Galileo Galilei School of Graduate Studies

Applied Physics Course

FIS/01

XVII Ciclo 2002–2004

Ph.D. Thesis

Modeling some Sources of

Thermal Noise through Finite
Element Analysis: Application
to the Virgo Interferometric
Antenna.

Candidate Advisor

Enrico Campagna Flavio Vetrano

Contents

1 Gravitational Waves 3
1.1 Propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.1.1 Theory of General Relativity . . . . . . . . . . . . . . . 4
1.1.2 The weak-field approximation . . . . . . . . . . . . . . 5
1.1.3 Effects on free-falling particles . . . . . . . . . . . . . . 6
1.2 Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 Radiation formula . . . . . . . . . . . . . . . . . . . . . 8
1.2.2 Sources . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3.1 Resonant detectors . . . . . . . . . . . . . . . . . . . . 10
1.3.2 Interferometric detectors . . . . . . . . . . . . . . . . . 11

2 Noise sources in interferometric detector 17

2.1 Optical read-out noises . . . . . . . . . . . . . . . . . . . . . . 17
2.1.1 Shot noise . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.1.2 Radiation pressure noise . . . . . . . . . . . . . . . . . 19
2.2 Thermal noise . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.3 Environmental noises . . . . . . . . . . . . . . . . . . . . . . . 21

I
II CONTENTS

2.3.1 Seismic noise . . . . . . . . . . . . . . . . . . . . . . . 21

2.3.2 Gravity gradient noise . . . . . . . . . . . . . . . . . . 22
2.3.3 Residual gas noise . . . . . . . . . . . . . . . . . . . . . 23
2.4 Other noises . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.4.1 Laser frequency noise . . . . . . . . . . . . . . . . . . . 24
2.4.2 Laser intensity noise . . . . . . . . . . . . . . . . . . . 24
2.4.3 Thermal lensing . . . . . . . . . . . . . . . . . . . . . . 24
2.4.4 Electric noise . . . . . . . . . . . . . . . . . . . . . . . 25
2.4.5 Scattered light . . . . . . . . . . . . . . . . . . . . . . 25
2.5 An example: the Virgo interferometer . . . . . . . . . . . . . . 26
2.5.1 A Virgo overview . . . . . . . . . . . . . . . . . . . . . 26
2.5.2 Virgo sensitivity curve . . . . . . . . . . . . . . . . . . 30

3 Theory of thermal noise 33

3.1 The fluctuation-dissipation theorem . . . . . . . . . . . . . . . 33
3.1.1 FDT in an n-dimensional system . . . . . . . . . . . . 35
3.1.2 FDT for a linear combination of coordinates . . . . . . 36
3.1.3 The harmonic oscillator and the loss angle . . . . . . . 38
3.2 Dissipation models . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.1 Structural damping . . . . . . . . . . . . . . . . . . . . 42
3.2.2 Thermoelastic damping . . . . . . . . . . . . . . . . . . 43
3.2.3 Superficial losses . . . . . . . . . . . . . . . . . . . . . 45
3.2.4 Recoil losses . . . . . . . . . . . . . . . . . . . . . . . . 46
3.2.5 Air losses . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.2.6 Dislocation propagation . . . . . . . . . . . . . . . . . 49
3.2.7 Impedance coupling . . . . . . . . . . . . . . . . . . . . 50
3.3 Thermal noise calculation methods . . . . . . . . . . . . . . . 50
3.3.1 Normal mode expansion . . . . . . . . . . . . . . . . . 51
CONTENTS III

3.3.2 Advanced mode expansion . . . . . . . . . . . . . . . . 54

3.3.3 Direct approach . . . . . . . . . . . . . . . . . . . . . . 56

4 Thermal noise of interferometers 61

4.1 Mirror-bulk thermal noise . . . . . . . . . . . . . . . . . . . . 62
4.1.1 Brownian thermal noise . . . . . . . . . . . . . . . . . 62
4.1.2 Thermoelastic noise . . . . . . . . . . . . . . . . . . . . 66
4.2 Mirror-coating thermal noise . . . . . . . . . . . . . . . . . . . 69
4.2.1 Brownian thermal noise . . . . . . . . . . . . . . . . . 69
4.2.2 Thermoelastic noise . . . . . . . . . . . . . . . . . . . . 71
4.2.3 Thermorefractive noise . . . . . . . . . . . . . . . . . . 73
4.3 Suspension thermal noise . . . . . . . . . . . . . . . . . . . . . 74
4.3.1 Pendulum thermal noise . . . . . . . . . . . . . . . . . 74
4.3.2 Vertical mode thermal noise . . . . . . . . . . . . . . . 76
4.3.3 Violin modes thermal noise . . . . . . . . . . . . . . . 77
4.3.4 Tilt and rotational modes thermal noise . . . . . . . . 78
4.4 The problem of new materials for Virgo . . . . . . . . . . . . . 78
4.4.1 Lowering the shot noise . . . . . . . . . . . . . . . . . 78
4.4.2 Lowering mirror-bulk thermal noise . . . . . . . . . . . 79
4.4.3 Lowering mirror-coating thermal noise . . . . . . . . . 82
4.4.4 Lowering suspension thermal noise . . . . . . . . . . . 85
4.4.5 Guidelines for new materials . . . . . . . . . . . . . . . 87

5 Using finite element analysis 91

5.1 The basis of finite element analysis . . . . . . . . . . . . . . . 91
5.1.1 The principle of virtual displacements . . . . . . . . . . 93
5.1.2 The elements . . . . . . . . . . . . . . . . . . . . . . . 94
5.1.3 The mesh . . . . . . . . . . . . . . . . . . . . . . . . . 96
IV CONTENTS

5.1.4 Checking the FEA results . . . . . . . . . . . . . . . . 97

5.2 FEA as a general tool in thermal noise studies . . . . . . . . . 99
5.2.1 Resonance modes extraction . . . . . . . . . . . . . . . 100
5.2.2 Shape parameter extraction . . . . . . . . . . . . . . . 101
5.2.3 Brownian thermal noise in complex systems . . . . . . 102
5.2.4 Parametric analyses . . . . . . . . . . . . . . . . . . . . 102
5.2.5 Analysis on thermoelastic noise . . . . . . . . . . . . . 103
5.3 Finite element model of a Virgo mirror . . . . . . . . . . . . . 103
5.3.1 The bulk . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.3.2 The coating . . . . . . . . . . . . . . . . . . . . . . . . 104
5.3.3 Magnets, markers and spacers . . . . . . . . . . . . . . 106
5.4 Application of FEA to a Virgo mirror . . . . . . . . . . . . . . 107
5.4.1 Resonance modes extraction . . . . . . . . . . . . . . . 108
5.4.2 Shape parameter extraction . . . . . . . . . . . . . . . 114
5.4.3 Brownian thermal noise calculation . . . . . . . . . . . 115
5.4.4 The effect of the markers . . . . . . . . . . . . . . . . . 120

6 Research on new materials for suspensions 123

6.1 Silicon properties . . . . . . . . . . . . . . . . . . . . . . . . . 123
6.2 Silicon fibres production . . . . . . . . . . . . . . . . . . . . . 127
6.2.1 Growth process . . . . . . . . . . . . . . . . . . . . . . 129
6.2.2 Seeding and growth procedure . . . . . . . . . . . . . . 129
6.3 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . 132
6.3.1 Loss angle measurement apparatus . . . . . . . . . . . 132
6.3.2 Profile measurement apparatus . . . . . . . . . . . . . 141
6.4 Measurement of the thermoelastic peak . . . . . . . . . . . . . 143
6.4.1 Standard measurement procedure . . . . . . . . . . . . 143
6.4.2 The etching process . . . . . . . . . . . . . . . . . . . . 146
CONTENTS V

6.4.3 Modeling the fibre . . . . . . . . . . . . . . . . . . . . 150

6.4.4 Phi measurements . . . . . . . . . . . . . . . . . . . . 157
6.5 Extraction of silicon thermo-mechanical properties . . . . . . . 160
6.5.1 Estimating the errors of the profilation . . . . . . . . . 161
6.5.2 Young’s modulus . . . . . . . . . . . . . . . . . . . . . 162
6.5.3 The coefficient of linear thermal expansion . . . . . . . 165
6.5.4 The coefficient of thermal conduction . . . . . . . . . . 165
6.6 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

7 Conclusions 171
VI CONTENTS
Introduction

Gravitational waves are a prediction of general relativity but have not yet
been directly detected.
Ground based interferometric antennae like Virgo, LIGO or GEO600 aim
at doing it in the next years.
Virgo sensitivity extends from a few Hz to a few kHz and is limited at low
frequency (around 5 Hz) by seismic and gravity gradient noise and at high
frequencies (over 500Hz) by shot noise. In the intermediate frequency range
there are two noise floors, due to pendulum thermal noise (below 100 Hz)
and to mirror thermal noise (over 100 Hz). Therefore, the comprehension
and the modelization of the thermal noise in its various aspects is of crucial
importance for Virgo and for any interferometer of the 2nd or 3rd generation
too.
The main problem is that the real components of the suspension system
are more complex than the ideal ones used for thermal noise analytical calcu-
lations: the symmetry of the problem can be very different from the one used
in the theoretical models. Therefore, at the level of sensitivity involved, even
small geometric deviations are important. It is necessary to have a flexible
tool for investigating thermal noise effects with the actual designed mirror
and mirror suspensions: finite element analysis.
Secondly, especially in the context of R&D activities, looking at cryogenic
interferometric detectors as a new workable solution, it is interesting to study

1
2 CONTENTS

new materials, characterizing ex novo their thermomechanical properties.

Silicon is a very promising candidate and it is worthwhile to build up an
experiment to test crystalline silicon fibers to be used as future suspensions.
Chapter 1

Gravitational Waves

Since the theory of gravitation was stated, many efforts have been made to
detect gravitational wave radiation. Gravitational waves come directly as a
solution of Einstein field equations [1] in the weak field approximation and
they are a spin-2 metric distortion that propagates at the speed of light.
Their emission takes place during catastrophic cosmic events that involve
large masses rapidly changing their distribution — e.g. coalescing binary
systems, asymmetric pulsars or the Big bang itself.
There are many experiments around the world that are active at the
moment but none of them has detected anything yet. Gravitational wave
emission and propagation processes are well theorised, and with the study on
the revolution period of the binary pulsar system PSR1913+16 the existence
of gravitational radiation was indirectly proved [2]. The problem of direct
detection lies on the extremely small coupling of the gravitational field. For
this reason all the experiments implemented for their detection — from the
resonant bar to the interferometer antennas — have to face great technical
problems to reach the needed sensitivities.

3
4 Chapter 1. Gravitational Waves

1.1 Propagation
1.1.1 Theory of General Relativity

According to the theory of general relativity space and time are merged to-
gether in a 4-dimensional manifold. The presence of masses on this man-
ifold causes its distortion and, on the other hand, the distortion of the
4-dimensional space governs the dynamics of the masses on it. This is the
kernel concept that is embedded in the Einstein field equations that can be
written as
Gµν = 8πGTµν , (1.1)

where Gµν is the Einstein tensor and Tµν is the energy-momentum tensor, rep-
resenting the distribution of mass and energy in the space-time. To identify
the Einstein tensor other basic definitions are needed: first of all the metric
tensor gµν , which defines the distance ds of two slightly separate points

ds2 = gµν dxµ dxν (1.2)

and it’s used to raise and lower the indexes of the other tensors; then the
affine connection Γµνλ
1
Γµνλ + g µα (gαν,λ + gαλ,ν − gνλ,α ) (1.3)
2
and the Riemann tensor defined using the affine connection

Rµναβ + Γµνβ,α − Γµνα,β + Γµγα Γγνβ − Γµγβ Γγνα . (1.4)

Introducing the Ricci tensor and the Ricci scalar

Rµν + Rαµαν , (1.5)

R + Rαα , (1.6)

the Einstein tensor is defined as

1
Gµν + Rµν − gµν R . (1.7)
2
1.1 Propagation 5

1.1.2 The weak-field approximation

Owing to the non-linearity of Einstein field equations, it’s very hard to find
general radiative solutions to them. In fact, any solution carries energy and
momentum that modify the second member of the equation itself. One ap-
proach can be to study the weak-field radiative solution which describes waves
carrying not enough energy and momentum to affect their own propagation.
This seems reasonable because any observable gravitational radiation is likely
to be of very low intensity. Suppose to have a metric gµν differing from the
Minkowskian one ηµν by a little quantity hµν so that

gµν = ηµν + hµν , (1.8)

with |hµν |  1; using ηµν to raise and lower all indexes and writing Ricci
tensor and affine connection up to the first order in h, the Einstein field
equations read

∂ λ ∂2 ∂
hµν − λ µ
h ν − λ ν
hλµ + µ ν hλλ = −16πGSµν , (1.9)
∂x ∂x ∂x ∂x ∂x ∂x

with
1
Sµν + Tµν − ηµν T λλ . (1.10)
2
Choosing the so called harmonic coordinate defined by the gauge

g µν Γλµν = 0 , (1.11)

that, to the first order in h corresponds to

∂ µ 1 ∂ µ
h ν = h , (1.12)
∂xµ 2 ∂xν µ

the field equations now read

hµν = −16πGSµν (1.13)

6 Chapter 1. Gravitational Waves

leading to the retarded potential solutions:

Sµν (~x0 , t − |~x − ~x0 |)
Z
hµν (~x, t) = 4G d3~x0 . (1.14)
|~x − ~x0 |
This solution describes a gravitational radiation, produced by the source Sµν ,
that propagates with unit velocity, i.e. the velocity of light. As r → ∞ the
retarded wave approaches a plane wave of the form
λ
hµν (x) = eµν eikλ x + c.c. , (1.15)

satisfying
kµ k µ = 0 , (1.16)

coming from (1.13) with S µν= 0, and

1
kµ eµν = kν eµµ , (1.17)
2
coming from (1.12). The symmetric matrix eµν is the polarization tensor.
For a monochromatic wave, propagating along the z axis (1.15) reduces to
 
0 0 0 0
0 h h 0
 
λ
hµν =  + ×  eikλ x , (1.18)
0 h× −h+ 0
0 0 0 0
where it is evident that there are only two independent components usually
called plus and cross polarization.

1.1.3 Effects on free-falling particles

What happens during the passage of a gravitational wave on the matter? If

we consider a particle at rest in the coordinate frame of the harmonic gauge,
using the geodesic equation it comes out that the particle is not subject to
any acceleration. So we have to consider the relative motion of two particles
and, consequently, we have to use the equation of geodesic deviation
d2 nµ µ dxα dxβ γ
− R αβγ n =0, (1.19)
dτ 2 dτ dτ
1.1 Propagation 7

where nµ is the vector connecting the two masses and xµ the position of
one of them. Considering only the first order in hµν and using the fact that
the masses will move with a velocity much smaller than the speed of light,
equation (1.19) brings to a solution

 1 
nµ (t) = nα (0) δαµ + hαµ , (1.20)
2

where we took the two particles at rest at t = 0. It is evident that the

relative displacement of the two masses oscillates periodically with the same
frequency of the overpassing gravitational wave. The effect is directly pro-
portional to the distance of the particles and to the amplitude of the wave.
In figure 1.1 the effect induced by a gravitational wave on a ring of particles
is shown either for a plus polarization, or for a cross one.

plus polarization

cross polarization

Figure 1.1: Effect on a ring of particles posed on the x-y plane due to the
passage of a gravitational wave coming along z axis with a plus polarization
or a cross polarization. It is shown that the effect on two chosen orthogonal
directions is strongly depending on the polarization of the wave.
8 Chapter 1. Gravitational Waves

1.2 Generation
1.2.1 Radiation formula

From rather complicated tensor calculations, in the approximation of slow

moving masses, it is possible to calculate the gravitational-wave power L
expressed in terms of the reduced quadrupole moment of the source Ijk . The
final result is
1 ... ...
L = h I jk I jk i , (1.21)
5
where the average is intended over several characteristic periods of the source
system and Ijk is defined as
Z  1 
Ijk + ρ xi xj − δij r2 d3 x . (1.22)
3
The gravitational wave radiation is a quadrupole radiation and so it is
impossible for spherically or axially symmetric sources to emit anything.

1.2.2 Sources

There are essentially three kinds of gravitational wave sources: burst sources,
periodic sources and stochastic sources.

Impulsive sources

These are sources that emit an impulsive signal in a very short time.
The main sources of this kind are supernovae, especially the supernovae
of type II, that are exploding in a non axial-symmetric way. For such an
event, called burst, at 10 kpc, the amplitude has been estimated to be of
the order of 10−20 at hundreds or thousands of Hz. A detection can give
information about explosion itself and also about the process of neutron-star
and black-hole birth.
Compact binaries in their last instants before coalescence, are supposed
to be impulsive sources too. For neutron-star neutron-star (NS-NS) binaries
1.2 Generation 9

the signal is expected to be at a few thousand Hz with h ∼ 10−21 at 20 Mpc.

Gravitational waves from NS-NS binaries carry important information about
the equation of state of the NS, the physics of the emission and the distance
from the source. If the binary is a black-hole black-hole (BH-BH) system the
peak of emission, for a source in our galactic halo, is expected to be at about
500 Hz with h ∼ 10−18 [3, 4]. Unfortunately, at the present time, no systems
of this kind are known.
For a very massive BH couple the gravitational wave emission moves down
in frequency to 0.01 mHz, far from the sensitivity band of ground based in-
terferometers and the signal cannot be considered impulsive anymore lasting
even thousands of seconds.
Among the impulsive sources one should consider also the BH formation
and the falling of matter on a massive BH. These events are expected to
happen in the center of galaxies in a frequency band from 1 mHz to 10 mHz.

Periodic sources

These systems, for their intrinsic constitution, are emitting gravitational ra-
diation at a well defined frequency for a consistent amount of time.
This is the case of compact binaries. The emission of gravitational waves
from such a system has been proved in an indirect way measuring the decay
of the revolution period of the two rotating stars [2]. In the first part of their
rotation the signal is periodic and of low intensity (h ∼ 10−22 ). The final part
of their rotation cannot be considered periodic anymore: we are in presence
of the so called chirp, that is classified under the burst sources.
Another group of periodic sources comes from pulsars among which there
are two kinds of candidates for gravitational waves emission: pulsars with a
quadrupolar momentum, i.e. pulsars that are slightly asymmetric, emitting at
a frequency that is twice the rotational one, and pulsars with no quadrupolar
10 Chapter 1. Gravitational Waves

momentum, but rotating along an axis that is not the axis of symmetry; in
this case the emission frequency is the rotational one. The amplitude for
these signals is of the order of 10−27÷10−24 .
Despite the little amplitude, these signals can overcome the detection
background thanks to their periodicity, which permits a long integration
time.

Stochastic sources

The origin of stochastic gravitational waves can simply be the summation

over a large amount of bursts or periodic sources.
But during the evolution of the universe various phenomena originating
stochastic signals could have taken place. This can be the case of the gravi-
tational radiation as an echo of the moment in which the gravitational waves
decoupled from the other elementary particles (this happened at about the
Plank time from the beginning of the universe)1 .
Other sources could be related to the consequence of the eventual phase tran-
sitions occurred during the evolution of the universe. One can imagine exotic
events like collisions of bubbles, collapses of cosmic strings or pressure fluc-
tuations as a cause of stochastic gravitational waves in the frequency region
below 1 mHz.

1.3 Detection
1.3.1 Resonant detectors

The first kind of gravitational wave detector developed in the last decades
are the resonant bars. There are many experiments of this kind all over the
world: the Italian AURIGA (Padova) [5], EXPLORER (CERN, Geneva) [6]
1
If the inflation occurred this radiation should be replaced by the gravitational waves
caused by quantum fluctuation.
1.3 Detection 11

and NAUTILUS (LNF, Rome) [7]; ALLEGRO (USA) [8].The detection sys-
tem can be schematized as two masses m at a distance L interacting through
a spring of elastic constant k. If a gravitational wave passes through the
system, the masses start oscillating, modulating their distance. The bar de-
tectors are long isolated cylinders — or spheres — made of a material with a
low viscous coefficient and free to oscillate in the longitudinal direction. The
role of the restoring force is played by the intrinsic elasticity of the material.
The measurement of bar-length variations is done through capacitive trans-
ductors. The response of the system is the best at the resonance frequency
of the bar, where the signal is maximally amplified. A gravitational signal
at this frequency is also stored for a long time: for this reason bar detectors
are considered systems with memory. The main limitation of this kind of
detector is the narrow frequency-window in which the bar is highly sensitive.
For frequencies out of the resonance band the signal induced on the bar is
almost null. The sensitivity is limited by the thermal elastic vibration and
by the noise of the transducers. For these reasons the detectors are made of
low loss material and are cooled to cryogenic temperatures.

