as the wavelength is tuned, ensuring that the
same longitudinal mode oscillates as the wavelength is varied.
where n is the refractive index of the material between the etalon plates (it is common to use a solid etalon in
this application). Hence we nd that,
L
c
< (2T)
c

osc
. (7.5)
Thus the use of an intra-cavity etalon allows SLM operation to be achieved in a cavity with a length increased
by a factor 2T than with no etalon. It may be shown that for N etalons the maximum length of the cavity can
be increased by a factor of (2T
1
) (2T
2
) (2T
3
) . . . (2T
N
) above that required for no etalon.
7.1.3 Ring resonators
SLM operation on homogeneously broadened laser transitions can be achieved by employing a ring resonator
and ensuring that oscillation only occurs in one direction around the ring; this prevents spatial hole-burning.
Uni-directional oscillation can be achieved by inserting a Faraday isolator within the ring as shown in Fig. 7.2.
7.2 Output linewidth
7.2.1 Theoretical limit
Suppose that SLM oscillation is achieved. What would be the linewidth of the laser output?
Since the laser cavity can be considered to be a Fabry-Perot etalon, we might expect that the linewidth
would be given by the FWHM of an etalon peak, i.e.
1/2
=
FSR
/T. However, this is not the case since
the nite linewidth of the etalon arises from the fact that the amplitude of a wave launched within the etalon
decays as it bounces between the mirrors; as such the wave is only stored within the cavity/etalon for a time
of order the cavity lifetime
c
, leading to a linewidth of order 1/
c
. For an oscillating laser, however, the losses
are balanced by the gain so, in eect, the cavity lifetime becomes innite. This would suggest that the laser
linewidth should be zero.
112 LECTURE 7. PRECISION FREQUENCY CONTROL

Re
Im
E

Re
Im
E

E
0
E
0
(a) (b)
Figure 7.3: Origin of the Schawlow-Townes linewidth. (a) In the absence of spontaneous emission, if a laser
oscillates on a single cavity mode the electric eld may be written as E(t) = E
0
exp(i[
L
t ]) where is an
arbitrary constant phase. This may be represented by a constant vector in an Argand diagram, as shown, where
the axes correspond to 1[E(t) exp(i
L
t)] and [E(t) exp(i
L
t)]. The eect of spontaneous emission is shown in
(b). After each spontaneously emitted photon is emitted into the oscillating mode, the tip of the electric eld
vector moves in a random direction by an amount E
sp
. The dots shown in (b) show the positions occupied by
the tip of the electric eld vector after a very large number of photons have been emitted spontaneously. The
inset in (b) shows the eect on the electric eld vector of the emission of 8 spontaneously emitted photons, and
in particular the net displacement, E

, along a direction,

, orthogonal to the original electric eld vector.
Notice that the electric eld vector of each spontaneously emitted photon has a dierent component along

,
and consequently on average the displacement E

is given by the result of a one-dimensional random walk

(along

) of varying step size. The resultant displacement causes the orientation of the electric eld vector to
rotate through an angle , as shown.
In fact the fundamental limit to the laser linewidth arises from the role of spontaneous emission within the
laser cavity. As we will see, the eect of spontaneous emission is a slow drift of the phase of the laser output,
which leads to a nite frequency width.
Supose that an oscillating mode within the cavity contains n photons. The energy of the mode is proportional
to n, or, equivalently, to E
2
0
where E
0
is the amplitude of the electric eld of the mode. Hence we have the
relation E
0

n.
If the cavity lifetime is
c
then photons are lost from the cavity at a rate n/
c
; these form the output
beam. Since photons produced by stimulated emission are perfectly coherent with each other, in the absence of
spontaneous emission the amplitude and phase of the electric eld at any point in the cavity would be constant
and could be written in the form E(t) = E
0
exp (i [
L
t ]). The amplitude of this eld could be represented
by a constant vector in the complex plane, as shown in Fig. 7.3.
Occasionally a photon will be emitted into the cavity spontaneously. This will have a random phase relative
to that of the stimulated photons, leading to a small shift in the phase (and also the amplitude, but this is small)
of the oscillating mode. Spontaneous emission therefore causes the phase of the mode to change randomly by
small amounts so that the phase of the mode diuses over time with a characteristic time
ST
, corresponding
to a frequency width
ST
= 1/
ST
.
We now estimate
ST
and hence the Schawlow-Townes linewidth. Since E
0

n we can write,
7.2. OUTPUT LINEWIDTH 113
E
sp
E
0
=

n
. (7.6)
If N photons are emitted spontaneously into the mode the change of the phase of the mode is
=
E