1.3.2 Interferometric detectors

A very promising technique to reveal gravitational waves involves the use of

large Michelson-Morley interferometers. There are currently 4 experiments
based on this idea: LIGO [9] (USA, with two 4 km long interferometers),
VIRGO [10] (Italy and France, 3 km), GEO600 [11] (United Kingdom and
Germany, 600 m), TAMA [12] (Japan, 300 m). The Michelson-Morley inter-
ferometer is a device that measures with high precision distance differences
along two chosen directions using the time travel of photons. A laser beam
passes through a beam splitter generating two twin beams each one traveling
back and forth along an arm thanks to the reflection on end mirrors. The
12 Chapter 1. Gravitational Waves

two beams recombine at the beam splitter and the signal is analyzed through
a photo detector. If a gravitational wave propagating along z with a plus
polarization passes through the interferometer that has its arms along the x
and y directions, the proper distance ds2 is given by

ds2 = −c2 dt2 + (1 + h(t))dx2 + (1 − h(t))dy 2 + dz 2 . (1.23)

For a ray propagating along the x direction, dy = dz = 0 and ds2 = 0, so it

follows
dx c
= ±p , (1.24)
dt 1 + h(t)
with the plus and minus signs referring to the two directions of propagation
along the arm. The mirrors in the x and y directions can be considered
as free falling, so, doing the calculation in the TT gauge, their position is
constant in time. Assuming that the light enters the arm of length lx at the
time t1 and arrives on the beam splitter at the time t0 , integrating equation
(1.23), it follows
t0
dt0
Z
2lx
p = . (1.25)
1 + h(t0 )
t1 c
In the first term of the equation it is possible to use the approximation of
h  1,
t0
1 t0
Z Z
 1 0  0
1 − h(t ) dt ' (t0 − t1 ) − h(t0 )dt0 . (1.26)
t1 2 2 t0 − cx
2l

Calling Ω the angular frequency of the light, the round-trip phase φx (t) is
2lx 1 t0
 Z 
0 0
φx (t) = Ωt1 = Ω t0 − − h(t )dt . (1.27)
c 2 t0 − 2lcx
From an analogous argument, considering the light traveling along the y axis,
2ly 1 t0
 Z 
0 0
φy (t) = Ωt2 = Ω t0 − + h(t )dt , (1.28)
c 2 t0 − 2lcy
where t2 is the entering time of the light in the y arm and ly is its length.
So, the phase difference of the two beams is:
2Ω(lx − ly )
∆φ(t) + φx (t) − φy (t) = − ∆φGW (t) , (1.29)
c
1.3 Detection 13

where
Z t0
∆φGW (t) = Ω h(t0 )dt0 (1.30)
t0 − 2l
c

and it has been used lx ' ly = l. The transfer function from the incoming
gravitational wave to the phase change of the Michelson interferometer is
then
2Ω  lω 
H(ω) = sin e−ilω/c , (1.31)
ω c

revealing that there is an optimal length lopt depending on the angular fre-
quency ωGW of the gravitational wave of interest
 
πc 300 Hz
lopt = = 250 km . (1.32)
2ωGW f

Delay line and Fabry-Perot cavity

From the equation (1.32) it is evident that an experiment that aims to reveal
gravitational waves at a few hundred Hz needs an interferometer with a
few hundred km arms! There are problems of cost and construction that
limits the length of the arms to few kilometers. But there are two methods
to enhance the effective optical length: delay-lines and Fabry-Perot cavities.
In the first case the light path is folded in the arms through a multiple
reflection between the two faced mirrors. The actual path is L= nl with l
the length of the arm and n the number of reflections that occur in the arm
before the light comes back to the beam splitter. Unfortunately there are
great difficulties in constructing an arm with n >10. Another drawback is
in the response function of the arm: owing to the delay-line configuration, it
presents a lot of zeros corresponding to the multiple reflections occurring to
the beam. In the second case Michelson arms are two Fabry-Perot cavities.
The parameter that defines the features of a Fabry-Perot cavity is the finesse
F defined as the ratio between the distance of two resonant peaks and the
14 Chapter 1. Gravitational Waves

width at half height of the resonant peak for the transmission coefficient:
√
π r1 r2
F = , (1.33)
1 − r 1 r2
where r1 and r2 are the reflection coefficients of the input and of the end
mirror of the cavity. The effect of such a cavity is to “store” the light,
simulating an enhancement of the optical path in the interferometer arms.
It is possible to define a storage time τs as the time required for the reflected
light to decrease its power of a factor 1/e once the laser is turned off, that is

L r1 LF
τs = ' . (1.34)
c 1 − r1 πc
A drawback of the Fabry-Perot cavity is that the heating of the laser on
the mirrors starts to be relevant and can cause distortion of the surface. It
is possible to calculate that the phase shift produced in the laser from a
displacement variation δL between the two mirrors is

∂φ 4r1
= k, (1.35)
∂L 1 − r1
with k=2π/λ and λ is the wave length of the laser light. For a displacement
∆L= 2Lh due to a gravitational wave, the power modulation at the output
of the interferometer is
   
2 4r1 2 4r1
Pout = Pin cos k∆L = Pin cos 2kLh . (1.36)
1 − r1 1 − r1
Comparing this equation with the one of a classic Michelson it is evident
4r1
that the visible phase shift is amplified of a factor 1−r1
. To maximize the
contrast2 , the working point for the interferometer is chosen by satisfying

π
k∆L = + nπ (n ∈ N) “dark fringe” , (1.38)
2
2
The contrast is defined as:
Pmax − Pmin
C= (1.37)
Pmax + Pmin
where Pmax and Pmin are the maximum and the minimum power read from the output
photodiode when the signal phase changes from 0 to 2π.
1.3 Detection 15

but in this condition the sensitivity of the interferometer is at its minimum:

∂Pout
Pout = Pin cos2 (k∆L) = 0 and =0. (1.39)
∂∆L

To avoid this situation a heterodyne phase modulation is used: through a

∂Pout
small asymmetry in the length of the two arms it permits to have ∂∆L
6= 0.
As for the detection a technique that goes under the name of Pound-Drever
technique is implemented. The phase of the incoming light in the interferom-
eter is modulated through a Pockel-cell, an electro-optic transducer whose
refraction index varies proportionally to the potential ∆V . This technique
allows to isolate the gravitational signal from any kind of noise external to
the interferometric cavity, translating the information in higher frequency
regions (usually radio frequency). The acquired signal is non-null only in
presence of real variation of the optical path of the light due to gravitational
signal or real displacement of the optic elements. For this reason this kind
of detectors are referred to as “null instruments”.
16
Chapter 2

Noise sources in interferometric

detector

In this chapter the main important noise-sources limiting the sensitivity of

a gravitational wave interferometer will be discussed in the frequency band
of interest. Throughout this thesis the strain sensitivity h will be used,
√ √
defined as h = L2 SX (in units 1/ Hz), where SX is the displacement power
spectrum of the quantity considered and L is the length of one interferometer
arm.

2.1 Optical read-out noises

2.1.1 Shot noise

In the previous chapter it was explained that the detection of gravitational

waves with interferometers is possible through the measurement of the power
variation at the output port. This corresponds to the count of the number
of photons that arrive at the photo-diode in the unit of time, because
~c
Pout = n , (2.1)
λ
with n medium number of photons coming on the photo-diode in the unit of
time, ~ Planck’s constant, c velocity of light and λ wave length of the laser.
Such a counting is governed by the Poisson statistic; the fluctuation σn of

17
18 Chapter 2. Noise sources in interferometric detector

the mean number is

σn 1
=√ . (2.2)
n n

For a Michelson interferometer, in the working point in which Pout = Pin /2,
this happens to be a simulated variation of the arm-length difference [15]

 −1 r
σn dPout ~cλ
δ∆L = Pout = , (2.3)
n d∆L 4πPin τ

where τ is the observation time interval. This phenomenon is called shot

noise and, for a Fabry-Perot interferometer in the dark fringe working point,
it takes the following form [16]:

s s  2
χ 4πc~λ f
δ h̃Shot (f ) = 1+ , (2.4)
8LF ηCPin fFP

where δhShot is the strain sensitivity contribution due to the shot noise,
cπ
fFP= 2LF is the frequency corresponding to the characteristic wave length
of the Fabry-Perot cavity of finesse F , C is the recycling factor, η is the
quantum-efficiency of the output photo-diode and χ is a corrective factor
taking into account the modulation1 .
The presence of a recycling factor C is due the positioning of a recycling
mirror, between the mode cleaner and the beam splitter (refer to figure 2.1 on
page 26). Since the interferometer is close to a null in the interference pattern,
nearly all of the light supplied would be reflected back towards the laser;
in other words the laser would not be properly impedance matched to the
interferometer. The recycling mirror, with a correctly chosen transmission,
improves the impedance matching allowing to reach the kilovolts needed on
the beam splitter from the tens of volts coming from the laser.

1
p
Bondu estimated [17] that for Virgo χ = 3/2.
2.1 Optical read-out noises 19

For high and low frequency regions w.r.t. fFP the shot noise reduces to:
 s
 χ 4πc~λ 1
 8LF ηCP ∝ F √CP , for f  fFP ,



in in
δ h̃Shot (f ) = s (2.5)
 χf 4π~λ f
∝√ for f  fFP ,



 2 ,
cηCPin CPin

showing, for low frequency, a constant behavior followed, over the Fabry-Perot
frequency, by a slope proportional to the frequency. Therefore this noise is
the dominant one at high frequencies. It is evident that going at high finesse
and power laser on the beam splitter (high CPin ) can be a solution to limit
this noise. But there are technical limitations that put constraints on F and
C. In fact introducing the optical loss l and the transmittance of the inner
mirror t one has:
2π
F 6 (2.6)
t + 2l

and
1
C6 , (2.7)
2neff l + lBS

where neff is the effective number of bounces of the laser in the Fabry-Perot
cavity (neff = 2F/π) and lBS is the loss of the beam splitter.

2.1.2 Radiation pressure noise

The mirror is pushed by the back-action of the reflected photons, whose

number is fluctuating according to the photon statistics. The radiation pres-
sure noise is the noise related to this uncertainty of the mirror position on a
macroscopic scale. In the case of a Fabry-Perot cavity, the radiation pressure
induced displacement is [18]
r
F ~CPin 1
δ h̃Rad = p . (2.8)
mLπ 2 f 2 πcλ 1 + (f /fFP )2
20 Chapter 2. Noise sources in interferometric detector

For high and low frequency regions w.r.t. fFP the radiation pressure noise
reduces to
 r √
F ~CPin F CPin
∝ , for f  fFP ,


2 2 2

δ h̃Rad (f ) = mLπ f rπcλ √ f (2.9)
 1 c~CPin CPin

 ∝ , for f  fFP ,
4mL2 π 2 f 3 πλ f3
showing a f −2 behavior at low frequencies, followed by a slope proportional to
f −3 over the Fabry-Perot frequency. It is evident that this noise dominates
at low frequencies and that can be limited decreasing the finesse and the
power on the beam splitter (CPin ). Unfortunately this is a requirement that
is opposite to the one found for the shot noise reduction.
Actually, shot noise and radiation pressure noise are strongly related
through a quantum-mechanical relation: the quantum noise δhQuant satis-
fies
q
δ h̃2Quant + δ h̃2Shot + δ h̃2Rad > 2 δ h̃2Shot δ h̃2Rad = 2δ h̃Shot δ h̃Rad . (2.10)

This is an obvious consequence, being the interferometer a Heisenberg mi-

croscope. The equality is verified when δ h̃Shot = δ h̃Rad and in this case the
quantum noise is called Standard Quantum Limit (SQL) and it is
s
1 2~
δ h̃SQL = √ . (2.11)
Lπf m η
The SQL represents an intrinsic limit for the sensitivity of an interferometer
and can be overpassed in a limited frequency band (obviously losing infor-
mation in other bands) using new techniques like the signal recycling [13] or
the light squeezing [14].

2.2 Thermal noise

Thermal noise constitutes the most important noise source in the band from
some dozen Hz to many hundred Hz. It represents the fluctuation of a macro-
scopic observable of the system at the thermodynamic equilibrium with the
2.3 Environmental noises 21

ambient, due to the thermal agitation of the microscopic elements constitut-

ing it. For this reason this noise comes out to be an unavoidable limit for
the precision of many measurements. In the interferometers this noise enters
the sensitivity in the following ways:

ˆ suspension thermal noise, which is responsible for displacement of the

mirrors center of mass. It has three different sources:

– pendulum mode thermal oscillation,

– vertical mode thermal oscillation,

– violin modes thermal oscillation;

ˆ mirror thermal noise (in the bulk and in the coating), which can be
divided into:

– Brownian thermal noise,

– thermoelastic noise,

– thermorefractive noise.

All these noise sources will be examined in detail in chapter 4.

2.3 Environmental noises

2.3.1 Seismic noise

Seismic noise is the main source of noise at low frequency (below 100 Hz).
In this band it is mainly due to human or geologic activity. Typically the
spectral amplitude has the form [23]
 2 2
−7 1 Hz m
x̃Seism (f ) ≈ 10 , for f > 0.01 Hz. (2.12)
f Hz

The damping of the seismic noise is obtained through the use of a me-
chanical oscillator. The transfer function H(f ) of an oscillator of mass m,
22 Chapter 2. Noise sources in interferometric detector

elastic constant k, under a viscous damping b is

f02
H(f ) = b
, (2.13)
f0 − f 2 + if 2πm
q
1 k
where f0 = 2π m
. At frequency f greater than the resonance

f02
H(f ) ≈ , for f  f0 , (2.14)
f2
that is, the oscillator acts as a filter for frequencies over the resonance fre-
quency. By the way, at the resonance an input signal is amplified of a factor
Q = 2πmf0 /b. So it is important for the system to be highly dissipative, i.e.
with a very low Q.

2.3.2 Gravity gradient noise

The Gravity gradient noise, also known as Newtonian noise, is the noise
due to the change in the gravitational attraction felt by the mirrors of the
interferometer. Variations in the position of masses in the vicinity of the
mirrors or in the ground or atmosphere density around the mirrors generate
a differential gravitational force that acts directly on the mirrors, bypassing
all the isolation systems. Obviously the interferometers are built in areas
far away from human activity — car, truck or train passages —; on the
other hand the atmospheric Newtonian noise is estimated to be well below
the sensitivity of the interferometers of this generation. The main important
contribution to the Newtonian noise is so due to the ground movements
induced by surface waves from the seismic activity.
Different models were done to estimate gravity gradient noise [19, 20],
all presenting similar behaviors. The most important variable that enters
the calculation is the local seismic activity model. According to Saulson’s
model [19]:
2Gρ x̃Seism
δ h̃Saul
GG = β (2.15)
πL f 2
2.4 Other noises 23

where ρ is the density of the ground around the mirror, x̃Seism is the power
spectrum density of the seismic noise and β is an order-of-unity parameter
that takes into account the seismic noise level for the particular moment un-
der consideration. In the next-generation interferometers, with the increased
sensitivity gravity gradient noise should be relevant in the interesting band
around 10 Hz. At the moment there are studies to limit this noise through the
construction of underground interferometers, with elliptical caves to displace
the mirrors in, to mainly annihilate also the effect of the seismic compression
wave propagating parallel to earth surface.

2.3.3 Residual gas noise

The gas molecules along the optical path affect the sensitivity of the inter-
ferometer through their refraction-index fluctuation. The effect is estimated
to be [15]
s r   23
2 V0 (n − 1)2 L p T0
δ h̃Gas = , (2.16)
L NA 2πu0 λ p0 T

where n is the refractive index of the gas, V0 is the volume of one gas mole
at the standard temperature T0 , NA is the Avogadro’s number, u0 is the
mean velocity of the gas molecules at the standard state and p and p0 are
the pressures at the temperature T and at the standard state respectively.
The typical required vacuum level for a kilometer class detector is ∼10−6 Pa.
In most of the constructed pipes there is a pressure of ∼10−8 Pa, taking a
good safety margin.

2.4 Other noises

There are also other noise sources depending on the characteristics of the
laser, on the readout electronics and on the vacuum apparatus.
24 Chapter 2. Noise sources in interferometric detector

2.4.1 Laser frequency noise

In a single Fabry-Perot cavity the spectral frequency noise δν̃ of the laser
light directly couples to the spectral displacement noise δ L̃ through a simple
relation:
δν̃
δ L̃ = L , (2.17)
ν
where L is the optical length of the cavity and ν is the frequency of the light;
thus with a shorter cavity the effect is less important. In the Fabry-Perot
Michelson interferometer, the frequency noise is subtracted optically at the
interference of the detection port and it appears as a signal through the
asymmetry ∆L of the two arms:

δν̃ ∆L
δ h̃Freq = 2 . (2.18)
ν L

2.4.2 Laser intensity noise

In a single Fabry-Perot cavity a laser intensity fluctuation in input is seen as

a displacement noise at the output and the locking of the cavity will compen-
sate for it as if it were a gravitational wave signal. In the case of a Fabry-Perot
Michelson interferometer this effect almost cancels out; it remains because
of a coupling with the residual displacement xRMS of the mirrors around the
resonance of the cavity, giving a strain equivalent noise δhInt :

2 δ P̃
δ h̃Int = xRMS (2.19)
L P

where P is the power and δ P̃ is the spectral power fluctuation.

2.4.3 Thermal lensing

If the power circulating in the cavities is increased in order to lower the

shot noise, as planned for future interferometers, a static distortion of the
mirror (thermal lensing) appears. This distortion must be corrected by using
2.4 Other noises 25

some thermal compensation through auxiliary laser, as in LIGO, or through

peripheral heated devices on the mirrors, as in GEO600. This noise can be
huge, depending on the finesse asymmetry [23].

2.4.4 Electric noise

The mirror position is changed by the seismic motion, especially at the res-
onance of the isolation system; for this reason the Fabry-Perot cavity length
has to be controlled and kept at its resonance through a feed-back force act-
ing on the mirror. The servo circuit and the actuator circuit for this control
can disturb the mirror position. On the one hand the servo circuit is de-
signed to have a sufficient gain for stable operation and suppression of the
other noises, but on the other also a high gain can introduce noise. Thus the
design of the servo-loop comes from a balance between a high gain and a low
introduced noise. Similarly the actuators should have enough dynamic range
and a sufficiently strong coupling. The actuator-circuit noise is typically of
√
the order of few nV/ Hz so, in order to make the actuator noise smaller
than the goal sensitivity, the uncontrolled motion of the mirror has to be
damped as much as possible using passive isolation or various active control
at different stages.

2.4.5 Scattered light

It can happen that photons of the beam are scattered by mirror imperfections
and bounce on the pipe walls getting different phases. If they recombine with
the laser beam a phase noise arises. To avoid this problem a set of mechanical
filters made by absorbing steel are usually placed along the pipe [21].
26 Chapter 2. Noise sources in interferometric detector

2.5 An example: the Virgo interferometer

It is interesting to see how the various noise levels discussed above reflect
on the sensitivity of an actual interferometer like the Virgo one, constructed
during the last years by a French-Italian collaboration in Cascina, near Pisa.

2.5.1 A Virgo overview

It will be helpful to have an overview of the whole interferometer with the

scheme showed in figure 2.1. The main important technical features of the

Figure 2.1: A schematic representation of the Virgo optical scheme.

Virgo interferometer are discussed in the following.

ˆ Suspension system. Virgo has developed a suspension system called

superattenuator [22] for suppressing the seismic noise transmission to
the mirror above 5÷10 Hz. This is a peculiar feature of this inter-
ferometer w.r.t. the ones of the other gravitational wave experiments.
The superattenuator is essentially a multi-stage pendulum about 10 m
tall, where each suspended mass is a special device acting as a vibra-
2.5 An example: the Virgo interferometer 27

tion low-pass filter in each of the 6 degrees of freedom. A scheme of a

superattenuator is shown in figure 2.2.

Figure 2.2: A schematic representation of the Virgo superattenuator. A chain

of 5 filters (with a horizontal resonant frequency fh = 0.45 Hz and a vertical
one fv = 0.4 Hz) is attached to the top of a big tripod, that acts as an inverted
pendulum (with fh = 30 mHz). The mirror is hanged through a device called
marionetta that allows a fine control of its position along the three relevant
degrees of freedom.

Shorter suspensions are used for the input bench, the mode cleaner and
the detection bench.
28 Chapter 2. Noise sources in interferometric detector

ˆ Laser. It is a Nd:YVO 4 emitting at λ = 1064 nm. Its output, in the

TEM00 mode, is about 20 W and its stability requirements (derived
from (2.18) and (2.19) assuming the measured xRMS of the residual
movements of the mirror) are:

– in power:
( √
δ P̃ 3×10−5 / Hz , for f = 10 Hz ,
< √ (2.20)
P 3×10−7 / Hz , for 100 < f < 1 kHz ;

– in frequency:
( √
δν̃ 10−4 / Hz , for f = 10 Hz ,
' √ (2.21)
ν 10−6 / Hz , for 100 < f < 1 kHz .

The laser, its stabilization apparatus and the phase modulator are not
under vacuum and they are placed on an optical table in a clean room.

ˆ Input bench. It is under vacuum and it holds the triangular refer-

ence cavity for the laser frequency pre-stabilization, the input and the
output mirrors of the mode cleaner and the main beam expansion and
alignment optotronics.

ˆ Mode cleaner. It is a triangular 144 m long cavity with the middle

mirror suspended to a short superattenuator and connected with the
two other mirrors by a dedicated pipe.

ˆ Power recycling. The recycling cavity is made by an additional sus-

pended mirror and the interferometer itself. The recycling factor is
about 50 and a power of about 1 kW goes on the beam splitter.

ˆ Fabry-Perot cavities. Four fused silica mirrors (35 cm in diameter

and 9.6 cm thick) form two 3 km long Fabry-Perot cavities with a fi-
nesse F=50. Input mirrors are made of Suprasil with a coating (10 cm
2.5 An example: the Virgo interferometer 29

in diameter, 2.25 µm thick) made of 12 alternating layers of SiO2 and

Ta2 O5 ; end mirrors are made of Herasil with a coating (28 cm in diame-
ter, 5.45 µm thick) made of 35 alternating layers of the same materials.
The optical requirements on the end mirrors are reported in table 2.1.

Property Value Units

reflectivity 0.99995 —
roughness 10 nm
coating loss <5 ppm
bulk loss < 0.7 ppm/cm

Table 2.1: Optical properties required for end mirrors.

ˆ Detection system. It is composed of a detection bench holding the

output mode cleaner and an optical table outside the vacuum vessel
holding the photodiodes (with a quantum efficiency η = 0.85).

ˆ Vacuum system. The vacuum vessel is composed of two parts: the

pipes (containing the optics) and the towers (containing the superatten-
uators). The pipes are 1.2 m in diameter and have a partial pressure of
10−9 mbar for hydrogen, 10−14 mbar for hydrocarbons2 and 10−10 mbar
for other gases. The ultra-high vacuum required in the pipe cannot be
achieved in the towers because of the devices outgassing; thus the tow-
ers are divided into two parts by a roof with a little hole for allowing
the passage of the wire holding the last stage of the suspension.

2
The limit on hydrocarbons is more stringent because of the capability they have to
stick on the mirror surfaces degrading their quality.
30 Chapter 2. Noise sources in interferometric detector

2.5.2 Virgo sensitivity curve

In figure 2.3 the main important noise contributions for the Virgo interferom-
eter are shown [23]. The sum of them, expressed in unit of strain (defined at
the beginning of this chapter), is the so called sensitivity curve. It is evident
that there are four frequency intervals characterized by a dominating noise
source over the others:

ˆ Seismic wall. Below 4÷5 Hz the seismic noise cannot be sufficiently

filtered by the superattenuators (see figure 2.2) and rapidly grows,
roughly with f −11 .

ˆ Pendulum thermal noise. It dominates approximately between 5

§
and 40 Hz and it goes as f −5/2 (see 4.3).

ˆ Mirror thermal noise. It is the dominant noise in the range 40÷500 Hz

and it goes as f (see § 4.1).
−1/2

ˆ Shot noise. Above 500 Hz it dominates increasing proportionally to

the frequency (see § 2.1.1).
2.5 An example: the Virgo interferometer 31

Figure 2.3: Virgo sensitivity curve.

32
Chapter 3

Theory of thermal noise

In this chapter a theoretical background for the comprehension and calcula-

tion of the thermal noise is given through the fluctuation-dissipation theorem.
The quality factor and the loss angle are introduced as essential quantities in
the characterization of the dissipative behavior of the materials. The basic
estimation methods of thermal noise are introduced and the evaluation of
the relevant thermal noise for interferometric gravitational wave detectors is
discussed.

3.1 The fluctuation-dissipation theorem

It is known that systems, left free to evolve in thermodynamical equilibrium

at a temperature T , show a fluctuation in the variables describing them. An
example is the Brownian motion of particles suspended in water [24]. The
particles follow complex trajectories that can be seen as fluctuations of the
position variable; the origin of this behavior resides in the collision processes
between them and the water molecules. The momentum transfer depends on
the velocity distribution of the molecules and therefore on the temperature
of the bath.
The same collisions are at the base of another process: the viscous damping.
A particle, moving in a fluid loses momentum by collisions with the other

33
34 Chapter 3. Theory of thermal noise

molecules, until it dissipates all its non-thermal kinetic energy. The above
examples show that it exists a relationship between the physical processes
responsible for the dissipation and fluctuation: they are originated from the
same mechanism. Having a tight link with the equilibrium temperature of
the system, the fluctuation takes the name of thermal noise.
The comprehension of the link between irreversible processes and fluctu-
ation allows to find quantitatively the power spectrum of the thermal noise
once the nature of the irreversibility is known. This result had been ob-
tained in a general way by Callen and Welton that stated it in the so called
Fluctuation-Dissipation Theorem (FDT) [25, 26, 27, 28]. Consider a linear1
and dissipative2 system in thermodynamical equilibrium. Let F (t) be an
external solicitation and Ẋ(t) the response of the system derived from the
equation of motion. In order to introduce the generalized force into this
equation of motion, the new term

Hint = −F (t)X(t) (3.1)

is added to the Hamiltonian of the system. The information of the dissipation

of the system is embedded in the impedance function Z(ω), defined as
Fe(ω)
Z(ω) + , (3.2)
X(ω)
ė

where the tilde stands for the Fourier transform. It is also useful to define
the admittance as
1
Y (ω) + (3.3)
Z(ω)
and the transfer function as
X(ω)
e 1 Y (ω)
H(ω) + = = . (3.4)
F (ω)
e iωZ(ω) iω
1
The linearity states that the dissipated power is quadratical w.r.t. the amplitude of
the solicitation.
2
A system is considered dissipative if, under an external solicitation, it absorbs energy
through irreversible processes.
3.1 The fluctuation-dissipation theorem 35

The FDT states that the power spectrum of the thermal noise for the quantity
X is
4kB T
SX (ω) = Re[Y (ω)] (3.5)
ω2
or, using (3.2) and (3.3), that the power spectrum of F is

SF (ω) = 4kB T Re[Z(ω)] , (3.6)

where kB is the Boltzmann’s constant and T is the temperature of the system.

Integrating the spectrum of (3.5) over frequency, one gets kB T /2 as it should
be from the equipartition theorem.
As an example it can be considered a resistor in an electric circuit. Calling
R the impedance of the resistor, the voltage V plays the role of F , the charge
the role of X and, from (3.5), it follows that the power spectral noise of the
tension is
V 2 (ω) = 4kB T R(ω) , (3.7)

that is the well known Johnson-Nyquist formula [29].

3.1.1 FDT in an n-dimensional system

The previous discussion can be extended to a system which has n coordinates

Xi . The interaction between the system and the world can be represented
by the generalized forces Fi introducing the new term

Hint = −Fi (t)Xi (t) (3.8)

in the Hamiltonian of the system. In this case the equation (3.2) becomes

Fei (ω) = Zij (ω)X

ė (ω) ,
j (3.9)

where now the impedance Zij is a matrix. Defining the admittance as the
inverse matrix of the impedance

Yij + Zij−1 , (3.10)

36 Chapter 3. Theory of thermal noise

the FDT in the form of the equation (3.5) is now written as

4kB T
SXi Xj (ω) = Re[Yij ] , (3.11)
ω2

where SXi Xj is the cross spectrum density, i.e. the Fourier transform of the
cross correlation function between Xi and Xj ; the cross spectrum for i = j is
identical to the power spectrum density of Xi . Equation (3.6) becomes

SFi Fj (ω) = 4kB T Re[Zij (ω)] , (3.12)

where SFi Fj is the cross spectrum density between Fi and Fj .

3.1.2 FDT for a linear combination of coordinates

In many practical situations it is necessary to evaluate the behavior of a

coordinate that is not a natural mode of the system but is easily measurable.
This can be the case of the surface of a mirror as seen by a laser beam profile.
A new coordinate Xnew of the n-dimensional system can be defined as

Xnew = Pi Xi , (3.13)

where Pi are arbitrary real constants. In this paragraph it is shown that the
power spectrum of Xnew can be derived using the FDT with no particular
effort, obtaining a formula of the same kind of (3.5), also in the case of a
continuous combination of coordinates. To allow the presence of a force Fnew
that drives only the momentum conjugate to Xnew the term to add in the
interaction Hamiltonian shall be of the form

Hint = −Fnew Xnew = −Fnew Pi Xi = −(Pi Fnew )Xi (3.14)

and, comparing this equation with (3.8), it is evident that the Hamiltonian
of the interaction of the 1-dimensional system described by Xnew can be
3.1 The fluctuation-dissipation theorem 37

reinterpreted as a Hamiltonian of a n-dimensional system where Pi Fnew plays

the role of Fi ; so, from the definition of Yij ,

X
ė = Y Fe = Y P Fe .
i ij j ij j new (3.15)

It is straightforward to define the admittance Ynew of the system w.r.t. the

new coordinate as

Xė
new Pi X
ė
i Pi Yij Pj Fenew
Ynew + = = = Pi Pj Yij , (3.16)
Fenew Fenew Fenew

where equations (3.13) and (3.15) were used. From equations (3.11) and
(3.13) it is possible to calculate the power spectrum density SXnew of the new
coordinate Xnew ,

4kB T
SXnew (ω) = Pi Pj SXi Xj (ω) = Re[Pi Pj Yij (ω)] (3.17)
ω2

and, using equation (3.16),

4kB T
SXnew (ω) = Re[Ynew (ω)] , (3.18)
ω2

that has the same form of equation (3.5) for the thermal noise of a 1-d system.
In a similar manner it is possible to derive the power spectrum of the force
fluctuation
SFnew (ω) = 4kB T Re[Znew (ω)] , (3.19)

where Znew+ Fenew /X

ė .
new

If X is a function of a continuous parameter r, i.e.

Z
Xnew = P (r)X(r)d3 r , (3.20)

all the previous calculations are still valid: the Hamiltonian of the interaction
is rewritten as
Z
Hint = − Fnew P (r)X(r)d3 r (3.21)
38 Chapter 3. Theory of thermal noise

and, comparing with the Hamiltonian of a generic system with a continuous

set of coordinates
Z
Hint = − F (r)X(r)d3 r , (3.22)

it is evident that Fnew P (r) plays the role of F (r), so

Z Z
X(r ) = Y (r, r )F (r) = Y (r, r 0 )P (r 0 )Fenew (r)
ė 0 0 e
(3.23)

and equation (3.16) becomes

ė 0 )d3 r0 Z
P (r 0 )X(r
R
X
ė
new
Ynew + = = P (r 0 )P (r 00 )Y (r 0 , r 00 )d3 r0 d3 r00 .
Fnew
e Fnew
e
(3.24)
Therefore the spectrum of the thermal fluctuation is still equation (3.18),
with Ynew given by (3.24).

3.1.3 The harmonic oscillator and the loss angle

The equation of motion of a harmonic oscillator of mass m is

mẍ + mω02 x = F (t) , (3.25)

where F (t) is the force acting on the mass and ω0 is the resonance of the
system. Passing in the frequency domain, the transfer function H(ω) defined
in (3.4) results to be
1
H(ω) = . (3.26)
m(ω02 − ω2)
There is no dissipation involved and the response of the system diverges when
ω = ω0 .
In a real system the elastic response to the external force F (t) is not
immediate and the elongation x is shifted in time w.r.t. the forcing term,
causing a deviation from the Hooke’s law that is called anelasticity. The effect
of this anelasticity is that a dissipation and a damping can occur, so that the
resonance peak becomes finite. To introduce the anelasticity in the model of
3.1 The fluctuation-dissipation theorem 39

the harmonic oscillator Hooke’s law is generalized. Hooke’s law says that the
stress σ needed to produce a relative displacement  is proportional to the
latter through a constant E, called Young’s modulus. A possible extension of
this law to the anelastic case is to assume that the proportionality constant
can be a complex number and can depend on frequency:

σ(ω) = E(ω)(ω) , (3.27)

where the coefficient E(ω) can be written as

E(ω) = E1 (ω) + iE2 (ω) = |E|eiφ(ω) . (3.28)

The phase angle φ takes the name of loss angle. If the deformation  is

(ω) = 0 eiωt , (3.29)

the relative stress is phase shifted of an angle φ w.r.t. deformation of the

anelastic body:
σ(ω) = |E|0 ei(ωt+φ(ω)) . (3.30)

If the effects of anelasticity are small, equation (3.30) reduces to

σ(ω) ' |E| 1 + iφ(ω) (ω) . (3.31)

From the previous formula it is evident that the introduction of anelasticity in

the harmonic oscillator is the same as defining a generalized elastic constant
with an imaginary part:

k = mω02 → mω02 (1 + iφ) (3.32)

and the transfer function of equation (3.26) becomes

1
H(ω) = . (3.33)
m(ω02 − ω 2 + iφω02 )
40 Chapter 3. Theory of thermal noise

It is possible to show that the loss angle at the resonance frequency is related
to the ratio between the dissipated energy in a cycle and the stored energy
in the oscillation. In fact, the maximum energy of the oscillator is [30]

1
Estor = Re[σ]Re[] (3.34)
2

and, for a deformation of the kind  = 0 eiωt , in the limit of low dissipation
(E2  E1 ), the maximum stored energy is

max 1
Estor = 20 E1 . (3.35)
2

The dissipated power is

Wdiss = F ẋ = Re[σ]Re[]
˙ = ωRe[σ]Im[] (3.36)

and, making the integral over a period, the dissipated energy is

Z T
T
Ediss + Wdiss dt = π20 E2 . (3.37)
0

The relation between the loss angle and the introduced quantities is then

E1 E max
φ−1 (ω0 ) ' (tan φ)−1 = = 2π stor
T
. (3.38)
E2 Ediss

It is usual to define a new parameter, the quality factor Q as

max
Estor
Q + 2π T
, (3.39)
Ediss

so that at the resonance Q = φ−1 (ω0 ).

As an example it is worthwhile to apply the anelastic model and the FDT
to the harmonic oscillator. The most simple case is the damped harmonic
oscillator with one degree of freedom; the damping is modeled by a loss angle
φ that is comprehensive of all the loss sources:
X
φ(ω) = φi (ω) . (3.40)
i
3.2 Dissipation models 41

The mechanical impedance of the system is

m
φω02 + i(ω 2 − ω02 )


Z= (3.41)
ω

and the FDT (3.5) allows to calculate the power spectrum of the thermal
noise
4kB T φω02
x̃2th = . (3.42)
mω (ω 2 − ω02 )2 + φ2 ω04
Note that if there is no dissipation (φ = 0) there is no fluctuation. In the limit
of far higher or lower frequencies w.r.t. the resonance frequency, equation
(3.42) reduces to

4kB T φ(ω)
x̃2th = , for ω  ω0 , (3.43)
mω ω0 + ω02 φ(ω)2
2

4kB T 2
x̃2th = ω φ(ω) , for ω  ω0 . (3.44)
mω 5 0

3.2 Dissipation models

Anelasticity can be described by a mechanical model [31] in which the system

is treated as the parallel between a spring of restoring constant k on one side
and a series of another spring with constant ∆k and a dash-pot with damping
constant c on the other side. The ideal body constructed in such a way is
called Debye’s solid. To calculate the effective restoring constant keff of the
solid it is possible to use a method based on a parallelism between mechanical
systems and electric circuits. In this way
 
∆k ωτ
keff ' k 1 + i , (3.45)
k 1 + ω2τ 2

where τ = c/∆k is the time constant that characterizes the response to the
solicitation. Comparing with (3.32)

∆k ωτ
φ(ω) = . (3.46)
k 1 + ω2τ 2
42 Chapter 3. Theory of thermal noise

This dependency of the loss angle w.r.t. frequency is called Debye’s peak.
Debye’s peak reproduces the effects of many kinds of irreversible processes
that happen in real bodies. The most important ones are examined in the
following paragraphs.

Internal losses

Consider a solid body with an internal structure. Anelastic dissipations

of acoustic energy in the lattice can occur in presence of dishomogeneities
of temperature field (see on page 43); also processes that involve lattice
phonons interaction can subtract energy [32]. In conductive solids it is not
negligible the presence of dissipative processes, due to phonon-electron in-
teractions, that occur also at very low temperature [33]. In the real lattice,
rich of impurities and dislocations, imperfections play the role of dissipation
sources. Irreversible processes due to defects in the lattice structure are var-
ious and not completely understood. They can involve point-like impurities,
wide dislocations (parallel or not to the crystallographic axes) and effects of
dislocation-impurity interactions [33].

3.2.1 Structural damping

Consider acoustic vibrations of an imperfect lattice. If it is vibrationally ex-

cited, the lattice can change its geometry toward more stable configurations.
Also local vibrations of the impurities can be energetically preferred. These
deformations and vibrations absorb acoustic energy from the mechanical vi-
brations causing their damping. For all these structural processes in the body
it can be imagined an activation energy ∆U . Therefore the time constant of
these anelastic processes depends on temperature and on activation energy
3.2 Dissipation models 43

following an Arrehnius’ exponential law:

∆U
τr = τ0 e kB T . (3.47)

With this value for τ , the loss angle φ follows a Debye’s peak:

∆k ωτr
φ(ω) = . (3.48)
k 1 + ω 2 τr2

The experimental data do not agree with the previous formula. In a wide
range of frequencies the loss angle for structural damping is almost constant.
To model this experimental behavior a dominant relaxation process with a
unique τ0 is imagined, but with activation energies distributed with a density
f (u) almost constant over a great interval. It follows that:
Z u2
∆k ωτr
φ(ω) = f (u)du
k u1 1 + ω 2 τr2

Z u2
∆k ωτr
' f (ū)
k u1 1 + ω 2 τr2

∆k

= f (ū)kB T arctan(ωτ1 − arctan(ωτ2 ) , (3.49)
k

where
u1,2
τ1,2 = τ0 e kB T . (3.50)

The loss angle φ(ω) obtained is poorly dependent on frequency, and the
flatness of the behavior increases with the widening of the interval [u1 , u2 ].

3.2.2 Thermoelastic damping

If the distribution of the temperature field of a vibrating body is not ho-

mogeneous, dissipation can occur. A mechanical vibration creates periodic
contractions and stretching in some regions; in the compressed parts tem-
perature increases and in the stretched ones it decreases. Under certain
44 Chapter 3. Theory of thermal noise

conditions a heat flux can be originated between two regions at a different

temperature; the reaching of the equilibrium requires a characteristic time
in which part of the vibrational energy is absorbed. Such a process is called
thermoelastic effect [34, 35].
The importance of the thermoelastic effect w.r.t. other energy-absorption
processes depends in a critical way on the geometric configuration of the
vibrating body and on the particular kind of vibration. In fact, if the two
regions with a different temperature are far away from each other, the time
necessary for the thermalization is great and compressions and expansions
occur in an adiabatic way. On the other hand if the two regions are near,
the thermalization is almost instantaneous and the vibration is isothermal.
The typical heat-migration time τth between two regions at a distance d
can be estimated starting from the equation of the heat diffusion [36]:

∂T κ 2 EαT0 ∂(∇ · u)
= ∇T− , (3.51)
∂t cV cV (1 − 2σ) ∂t

where cV is the specific heat per unit volume3 , κ the thermal conductivity,
E Young’s modulus, T the temperature, α the coefficient of linear thermal
expansion and u the elastic deformation. It results that

cV
τth ' d2 . (3.52)
κ

Defining ωth = 2π/τth , if the vibrating angular frequency is ω  ωth compres-

sion and expansion occur adiabatically; on the other hand, if ω  ωth the
process it isothermal. The thermoelastic effect will be maximum if ω and ωth
are of the same order. Therefore Debye’s peak is centered in ωth and has the
form
ωωth
φ(ω) = φ0 2
, (3.53)
ω2 + ωth
3 J
It is measured in m3 K .
3.2 Dissipation models 45

where the constant coefficient φ0 is

α2
φ0 = T E . (3.54)
cV

Vibrating objects with a small section w.r.t. their length, as fibers and strings,
typically show important thermoelastic peaks at the first resonance modes.
This effect, vice versa, is not observable in thick objects for which ωth is small
w.r.t. the modal frequencies.

3.2.3 Superficial losses

Lattice defects and impurity are not distributed in a uniform way in the
volume of the body. The region near the surface of the body is continually
exposed to damages due to chemical or mechanical contacts: it is reasonable
to think that most of the dissipation occurs in the external shell. In fact, in
particular geometric configurations, materials with very low values of bulk
loss angle appear to have greater losses because of superficial effects. The
loss angle φtot including also these effects can be introduced as

∆Ebulk + ∆Esurf
φtot = , (3.55)
Etot

being Etot the total energy stored in an oscillation, ∆Ebulk and ∆Esurf the
absorbed energy in the volume and in the surface respectively. To describe
the dependency of the superficial loss on the the geometry of the body it
suffices [37] to assume that ∆Esurf is proportional to the surface of the body
and ∆Ebulk to the volume. In this case it is possible to write:

∆Esurf S
= µds . (3.56)
∆Ebulk V

The coefficient µ takes into account the relative amount of elastic deformation
occurring at the surface — depending on the system geometry and oscillation
mode-shape — and ds , called dissipation deepness, has the dimension of a
46 Chapter 3. Theory of thermal noise

length and measures the entity of the surface dissipation. Because Etot' Ebulk ,
defining φbulk = ∆Ebulk /Ebulk , it follows

∆Ebulk ∆Esurf S

φtot = + = φbulk 1 + µds . (3.57)
Etot Etot V

For cylindrical fibers of diameter d, oscillating transversally to the axis of the

fiber, µ = 2 and
ds

φtot = φbulk 1 + 8 . (3.58)
d
The superficial effects are relevant when ds . d. The deepness ds can be
theoretically calculated [38] and, if Young’s modulus is constant in the body,
Z h
1
ds = φ(n)dn , (3.59)
φbulk 0

where h is the thickness of the superficial zone of dissipation. If the loss angle
is constant w.r.t. the deepness it follows that

ds
φsurf ' φbulk . (3.60)
h

External losses

The three processes examined above take place inside the body determining
its intrinsic anelasticity. On the other hand, energy losses can involve the
ambient in which the body is settled. In the following the attention is focused
on these external sources of losses.

3.2.4 Recoil losses

A vibrating body tends to transmit its oscillation to the supporting frame.

In the ideal case the frame is infinitely massive and rigid, but, in reality, it is
flexible and its mass is finite and so its vibration can be a dissipation source.
It is possible to modelize this situation as in figure 3.1. Let m be the mass of
the oscillating body connected with the supporting frame of mass M through
3.2 Dissipation models 47

M m
kM km

Figure 3.1: Schematization of the model done for calculating recoil losses.

a spring of elastic constant km . The frame is also joined to a perfectly rigid

environment through another spring of elastic constant kM . The dissipations
on the masses m and M are included by means of φm and φM respectively.
The elastic constants to use in the model become

2 2
km = mωm (1 + iφm ) and kM = M ωM (1 + iφM ) , (3.61)

where ωm and ωM are the proper vibration frequencies of the body and of
the frame. If the mass m is acted by a force F , from the equations of motion
it follows
 
2 2
F 2  ω (1 + iφm )
= −mωm −1 + 2 − iφm + ω2  ,
xm ωm µ ωM
2 (1 + iφM ) + 1 + iφm − µ ω2
2
m ωm
(3.62)
where µ = M/m. After little manipulation the last expression can be written
as
F 2
= keff + imωm φeff (3.63)
xm

and the total dissipation is determined by the effective loss angle

2 2
ωm ωM
φeff = φm + φM 2 2 )2
. (3.64)
µ(ωM − ωm

Therefore this kind of loss has a tiny effect if µ  1; on the other hand, near
the resonance of the system, the recoil losses are not negligible.
48 Chapter 3. Theory of thermal noise

3.2.5 Air losses

If a body is vibrating in a fluid, it undergoes viscous friction causing dissi-

pation. This loss source can be easily reduced keeping the oscillation under
vacuum. The friction in a rarefied gas is well modeled by a force proportional
to the velocity of the body:
F = −bẋ . (3.65)

In this case the transfer function defined in (3.4) is

1
H(ω) = , (3.66)
m(ω02 − ω 2 + i ωb
m
)

from which, by comparison with (3.33), the expression for the loss angle due
to gas friction can be derived:

b
φgas (ω) = ω. (3.67)
mω02

If the pressure is sufficiently low the mean free path of molecules is larger
than the typical dimensions of the oscillator and the dissipation caused by
the momentum transfer between the oscillator and the molecules is larger
than that caused by the viscosity of the gas. In this case, calling ρgas the
gas density and v̄ the mean velocity of the gas molecules, it can be shown
that [15, 40]:
1
b = ρgas Av̄ , (3.68)
4
where A is the surface of the body in which the friction occurs; therefore, in
general4
ρgas Av̄
φgas = ω. (3.70)
4mω02
4
For example, in the case of a cylindric vibrating fiber of diameter d, (3.70) leads to
v̄ρgas
φgas (ω) = ω. (3.69)
2ρdω02
3.2 Dissipation models 49

For example, a lower limit on the Q-value limited by the air damping on a
cylinder is expressed as
ρω0
Qgas = d √ , (3.71)
n mmol kB T
where d, ρ and ω0 are the size, the density and the angular resonant frequency
of the oscillator respectively. The values n and mmol are the numerical density
and mass of the gas molecules.

3.2.6 Dislocation propagation

In a mono-crystal lattice there can be a lot of defects: holes in the periodic

structure, impurities, dislocations. Mechanical vibrations or thermal excita-
tion can cause irreversible movements or vibrations of the dislocation lines
connecting the impurities. In this way a part of the acoustic energy of the
vibration can be absorbed. The Granato-Lücke theory [41] predicts a loss
angle for the ultrasounds band according to the following law:

φ(ω) ∝ ΩL4 ω , (3.72)

where Ω is the dislocation density, and L the mean length of a dislocation.

In the low frequency region a vibration of a dislocation of a certain length
L generates a Debye’s peak. Experiments have found that φ has an almost
constant behavior w.r.t. frequency and therefore it is necessary to suppose a
continuous distribution of length L that makes the theory fit the experimental
§
data as explained at the end of 3.2.1.
Therefore the acoustic waves dissipation theories for dislocation propa-
gation are still in progress. At the moment the theoretical formulas include
hardly measurable parameters (e.g. density of dislocation and Burger’s vec-
tor) or arbitrary relaxation-time distributions. One thing is definitely sure:
the dislocation losses are proportional to the dislocation density and can be
reduced with processes such as the baking of the material.
50 Chapter 3. Theory of thermal noise

3.2.7 Impedance coupling

In the case of longitudinal acoustic waves in a bar the impedance can be

written as
p
Z=S ρE , (3.73)

where S is the section of the bar, ρ its density and E its Young’s modulus. To
avoid acoustic wave radiation at the contact surface of two connected bars,
the impedances shall be chosen to be very different from each other. In the
case of contacts between solids of an arbitrary shape it is very difficult to
quantify the impedances of the two bodies but the material choices that give
√
very different values for ρE seem to give good results.
To limit the acoustic wave radiation, it seems reasonable to reduce the
contact surface. For example in the case of contact between a cylinder of
length L and radius r, and a plane parallel to its axis the surface is given by:
s
4F lr 1 − ν12 1 − ν22
 
S= + , (3.74)
π E1 E2

where E1 and ν1 are Young’s modulus and Poisson’s ratio of the cylinder, E2
and ν2 the ones of the solid plane and F the force exerted on the surface.
Typically the material choice does not allow a great variation of Young’s
modulus and so it is the cylinder radius reduction that minimizes the contact
surface.

3.3 Thermal noise calculation methods

To apply the FDT (3.18) in calculating the thermal noise power spectrum,
it is necessary to know the real part of the admittance of the system in a
broad frequency band. This is usually hard work because the imaginary part
is much bigger than the real one [42, 43, 44]. Thus, to calculate Re[Y (ω)]
there are different methods that try to overcome this difficulty. Chronolog-
3.3 Thermal noise calculation methods 51

ically, the first one used was the so called “normal-mode expansion” that
derives Re[Y (ω)] from the measured Q-values of each mode of the resonator,
supposing homogeneity of the losses. A natural extension of this method is
the so called “advanced mode expansion” introduced recently to take into
account the inhomogeneous losses on the resonator [45]. This method gives
a clear physical interpretation of the losses processes and it’s substantially
in agreement with the nowadays used techniques called “direct approaches”
that do not require any mode decomposition [46, 16, 12, 18].