E
0
, (7.7)
where E

is the is the component of the vector sum of the electric elds of the N photons along the

direction.
The value of E

depends on the phases of the spontaneously emitted photons. Since these are uncorrelated,
we may use the results of the theory of a one-dimensional random walk of variable step size. These state that
for N steps in the random walk i.e. after the emission of N spontaneous photons the mean amplitude and
mean square amplitude along

are given by:
E

= NE
sp
(7.8)
E
2

= NE
2
sp
, (7.9)
where E
sp
and E
2
sp
are the mean amplitude and mean square amplitude along

of a single spontaneously
emitted photon. Since for spontaneously emitted photons E
sp
is the same, but their phase can take any value
with equal probability, it is easy to show that
E
sp
= 0 (7.10)
E
2
sp
=
E
2
sp
2
, (7.11)
and hence that after N photons have been emitted into the mode:
= 0 (7.12)

2
=
E
2

E
2
0
=
N
2
_
E
sp
E
0
_
2
. (7.13)
If photons are spontaneously emitted into the mode at a rate R
sp
, then after a time the root mean square
value of will be

rms
=
_

2
=
_
R
sp

2 n
, (7.14)
where we have used eqn (7.6).
We can say that the uncertainty in the phase of the mode is signicant once
rms
/2, and hence we may
write,
_
R
sp

ST
2 n


2
,
114 LECTURE 7. PRECISION FREQUENCY CONTROL
from which we deduce,

ST
=
1

ST

_
2

_
2
R
sp
2 n
. (7.15)
To nd the rate R
sp
at which photons are emitted spontaneously into the mode we note that the rate of
spontaneous emission into a mode is equal to 1/ n of the rate of stimulated emission from the upper laser level.
Further, under steady-state conditions the rate of stimulated emission from the upper level must equal n/
c
,
i.e. the rate at which photons leave the cavity. Hence R
sp
= 1/
c
and so,

ST

_
2

_
2
1
2 n
c
. (7.16)
It is convenient to write n in terms of the output power P of the laser. Since the cavity loses photons at a
rate n/
c
, we may write P = ( n/
c
)
L
and hence,

ST

2

L
P
1

2
c
= 8

L
P

2
c
(7.17)

ST

4

h
L

2
c
P
. (7.18)
A more complete derivation shows that,

ST
=
_
N
2
N

_
th
h
L

2
c
P
, (7.19)
where (N
2
/N

)
th
is the ratio of the population density in the upper laser level to the population inversion
density under the operating conditions of the laser. Typically this ratio is close to unity.
7.2.2 Practical limitations
The Schawlow-Townes linewidth is usually extremely small (very often much less than 1 Hz!). In the absence of
active frequency-locking techniques the linewidth of even a single-mode laser will be much greater than
ST
.
A change in cavity length will change the frequency by an amount,

p
=
L
c
L
c
. (7.20)
A wide variety of processes (vibration, thermal eects, changes in ambient pressure etc.) can cause the cavity
length to change, and hence the output frequency to drift or change.
To give just one example, if the coecient of thermal expansion of the material on which the cavity mirrors
is mounted is
T
, then a change in temperature of T gives,

p
=
T
T. (7.21)
For invar (a steel alloy with very low coecient of thermal expansion of
T
10
6
) then to keep
p
below,
say, 1 MHz requires temperature stabilization to better than 2 mK for a laser operating in the visible region of
the spectrum.
7.3. FREQUENCY LOCKING 115
laser
Faraday
isolator
frequency
locking
system
beam
splitter
to expt.
beam
splitter
phase
modulator
absorption
cell
saturating
beam
()
mirror
probe beam
(, ) +_
photodiode