3.3.1 Normal mode expansion

In the following it will be described how to derive the thermal noise of an

oscillator through the method of the normal-mode expansion. An observable
physical quantity X of a system is
Z
X(t) = u(r, t) · P (r)dV , (3.75)

where u(r, t) is the displacement vector of the system at the location r at

time t and P (r) is a vector weighting function defining the physical quantity
to be observed. For example, for thermal noise fluctuation of the internal
modes of interferometers’ mirrors, r is a 2-dimensional vector on the surface
of the mirror and P (r) is the laser beam profile. The power spectrum den-
sity of X, SX , is obtained from the equation (3.18). In order to calculate
the admittance in (3.18), the generalized force F (t)P (r) is applied on the
system, leading, for the system without dissipation, to the following equation
of motion:
∂2u
ρ 2 + L(u) = F (t)P (r) , (3.76)
∂t

where ρ is the density and L(u) a linear operator representing the elastic re-
sponse of the system. The solution of (3.76) can be viewed as a superposition
52 Chapter 3. Theory of thermal noise

of basis functions w(r):

X
u(r, t) = wn (r)qn (t) , (3.77)
n

where wn satisfy the eigenvalues problem

L wn (r) = ρωn2 wn (r) ,

(3.78)

ωn being the angular resonant frequency of the n-th resonant mode of the
system. The wn (r) are orthogonal:
Z
ρ(r)wi (r) · wj (r)dV = mi δij , (3.79)

and can be normalized imposing

Z
wn (r) · P (r)dV = 1 . (3.80)

The parameter mn is called effective mass of the n-th mode and δij is Kro-
neker’s symbol. The function qn (t) in (3.77) represents the time evolution of
the n-th mode. To derive the equation of motion for qn , the expression (3.77)
for u is substituted in (3.76). Multiplying (3.76) by wn and using (3.79) and
(3.80), it follows
mn q̈n (t) + mn ωn2 qn (t) = F (t) . (3.81)

Therefore, the time evolution of the n-th mode is the same as that of a
harmonic oscillator of mass mn and angular resonant frequency ωn with an
external force F (t) acting on it. Putting (3.77) in (3.75) and using (3.80) it
follows
X
X(t) = qn (t) , (3.82)
n
showing that the observable coordinate X can be simply described as a su-
perposition of the motions of the harmonic oscillators qn . Moreover, the
kinetic energy of the system Ekin is expressed as
Z
1 X1
Ekin = ρ|u̇(r, t)|2 dV = mn |q̇n |2 , (3.83)
2 n
2
3.3 Thermal noise calculation methods 53

showing that it is the sum over all the harmonic oscillators energies mn |q˙n |/2.
To evaluate the admittance function Y (ω), the equation of motion (3.81) is
rewritten in the frequency domain, including the dissipation:

−mn ω 2 q˜n + mn ωn2 1 + iφn (ω) q̃n = F̃ ,

(3.84)

where φ(ω) is the loss angle. From the previous equation and (3.82), Y (ω)
is described as
P
X(ω)
ė q̃n X 1
Y (ω) + = iω n = iω
. (3.85)
Fe(ω) Fe(ω) n
−mn ω + mn ωn2 1 + iφn (ω)
2

The power spectrum density of X, SX can now be derived using (3.18) and
(3.85):
4kB T X ωn2 φn (ω)
SX (ω) = 2 − ω 2 )2 + ω 4 φ2 (ω)

. (3.86)
ω n
m n (ω n n n

Therefore the thermal motion of the system is the sum of the harmonic
oscillators of the normal-mode expansion. Equation (3.86) allows to calculate
the thermal noise from the angular resonant frequency ωn , the effective mass
mn and the loss angle φn (ω) of each mode. The angular resonant frequency
and the displacement of the mode wn are obtained from the eigenvector
problem (3.78). The effective mass mn is calculated from the wn found
and equation (3.79). The loss angle is derived from the experiments but
its measurement in a wide frequency range is commonly difficult; thus, it is
usual to estimate it from the Q-value on the resonant frequencies according
to the following relation:
1
Qn = ; (3.87)
φn (ωn )

this formula is frequently used to estimate the thermal noise of interferome-

ters through the structural damping model.
54 Chapter 3. Theory of thermal noise

3.3.2 Advanced mode expansion

Theoretical researches of the nineties [47, 48, 49, 46] have shown that the pre-
dictions on thermal noise made by the normal-mode expansion were wrong
in the case of oscillators with inhomogeneous losses. The reasons of this
disagreement come clear if one considers the more accurate mode expansion
called “advanced mode expansion” [45]. The difference between this expan-
sion and the traditional one relies on the introduction of the dissipation term
in the equation of motion from the very beginning and not after the mode
decomposition.
In the case of viscous damping the equation (3.76) becomes now

∂2u ∂u
ρ 2
+ ρΓ(r) + L(u) = F (t)P (r) , (3.88)
∂t ∂t

Γ(r) being the coefficients of the friction forces. Following the same pro-
cedure as in the previous section and considering the dissipation term as a
perturbation, the derived equation for the n-th mode is
X
−mn ω 2 q˜n + mn ωn2 1 + iφn (ω) q̃n +


iαnk (ω)q̃k = F̃ , (3.89)
k6=n

where αnk is defined by

Z
αnk = ω ρ(r)Γ(r)wi (r) · wj (r)dV = αkn (3.90)

and
αnn (ω)
φn (ω) = . (3.91)
mn ωn2
The last term of the left-hand side of (3.89) does not exist in the formula
obtained from the traditional mode expansion. It contains the contributions
to one mode due to the coordinates of all the others, i.e., it accounts for the
coupling between modes. The normal mode expansion gives results consistent
with the advanced mode expansion only if all the αnk (n 6= k) vanish. The
3.3 Thermal noise calculation methods 55

comparison between (3.90) and (3.79) tells that this happens if

Γ(r) = Γ , (3.92)

that is, if the losses are homogeneously distributed over the oscillator.
It is possible to apply the advanced mode expansion to a body with
inhomogeneous structural damping. The equation of motion for an elastic
isotropic body is
∂ 2 ui ∂σij
ρ − = F (t)Pi (r) (3.93)
∂t2 ∂xj
where ui and Pi are the i-th components of the displacement u and of the
weighting function P , and the stress tensor is
 
E0 σ
σij = uij + ull δij , (3.94)
1+σ 1 − 2σ

where E0 is Young’s modulus, σ Poisson’s ratio and uij the strain tensor. To
take into account the structural damping the stress tensor is rewritten in the
frequency domain as

 
E0 1 + iφ(ω, r) σ
σij = uij + ull δij , (3.95)
1+σ 1 − 2σ

where φ is the loss angle. Following the same procedure as in the case of
viscous damping, the derived equation for the n-th mode is exactly equation
(3.89), with φ as in (3.91) and αnk now defined by
Z  
E0 φ(ωr) σ
αnk + wn,ij wk,ij + wn,ll wk,ll dV = αkn , (3.96)
1+σ 1 − 2σ

where wn,ij is the strain tensor of the n-th mode. The condition of not having
coupling terms between the modes now becomes

φ(ω, r) = φ(ω) . (3.97)

This implies that the traditional mode expansion is valid when the damping
is homogeneous.
56 Chapter 3. Theory of thermal noise

The advanced mode expansion relies on the approximation that the dis-
sipation could be treated as a perturbation. This happens if the following
condition on the distance between the resonant peaks is satisfied:

mn mk 2
|α|2nk  |ωn − ωk2 |2 . (3.98)
4

It is worthwhile to think about the physical reason for which an inho-

mogeneous distributed loss can cause mode coupling. Consider the decay of
a single resonant mode. When the loss is homogeneous the phase of its de-
cay does not depend on the position and so the displacement shape does not
change w.r.t. time. If the loss is concentrated in some regions, a phase lag will
appear in those places and, as a consequence, the shape of the displacement
will change, i.e., other modes will be excited.

3.3.3 Direct approach

Nowadays the most used technique to calculate the thermal noise is the
direct method. Its difference from the previous ones relies on a computation
that leaves aside a modal decomposition. There are many different ways of
implementing this computation. In the following the main analytical one, by
Levin [46] will be reviewed, together with the numerical dynamic approach
firstly used by Numata [18].

Levin’s method

In 1997 Y. Levin [46] proposed a new technique in calculating mirror thermal

noise. The novelty of his approach consists essentially in restating the FDT
for a continuous combination of coordinates (3.18) in an easier to calculate
form. Consider the surface of the mirror invested by the laser beam which
has a profile given by the weighting function P (r). The read-out variable
3.3 Thermal noise calculation methods 57

will be
Z
Xnew (t) = P (r)u(r, t)d2 r , (3.99)

where u(r) is the displacement of the mirror surface in the direction of the
§
beam propagation axis. At the end of 3.1.2 it was shown that, if we have to
deal with a coordinate Xnew that is a continuous combination of coordinates,
the thermal noise power spectrum will take the form
4kB T
SXnew (ω) = Re[Ynew (ω)] , (3.100)
ω2
where Ynew is the admittance associated with the new coordinate as defined
in (3.24). Therefore, the calculus of the thermal noise reduces to the calculus
of the real part of the admittance. Levin suggested an easy way to calculate
Re[Ynew ]. Suppose to apply on the mirror surface an oscillatory continuous
force F (r, t) that mocks the profile P (r) of the laser beam:

F (r, t) = F0 cos(ωt)P (r) ; (3.101)

§
this force, as explained in 3.1.2, corresponds to a generalized force

Fnew (t) = F0 cos(ωt) , (3.102)

that drives only the momentum conjugate to the Xnew defined in (3.99);
therefore, using the definition of Ynew , it is possible to calculate the mean
dissipated power

hWdiss i = hRe[Fenew X
ė ]i =
new

2
= hRe[Fenew Ynew ]i =

= F0 hRe[cos2 (ωt)Ynew ]i =
1
= F02 Re[Ynew ] (3.103)
2
and to rewrite the fluctuation dissipation theorem (3.18) in the form
8kB T hWdiss i
SX (ω) = . (3.104)
ω2 F02
58 Chapter 3. Theory of thermal noise

Also Nakagawa [16] solved the same Levin’s problem in a different formalism,
using Green’s function, but arriving at exactly the same result. It is worth to
mention that Tsubono [45] developed an interesting method based on matrix
formalism, that is quite simple to apply but is limited to analyze system that
extends mainly in only one dimension.

Numerical dynamic approach

This method was firstly introduced by Numata [18]. To avoid all the problems
related to an analytical calculation, he decided to perform a finite element
analysis of the system. In this way he could solve the equation of motion
numerically and then apply the FDT theorem in the form of (3.104). He
was not forced to do any static approximation and so the problem of finding
the mirror thermal noise was solved even in the frequency region near mirror
resonances. Moreover, in his analysis, he could use models of the mirror
including separately the magnets and the coating, performing calculations
that would have been difficult to do with the analytical method. Basically,
the steps of the method are the following:

ˆ do a geometric model of the complete system and mesh 5

it, in order
to have a model with not too many elements but with a good mesh
definition in the smallest and in the most solicited parts,

ˆ apply boundary constraints and an oscillatory force on the mirror sur-

face mocking the laser beam profile,

ˆ perform a harmonic analysis at the desired frequency,

ˆ extract from the solved model the dissipated energy multiplying, ele-
ment by element, the stored strain energy by the loss angle,
5
The meaning of the words “mesh” and “elements” will be clarified in chapter 5.
3.3 Thermal noise calculation methods 59

ˆ apply equation (3.104) in order to extract the thermal noise for the
system at the chosen frequency,

ˆ redo the last three steps for all the desired frequencies.
In chapter 5 it is shown that the potential of finite element analysis is
highly enhanced if the model is performed with a particular care in driving
the mesh and applying the pressure profile: this allows to perform harmonic
analyses on a complete model of the actual mirror (never done before) and
opens the field to faster and more precise parametric analyses.
60
Chapter 4

Thermal noise of
interferometers and some
relevant problems for Virgo

In this chapter, after a theoretical review of the various sources of thermal

noise for interferometers, just listed in § 2.2, a discussion is presented on
the relevant problems a Virgo-like interferometer has to deal with, especially
looking at future developments towards a better sensitivity. A review of
the studies performed in the past years for minimizing the loss angle in
mirrors, mirror coatings and suspensions is then presented, identifying the
most relevant causes of loss, trying to modelize the dissipation processes and
to measure the losses on bulk, membrane and fibre samples.
The thermal noise limits the interferometer sensitivity in a frequency band
that ranges from few Hertz to about 1 kHz.
This noise is intimately related to the mechanics of the suspension system
used for the mirrors and to the geometry and constitution of the mirror
itself. In interferometer detectors the mirrors are suspended at the end of
a chain of oscillators for damping as much as possible the horizontal and
the vertical motion of the ground together with the thermal noise coming
from the upper stages. In fact, over the lower characteristic frequency of

61
62 Chapter 4. Thermal noise of interferometers

the resonators, oscillations are filtered and suppressed. Therefore, the main
sources of thermal noise are the components of the mirror-suspension last
stage. Firstly our attention will be focused on the mirror and its behaving
under the impinging laser beam and then on the suspension-system resonator.

4.1 Mirror-bulk thermal noise

In this section the theoretical model of the bulk contributions to the mir-
ror thermal noise will be reviewed. The theoretical efforts done in the last
decade can be referred to three main branches: Brownian thermal noise,
thermoelastic noise and thermal lensing.
Brownian thermal noise can be interpreted as a fluctuation of the mirror
surface position coming from the mirror recoil from its internal phonons or,
using the FDT, as fluctuations induced by a structural damping.
Thermoelastic noise is intended as a noise that comes from the coupling
of thermal fluctuations with displacement fluctuations thanks to a non-null
coefficient of thermal expansion.
Finally, thermal lensing is caused by the coupling of thermal fluctuation
to phase fluctuations of the light (and therefore to measured displacement)
thanks to a non-null coefficient β = dn/dT , where n is the refraction index
and T is the temperature.

4.1.1 Brownian thermal noise

Normal mode expansion

This section is a short review of the results of the modal-expansion analysis

§
(presented in 3.3.1), in the approximation of a free mirror with an impinging
laser beam with a Gaussian profile. A detailed calculation for the Virgo
mirrors with this procedure was done by Bondu and Vinet [50]. The basic
4.1 Mirror-bulk thermal noise 63

idea is to solve with a harmonic analysis the equation of the acoustic-wave

propagation in the continuous media:

ρ∂t2 u = µ∇2 u + (λ + µ)∇(∇ · u) , (4.1)

where u is the displacement of each point of the solid and λ and µ are the
Lamé coefficients. Passing to cylindric coordinates, A.E.H. Love [51] found
three independent solutions (two transversal and one longitudinal) that can
be mixed as proposed by J.R. Hutchinson [52] to satisfy all the constraints
to be imposed on the external surfaces of the mirror, that is considered free.
It results that all the constraints can be satisfied only for a set of particular
frequencies. Three numbers, n, ξ and m, identify the various solutions (one
for each frequency): n represents the number of nodal diameters of the mode
under consideration (it enters through the Bessel functions Jn of which the
solutions are a superposition), ξ is a boolean variable representing the parity
(1 or 0 depending on if the opposed mirror faces oscillate on phase or not)
and m is the order number (that plays a role very similar to n but in the
radial direction instead of the azimuthal one).
√
If a Gaussian beam of waist w0 = 2r0 is impinging on the center of the
mirror, it sees a displacement equal to
Z 2
1 − r2
∆z = uz (r, φ, z = 0) 2 e r0 rdrdφ , (4.2)
S πr0

where S is the surface of the mirror, located at z = 0. It can be shown

that n > 1 implies ∆z = 0. Therefore, only the modes with n = 0 (that are
axisymmetric and referred to as “drum modes”) are a source of noise. Given
a certain displacement ∆z the effective mass mi can be defined through the
relation
1
mi ωi2 ∆z 2 + E , (4.3)
2
64 Chapter 4. Thermal noise of interferometers

where i is the index of the mode (the set n, ξ, m) and E is the energy of the
mode, calculated from
Z 
λ 2 
E= uii (r) + µu2ik (r) dr3 (4.4)
V 2

where the integral is intended over the whole mirror volume V and uik (r)
are the components of the strain matrix. The thermal noise can be calcu-
lated using (3.86), where the summation has to be done over a number of
modes that assures a good convergence of the series. This is a critical point
of this method, because it can happen that there is not such a frequency
and, moreover, even if the series converges, it can bring to a wrong result
due to the coupling between the various modes produced by the presence
of inhomogeneous losses (that are not included in the model as explained in
§ 3.3.2). For the Virgo mirrors, supposing a bulk loss angle φ
B not depending
on frequency, they calculated the Brownian thermal noise for the bulk to be

T 100 Hz φB m2
Brown
Sx,B = 3 × 10−38 . (4.5)
300 K f 10−6 Hz

Advanced mode expansion

Unfortunately there are no calculations done on mirrors using this method

(see § 3.3.2) and Yamamoto himself, in his work on it [45], says that the
analysis is not performed with this method because it would be necessary
to take into account too many modes for having a good convergence of the
series.

Levin’s method

§
Following the idea presented in 3.3.3 Levin [46] calculated the thermal noise
for a mirror with homogeneously distributed losses, so that Young’s modu-
lus E 0 = E 1 + iφ(ω) can be used. If this is the case, using (3.38) it is


4.1 Mirror-bulk thermal noise 65

straightforward to calculate the average dissipated power hWdiss i under the

application of an oscillatory force:

max
hWdiss i = Estor ωφ(ω) , (4.6)

max
where Estor is the energy of the elastic deformation of the mirror at the
moment of maximal strain configuration. Levin supposed that the mirror
is an infinite half-space and that the force F0 is a static one (a reasonable
hypothesis because the first resonance mode of the mirror is far higher than
the region of interest). With these assumptions he obtained

max 1 − σ2
Estor = √ F02 , (4.7)
2 2πEr0

where σ is Poisson’s ratio of the material. Assuming φ(ω) = φB , from (3.104)

it follows:
Brown 4kB T 1 − σ 2
Sx,B (ω) = √ φB . (4.8)
ω 2πEr0
As for the homogeneous case an identical result was reported by Bragin-
sky [53], calculating the real part of the admittance of the system and using
(3.18). The calculation was also made by Bondu et al. [54], Nakagawa et
al. [55] and reviewed by Liu and Thorne [56] in the case of a finite size mirror.
Their result is
Brown 8kB T
Sx,B (ω) = φB (ω)(U0 + ∆U ) , (4.9)
ω
where U0 and ∆U are infinite series of Bessel function with coefficients de-
pending on the material properties and on mirror and beam-spot size dimen-
sions. If compared with the result in the approximation of infinite mirror
mass, this calculation gives a correction that, e.g. in the case of the new
LIGO-II mirrors, starts to be greater than 10% for beam spot sizes greater
than 2.5 cm. Considering that there is the project to use flat beams, this cor-
rection becomes crucial in estimating the Brownian thermal noise correctly.
66 Chapter 4. Thermal noise of interferometers

Numerical dynamic approach

Following the method described in § 3.3.3 Numata calculated the thermal

noise using a finite element analysis. He performed the simulation [18] for the
BK7 mirror used in his experiment (7 cm in diameter, 6 cm thick) in three
separate cases: homogeneous losses, losses coming only from the coating,
losses coming only from the magnets. His results are in good agreement with
analytical approximated calculations [57]. He showed that the relative error
is about 3% in the case of coating loss and that the magnets contribution
can be neglected.

4.1.2 Thermoelastic noise

The thermoelastic noise is generated by temperature fluctuations that couple

with displacement thanks to the non-null coefficient of thermal expansion α.
Energy is dissipated because of the thermal flux induced by the generated
temperature field.
Temperature fluctuations can be originated either by intrinsic thermody-
namical fluctuation, or by laser-photon absorption.

By thermodynamical fluctuations

In the field of the thermoelastic noises, thermodynamical fluctuations are the

FDT-counterpart of the thermoelastic damping process. The mechanism of
this kind of dissipation was explained in § 3.2.2. To calculate this noise for
the mirror it is necessary to solve the equation of elasticity [30] for a field of
deformation u

1−σ 1 − 2σ
∇(∇ · u) − ∇×∇× u = α∇δT , (4.10)
1+σ 2(1 + σ)
4.1 Mirror-bulk thermal noise 67

together with the thermal conductivity equation for the temperature dis-
homogeneity δT

∂δT κ αET ∂(∇ · u)

− ∇2 δT = − , (4.11)
∂t cV cV (1 − 2σ) ∂t

where E, σ, α, κ and cV are Young’s modulus, Poisson’s ratio, the coefficient

of linear thermal expansion, the thermal conductivity and the specific heat
per unit volume of the mirror.
Braginsky et al. [53] solved the problem using the fluctuation-dissipation
theorem in the form (3.18), approximating the mirror with an infinite half-space
and considering that the frequency ω is far higher w.r.t. the typical frequency
of the heat flow
1 κ
ωc = = 2 , (4.12)
τth r0 c V
where r0 is the beam radius and τth was defined in (3.52). This is the so
called adiabatic limit in which the second term on the right hand side of
(4.11) is neglected. They found that the displacement power spectrum is
 2
th-dα 8kB 2 αT 1
Sx,B = √ (1 + σ) κ 3 2
. (4.13)
2π cV r0 ω

Moreover they solved the same system of coupled equations directly calculat-
ing the displacement fluctuations and using the Langevin approach (without
using the FDT). On the mirror they applied a heat source — the right side of
(4.11) — coming from the temperature fluctuations satisfying the following
thermodynamical relation:

kB T 2
hδT 2 i = . (4.14)
VcV

The solution they found is exactly the same as (4.13), therefore he demon-
strated that thermodynamical fluctuations of temperature are the physical
source of fluctuations deduced by FDT based on thermoelastic damping.
68 Chapter 4. Thermal noise of interferometers

Liu and Thorne [56] solved the same problem with the same approxima-
tions applying the FDT in Levin’s form (3.104) and using
D T dS E DZ k E
T 2
Wdiss = = (∇δT ) dV , (4.15)
dt T

where S is the entropy of the system and the temperature perturbation δT

comes from the law of adiabatic temperature change [30]

αET ∇ · u
δT = − . (4.16)
cV (1 − 2σ)

They achieved the same solution (4.13). They also made the calculations for
a finite cylindrical test mass finding the correction factor w.r.t. the infinite
test mass solution. It is worth noting that the correction for a wide range of
beam-spot size is not greater than 10% if the ratio between the radius and
the thickness of the cylinder is near unity.
Cerdonio et al. [58] extended the Liu analysis releasing the assumption of
the adiabatic limit, i.e. performing the analysis that is valid also at low tem-
peratures and/or small beam spot size. Equation (4.13) has to be multiplied
by a correction factor Ω2 J(Ω) with1

ω
Ω+ , (4.17)
ωc
2
r ∞ ∞
u3 e−u /2
Z Z
2
J(Ω) = du dv 2
(4.18)
π3 0 −∞ (u + v 2 ) (u2 + v 2 )2 + Ω2
and ωc as defined in (4.12). For ω  ωc this factor is equal to unity but for
th-dα
ω  ωc goes as ω 3/2 and then Sx,B ∝ ω −1/2 .

By photothermal fluctuations

This effect is called dynamic photo-thermal noise2 and it is due to the tem-
perature fluctuation induced by random absorption of optical photons in
1
Formula (4.18) in [58] had a wrong factor as noted by Numata [18].
2
It is also referred to as photo-thermal shot noise.
4.2 Mirror-coating thermal noise 69

the surface layer of the mirror. Thanks to a non-null coefficient of thermal

expansion these fluctuations couple with displacement fluctuations.
Braginsky et al. [53] assumed that the coating thickness of the mirror is
much smaller than the beam radius, and, in the adiabatic limit, found the
following expression for the power spectral displacement noise:
 2
ph-th 2 2 α 1
Sx,B α = 2 (1 + σ) 2
~ω0 hWabs i 2 , (4.19)
π c V r0 ω

where the mean of the absorbed power Wabs is intended over a period π/ωgw
and ~ω0 is the typical energy of the absorbed photons.
Cerdonio et al. [58] extended the calculation releasing the assumption of
the adiabatic limit finding:
 2
ph-th 2 2 α
Sx,B α = 2 (1 + σ) ~ω0 hWabs iK(Ω) , (4.20)
π κ

with
∞ ∞ 2
u2 e−u /2
Z Z
1
K(Ω) = du dv 2 . (4.21)
π 0 −∞ (u + v 2 )(u2 + v 2 + iΩ)

Equation (4.20) reduces to (4.19) for ω  ωc but for low frequencies it gives
lower values w.r.t. the adiabatic prediction and it displays a behavior almost
independent of frequency.

4.2 Mirror-coating thermal noise

As done in the mirror-bulk thermal noise section the review is logically di-
vided into three parts: Brownian thermal noise, thermoelastic noise and
thermorefractive noise.

4.2.1 Brownian thermal noise

Following their formalism [16], Nakagawa et al. [57] calculated the dissipation
induced by an inhomogeneous loss distribution on a half-infinite mirror due
70 Chapter 4. Thermal noise of interferometers

to the fact that loss angle in the bulk material φB is different from loss angle
in the coating φC :

Brown 4kB T (1 + σB )(1 − 2σB ) dC

Sx,C = φC , (4.22)
ω πEB r02

where EB , σB and dC are the bulk Young’s modulus, the bulk Poisson’s ratio
and the coating thickness respectively. It is worth noting that there is a
linear dependence on 1/r02 as predicted by the scaling argumentation done
by Levin [46] and that the result is exactly the same if performed using
Levin’s method [46, 57].
Harry et al. [38], using Levin’s method, generalized this result for a non-
isotropic coating loss φC
 
δUk dC δU⊥ dC
φC = φk + φ⊥ , (4.23)
U U

where δUk and δU⊥ are the surface integrals of the coating energy-density
components along the parallel and perpendicular direction w.r.t. the surface,
φk and φ⊥ are the associated coating loss angles and U is the total energy
stored in the bulk. For a thin coating, provided that σC ' σB , EC + ECk '
EC⊥ and σC + σCk ' σC⊥ , they found3

1 dC EC (1 + σB )(1 − 2σB )2 + EB σC (1 + σC )(1 − 2σB )


φC = √ φk +
2π r0 EB (1 + σC )(1 − σC )(1 − σB )

EB (1 + σC )(1 − 2σC ) − EC σC (1 + σB )(1 − 2σB )
φ⊥ (4.24)
EC (1 − σC )(1 + σB )(1 − σB )
and therefore
Brown 4kB T (1 − σB )2
Sx,C = √ φC . (4.25)
ω 2πEB r0
Equation (4.25) is valid provided that the coating loss occurs in the coating
materials themselves and not for rubbing in the bulk/coating or layer/layer
3
It is worth to note that Harry [59] recently performed some calculations to account
for eventual anisotropies in Young’s modulus and Poisson’s ratio of the coating.
4.2 Mirror-coating thermal noise 71

interfaces. At the first order in σB and σC formula (4.25) reduces to

 
Brown 2kB T dC EC EB
Sx,C = (1 − 2σB )φk + σC (φk − φ⊥ ) + φ⊥ . (4.26)
ω πEB r02 EB EC

If φk ' φ⊥ , then the lowest coating-induced thermal noise occurs when

Young’s modulus of the coating is matched to that of the substrate.

4.2.2 Thermoelastic noise

Temperature fluctuations (that couple with displacement thanks to αC 6= 0)

can be originated either by intrinsic thermodynamical fluctuation, or by laser-
photon absorption.

By thermodynamical fluctuations

The problem was extensively treated by Braginsky and Vyatchanin [60].

Firstly, using the Langevin approach, they solved the problem for a half-
0
infinite mirror with an expansion coefficient αC + αC−αB for the coating and
αB0 = 0 for the bulk; the bulk and the coating are treated as if they had the
same elastic constants. Taking into account only stresses produced by the
coating due to non-uniform temperature distribution, they obtained
√ 02 2
 2
th-dα 4 2 2 αC kB T dC 1
Sx,C = (1 + σB ) √ √ . (4.27)
π κB cVB r0 ω

Secondly, using the FDT theorem, they calculated a formula for the case of
different elastic parameters between the bulk and the coating: it suffices to
use the same formula (4.27) with4
 
0 αC 1 + σC EC (1 − 2σB ) cV
αC = + − C , (4.28)
2αB (1 − σC )(1 + σB ) EB (1 − σC ) cVB

They generalized formula (4.27) and (4.28) for multilayer coating [60] also.
Thirdly, using the FDT approach developed by Liu and Thorne [56], they
4
The formula in [60] has an error, successively corrected by [61] as reported.
72 Chapter 4. Thermal noise of interferometers

calculated the finite test mass correction factor that, e.g., is about 1.6 in the
case of LIGO-II.
Recently Fejer et al. [61] performed the same calculation in the case of dif-
ferent thermoelastic properties of coating and substrate (other than thermal
expansion) and out of the low-frequency region.

By photothermal fluctuations

In literature this effect was not given so much attention because of its small-
ness. An estimate of the behaving of the effect was made by Rao [62]. He
studied two different cases depending on the values assumed by the thermal
conductivity in the bulk κB and in the coating κC .

Figure 4.1: Heat flow simulation on a Figure 4.2: Heat flow simulation on a
coated mirror with a highly conduc- coated mirror with a low conductive
tive bulk. bulk.

If κB  κC , the heat generated by photons absorption flows in the coating

essentially in one dimension (see figure 4.1). Following the same approach of
Braginsky et al.[53] he found that

2

αC
2
1
 √ 4
ph-th −dC iωcVC /κC
Sx,C α = 2 ~ω0 hWabs i 2 1 − e , (4.29)
π cVC r02 ω

where
1
αC = (α1 d1 + α2 d2 ) , (4.30)
d
4.2 Mirror-coating thermal noise 73

calling α1 , α2 and d1 , d2 the coefficients of thermal expansion and the total

thickness for the two coating materials. At high frequencies equation (4.29)
takes the same dependency form as (4.19) and at low frequencies approaches
zero.
If κB  κC the heat flow behaves like shown in figure 4.2. Following a
technique similar to [60], he solved the heat flow in a uniform half-space, com-
puting the average temperature for a layer near the surface with a frequency-
p
dependent scale height determined by rt = κ/(ρcV ω). He found
 2
ph-thα 1 αC dC ~ω0 hWabs i 1
Sx,C = 2 , (4.31)
π r02 cVC κC ω
that shows a behavior like 1/ω.

4.2.3 Thermorefractive noise

The thermorefractive noise is generated by temperature fluctuations that

couple with phase fluctuations of the laser (and therefore with measured
displacement) thanks to the non-null coefficient β = dn/dT , where n is the
refraction index. Temperature fluctuations can be originated either by in-
trinsic thermodynamical fluctuation, or by laser-photon absorption.

By thermodynamical fluctuations

Following the same approach as in [53], Braginsky et al. [63] calculated the
thermorefractive noise induced by thermodynamical fluctuation of tempera-
ture. Considering the frequency range (that is typical for the ground inter-
ferometer) in which the detection frequency ω satisfies
q
κC /cVC
dC  √  r0 , (4.32)
ω
where dC is the coating thickness and r0 the beam radius, they calculated
√  2
th-dβ 2 kB βeff λT 1
Sx,C = √ √ , (4.33)
π cVC κC r0 ω
74 Chapter 4. Thermal noise of interferometers

where5
n1 n2 (β1 + β2 )
βeff = (4.34)
4(n21 − n22 )
and β1 = dn1 /dT , β2 = dn2 /dT , the indexes referring to the two materials
alternating in the coating layers.

By photothermal fluctuations

This effect is numerically much smaller than the other thermorefractive noise,
just discussed here before, and it will not be reviewed (see, e.g., [64] for
details).

4.3 Suspension thermal noise

Usually the horizontal damping of the test mass is obtained through simple
pendula and the vertical one through springs or blades. These kinds of
filters introduce noise in each degree of freedom of the whole structure they
enter, according to the fluctuation dissipation theorem. This problem has
been theoretically dealt with since the early nineties [65, 66, 67] and can be
divided into four separate contributions: the pendulum thermal noise, the
vertical thermal noise, the violin-modes thermal noise and the tilt/rotational
mode thermal noise.

4.3.1 Pendulum thermal noise

The suspension wires, together with the suspended mass, constitute a pen-
dulum with a proper resonance frequency determined by the restoring elastic
force of constant ke (1 + iφe ) and by the gravity force acting as a spring of
constant kg = mg/l, that is lossless. The viscous damping of the air is made
negligible by the low pressure involved (10−9 mbar) and so the equation of
5
The formula reported in [63] was wrong and was successively corrected [60] as it is
reported.
4.3 Suspension thermal noise 75

motion can be written as

mẍ = −kg x − ke (1 + iφe )x , (4.35)

which is equivalent to
ke

mẍ = −(kg + ke ) 1 + i φe x ; (4.36)
kg + k e
therefore the effective loss angle is equal to
ke ke
φeff = φe ' φe , (4.37)
kg + ke kg
being usually ke  kg . The factor ke /kg is called “dilution factor” and it is
evident that, thanks to it, the effective losses of the pendulum system are
much lower than the internal φe . The elastic constant ke of a wire is given
by [39]
1 √
ke = T EI , (4.38)
2l2
where l is the wire length, T is the tension applied on the wire, E is Young’s
modulus of the material and I is the superficial moment of inertia of a wire
cross-section. In the last-stage suspension different solutions can be adopted
in the way the test mass is suspended, depending on how many wires are
used. To reduce the dilution factor it is convenient to have T and I as low
as possible. Usually the wires are loaded at a defined ratio c (c ' 0.3÷0.4) of
their breaking stress σb . Therefore, if the mirror is suspended with a certain
load in a multi-wire configuration, the tension on each wire is reduced and
the radius of the wire can be reduced accordingly. If there are N wires for a
mirror of mass m, the requirement for being at the working tension is
mg
= cσb (4.39)
N πr2
and so the elastic constant is
r r
N√ N mg πr4 1 m3 g 3 E
ke = 2 T EI = 2 E = 2 , (4.40)
2l 2l N 4 4l cσb Nπ
76 Chapter 4. Thermal noise of interferometers

where (4.39) was used to explicitly show the dependence on m and N . The
loss angle is then r
1 mgE
φeff = φe (4.41)
4lcσb Nπ
and this value can be used to evaluate the pendulum thermal noise using
p
formula (3.42) with ω0 = g/l:

4kB T φeff ω02

Sxpend = . (4.42)
mω (ω 2 − ω02 )2 + φ2eff ω04

Since the band of interest is over the resonance frequency (from a few Hertz to
few a hundred Hertz) the approximated formula (3.44) can be used, showing
that the pendulum thermal noise scales as m−1/2 and N −1/2 .

4.3.2 Vertical mode thermal noise

Owing to the non uniform gravitational field generated by the earth, local
vertical directions at both ends of the arms are not parallel to each other.
Therefore, a vertical motion of the mirrors gives a change in the cavity length
that can mock a gravitational wave signal. The coupling between the vertical
motion and the horizontal one is exactly equal to the angle under which an
arc of length Larm /2 is seen from the center of the earth (figure 4.3). For this

L arm

θ 0
R

Figure 4.3: The vertical directions at the ends of the arm are not parallel
and give origin to a coupling between the vertical and the horizontal dis-
placements of the mirrors.
4.3 Suspension thermal noise 77

reason a part of the thermal noise associated with the vertical oscillation is
seen on the horizontal direction according to the following formula:
4kB T φωv2
Sxvert = θ0 , (4.43)
mω (ω 2 − ωv2 )2 + φ2 ωv4
in which ωv is the resonance frequency of the vertical motion of the system
whose value is given by
πr2 E/l gE
ωv2 = = , (4.44)
m/N cσb l
where l is the wire length, E and σb are Young’s modulus and the breaking
stress of the material and (4.39) was used to show explicitly that there is no
dependence on the mass m of the mirror and on the number N of suspension
wires.

4.3.3 Violin modes thermal noise

The wires suspending the mirror present numerous resonances relative to the
normal modes of oscillation called violin modes. These oscillations show up
each time a wire is under a certain tension. If there are N wires of density ρ,
Young’s modulus E, radius r and length l, cross-sectional inertia momentum
I, holding a mirror of mass m, considering that the load on each wire is
T = mg/N and the linear density of the wire is ρlin = πr2 ρ, the resonance
angular frequencies will be [39]
s s
nπ T nπ mg/N
ωn = (1 + a) = (1 + a) , (4.45)
l ρlin l πr2 ρ
where r s
2 EI 2 Eπr4 /4
a= = , (4.46)
l T l mg/N
and n is the mode number under consideration. The violin mode contribu-
tions to thermal noise can be evaluated using [39]:
4kB T 2ρr2 l X 1 ωn2 φn
Sxviol = N , (4.47)
ω πm2 n2 (ω 2 − ωn2 )2 + φ2n ωn4
78 Chapter 4. Thermal noise of interferometers

where the loss angle φn of each mode is given by

aφ(ωn )
φn = . (4.48)
1 + a4 n2 φ2 (ωn )

Therefore, the contribution of this noise to the sensitivity curve is limited to

the various resonance frequencies of the suspension system.

4.3.4 Tilt and rotational modes thermal noise

The term “tilt mode” is referred to the oscillation of the suspended mirror
around a horizontal axis perpendicular to the laser beam, while “rotational
mode” is referred to the oscillation around a vertical axis.
The angular displacements θ induced by these modes are expressed by
formulas similar to (4.42) where the appropriate angular frequency and loss
angle should be used. These modes have no effect on the sensitivity at the
first order in θ because the induced optical path difference is δL ∼ θ2 .

4.4 The problem of new materials for Virgo

Having a look at figure 4.4, it is possible to make some considerations on the

perspective to improve Virgo sensitivity fighting the various noise floors in the
relative frequency bands. As already said in § 2.5.2, the main contributions
limiting the sensitivity from higher to lower frequencies are the shot noise,
the mirror thermal noise and the suspension noise. As regards the seismic
noise, Virgo can already be considered an advanced gravitational detector,
§
because of the superattenuator system (see 2.5.1).

4.4.1 Lowering the shot noise

Shot noise limits the sensitivity for frequencies over 500 Hz.
Looking at (2.4) it is evident that a direct way of lowering the shot noise
floor is to use a higher power laser. In a provisional draft about an advanced
4.4 The problem of new materials for Virgo 79

Figure 4.4: The Virgo sensitivity curve with all its main noise contributions
(taken from [23]).

√
Virgo [64] a laser of 100 W is foreseen, gaining a factor 5 in sensitivity.
Apart from the technical problems of going to further higher powers, there
is a limitation imposed by the radiation pressure noise. According to equa-
tion (2.8) this noise increases as the square root of the power impinging on
the beam splitter. Fortunately it has a dependence on the inverse of the
suspended mass also, so that it can be convenient to go toward higher loads.

4.4.2 Lowering mirror-bulk thermal noise

In the frequency region between 50 Hz and 500 Hz the dominating noise is

§
the Brownian thermal noise of the mirror substrate (see 4.1.1).
According to (4.5) or (4.9) this noise is proportional to the loss angle φB
of the mirror substrate. This is why during the last years a lot of efforts
have been done in order to measure losses in various substrates and trying
80 Chapter 4. Thermal noise of interferometers

to estimate the intrinsic loss in the frequency region of interest. Hereafter a

short list of recent results is furnished.
Rowan et al. [68], using metal-wire single-loop suspensions, have measured
a loss factor of 3.7×10−9 for HEM sapphire, grown by a method suitable for
producing test masses of the required size and quality.
Numata et al. [69, 70, 71] performed accurate series of measurements on
fused-silica bulk-samples using a nodal support system (a horizontal mirror
on a ruby ball touching the center of the mirror face). They found values
for φB ranging from 2.5×10−8 to 1.5×10−6 , demonstrating that the nodal
technique is a valid one in this kind of high precision measurements and
that annealing helps to increase the quality factor. They also verified that
the losses have a behavior that is roughly proportional to frequency. The
understanding of this phenomenon is crucial in estimating the quality factor
in the 50÷500 Hz band, which usually is some orders of magnitude lower than
the frequency-region of measurements on samples. Willems and Busby [72]
suspended sapphire and fused silica samples with the same method as [68];
the wires were mechanically polished and greased with lard. They obtained
phi down to 4×10−9 for sapphire and 8.3×10−9 for fused silica.
Smith et al. [73] measured bulk loss angles on fused silica masses attached
to fused silica suspensions in a configuration adopted in the last 2 stages of
GEO 600 finding loss angles down to 2.6×10−7 .
Penn et al. [74], relying on measurements performed on several fused silica
samples of various shapes, proposed a heuristic model of the loss angle φ as
a function of both frequency and volume to surface ratio:
φ(f, V /S) = φsurf + φbulk + φth−el =
(4.49)
= C1 (V /S)−1 + C2 f C3 + C4 φDebye (f ) ,

where φDebye (f ) is a function embedding the typical Debye peak described

by equation (3.53). The best fit for the four parameters gave, in the case of
4.4 The problem of new materials for Virgo 81

Suprasil 2 and Suprasil 312, the results shown in table 4.1. It is worth to note

Type C1 C2 C3 C4
2 8.6×10−9 7.1×10−12 0.8 1.0
312 7.1×10−9 4.6×10−12 0.8 —

Table 4.1: Best fit values of the four parameters of the function described in
(4.49) for Suprasil 2 and Suprasil 312 samples.

that the exponent of the explicit frequency dependency is 0.8, which gives
a strong frequency dependence of the loss angle. This value is in agreement
with the value found by Wiedersich et al.(' 0.78 [75]) with light scattering
experiments on silica glass. A plot of φ(f, V /S) is shown in figure 4.5. For

Figure 4.5: A plot of the best fit to data for the 2-d function described
by (4.49), modeling with 4 parameters the loss as a superposition of bulk,
surface and thermoelastic losses.

frequencies around 200 Hz and for volume-to-surface ratio of the order of

1 cm (for a mirror this number can be up to 10 cm) the estimated loss angle
is further lower w.r.t. all the available measurements.
82 Chapter 4. Thermal noise of interferometers

Penn et al. [76] and Ageev et al. [77], following an experimental setup of
the same kind of Gretarsson et al. [37], performed some measurements on
annealed fused silica rods, finding values down to 5×10−9 for the loss angle.

4.4.3 Lowering mirror-coating thermal noise

The problem of treating thermal noise in the coating is not trivial because
it requires the knowledge of modeling the effects of non-homogeneously dis-
tributed losses (coating loss angles are usually higher than bulk loss angles).
Yamamoto [45] clarified definitely the unreliability of the normal-mode ex-
pansion method (see § 3.3.1) in this kind of problems and proposed an ad-
vanced modal expansion method (see § 3.3.2) as a solid theoretical tool. By
the way, from the practical point of view, it was two years before, with
§
Levin [46] and his direct method, (see 3.3.3) that actual calculations started
to be possible.
The first ones who demonstrated the invalidity of the normal-mode ex-
pansion experimentally were Yamamoto et al. [79] with measurements on
an inhomogeneous slab. Their theoretical predictions [80] for mirror thermal
noise with inhomogeneous losses (performed using Levin’s method) were con-
firmed by their experimental work [81, 82] on an aluminum hollow cylinder to
which they could apply an arbitrary loss distribution through eddy currents
induced by strong magnets. They studied two distinct cases (also with the
help of FEA): losses concentrated in the front face of the mirror (the laser
side) and losses in the back face, demonstrating the importance of the loss
distribution w.r.t. the position of the impinging laser beam.
The results reviewed in the previous paragraph are extremely encourag-
ing, in the frame of lowering the mirror-bulk thermal noise, but they imme-
diately open the field to other battles, this time against the noise induced by
Brownian and thermodynamical fluctuations in the coating.
4.4 The problem of new materials for Virgo 83

By Brownian fluctuations

Calculating the ratio of the Brownian thermal noise for the bulk, equation
(4.8), to the one for the coating, equation (4.22), it follows
Brown
5 × 10−5
r r
Sx,B π 1 − σB2 φB r0 π φB r0
Brown
= ' ' , (4.50)
Sx,C 2 (1 + σB )(1 − 2σB ) φC dC 2 φC dC φC
where, in the last passage, the values φB ' 5 × 10−9 , r0 ' 5 × 10−2 m, dC '
5×10−6 m have been used. It is evident that, if φC is higher than 5×10−5 , the
Brownian thermal noise of the coating starts to be a serious problem. This is
why a lot of researchers focused their efforts on measuring the intrinsic loss
of the coatings.
The effect of optical coating deposition on fused silica slabs was studied by
Gretarsson et al. [83]. In their work they stressed the importance of keeping
under control dissipations due to surface losses (see § 3.2.3). They showed
that a variety of surface treatments (washing with detergent, acetone, KOH,
or HF etching) produced no change in the condition of the surface or made
it a little worse. They found that the intrinsic loss angle of the standard
SiO2 /Ta2 O5 coating was φC ' 4×10−4 .
Crooks et al. [84] performed measurements on Al2 O3 /Ta2 O5 coatings de-
posited on two fused silica mirrors. They found an excess loss of unknown
source: it was neither due to the inhomogeneity of the coating, nor due to the
§
possible difference among φk and φ⊥ in the coating (see 4.2.1), nor due to
the friction in the mirror suspension points. By the way, with a multivariate
regression analysis they got φC = (6.4±0.6)×10−5 .
In a companion paper Harry et al. [38], following the model reviewed
in § 4.2.1, tried to measure φk of SiO2 /Ta2 O5 coating on three fused silica
samples. They found φk = (4.2±0.3)×10−4 for coatings deposited on com-
mercially polished slides and φk = (1.0±0.3)×10−4 for coating deposited on
a superpolished disk.
84 Chapter 4. Thermal noise of interferometers

Penn et al. [85] investigated the sources of loss in SiO2 /Ta2 O5 coatings on
polished and annealed fused silica samples of various thickness. Their series
of measurements give strong evidence that losses due to coating/bulk and
layer/layer interfaces are negligible w.r.t. losses associated to the intrinsic φ
of the material constituting the coating.

By thermodynamical fluctuations

As for the thermal noise induced by thermodynamical fluctuations on the

coatings, the situation can be even more critical. Equations (4.33) and (4.27)
do not depend on the intrinsic loss of the coating and so the research should
move on quite a new field, exploring the possibility of using different materials
in order to achieve the aimed sensitivities.
To have an idea of the relative weight of the thermorefractive contribution
and the thermoelastic one in the coating losses, the square root of the ratio
between (4.33) and (4.27) can be used:
v s
u th-dβ
u Sx,C 2 2
βeff λ βeff λ
t
th-dα
' 2 ' , (4.51)
Sx,C 4(1 + σB )α0 C d2 2α0 dC

which, in the case of SiO2 /Ta2 O5 coatings on a fused silica substrate, is

around6 2÷3. It is evident that relative weight strongly depends on the
thermoelastic characteristic of the materials involved. From a theoretical
point of view, good models trying to take into account the differences among
the thermoelastic coefficients of the bulk and the various layer materials are
§
being developed (see 4.2). To have an idea of the variety of the out-coming
effects, figure 4.6 can be considered: following thermoelastic predictions by
Fejer et al. [61], it is evident how different the effect of the same type of
coating (e.g. Al2 O3 /Ta2 O5 ) can be in dependence of the substrate it is applied
6
This factor was initially found to be about 7 time smaller because of the used value
αTa2 O5 = −4×10−5 [86] instead of αTa2 O5 = 4×10−6 [87, 88].
4.4 The problem of new materials for Virgo 85

Figure 4.6: Calculated thermoelastic noise [61] of coatings of Al2 O3 /Ta2 O5

or SiO2 /Ta2 O5 , applied to silica or sapphire samples. Dashed lines are from
the complete model, solid lines from the low-frequency approximation.

to.
On this research line there is the work by Baker et al. [89] about the depen-
dence of hardness on density of Ta2 O5 layers, through Rutherford backscat-
tering spectroscopy and nanoindentation studies. Tien et al. [87] performed
simultaneous measurements of αTa2 O5 and ETa2 O5 and Braginsky et al. [88]
performed measurement of αTa2 O5 , both groups using phase shifting interfer-
ometry.
There is a clear need to carry out a series of experiments, in which coat-
ing parameters are systematically varied as this will allow the source of the
coating losses to be investigated.

4.4.4 Lowering suspension thermal noise

In the frequency region up to 50 Hz the dominating noise is the thermal noise

of the suspension (see § 4.3). Among the various suspension thermal noise
contributions, pendulum thermal noise is by far the main important one.
86 Chapter 4. Thermal noise of interferometers

Looking at (4.42) and (4.41) it is evident that it is convenient to suspend

heavy masses in a multiple-wire configuration using materials with a high
breaking stress and low internal loss angle. Researches on materials with
good properties in the past years proceeded together with the studies on the
clamping and on the identification of possible excess losses.
The first measurements testing the theoretical models were performed by
Gillespie and Raab [90, 91], giving encouraging results.
Cagnoli et al. [92] focused the attention on the geometry of the clamp for
the suspensions. They obtained a significant decrease of losses due to the
stick-and-slip mechanism on the clamp surfaces by increasing and localizing
the clamping pressure on the heads of the suspension wires. Their work gave a
strong impulse to investigate suspension techniques different from clamping,
like welding [93] and silicate bonding [94].
Ageev et al. [95, 96] studied excess losses connected with loaded suspen-
sion wires made of steel and tungsten. They found that spurious increases
of the noise floor and isolated noise peaks (which could not be explained by
a mechanical shot noise model [92]) are much less present in steel wires than
in tungsten wires. A recent study by Gretarsson and Saulson on tungsten
and silica fibres [97] demonstrated that such kind of effects are negligible.
Cagnoli et al. [98] performed internal friction measurements for the nor-
mal modes of circular fibres of different materials demonstrating that the loss
angle in fused silica is 2 orders of magnitude better than the one in metallic
wires.
Gretarsson and Harry [37] performed accurate measurements on fused
silica fibres reducing clamp losses thanks to the introduction of an isolation
bob between the sample and the clamp. They noted that surface losses
dominated at diameter values less than about 1 mm.
4.4 The problem of new materials for Virgo 87

Experiments on all fused-silica massive pendula [99, 100] gave values for
the total loss angle around 10−8 .
Ribbons seem to be a good alternative to fibres [101] and recently [102]
it has been shown that, with the correct aspect-ratio choice (to minimize
thermoelastic damping) they can perform even better than fibres.