phase
modulator
reference
cavity

photodiode
/4
plate
polarizer
(a)
(b)
(c)
beam
splitter
to expt.
beam
splitter
to expt.
Figure 7.4: (a) Schematic diagram of the arrangement used to frequency lock a laser. (b) Frequency-locking to
a Doppler-free absorption line. (c) Frequency-locking to a reference cavity.
7.3 Frequency locking
Under practical conditions in order to achieve the narrowest linewidths, and to prevent drift, it is necessary to
lock the frequency of a laser to a stable reference frequency. In principle this is straightforward:
1. The frequency of the laser to be locked is compared with some xed frequency
ref
.
2. Any dierence in frequency is converted to an error signal, which increases with the magnitude of the
frequency error and diers in sign according to the sign of (
ref
).
3. The frequency of the oscillating mode is adjusted usually by changing the length of the laser cavity
so as to minimize the magnitude of the error signal, thereby minimizing the frequency error.
This is usually achieved by picking o a small portion of the laser beam and modulating its frequency so
that phase-sensitive detection (PSD) techniques can be employed. Figure 7.4 shows how this is achieved in the
general case, and for the cases of locking to an absorption line or an external cavity.
Form of error signal
We now consider the form of the error signal for the general case of a frequency-modulated wave passing through
(or reecting from) a reference system with a frequency-dependent amplitude response a() which is strongly
peaked in the region of the reference frequency
ref
. The amplitude of the wave leaving a frequency modulator
with a time-dependent amplitude transmission exp(i sin t) may be written as,
116 LECTURE 7. PRECISION FREQUENCY CONTROL
3 2 1 0 1 2 3
1
3 2 1 1 2 3

( -
ref
) / ( -
ref
) /
error
signal
(a) (b)
Figure 7.5: Frequency-locking to a Lorentzian response function centred at
ref
with a full-width at half maxi-
mum . (a) the amplitude response function a(); (b) the amplitude of the error signal S

(t) as a function
of the output frequency of the laser.
E(t) = E
0
exp (it) exp(i
0
)

m=
J
m
() exp(imt), (7.22)
where is proportional to the phase modulation introduced by the modulator. If the amplitudes of the sidebands
other than m = 1 may be ignored, we have
E(t) = E
0
exp(i
0
) J
1
() exp [i( + )t] +J
0
() exp (it) +J
1
() exp [i( )t] , (7.23)
where we have used the fact that J
1
() = J
1
(). After passing through the reference system the amplitude
of the radiation is
E(t) = E
0
exp(i
0
) J
1
()a( + ) exp [i( + )t] +J
0
()a() exp (it) +J
1
()a( ) exp [i( )t] ,
(7.24)
It may be shown that for a square-law detector, i.e. one which detects a signal S(t) E(t)E (t), the signal at
the modulation frequency is given by,
S

(t) [a()[
[a()[

cos t. (7.25)
As shown in Fig. 7.5 the error signal is proportional to the derivative of [a()[, and hence the sign of the error
signal depends on whether is greater or less than the reference frequency
ref
.
7.3.1 Locking to atomic or molecular transitions
Locking to atomic or molecular transitions provides the best long-term stability and gives highly reproducible
results: the frequency reproducibility is typically found to be 1 part in 10
11
.
7.3. FREQUENCY LOCKING 117
( -
0
) /
D
0.1
0.0
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
-2 -1 0 1 2
T
r
a
n
s
m
i
s
s
i
o
n

H
Figure 7.6: Calculated frequency dependence of the transmission of a weak probe through a sample of Doppler
broadened atoms with (solid line) and without (dashed line) a counter-propagating saturating beam with the
same frequency as that of the probe beam. Note that the main absorption has a frequency width of order the
Doppler width
D
, but the spike of reduced absorption has a width approximately equal to the homogeneous
width of the atoms,
H
. For the purposes of illustration, in this calculation
H
was set equal to 0.1
D
; in
practice the ratio of the homogeneous and to inhomogeneous linewidths would usually be very much smaller.
The most convenient transitions suitable as reference frequencies occur in atoms or molecules in the gaseous
state, and consequently it is necessary to use Doppler-free techniques to eliminate Doppler broadening, as
illustrated in Fig. 7.4(b). In this approach the laser beam is divided into a strong saturating beam and a
weak probe beam. The saturating and probe beams pass through the absorption cell in an exact, or near,
counter-propagating geometry, and the intensity of the transmitted probe beam is measured by a photodiode.
This is the same arrangement used in saturation absorption spectroscopy.
In order to determine the form of the error signal we must nd the frequency response a() of the reference
system, which in this case is the frequency dependence of the amplitude transmission of the probe beam.
Now, in the absence of the saturating beam the transmitted amplitude will be large for frequencies such that
[
0
[
D
, and it will be small for detunings [
0
[ comparable to or small compared to
D
. This
behaviour is shown in Fig. 7.6.
Since they propagate in opposite directions, the saturating beam and the probe beam usually interact with
dierent velocity classes. As such, for frequencies such that [
0
[
H
the absorption of the probe beam
is unaected by the presence of the saturating beam. However, for frequencies
0