4.4.5 Guidelines for new materials

To summarize, there are a number of constraints which affect the selection

of materials for test masses, coatings and suspensions of long baseline gravi-
tational wave detectors. In addition to the common feature of having a very
low loss angle at room temperature,

the materials for the bulk should:

– have very low optical loss at the wavelength of the laser light to
minimize the heat deposited,

– have high thermal conductivity and low thermal expansion to min-

imize mechanical distortion when heat is deposited in it by the
laser beam,

– have a low rate of change of refractive index with temperature to

minimize thermal lensing effects for transmitted laser beams,

– be suitable for polishing and coating to sub-angstrom surface rough-

ness,

– be capable of being produced in suitable size, to reduce radiation

pressure noise,

– have oxidised aluminum or silicon in their makeup so that they

can be hanged to suspension through silicate bondings;

the materials for the coating should:

88 Chapter 4. Thermal noise of interferometers

– have thermoelastic properties that match with the substrate ones,

to minimize mechanical distortion when heat is deposited in it by
the laser beam,

– have a low value of change of refractive index w.r.t. temperature

to minimize thermorefractive noise,

– be capable of being deposited in quarter-wavelength layers;

the materials for the suspension should:

– have high thermal conductivity and low thermal expansion to min-

imize the thermoelastic damping,

– be capable of being produced in suitable shapes (ribbons or fibres),

– have oxidised aluminum or silicon in their makeup so that they

can be attached to test masses through silicate bondings.

Good candidate materials for mirror bulks can be very pure glasses such as
fused quartz and fused silica or single-crystal materials such as sapphire and
silicon. For the coating there are researches on Al2 O3 /Ta2 O5 , SiO2 /Ta2 O5
and SiO2 /TiO2 -doped Ta2 O5 .
In this complex research context, finite element analysis can be a very
useful tool in various steps of the modelization as it will be shown in chapter 5
through the work done by the candidate.
As far as suspensions are concerned, fused silica can be a good material:
a lot of work has been done on it, and presently there is already a planning
for bringing Virgo to an upgraded detector, called Virgo+, in which C85 steel
wires will be replaced by fused silica fibres.
Silicon also is a very promising candidate: in chapter 6 its interesting
features will be described, together with the experiments performed by the
candidate in order to measure the loss angle of crystalline fibres, aiming at
4.4 The problem of new materials for Virgo 89

the possibility of using this new material for the suspensions of future 3rd
generation GW interferometric detectors.
90
Chapter 5

Using finite element analysis

In the general framework of calculating thermal noise in composite system,

the finite element analysis (FEA) can play a central role. The power of FEA
resides on the possibility of facing many different problems connected with
the behavior of complex systems. In many steps of the calculations that
are necessary toward a complete thermal noise evaluation, FEA can give its
support and, in some cases, it can allow to undertake problems otherwise
impossible to solve.
In this chapter, first of all the FEA basis is introduced examining strong
points and limitations of this approach. Then it will be discussed the appli-
cation of FEA to an actual model of a Virgo mirror for calculating thermal
noise with a direct method.

5.1 The basis of finite element analysis

The finite element method is a discretization process that, through the use of
a mathematical model and numerical calculus techniques, allows the study
of particularly complex problems. A solution is sought on the basis of funda-
mental equations governing the phenomena under investigation and through
apposite approximations of the variables involved. The basic idea of FEA is
quite simple and two separate steps can be identified:

91
92 Chapter 5. Using finite element analysis

ˆ the discretization in which the structure is divided into finite elements

with dimensions that bring to approximate the stress and strain fields
within a certain error;

ˆ the building of the structure in which all the single elements are assem-
bled together, to solve the complete system so that

– the stress and strain fields satisfy the continuity requirements on

elements interfaces,

– the internal stresses are in equilibrium, considering the applied

loads and the boundary conditions.

In mechanical analyses of structures FEA can be used to face many different

problems: linear and non-linear elastic, plastic, viscoplastic either static or
stationary or dynamic. For a generic structural problem — whatever the
geometry, the loads and the materials involved— three main conditions are
always present:

ˆ equilibrium condition:
it relates the stress to itself and the applied forces. If the displacements
are small, the equilibrium equations are linear;

ˆ compatibility:
it relates the strain to the displacements, in a geometric way. If the
displacements are small the compatibility equations are linear;

ˆ constitutive relation:
it is an empirical law, as the elastic or the thermal one, that defines
the constitutive equation of the system.

The software looks for a displacement field satisfying the above conditions.
The found solution is not an exact one, but it gives a good approximation
5.1 The basis of finite element analysis 93

of the real one1 , in the analysis of problems for which obtaining the solution
with the classical methods in an analytic way is, when possible, very difficult.

5.1.1 The principle of virtual displacements

The method adopted inside the FEA for solving the equilibrium equations is
based on the principle of virtual displacements (PVD). The PVD states that
if the system is subjected to a displacement field (arbitrary but compatible
with the vincula) w.r.t. the equilibrium configuration, the work done by the
internal forces equals the work done by the external ones. For a generic solid,
subjected to the volume force Fi and to the surface force Pi , this means
Z Z Z
σij uij dV = Fi ui dV + Pi ui ds , (5.1)
V V S

where ui , σij and uij are the displacements, the stress matrix and the strain
matrix of the system respectively. Introducing in the previous equation the
compatibility relation between the strain uij and the displacements ui

1
uij = (∂i uj + ∂j ui ) (5.2)
2

and using the Gauss theorem in the form2

Z Z


σij ∂i uj + uj ∂i σij dV = σij uj ni ds , (5.3)
V S

it follows
Z Z



ui ∂j σij + Fi dV − ui σij nj − Pi dS = 0 , (5.4)
V S

where ni are the components of the unit vector normal to the surface of inte-
gration. Assuming u(r) is an arbitrary everywhere-continuous function, its
1
The accuracy of such an approximation depends on the refinement of the mesh as it
will be explained in § 5.1.4.
2
An important requirement for the theorem to hold is that ∂i uj and ∂i σij are finite
inside the volume V of integration, i.e. the fields σ and u are continuous.
94 Chapter 5. Using finite element analysis

coefficients must vanish in V and S to satisfy the previous relation, bringing

to the equilibrium equations

∂i σij + Fj = 0 in V and (5.5)

ni σij = Pj in S. (5.6)

The same equations can be obtained also imposing the system to be in a

minimum of the potential energy. However, especially in the context of FEA,
the PVD method reveals its power.

5.1.2 The elements

It is simple to solve integrals of known functions if an approximated method

is used. This is the knot of the finite element method when it is used in
a structure on which the displacements field is characterized by continuous
but unknown quantities; the first step is to divide the structure into parts
— finite elements — so tiny that they can suitably mock the real variation
of the involved quantities.
The integral necessary to solve the structural problem can be substituted
by a sum over functions evaluated on the elements. Therefore the finite
element method can be simply viewed as a strategy to numerically solve
integrals otherwise too complex — if not even impossible — to solve in a
closed form. Moreover, for solving the integral, it is necessary to hypothesize
the form of the displacements in the interior of each element: form functions
are used to interpolate these values to obtain an everywhere defined solution
to be used in the algorithm. Generally a simple kind of form function is
used (linear or quadratic) and the precision of the solution is assured by the
refinement of the mesh.
5.1 The basis of finite element analysis 95

Element types

There are many types of elements that can be used, depending on the geom-
etry of the system and on the kind of analysis that has to be performed.
The first thing characterizing an element is the dimensionality of the
space in which it is immersed: it can be a 2-D or a 3-D space.
A 2-dimensional element is useful in problems in which the structure
extends essentially on a plane and there are no stresses in the orthogonal
direction. It can also be used for modeling structures in which there is a
rotational or translational symmetry along one axis. In these kinds of model
the analysis is performed on one section only, introducing ad hoc boundary
conditions to take into account the actual behavior of the whole structure.
In all the other cases — and this also happens in the work performed in
this thesis — elements in 3 dimensions are used. They divide into various
categories depending on the dimensionality of the internal structure of the
element itself:

ˆ line elements. They are used to model strings, bolts, preloaded bolts,
tubular members or any long, slender members where only membrane
and bending stresses are needed. They have 2 or 3 nodes.

ˆ shell elements. They are used to model thin panels or curved surfaces.
The definition of thin depends on the application but, as a general
guideline, the major dimension of the shell structure should be at least
10 times its thickness. They have a rectangular shape — with 4 or 8
nodes —, that can degenerate to a triangular one.

ˆ solid elements. They are used for structures which — because of ge-
ometry, materials, loading or detail of required results — cannot be
modeled with simpler elements. They are also used when the model
96 Chapter 5. Using finite element analysis

geometry is transferred from a 3-D CAD system and a large amount of

time and effort is required to convert it to a 2-D or shell form. They
have a brick shape — with 8 or 20 nodes —, possibly degenerating into
tetrahedrons, pyramids and prisms, or a tetrahedral shape — with 8
nodes —.

Each element is associated with a set of values that defines the degrees of
freedom w.r.t. which the FEA has to be solved. A thermal element type,
for example, has one d.o.f., which is the temperature, whereas a structural
element type may have up to six d.o.f., i.e. three translations and three
rotations. The most commonly used element types of the finite element
software ANSYS R 8.1 [103] are reported in table 5.1. All the reported elements

2-D Solid 3-D Line 3-D Shell 3-D Solid

BEAM4 SHELL63 SOLID45

Linear PLANE42∗
BEAM188 SHELL181∗ SOLID185∗
SOLID95
PLANE82∗ SHELL93
Quadratic BEAM189 SOLID92
PLANE2 SHELL99
SOLID186∗

Table 5.1: Commonly used structural element types of ANSYS R software.

Stars indicate elements suitable for non-linear analyses.

are good for describing the behavior of linear materials; element types with a
star must be used for non-linear analysis. Elements PLANE2 and SOLID92
have a triangular and a tetrahedral shape respectively. Element SHELL99 is
designed for shells composed of a great number of layers (up to 250).

5.1.3 The mesh

Half of the efforts in doing a FEA concerns the realization of a good finite
element model.
5.1 The basis of finite element analysis 97

After choosing the various element types and after defining the actual
properties of the involved materials, the model has to be divided into a sort
of grid, called mesh: each element is assigned a definite part of space and
close elements are related to each other in the required way. In the last
years a lot of improvements have been done in order to supply routines that
automatize as much as possible this step of the modeling. The program has
a specific engine, the mesher, that decides on an optimal disposition for the
elements.
It is up to the user to pilot the mesh toward the desired configuration. If
the mesher is run without any particular setting, it will generally construct
a mesh of nearly identical in volume elements by using tetrahedral ones for
the volumes and triangular ones for the areas. The user has many ways of
modifying the building of the mesh, in order to optimize the extraction of the
needed quantities, to simplify the application of boundary conditions and to
reach the required level of precision for output results. He can choose that
elements volumes are in a particular range and that lines or areas are divided
into elements only in a certain number of parts or following a defined scheme.
He can also modify the mesh after its constitution choosing to refine it at
some locations.

5.1.4 Checking the FEA results

The hypothesized form functions introduced in § 5.1.2 could not represent

what is happening really and approximation errors can occur, independently
of what kind of element has been chosen for the FEA. Out of balance forces
can appear among the applied loads and the reaction of the elements. This
is why it is necessary to do ”sanity check” to make sure the solution is
acceptable.
Generally it is very useful to verify if the results agree with hand calcu-
98 Chapter 5. Using finite element analysis

lation or experimental data, especially for the displacement solution, from

which all the other results are derived. If this kind of calculation is not
possible on the actual structure, it is possible to perform it on a model that
embeds all the physical phenomena under consideration but that is simplified
in the geometry (fewer objects of simpler shape).
The reaction forces should balance the applied loads and the mesh should
be adequately refined. Criteria for understanding if the mesh is correct are
described in the following.

Mesh error

The residual error can be physically interpreted as a continuous and artifi-

cial load — or constraint — that forces the element to maintain, e.g. in a
structural analysis, the approximate deflection. If it is imagined to gradu-
ally remove this constraint and to leave the system moving toward the true
equilibrium position, a positive work would be exerted by the applied forces.
Therefore the elastic potential energy of the system increases releasing the
constraints: the deformation energy of a structure, when it is an approximate
solution obtained using the PVD, is less than the real deformation energy.
The solution is called a lower bound solution and the structure can appear
more rigid than the real one. By increasing the degrees of freedom, a con-
vergence of the results toward the real value occurs. That phenomenon is
referred to as mesh convergence. It is important to notice that the limit of
the real value is on the mean of the total energy and not on the tension and
displacements in a point; locally the tension could be higher than the real
one.
The usual assumptions in FEA lead to a displacement field that is con-
tinuous on the elements and a stress field that is discontinuous. The element
stress σ̂ are obtained using the value inside the chosen element; the nodal
5.2 FEA as a general tool in thermal noise studies 99

stress σ ∗ are obtained as a mean among all the values of the elements con-
curring at the chosen node. The difference

∆σ = σ ∗ − σ̂ , (5.7)

among element and nodal stresses is a value that can be easily computed
in the post-processing and it gives a good indication of the suitability of
the current mesh refinement. To define an error ei of the mesh on the i-th
element3 it is possible to integrate the deformation work done by ∆σ on the
i-th element volume Vi :
Z
1
2
kei k + ∆σij E −1ik ∆σkj dV , (5.8)
2 Vi

where Eij is the elasticity matrix of the material. Therefore an estimation of

the percentage error etot on a series of selected elements is given by4
 12
kei k2
 P
i
etot = 100% P 2
P 2
, (5.9)
i kui k + i kei k

where kui k2 is the deformation energy of the i-th element. As a general rule
of thumb, etot should be under 10%. The elements of the structure where
this threshold is overtaken are good candidates for a mesh refinement5 .
Another way to check mesh adequacy is to plot the element stresses and
look for elements with high stress gradients.

5.2 FEA as a general tool in thermal noise

studies

Finite element analysis is a very useful tool in solving several theoretical and
experimental problems in the framework of thermal noise studies. A list of
various cases in which FEA can be used is here shown.
3
This value in ANSYS R is called SERR.
4
This value in ANSYS R is called SEPC.
5
Since the quantity etot is always highest at stress singularities, it is necessary to uns-
elect the relative elements before investigating the error distribution.
100 Chapter 5. Using finite element analysis

5.2.1 Resonance modes extraction

Frequency and shape of resonance modes of a system can be extracted per-

forming a modal analysis on its finite element model. There are many prac-
tical situations in which such kind of analysis can be very useful.

Ring-down experiments

In a class of experiments aiming at measuring quality factors of materials,

the ring-down technique is used. The quality factors Q are extracted using
Q = ωr τ /2, performing measurements on the time decaying constant τ in ring-
down of oscillations at the various resonance frequencies ωr of the system.
Therefore, especially for quite complex systems, for which the analytical
calculation is not straightforward, it is useful to know the exact values of
the resonances in order to excite them without losing time in looking for
them. In fact, if the quality factor of the resonator is very high the spectral
amplitude of its resonances is very sharp and it can be difficult finding them
with a white noise or a swept-sine excitation.
The strain energy E of each mode can be expressed as [30]
Z Z
1 2 3 1
E = µ (uik − δik ull ) d r + K u2ll d3 r = Esh. + Eh.c. , (5.10)
V 3 2 V

where the integrals have to be performed on volume V of the sample. The first
addendum of (5.10) is due to a pure shear (distortion without volume change)
and the second to hydrostatic compression (volume change without shape
modification). The quantities K and µ are called modulus of hydrostatic
compression and modulus of rigidity respectively and they are related to
Young’s modulus E and Poisson’s ratio σ by
E
µ= , (5.11)
2(1 + σ)
E
K= . (5.12)
3(1 − 2σ)
5.2 FEA as a general tool in thermal noise studies 101

It can happen that the quality factor is related to whether the energy is
stored in compression or shear. With a FEA it is possible to compute, from
the strain distribution of each mode, the compression part and shear part of
the strain energy.

Interferometer read-out

In interferometer mirrors like the ones used in Virgo only certain modes give
a non-zero contribution to the effective displacement read through the laser
beam. The gaussianity of the laser-beam profile makes the interferometer
almost insensible to mirror surface movements relative to modes that have
radial node-lines: during an oscillation the shortening of the light path hap-
pening on one side of the nodal line is exactly compensated by the lengthening
happening on the other. Therefore it is important to identify the modes that
can represent, with their excitation, a noise for the interferometer.

§
In 5.4.1 there will be shown the analysis done with a FEA on the resonances
of a mirror like the one used in Virgo.

5.2.2 Shape parameter extraction

As explained in § 3.2.3, superficial losses could be dominant in ring-down

experiments and they depend on geometries and mode shapes. The size of
the effect is described by (3.57), in which a correct calculation of µ becomes
crucial for a precise estimate of the superficial losses. According to [37]
R
V S u2ik (r)d2 r
µ= R 2 , (5.13)
S V uik (r)d3 r

where S, V and uik (r) are the surface on which the dissipation occurs, the
volume of the sample and the strain induced by the considered mode at the
position r on the sample. This calculation, especially for higher modes and
102 Chapter 5. Using finite element analysis

for complex sample shapes, can be hard to perform. In § 5.4.2 FEA will be
used to calculate the parameter µ in the case of a mirror like the one used in
Virgo.

5.2.3 Brownian thermal noise in complex systems

§
Nowadays, as it was explained in 3.3.3, the most used technique for thermal
noise calculation is Levin’s direct approach. Numata [18] was the first to
use this approach in combination with a FEA performed to calculate the
mean dissipated energy to be used in Levin’s formula (3.104). He did the
analysis for an axisymmetric mirror including the loss of a 1-layer coating
§
and, separately, of the magnets. In 5.4.3 there will be shown how to perform
a complete analysis of the mirror as a whole, including the presence of a multi-
layer coating, magnets, spacers and markers. The result obtained for the
north-input (NI) and the north-end (NE) mirrors of the Virgo interferometer
will be shown. It is the first time that such an analysis has been performed.

5.2.4 Parametric analyses

Thanks to detailed finite element models like the one developed hereafter
for the mirror, it is possible to perform calculations previously impossible,
investigating behaviors of the thermal noise budget w.r.t. arbitrarily chosen
variables. That is very important because, on following the same scheme of
finite element model creation, it will be possible, during the design phase of
the future mirror, to chose the most suitable configuration for minimizing
thermal noise contributions.
As an example, in § 5.4.4 there will be analyzed the effect of markers
position on the thermal noise of the silicate bonding layer used to attach
markers on the mirror surface.
5.3 Finite element model of a Virgo mirror 103

5.2.5 Analysis on thermoelastic noise

In § 3.2.2 the thermoelastic damping process has been introduced and in

§ 4.1.2 and § 4.2.2 its effect on the bulk and the coating thermal noise has
been reviewed. Following the method introduced by Liu and Thorne [56] and
§
described in 4.1.2 on page 67, the thermoelastic noise can be evaluated once
the average dissipated power Wdiss is known. It is possible to calculate Wdiss
through a finite element analysis using thermal elements associated with the
structural ones but, because of the complexity of the problem and the lack
of a specific ANSYS R element type, an actual calculation on realistic mirrors
has not been performed yet.
In chapter 6 it is shown how a FEA can enter the study of thermoelastic
noise in crystalline silicon fibres.

5.3 Finite element model of a Virgo mirror

The first step of a FEA on a Virgo mirror is the construction of a finite

element model. The ANSYS R 8.1 [103] finite element software was used and
the analysis is performed on the mirrors of the north-arm Fabry-Perot cavity
which will be referred to as north input (NI) and north end (NE). For a
complete modelization, all the components of the mirror used in the actual
interferometer should be implemented: bulk , coating, magnets, markers and
spacers.

5.3.1 The bulk

It consists of a single fused silica cylinder 350 mm in diameter and 96 mm

thick for both NE and NI mirrors. NI mirror is made of Suprasil and NE of
Herasil. The lateral surface is filed in order to obtain rectangular flat areas
for spacer settlement. The volume is modelized with tetrahedral SOLID92
104 Chapter 5. Using finite element analysis

elements created with a free-mesh after the mesh of all the other components.
In this way the mesher is forced to create transition zones between the high-
definition regions and the low-definition ones.

5.3.2 The coating

It consists of alternating layers of Ta2 O5 and SiO2 , materials with high and
low refractive indexes respectively. On the NI the coating diameter is 100 mm
and on the NE it is 280 mm because the transversal dimension of the resonant
mode increases along the Fabry-Perot cavity reaching its maximum at the
NE mirror. The composition of the two coatings for the mirrors are reported
in table 5.2. The NI has four extra layers on the surface; their specifications
are reported in table 5.3. The mesh is driven in a particular shape in order

# of layers
Type Thickness
NI NE
High 130.4 nm 17 4
Low 182.2 nm 18 4

Table 5.2: Coating layer constitution for the two mirrors.

NI extra layers

Type Thickness # of layers

High 100.8 nm 1
Low 252.7 nm 1
High 121.0 nm 1
Low 539.1 nm 1

Table 5.3: Characteristics of the four extra layers of the NI mirror.

to simplify and optimize the application of a pressure with a Gaussian profile.

First of all a series of circular lines is created with a definite spacing factor,
5.3 Finite element model of a Virgo mirror 105

in order to have a better spatial resolution in the central area. These lines,
together with 4 other radial straight lines, identify a set of areas (figure 5.1 a).
Secondly the mesher is forced to follow a definite pattern of line division for
the node creation (figure 5.1 b). Finally the mesh is created using SHELL99
elements (figure 5.1 c). In the central part of the coating the mesh is simply

Figure 5.1: Steps of the mesh-driving procedure followed for creating the
shell element of the coating on the mirrors. Area creation (a), lines division
forcing (b), mesh with SHELL99 elements (c).

completed with triangular elements in order to have only 7 extra nodes (see
figure 5.2).

Figure 5.2: A magnification of the central part of the coating mesh. Only 7
nodes are added to complete the coating model.
106 Chapter 5. Using finite element analysis

5.3.3 Magnets, markers and spacers

Figure 5.3: A view of the various components on the mirror: magnets (in
blue), markers (in red) and spacers (in green).

Magnets: they are four elements (blue colored in figure 5.3) attached
to the backside of the mirror and they consist of two parts: a cylin-
dric support made of SiO2 and a cylindrical magnet (simulated with a
material with the same properties as C85 steel). A view of a meshed
magnet is shown in figure 5.4 a.

Markers: they are located on the front-side of the mirror, external to

the coating and they are four simple cylinders made of SiO2 (red colored
in figure 5.3). A view of a meshed marker is shown in figure 5.4 b.

Spacers: they are four prisms (green colored in figure 5.3) with a
triangular base, attached along a lateral face on the flattened area of the
mirror lateral side. A view of a meshed spacer is shown in figure 5.4 c.
5.4 Application of FEA to a Virgo mirror 107

Figure 5.4: A view of a meshed magnet (a), marker (b) and spacer (c).

Silicate bonding areas: magnets, markers and spacers are attached

to the bulk by means of the silicate bonding process. The interface
region is modeled with SHELL99 elements formed by a 100 nm thick
layer of SiO2 .

The characteristics of the materials used are reported in table 5.4 A view of
the complete NI mirror is shown in figure 5.5.

Material ρ [kg/m3 ] E [GPa] σ

SiO2 2202 73.2 0.164

Ta2 O5 8316 140 0.23
C85 7900 200 0.3

Table 5.4: Characteristics of the materials used in the simulation: density ρ,

Young’s modulus E and Poisson’s ration σ.

5.4 Application of FEA to a Virgo mirror

§
Hereafter, using the model of the Virgo mirror described in 5.3, there will
be a review of the FEA’s performed by the candidate in the various research
108 Chapter 5. Using finite element analysis

Figure 5.5: A view of the complete finite element model of the NI mirror.

§
areas anticipated in 5.2.

5.4.1 Resonance modes extraction

A modal analysis has been performed on the finite element model of the
§
mirror described in 5.3. A plot of the found resonances w.r.t. the sequential
mode number is shown in figure 5.6 to give an idea of the plethora of modes
that characterizes this system. It is interesting to analyze the shape of the
resonance modes.
The modes can be catalogued depending on oscillation-directions of the
system points (along x̂, ŷ, ẑ working in Cartesian coordinates or r̂, ϑ̂, ẑ in
cylindric ones or r̂, ϑ̂, ϕ̂ in spherical ones). The choice of the coordinate sys-
tem depends on the easiness of the mode-pattern description. If not specified
otherwise, the cylindric coordinate system will be used. Each pure mode is
5.4 Application of FEA to a Virgo mirror 109

Figure 5.6: Resonance mode frequency of the NI mirror versus the sequential
mode number, plotted for the first 356 modes.

labeled with a capital letter, indicating the oscillation direction, and by a set
of three numbers indicating the behavior of the oscillation (which is simply
related to the number of the nodal planes) along the three coordinate direc-
tions. For example, Zr,ϑ,z refers to a mode along ẑ with r nodal cylindric
surfaces ⊥ r̂, with ϑ nodal radial half-planes6 (⊥ ϑ̂) and z nodal planes ⊥ ẑ.
In 2-D and 3-D, objects generally do not produce pure modes, because
they couple with modes along orthogonal directions having other kinds of
symmetry. For example (see figure 5.7), consider the first mode of the mir-
ror in cylindrical coordinates called butterfly mode: it is practically a pure
Z0,2,0 but it is slightly coupled with R0,2,1 and Θ0,2,1 modes, which are along
orthogonal directions w.r.t. ẑ and with a (r = 0, ϑ = 2, z = 1) symmetry.
Sometimes the coupling is very high and all the pure concurring modes are

6
The nodal surfaces for Rr,ϑ,z and Θr,ϑ,z are the same as for Zr,ϑ,z with the exception
of r label of Θ, which corresponds to r/2 cylindrical surfaces for r even and to (r −1)/2
cylindrical surfaces for r odd.
110 Chapter 5. Using finite element analysis

Figure 5.7: The first mode of the NI mirror in a three-quarter view (a) and
in a front view (b). It is evident that the motion is mainly along ẑ but that
it slightly couples with movements along r̂ and ϑ̂.

strongly present. See, e.g., the 6th mode (figure 5.8): the two pure modes
R0,2,0 and Θ1,2,0 are both present as shown in the diagram of figure 5.9 a.

Figure 5.8: Three-quarter view of the 6th mode of the NI mirror.

Whatever the coupling is, the resulting symmetry is the composition of

5.4 Application of FEA to a Virgo mirror 111

+ =

a) R 0,2,0 Θ1,2,0 R 0,2,0+ Θ1,2,0

b) 2 planes 2 planes 5 axes

1 cylinder

Figure 5.9: Diagram of the superimposition of the R0,2,0 and Θ1,2,0 pure
modes forming the 6th mode. Nodal surfaces are indicated with thick lines,
nodal axes with thick points (a). The composition of the nodal surfaces is
shown, giving birth to 5 nodal axes (b).

the pure-mode symmetries. In figure 5.9 b the composition of the 2 pure-

modes symmetries is shown for the 6th mode. The intersection among the
mode nodal-surfaces is taken, resulting in 5 nodal axes.
In table 5.5 the pure modes constituting the first 41 modes of the FEA
mirror are reported (the bold font indicates the dominant ones). Some modes
appear on the same row because they go into each other7 with a rotation
along ẑ.

7
This is not strictly true because of the presence of magnets, markers and spacers that
break this symmetry. By the way, the the modes are so slightly different from each other
that the two frequencies are nearly equal and it makes sense to consider them part of the
same doublet.
112 Chapter 5. Using finite element analysis

Mode Eh.c.
Pure modes Frequency zout Name
number Etot
1,2 R0,2,1 Θ1,2,1 Z0,2,0 3879, 3883 .106 7×10−4 aB2
3 R0,0,1 — Z1,0,0 5544 .736 1 aD1
4,5 R0,3,1 Θ1,3,1 Z0,3,0 7549, 7552 .197 1×10−6 aB3
6,7 R0,2,0 Θ2,2,0 Z0,2,1 8042, 8048 .019 3×10−5 sB2
8,9 X1,1,0 Y2,0,0 Z0,1,1 9021, 9023 .458 5×10−8
r,ϑ,z
10,11 X2,4,1 Y1,1,1 Z 1,1,0 10121, 10124 .673 2×10−5
12 R0,0,0 — Z1,0,1 10379 .737 7×10−2 sD1
13,14 R0,4,1 Θ1,4,1 Z0,4,0 11308, 11321 .207 2×10−3 aB4
15,16 R0,3,0 Θ2,3,0 Z0,3,1 12297, 12299 .103 7×10−8 sB3
17,18 R0,2,0 Θ3,2,0 Z0,2,1 13900, 13926 .566 2×10−6
19,20 R1,2,1 — Z1,2,0 14434, 14465 .594 8×10−4
21,22 R0,5,1 Θ1,5,1 Z0,5,0 15072, 15079 .205 2×10−5 aB5
23 R1,0,1 — Z2,0,0 15606 .651 5×10−1 aD2
24,25 R0,4,0 Θ2,4,0 Z0,4,1 15938, 15961 .154 1×10−2 sB4
26 R1,0,0 Θ2,0,0 — 17647 .000 4×10−6
27,28 R1,3,1 — Z1,3,0 18412, 18415 .587 3×10−5
r,ϑ,z
29,30 X2,4,0 Y1,1,0 Z1,1,1 18620, 18622 .546 3×10−8
31,32 R0,6,1 Θ1,6,1 Z0,6,0 18788, 18810 .194 6×10−4 aB6
33,34 R0,3,0 Θ3,3,0 Z0,3,1 19106, 19115 .406 1×10−6
r,ϑ,z
35,36 X2,2,1 Y1,1,1 Z 2,1,0 19172, 19173 .516 3×10−5
37,38 R0,5,0 Θ2,5,0 Z0,5,1 19360, 19363 .179 8×10−7 sB5
39 R0,0,1 Θ0,0,1 — 19719 .001 2×10−6 aT1
r,ϑ,z
40,41 X1,1,1 Y2,0,1 Z1,1,1 21381, 21389 .285 5×10−6

Table 5.5: The first 41 mode frequencies of the FEA mirror are listed together
with the pure modes forming them (the bold font indicates the dominant
ones). The modes appearing in pairs are doublets, connected by simple
rotations along ẑ. The 4th column reports the ratio between hydrostatic
compression energy Eh.c. and total energy Etot of√the mode. In the 5th column
the effective read-out displacement per unit of Etot is shown, referring it to
3rd -mode displacement. The last column indicates the common name of the
mode: “B”, “D” and “T” stand for “butterfly”, “drumhead” and “torsional”
respectively; “a” and “s” stand for asymmetric and symmetric.
5.4 Application of FEA to a Virgo mirror 113

Ring-down experiments

Frequencies obtained with FEA have to be considered reliable within 1 ‡;

therefore they are very useful as a guide to find the resonances to be excited
in a ring-down experiment. In table 5.5 the frequencies of the first 41 modes
of the mirror are reported as obtained with a modal analysis performed with
ANSYS R.
§
As explained in 5.2.1, if the damping of the modes can depend on how
the energy is distributed between shear and compression, it is relevant to
calculate these quantities for each mode. Shear energy Esh. and hydrostatic
compression energy Eh.c. are obtained by the integrals over mirror volume
defined in (5.10). Obviously in FEA these integrals reduce to a sum of
products among volumes Vn and strain components uij,n of each element:
X 1 X 1
Esh. = µ (uik,n − δik ull,n )2 Vn = µ (u2ik,n − u2ll,n )Vn , (5.14)
n
3 n
3

1 X 2
Eh.c. = K ull,n Vn , (5.15)
2 n

where µ and K are defined in (5.11) and (5.12). These values were calcu-
lated from ANSYS R output, using AWK [104], a pattern-directed scanning and
processing language.
In table 5.5 the ratio between Eh.c. and the total energy of the mode is
reported. This value is almost zero (energy completely stored in shear) for
the torsional modes (e.g. 26th and 39th ), it is below 0.2 for the butterfly modes
and it is distributed around 0.6 for the others, with the highest values of over
0.7 for the first drumhead modes, indicating a great hydrostatic compression.

Interferometer read-out

As regards the influence of mirror resonances on interferometer read-out, it

is possible to calculate, for each mode, the neat displacement read by the
114 Chapter 5. Using finite element analysis

Gaussian laser beam at the maximum of oscillation. The beam profile is

simulated with forces Fn acting on the mirror-coating nodes reproducing
a Gaussian pressure-profile as it will be explained in § 5.4.3. The effective
displacement zout , normalized to the displacement of the 3rd mode, is obtained
for each mode by the coating-nodes displacements zn using
P p
zn Fn Etot,3
zout = √ P , (5.16)
Etot zn,3 Fn,3

where Etot is the total energy stored in the mode and the index “3” refers
√
to the 3rd mode. The factor Etot is necessary to consider modes as having
the same stored energy; the square root takes into account that energy is a
quadratic form of displacements.
The obtained values, referred to the 3rd -mode displacement, are listed in
table 5.5: it is evident that among the first 41 modes the most contribut-
ing ones are the three drumhead aD1, sD1, aD2, followed — two orders of
magnitude further down — by the butterfly modes.

5.4.2 Shape parameter extraction

In § 5.2.2 the importance of the shape parameter µ defined by (5.13) has

been stressed. A FEA has been performed to calculate this parameter from
the element strain distribution.
The particular conformation of the actual mirror forces to consider two
distinct sources of superficial damping: the coating surface and all the other
uncoated surfaces. Therefore (3.56) now becomes
c uc
∆Esurf ∆Esurf ∆Esurf Sc S uc
= + = µc dsc + µuc dsuc . (5.17)
∆Ebulk ∆Ebulk ∆Ebulk V V

where the notation is the same as in (3.56) and the superscripts “c” and
“uc” refer to coated and uncoated surfaces respectively. In order to calculate
µc and µuc with formula (5.13), superficial strain energy density for coated
5.4 Application of FEA to a Virgo mirror 115

and uncoated surfaces is needed. ANSYS R software does not allow access to
superficial densities directly, but for the coated surface the strain is almost
constant through the coating SHELL elements, in the direction normal to
the surface, so that
R R
u2ik (r)d2 r u2ik (r)d3 r
P 2
uik,n
Sc
' Vc
= Pn , (5.18)
Sc Vc n Vn

where the integral in the numerator has moved from the coating surface to
the coating volume and finally it has become a simple sum over the coating
elements. As regards the uncoated surfaces, the finite element model of the
mirror has been slightly modified covering them with a layer of fused silica
0.1 µm thick.
In table 5.6 the calculated µc and µuc parameters for the first 41 modes
of NI and NE mirrors are reported. The values for µc are more variable
(from 0.0 to 6.7) w.r.t. the ones for µuc (from 0.4 to 2.3). This is due to
the fact that µc is extremely sensitive to modes that, for their particular
shape, have a lot of energy in the center of the mirror; on the other side µuc
mediate on a larger surface, with parts either near or far w.r.t. the center of
the mirror. Moreover, focusing on µc , low values (below 2) for the NI mirror
systematically correspond to higher values for the NE one and vice-versa,
except for the 10th and 11th modes. This is related to the different coating
radius (28 cm for NE and 10 cm for NI) that allows the parameters to be
more or less sensitive to the energy stored in the center or the peripheral
parts of the mirror.

5.4.3 Brownian thermal noise calculation

Brownian thermal noise calculation for the NI and NE mirror of Virgo ex-
periment has been performed using FEA. The formula (3.104), reported also
116 Chapter 5. Using finite element analysis

Mode µc µuc
number NI NE NI NE
1,2 6.70 4.68 1.89 1.29
3 10.50 5.26 1.68 1.11
4,5 1.02 3.71 2.06 1.49
6,7 4.46 2.03 0.60 0.37
8,9 0.28 1.22 1.06 0.87
10,11 2.68 4.58 1.55 0.92
12 2.77 1.79 0.70 0.49
13,14 0.16 3.05 2.13 1.65
15,16 1.03 1.94 0.80 0.51
17,18 0.83 0.82 1.26 1.15
19,20 1.49 3.88 1.48 0.93
21,22 0.03 2.57 2.22 1.81
23 5.18 4.13 1.32 0.80
24,25 0.20 1.85 0.98 0.69
26 0.20 1.83 0.66 0.37
27,28 0.55 3.17 1.48 1.01
29,30 0.96 1.64 0.56 0.34
31,32 0.00 2.20 2.32 1.96
33,34 0.26 0.57 1.32 1.23
35,36 3.17 1.99 0.69 0.45
37,38 0.04 1.75 1.16 0.88
39 0.00 0.00 0.73 0.73
40,41 0.34 0.49 1.06 0.99

Table 5.6: Shape parameters for the coated (µc ) and uncoated (µuc ) surfaces
of NI and NE mirror for the first 41 modes.

here for clarity,

8kB T hWdiss i
SX (ω) = , (5.19)
ω2 F02

gives the power spectrum of the Brownian thermal noise SX (ω) from the
average dissipated power Wdiss by the mirror subjected to a pressure distri-
bution (of total force F0 ) that mocks the laser power profile, as explained in
5.4 Application of FEA to a Virgo mirror 117

§ 3.3.3.
Virgo laser has a total power P0 = 20 W, distributed with a Gaussian
profile P (r, ϑ) expressed by
2
P0 − r 2
P (r, ϑ) = 2 e r0 , (5.20)
πr0
√
where r0 is the beam radius related to the beam waist w0 by w0 = 2r0 (on
the NI w0 = 2 cm, on the NE w0 = 5.5 cm). The coating has been meshed, as
§
explained in 5.3.2, in order to simplify as much as possible the application
of the pressure profile, that is simulated with a set of forces acting on the
corner nodes of the SHELL elements constituting the coating itself. These
forces are scaled according to the effective area corresponding to each node
on which they act. The parameter F0 is simply obtained by the sum of all
the applied forces.
ANSYS R software allows a harmonic analysis: all the applied forces are
varied periodically with a function sin(2πf t) on a set of defined frequencies
f and the solution of the dynamic problem is stored, for each frequency, in a
result file. From (3.38) it follows that average the dissipated power hWdiss i is
X X
max
hWdiss i = hWdiss,i i = 2πf Estor,i φi , (5.21)
i i

where the sum is intended over the various mirror parts (bulk, coating, mag-
max
nets, markers, spacers) for which Estor,i is the strain energy at maximum
oscillation and φi is the loss angle, reported in table 5.7. Loss angle val-
ues for the mirror bulk are taken from [105] and for the coating from [106].
Magnets, markers and spacers dissipate through the silicate bonding layer
connecting them to the mirror surface; silicate bonding loss angle is taken
from the measurements done in [107]. Formula (5.21), plugged in (5.19),
gives the total power spectrum of Brownian thermal noise as a sum of the
various mirror-part contributions. In figure 5.10 these quantities are shown
118 Chapter 5. Using finite element analysis

NI mirror NE mirror Silicate

Coating
bulk bulk bonding
Material Suprasil Herasil SiO2 /Ta2 O5 SiO2
φ 9.6×10−7 5.5×10−7 4×10−4 10−2

Table 5.7: Material and loss angle of the various mirror parts contributing
to dissipation.

in strain sensitivity units8 . For comparison it is plotted the prediction of

Figure 5.10: Brownian thermal noise for NI mirror. Contributions for the
various mirror parts are highlighted by different colors. The yellow line in-
dicates the prediction for and uncoated finite-size mirror made by Bondu,
Hello and Vinet (with Liu and Thorne correction).

Bondu, Hello and Vinet [54] corrected by Liu and Thorne [56] for an un-
coated finite size mirror (see (4.9) on page 65): simulated Brownian thermal
noise follows the theoretically predicted trend, being proportional to ω −1/2 .
8
It implies that, for a single-mirror noise, it is plotted the equivalent noise of the
interferometer having, in the two Fabry-Perot cavities, 4 mirrors identical to it. For the
definition of strain sensitivity units see on page 17.
5.4 Application of FEA to a Virgo mirror 119

Figure 5.11: Brownian thermal noise for NI mirror in a linear frequency

scale from 25 kHz to 45kHz. Contributions for the various mirror parts are
evidenced with different colors.

Coating noise is about one order of magnitude lower w.r.t. bulk noise and,
among the other contributions, the most relevant is the one due to markers.
The big peaks correspond to the three drumhead modes (aD1, sD1 and aD2
with the notation of table 5.5); this is in agreement with the high values cal-
culated on those modes for the read-out displacement zout , listed in table 5.5.
An enlargement of the frequency region from 25 kHz to 45 kHz is shown in
figure 5.11. It is evident that with this analysis it is possible to compute
directly the Brownian thermal noise in near-to-resonance regions, including
the coupling effects induced by the inhomogeneous loss distribution.
In figure 5.12 it is shown a comparison among NI and NE mirror ther-
mal noise in interferometer strain units. The factor of about 2 in the bulk
contributions is due to the different loss angle of NI w.r.t. NE mirror (see
120 Chapter 5. Using finite element analysis

Figure 5.12: Brownian mirror thermal noise comparison among NI and NE

mirrors for bulk and coating.

table 5.7). The coatings are contributing similarly with a little difference due
mainly to their diameters (10 cm for the NI and 28 cm for the NE).

5.4.4 The effect of the markers

Calculations of the same kind as the ones described in the previous section
have been performed on NI mirror in order to see the effect of marker position
on Brownian thermal noise.
Markers are settled at a radial distance of 155 mm from the mirror center.
This distance d has been varied from 155 mm down to 62 mm (a position in
which they are almost touching the coating). A harmonic analysis has been
performed in each configuration calculating the contribution of the markers
to the Brownian mirror thermal noise at 100 Hz and at 600 Hz.
The results are shown in figure 5.13. It is evident the increment in thermal
5.4 Application of FEA to a Virgo mirror 121

Figure 5.13: Marker contribution to Brownian mirror thermal noise as a

function of the distance of the markers from the center. The analysis has
been performed for two frequencies inside the Virgo band.

noise budget as the distance of the markers w.r.t. the center of the impinging
beam decreases. This is in agreement with what expected because a source
of loss nearer to the mirror axis can influence the read-out easier. The de-
pendence is roughly linear in 1/d6/5 and is not influenced by the inspected
frequency.
122
Chapter 6

Research on new materials for

suspensions: characterization of
silicon crystalline fibres

This chapter begins with a description of the interesting features of silicon

as a material for mirror suspensions in interferometers of the 3rd generation.
Then there will be the description of the experiment performed by the can-
didate in order to measure the loss angle of crystalline fibres. Finally, there
will be an outline of the foreseen work on the same research line.

6.1 Silicon properties

Silicon thermal and mechanical properties are extremely favorable to reduce

thermal noise in the optics suspensions of a GW interferometric detector.
In the frequency range comprised between the pendulum resonance and the
first violin mode of the suspension fibres, the power spectral density of the
thermal noise displacement of a suspended mirror Sxpend can be deduced by
the formulas (4.42) and (4.41) with the approximation ω  ω0 :
r
pend kB T g gE
Sx = 5 2 φ(ω) . (6.1)
ω l cσb mN π
where the loss angle φ(ω) represents the sum of all the dissipative processes
that occur in the material (structural and thermoelastic losses) plus an ef-

123
124 Chapter 6. Research on new materials for suspensions

fective loss angle for the losses associated with the connecting elements such
as clamps, break-off points or chemical bonded pieces. Silicon is expected to
have large bulk tensile strength (about 7 GPa, dominated by surface effects
that can decrease the value down to about 200 MPa [108]) and comfortably
low intrinsic loss angle — φi (300 K) ' 2.8 × 10−8 , φi (77 K) ' 5 × 10−9 and
φi (4.2 K) ' 6×10−10 [36, 109, 110] —. The behavior of the linear thermal
expansion coefficient α and of the thermal conductivity κ are peculiar1 as
displayed in figure 6.1.

Figure 6.1: Linear thermal expansion coefficient and thermal conductivity

of silicon; the two temperatures (about 18 K and 123 K), where the thermal
expansion coefficient vanishes, and the peak of the thermal conductivity at
low temperature are well evident.

The thermal expansion coefficient decreases with temperature until it

vanishes at about 123 K, it is negative in the temperature interval 18÷123 K
and almost zero at lower temperatures. This means that the thermoelastic
dissipation decreases sharply with temperature and becomes negligible w.r.t.
1
All the thermo-mechanical properties of silicon reported in this paragraph are taken
from [111].
6.1 Silicon properties 125

the structural losses in the range 120÷130 K and below 20 K as reported in

figure 6.2.

Figure 6.2: The amplitude of the linear thermoelastic loss angle in a silicon
fibre, computed using (3.54) and [111] is shown with the solid line. Ideally, at
the two temperatures where the thermal expansion coefficient vanishes φ0 is
null. The expected temperature dependence of thermoelastic peak frequency
1/τth in a 560 µm diameter silicon fibre is also shown (dashed line). The
increase at low temperatures should contribute to reduce the thermoelastic
dissipation in the suspension.

The thermal conductivity κ of silicon increases at low temperatures push-

ing the thermoelastic peak toward higher frequencies according to equations
(3.53) and (3.52). After reaching a maximum, the thermal conductivity
drops as the phonon mean free-path becomes larger than the square root of
the typical cross-section of the sample (see the red curve in figure 6.1). Note
that the thermal conductivity varies greatly with the doping of the material
and with the concentration of the lattice defects, so that the red curve of
figure 6.1 ought to be taken as a qualitative representation of κ(T ).
It is worth to compare the thermoelastic losses in different materials.
Almost all the present GW interferometric detectors use steel wires to sus-
126 Chapter 6. Research on new materials for suspensions

pend the main optics. This choice provides a convenient solution to keep
in position (safely and easily) very expensive optics. In Virgo the wire steel
(C85) [98] and the suspension design [112] have been selected to optimize the
thermal noise control. The intrinsic loss angle of C85 has been measured to
be about 2×10−4 at the best, while many crystalline materials (like sapphire
or YAG) have intrinsic loss angles well below that value. However, these ma-
terials have not been considered for low thermal noise suspensions because
of their large thermal expansion coefficient, which enhances their thermoe-
lastic dissipation according to (3.53) and (3.54). The thermoelastic losses of
different materials are compared in figure 6.3. For instance the advantage in

Figure 6.3: Expected thermoelastic peaks at room temperature (T=300 K)

in fibres (560 µm in diameter) made of: C85 steel (solid curve) [98], sapphire
(dashed curve) [111], fused silica (dotted curve) [98, 111] and silicon (dash-
dotted curve) [111].

using sapphire fibres is clearly nullified by the dominant thermoelastic loss

angle at low frequency. On the contrary, fused silica has an intrinsic loss
6.2 Silicon fibres production 127

angle below 10−7 ÷10−8 (depending on the type of fused silica) and, more
notably, its thermoelastic dissipation is suppressed by a small thermal ex-
pansion coefficient. Finally, figure 6.3 suggests that another material could
be selected to realize low thermal noise suspension, namely crystalline sili-
con. In fact, silicon does have a large thermoelastic peak, but for very high
frequencies. This is due to the high thermal conductivity of silicon. As a
result, at low frequencies, where thermal noise in the suspensions dominates
the noise budget of GW detectors, the thermoelastic dissipation in silicon
fibres could be even lower than in fused silica fibres. This effect, together
with its low intrinsic loss angle [109], singles out silicon as one of the most
promising materials for low thermal noise suspensions in GW detectors.