H
, both beams
interact with the same (v
z
0) class of atoms and hence the absorption experienced by the probe beam is
reduced. This leads to a very narrow feature in the transmission of the probe beam, as shown in Fig. 7.6 which
can be used to lock the frequency to
0
.
7.3.2 Locking to an external cavity
Locking to the resonant frequency of an external cavity provides better short-term stability than locking to an
atomic transition since the error signals are large and hence have a high signal-to-noise ratio. There are many
ways in which this can be done, but the most widely used is the Pound-Drever-Hall technique illustrated
schematically in Fig. 7.4(c).
118 LECTURE 7. PRECISION FREQUENCY CONTROL

m

(m+1)
|
s
(

)
|

a
n
d

A
r
g
[
s
(

)
]
0
-0.2
-0.4
-0.6
-0.8
-1.0
-1.2
-1.4
-1.6
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6

0
-1
-2
1
2
-0.2 -0.1 0 0.1 0.2
( -
m
)/
FSR
E
r
r
o
r

s
i
g
n
a
l

capture

FSR

lock
Figure 7.7: The Pound-Drever-Hall technique. (a) shows the amplitude (solid line) and phase (dashed line) of
the beam reected from a cavity as a function of frequency, where
m
and
(m+1)
are two adjacent resonant
frequencies of the cavity. (b) shows the error signal as a function of frequency near one of the resonant
frequencies. Notice that the capture range of the error signal is a signicant fraction of the free spectral range,

FSR
, of the cavity, but that the error signal has a slope of very large magnitude near the resonant frequency.
The Pound-Drever-Hall technique utilizes the frequency-dependent reection from a high-nesse cavity.
2
The advantage of using the signal reected from the cavity is that the error signal has a much wider capture
range than that which would be obtained by monitoring the transmitted beam.
To see this we note that the cavity acts as a Fabry-Perot etalon with a transmission which is sharply peaked.
As shown in Fig. 7.5(b) when used in transmission the resulting error signal would only extend over a frequency
range of order the cavity linewidth
c
, and this must be narrow for accurate locking. This behaviour may
also be considered in the time domain; a narrow linewidth cavity corresponds to a long cavity lifetime
c
and consequently the transmitted beam does not respond to uctuations in the beam which occur in a time
signicantly shorter than
c
, or, equivalently, to frequency shifts greater than of order 1/
c

c
.
The beam reected from a reference cavity behaves dierently since it is comprised of a component reected
directly from the rst mirror plus leakage of radiation circulating within the cavity. The response function is
more complex, and leads to an error signal which has a very wide capture range, yet a narrow central feature
which can be used to lock the frequency to a tight tolerance. Fig. 7.7(a) shows the amplitude and phase of the
2
In this application cavity nesses up to 2 10
5
may be obtained by using super-polished mirrors.
7.3. FREQUENCY LOCKING 119
signal reected from a reference cavity. It may be seen that the reected amplitude is approximately unity apart
from sharp dips in the region of the resonance frequencies of the cavity; this behaviour is to be expected from
our understanding of the transmission of a Fabry-Perot etalon. Fig. 7.7(b) shows the error signal derived from
a beam modulated at a frequency . The error signal is large, and has the correct sign, for frequencies within
approximately of the resonance frequency. At the same time the central feature has a much smaller width,
of order
1/2
=
FSR
/T, where T is the nesse of the reference cavity, allowing narrow frequency-locked
linewidths to be obtained. This approach can be used to achieve extremely narrow linewidths if the reference
cavity is isolated against mechanical and thermal uctuations and drift.