6.2 Silicon fibres production

Silicon fibres were grown in a dedicated research facility of an INFM Pisa

laboratory. There, a crystal growth furnace that uses the so-called micro
pulling down technique (µ-PD) was set up. Basically the µ-PD method in-
volves downward pulling of a crystal fibre through a micro nozzle placed at
the bottom of the crucible as shown in figure 6.4. The method allows to
grow crystals in shape of fibres, rods or ribbons with diameters ranging from
0.15 mm to 5 mm at widely variable pulling rates [113].
The melt is placed in crucibles made of materials stable at temperatures
as high as the melting point of the target crystalline material. The crucibles
are heated using a radio-frequency (RF) generator. At the first stage a crystal
fibre seed, produced from previous experiments or cut from a bulk crystal
of the corresponding material, is inserted into the crucible orifice. The seed
is then pulled downward using a precise pulling mechanism. Normally the
surface forces do not allow the melt to separate from the bulk material.
128 Chapter 6. Research on new materials for suspensions

Figure 6.4: Schematic diagram for the µ-PD growth apparatus (hot zone
part).

Therefore the melt passes through the nozzle at the bottom of the crucible
and the new fibre is grown. Application of an after heater allows adjustment
of the appropriate temperature gradients under the crucible and therefore
regulation of the position of the solid-liquid interface in the vicinity of the
crucible tip.
The shape and location of the growth interface are among the most im-
portant parameters determining the quality and uniformity of the resulting
crystal. Therefore special attention is normally paid to monitor and detect
the spatial distribution of the temperature gradient in the vicinity of the
phase boundary. CCD camera and monitor are considered to be very good
tools to view the solid-liquid interface and the meniscus region.
The furnace can be evacuated and an inert atmosphere can be inserted.
It has been tested with several materials and about 20 silicon fibres have
already been grown.
6.2 Silicon fibres production 129

6.2.1 Growth process

The starting material was a 5-N-purity silicon in small pieces of a few mil-
limeters. These pieces were inserted in a vitreous carbon crucible about
27 mm high and 18 mm in diameter; it was placed on an after heater of the
same material and everything was held by a zirconate pedestal in a vertical
alumina ceramic tube and was heated using a RF generator. Cylindrical
heat shields made of alumina or zirconate ceramics were placed around the
crucible as shown in figure 6.4.
The calibrated orifice made in the bottom of the crucible was about
0.5 mm in diameter. The crucible temperature was controlled by the power of
the RF coil which is about 80 mm long and 70 mm in diameter with 8 wind-
ings and the maximum operating furnace temperature is around 2400 K. To
avoid the oxidation of the crucible, the fibre was grown in Ar gas of 5-N
purity. Visual observation of the meniscus region, solid-liquid interface and
crystal growing was made by a CCD camera and monitor.

6.2.2 Seeding and growth procedure

Several Si single crystals with different lengths and diameters were grown.
As a first step, a < 100 > oriented thin cut from a Si disk was used as a seed.
The crystals were grown at various pulling rate in the range 0.3÷2 mm/min
and were 0.4÷3 mm in diameter and 40÷310 mm in length.
The growth of crystal fibres is affected by vibrations which become es-
pecially intense at longer length. It was found that the length of crystal
oscillations also depends on the fibre diameter [114]. Disconnections of the
fibre growing from the molten zone were never observed, but the growth pro-
cess was not perfectly stable and some abrupt changes in diameter and/or
temperature happened, probably due to instabilities of the RF generator and
130 Chapter 6. Research on new materials for suspensions

Figure 6.5: Image of the fibre during the growth process.

to reactions between silicon and the crucible that also prevented a good con-
tact between the seed and the melt. Anyway, for most of their length, fibres
show good quality. An image of the growth process for a sample is shown in
figure 6.5.
Some typical Si crystal fibres, grown with different diameters and lengths,
are shown in figure 6.6. The crystal orientation of the fibre was determined
using Laue X-ray diffraction method. From this measurement it is pos-

Figure 6.6: Grown Si crystal fibres about 0.4 mm in diameter, 17 cm and

21 cm long.

sible to say that all the good quality parts of the fibre inspected showed
single crystalline character (figure 6.7), but it was found that the orientation
6.2 Silicon fibres production 131

Figure 6.7: Laue simulation for Si.

Figure 6.8: Absorption spectrum for Si.

changes along the fibre length in every point in which the diameter is not
stable. As a result the fibres are not single crystals along the whole length
132 Chapter 6. Research on new materials for suspensions

but they are composed of several single-crystal parts.

Absorption investigations, which did not show the presence of any con-
taminants within the sensitivity of the apparatus, were also performed (see
figure 6.8).

6.3 Experimental setup

In this paragraph there will be a description of the two experimental appa-

ratuses for measuring the loss angle of excited fibres and for taking a profile
of their shape to be used in the subsequent stage of data analysis.

6.3.1 Loss angle measurement apparatus

The experimental apparatus for measuring the quality factor of the silicon
fibres produced as explained in the previous paragraph, is made of 4 func-
tionally independent parts. The first one allows the clamping in the desired
configuration; the vacuum chamber reduces the air damping to acceptable
values; the electrostatic actuator and shadow-meter permit to manage the
measurement. In figure 6.9 there is a scheme of the whole apparatus. A
description of each of these subsystems will follow.

Vacuum system

The measured loss angle can significantly be affected by the presence of air
around the fibre. In § 3.2.5 it was explained how the loss angle increases
for viscous and molecular air damping. The best way to avoid this effect
is to reduce air pressure: the stand with the clamping system of the fibre
was put on a bench inside a vacuum chamber equipped with bull’s eye and
a pass-through connector.
The vacuum is realized through the action of two pumps arranged in
cascade. A rotative pump brings the internal pressure (measured by a con-
6.3 Experimental setup 133

Figure 6.9: A scheme of the phi-measurement apparatus. Inside a vacuum

chamber (d) there is a stand supporting the clamp (a), the excitation system
(b) and the sensing shadowmeter (c).

duction manometer) to values around 6 · 10−2 mbar; under this condition of

pressure the turbo-molecular pump — whose rotor reaches 600 rpm in about
15 min — can be started up. Thanks to the turbo-molecular pump it is
possible to reach a pressure around 10−6 mbar in about a day. Under this
condition the air friction is negligible (the associated loss angle is less than
10−7 ).
It is worth to note that the two pumps are linked with the vacuum cham-
ber through semi-rigid tubes; for that reason the vibrations induced by the
rotation of their mechanical parts are transferred to the stand: the vibra-
tion of the fibre will present noise structures, especially at 600 Hz and 50 Hz,
which are the two pumps rotational frequencies.
134 Chapter 6. Research on new materials for suspensions

Figure 6.10: The stand of the suspended fibre is shown. In the upper part
there is the clamp, in the middle the excitation system and at the bottom
the sensing shadowmeter.

The stand and the clamp

During the measurement the fibre is in a vertical position being clamped at

the upper end to a rigid stand (figure 6.10).
The excitation and the read-out system can be moved along the fibre axis
and fixed to the stand; the fibre is clamped with two aluminum blocks and
can be up to 27 cm long out of the clamp.
The fibre clamping is of the utmost importance for a good measurement:
if the sample is softly tightened, it can move among the blocks and dissi-
pate energy, if it is tightened too much it can be damaged on the surface,
§
generating other losses. Moreover, as explained in 3.2.4, even if the clamp-
ing system is extremely rigid, energy losses can occur due to the recoil of
6.3 Experimental setup 135

the whole structure. Although the mass of the stand is considerably greater
w.r.t. the fibre mass (µ  1) it is better to avoid measurements near the
resonance modes of the structure (of the order of kHz).

The excitation system

The fibre can be excited at the desired frequency; this is useful, both in a
preliminary stage to determine the resonance frequencies and in the mea-
surement stage to excite the known modes.
The excitation makes use of the polarization of the dielectric material
~ sinu-
that forms the fibre; this is dipped into a high-gradient electric-field E,
soidally variable in time:

~ x, t) = E
E(~ ~ 0 (~x) sin ωt . (6.2)

The fibre polarizes and, on its surfaces, a density charge σp proportional to

the polarization P~ is induced :

σp = P~ · n̂ = 0 (r − 1)E
~ · n̂ , (6.3)

where n̂ is the versor of the fibre surface, 0 and r are the vacuum and
relative dielectric constants. The effect of the electric field is to generate a
local force per unit surface:

F~ ~ = 0 (r − 1)E02 sin2 (ωt)(Ê0 · n̂)Ê0 .

= σp E (6.4)
∆A

The excitation is proportional to the square of sin(ωt), and so, having sin2 (ωt)
a period of π, the effective angular frequency is 2ω. To avoid this discrepancy
between the frequency of the field and of the induced excitation, a constant
~ c is over-imposed. The exciter is composed by 2 electrodes in shape
field E
of compenetrating combs, in a way that two consecutive teeth are linked to
opposite poles of the generator. The fibre is put near the plane of the combs
136 Chapter 6. Research on new materials for suspensions

(figure 6.11), among two teeth: here the gradient is almost maximum and
parallel to the plane itself.

Figure 6.11: Photo of the actuator made by a comb capacitor.

In this way the fibre is forced to oscillate in the direction more convenient
for the detection. The signal used for the excitation has a peak-to-peak
amplitude of 700 V, to which a continuous voltage of 400 V is over-imposed.

The read-out system

The movements of the excited fibre are detected following the displacements
of the fibre shadow, projected by a LED diode on a double photodiode.The
photodiode is separated in two elements by an insensitive gap of about
500 µm; each element is about 5 mm wide. The two photodiodes were ini-
tially placed as shown in figure 6.12 a with the gap parallel to the direction
of the fibre axis. In this configuration the allowed oscillating region for a
good detection is restricted to a little slice, about two fibre diameters wide:
the detection is made extremely difficult, especially for the low frequency
resonance modes for which the oscillating amplitude is quite high. It was
decided to tilt the photodiodes of about 45 degrees and thus the detecting
region is much bigger as shown in figure 6.12 b. If the fibre is exactly placed
at the center of the gap, the difference of the signal from the two elements
6.3 Experimental setup 137

a) b)

Figure 6.12: Two different configuration for the read-out system are shown
with the photodiode gap parallel (a) and tilted (b) w.r.t. the fibre axis. The
gray areas highlight the allowed sensitive regions for a good detection.

is zero; it becomes negative or positive if the fibre moves and its amplitude
and frequency mock the fibre oscillation.
It is worth to note that the read-out system detects the movements of
the fibre at only one point, and gives no information on the whole motion
of the fibre. It can happen that, for certain resonance modes, the detection
point shows small or even null displacements (if it is a nodal point or if the
movements are on the same direction of the LED light); in this case, the
relative frequency does not admit a Q-factor measurement.
A sketch of the output line is shown in figure 6.13. The two photodiodes
are put in a push-pull configuration in order to subtract their signal and the
difference is filtered and amplified.

Vout

Figure 6.13: Sketch of the acquisition system circuit.

138 Chapter 6. Research on new materials for suspensions

The cables with the output voltage from the photodiodes come out of the
vacuum chamber by means of a feed-through; inside the chamber they are
bound by aluminum foils put at a reference voltage, to shield them from the
electro-magnetic noises induced by the surrounding electric currents (espe-
cially the ones in the high voltage cables of the excitation system).
The LED diode is fed from outside; the power supply is stabilized to
reduce at minimum the fluctuation introduced in the reading. The power
voltage is tunable and it is around 6 V.

Data acquisition

The output of the filter is referred to the ground of the system and goes on a
single ended port of the computer acquisition board. The board is driven by
a LabVIEW R 6.0 [115] program that samples the signal and converts it with a
16-bit ADC. The program allows to choose the internal clock frequency of
the acquisition (number of sample per second); this value has to be chosen
high enough to avoid aliasing in the acquired signal.
In order to measure the fibres quality factor, several programs have been
realized with LabVIEW R. In each of them the basic part is the sampling
system and the digitalization of the measurement (DAQ) that acquires N
samples at a frequency fc in a time N/fc .
In a centering program the signal is acquired continuously in time and
followed with a real-time plot. This is useful to accurately position the fibre
in front of the photodiode and to verify the whole acquisition system.
A test program acquires the signal in subsequent sets of N samples and
filters them with a digital band-pass filter around a chosen frequency. The
signal is followed in the excitation phase at the given frequency ωr and in the
subsequent phase of excitation interruption; the behaving of the oscillation
allows to verify the actual presence of a resonance. If this is not the case,
6.3 Experimental setup 139

when the excitation is suspended, the signal abruptly decreases instead of

decaying exponentially.
The program that manages the standard measurement enables the DAQ
procedure to start as soon as the excitation of the fibre is interrupted.
The acquisition extends for a time at least of the order of the characteristic
damping time τ :
2Q
τ= . (6.5)
ωr

With the acquired samples vector, the program extracts the Fourier spectrum
through the FFT algorithm (Fast Fourier Transform), identifies the peak of
the excited resonance and calculates the frequency finding the centroid with
a Gaussian fit.
The sampled signal is then filtered sharply around the resonance fre-
quency with a digital filter in order to isolate that Fourier component only.
This component is shown in a plot and its amplitude shall follow an expo-
nential damping.
To obtain an estimate of the quality factor of the single measurement the
program makes a Hilbert transform x̂ of the filtered signal x(t):
Z +∞
1 x(s)
x̂(t) = − ds . (6.6)
π −∞ t−s

If the Fourier transform is referred to with a superscript tilde, the following

identity

x̂(ω)
e = −i sign(ω)x̃(ω) (6.7)

holds. Thus the Hilbert transform yields a dephase of −π/2 for the positive
frequency components of x(t), leaving their amplitude unaltered. Assuming,
as it is for the unfiltered signal, that

t
x(t) = Ae− τ sin(ωr t) , (6.8)
140 Chapter 6. Research on new materials for suspensions

the Hilbert transform is

t
x̂(t) = −Ae− τ cos(ωr t) . (6.9)

The quantity
Z(t) = x(t) + ix̂(t) (6.10)

is defined pre-envelope of x(t). The program, once x̂(t) has been extracted,
defines Z(t) as in (6.10) and gets the envelope of the oscillation calculating
the squared modulus of the pre-envelope

2t 2t
|Z(t)|2 = A2 e− τ cos2 (ωr t) + sin2 (ωr t) = A2 e− τ


(6.11)

To determine the quality factor it is necessary to measure τ (see (6.5)); thus

it suffices to calculate the logarithm of the envelope:

t
R(t) = ln(|Z(t)|2 ) = 2 ln(A) − 2 , (6.12)
τ

The output signal should be a straight line; the program performs a linear
fit, giving the best value for the slope γ = −2/τ . This quantity is used to
have an estimate of Q:
ωr
Q=− . (6.13)
γ
It is worth to note that the initialization procedure of the filter is not imme-
diate. The filter algorithm proceeds through the computation of the convolu-
tion product between the signal and the transfer function in the time domain;
the filter shall be filled with a samples vector in order to work properly. The
creation of this vector needs a while, depending on the passing band; the
beginning of the acquisition is jeopardized and thus a samples set in which
R(t) is linear is selected for the fitting procedure.
It can happen that the characteristic decay time τ is of the order of
hundreds of seconds, or more. Under those conditions the number of samples
6.3 Experimental setup 141

can be very high. On the other hand, it is not possible to decrease the
acquisition frequency at will, because of aliasing. To solve this problem a
lock-in amplifier can be inserted before the acquisition board. The lock-in
multiplies the input, at high frequency ωin , by a generated sinusoidal signal at
a frequency ωg ' ωin . The output results to be a superposition of two signals
at frequencies ωlow = |ωin −ωg | and ωhigh = ωin +ωg . The lock-in includes also
a low-pass filter that selects the ωlow component only. The amplitude decay
of this signal has the time constant of the original signal and the Q-factor
measurement can proceed as explained before, filtering again around the
resonance frequency, if necessary.
This method is valid if τ is high enough. In fact, to see the oscillation
damping, several oscillations in the time (τ ) are needed. Thus, if τ is short,
ωlow should be high. This reduces the usefulness of the lock-in and introduces
in the output-signal spectrum the input noise, in a band of 2ωlow width,
centered around the frequency ωg . Therefore it can happen, especially at
high frequencies, that the standard procedure is preferable.

6.3.2 Profile measurement apparatus

The produced silicon fibres do not have a circular-cylindrical shape: the

diameter changes along the fibre axis and the section is more similar to an
ellipse indeed, that can also have a variable axes orientation. In order to
make a more realistic model of the fibres it was decided to construct an
apparatus to perform a measurement of the diameter changes. The main
axis of the fibres, on the other hand, is straight (this is a consequence of the
growing procedure) and there was no need to model its bending. These data
§
are used to reconstruct a 3-D model of the fibre as explained in 6.4.3.
The fibre to be measured is put on a stand in the same configuration
as it is clamped during the φ measurements. The stand is settled on a rail
142 Chapter 6. Research on new materials for suspensions

to allow sliding along the fibre axis by means of a long rotating screw fixed
w.r.t. the optical bench as shown in figure 6.14; the structure provides also
the possibility of rotating the fibre of a chosen angle around its axis.
The sensing part of the apparatus consists of a shadowmeter realized
with a 635 nm He-Ne laser of 1 mW power and a 5 mm×5 mm photodiode
that receives the shadow projected by a little segment of the fibre. The
light from the laser is focused to obtain a plane wave front and a sufficiently
large beam spot size. The photodiode is shielded and receives the laser light
passing through a little fissure only. This allows diameter measurements at
a particular chosen position along the fibre axis. Before being shielded the
laser light impinges on a beam splitter: a part of it goes towards the fibre
and the remaining part is focused on another photodiode. The signal from
the latter is used as a reference to compensate for laser power instabilities:
the acquired signal is taken as the ratio between the shadowmeter signal and
the reference one.
Another shadowmeter system, similar to the previously described one, is
used for monitoring the displacement of the fibre. As the stand moves on
the rail, a rotating shield alternatively passes on the beam, generating an
oscillating signal on the photodiode. This signal is used as a trigger for the
acquisition.
The three signals are amplified and sent to an acquisition board on a
computer (see figure 6.15). The board is driven by a LabVIEW R program that
samples the signals at 10 kHz and converts them with a 16-bit ADC. As
a first step the signals are modulated using a filter around the modulation
frequency of 1360 Hz. The signal coming from the trigger photodiode is
analyzed with a program that identifies the peaks due to the rotating shield.
In coincidence with the peaks times, a routine performs a mean over 200
6.4 Measurement of the thermoelastic peak 143

samples of the signals coming from the detecting photodiode and from the
reference one, and calculates the ratio between them. The results are stored
in a file.
trigger
photodiode

laser
fibre
reference
photodiode

laser detection
beam
photodiode
splitter
lenses

clamp

sliding
stand

Figure 6.14: Sketch of the profilation system setup.

6.4 Measurement of the thermoelastic peak

6.4.1 Standard measurement procedure

The measurement proceeds following the general scheme as described be-

low. Before starting the thermoelastic peak measurement it is necessary to
perform an acquisition of the fibre profile.
A calibration sample, consisting of four wires of well known sections, is
clamped on the stand of the profile apparatus. The measurement of the wires
diameters is performed making the stand slide on the rail by means of the
long turning screw and using the acquisition program described in § 6.3.2.
144 Chapter 6. Research on new materials for suspensions

trigger
photodiode
laser ampl
a)
b)
reference ampl DAQ file
photodiode

c)
mod laser ampl
beam detection
splitter photodiode

Figure 6.15: A sketch of the acquisition system for profile measurement is

shown. The signals from the three photodiodes are amplified and come to
the acquisition system. Using the signal (a) as a trigger, the ratio between
the main signal (c) and the reference one (b) is performed and stored in a
file.

The data, stored in a file, are used for a calibration of the signal from the
sensing photodiode.
Then it is necessary to accurately clean the surface of the fibre with iso-
propilic alcohol, to remove impurities and chemical volatile residuals, which
could compromise the Q-factor measurement and the efficiency of the vacuum
system.
The fibre is filed at one end on two opposite sides as shown in figure 6.16.
In this way the contact surface is increased allowing a tighter clamping and
lowering the probability of breaking the fibre in the clamping process. A
total force acting on the clamp is chosen and is applied using a simple device
based on scales and a press.
The fibre, together with its clamp, is put on the stand of the profile
apparatus. Profile measurements are performed as just described for the
calibration procedure and fibre diameters are measured for several rotation
angles around fibre axis, usually every 30◦ . The produced data are used
inside a finite element analysis (FEA) to reproduce an accurate model of the
6.4 Measurement of the thermoelastic peak 145

§
fibre as explained in 6.4.3.
Then the fibre is ready for the φ measurement and it is settled in the
vacuum chamber and blocked on the stand together with the clamp. Using
the centering program described in § 6.3.1, the position read-out system is
set in order to center the fibre shadow w.r.t. the gap between the two sensing
photodiodes. The vacuum chamber is closed and the air pressure is lowered
as described in § 6.3.1. In this phase it is important to avoid activating the
excitation system: around a pressure of 1 mbar the gas in the chamber is at
a critical pressure for the production of electric discharges.
At a level below 10−5 mbar the measurement can start. Preliminarily, it
is necessary to identify the various resonance mode frequencies of the fibre.
A power-spectrum analyzer is used to display the frequency response of
the fibre to a white noise excitation. The analyzer shows a mean over several
acquisitions and the visible peaks reveal the presence of resonances. The
FEA model gives these frequencies with a good level of precision, except for
a scale factor very near to unity. This can be very useful in speeding up the
procedure: once the first two modes are identified, it is easy to find all the
others, exciting with a sweeping frequency near the expected ones.
§
The test program described in 6.3.1 allows a direct check of the damping
behavior of the found frequencies.
When a set of frequencies is available, for each one a Q-factor measure-
ment is performed exciting the oscillation and then interrupting the excita-
tion abruptly. The subsequent damping motion is detected and recorded as
a file for possible further analysis. The standard measurement program gives
an estimate of the Q-factor. For each frequency the measurement is repeated
several times and the mean of the results is taken as the final value with an
error equal to the standard deviation of the set. This error ranges from 1%
146 Chapter 6. Research on new materials for suspensions

to 5%; the contribution coming from the fit is only al the level of 1%.

a) b)

Figure 6.16: The head of the cylindrical fibre (a) is filed on two sides in order
to allow a better clamping (b).

6.4.2 The etching process

It is possible to operate on the fibres with a chemical process called etching.

It consists essentially of a bath in a chemical solution that acts on the surface
of the fibre removing atoms layers.
It can be very useful if it is necessary to polish the surface from contam-
inants or unwanted oxides. Their presence can jeopardize the measurement
of the fibre Q-factor hiding the real loss angle of the material as described in
§ 3.2.3.
Another advantage of the etching process is that it allows to deeply act
on the fibre shape.
On the one hand it is possible to lower the fibre diameter for exploring
different regions of the thermoelastic noise, whose frequency peak depends
on the diameter. In fact, there are problems in producing fibres with diame-
ters below 500 µm, because of instabilities and technical difficulties occurring
during the micro-pulling down process.
6.4 Measurement of the thermoelastic peak 147

On the other hand the form of the profile can be shaped in order to
reduce clamp losses or correct macroscopic defects. In fact, the etching can
be realized only on certain parts of the surfaces, shielding the rest with a
layer of enamel — like the nail varnish —.
It is possible to shield the head of the fibre: the obtained bigger head allows
a better clamp — the bigger the surface, the more tightened the clamp can
be — and shifts the most solicited parts of the fibre to a region that is far
from the clamp itself. The latter condition is important because the heavily
solicited parts near the clamp can introduce extra friction on clamp surfaces
and transmit energy to it affecting the loss angle measurement.
It is also possible to act locally on different segments of the fibre in order to
correct any defects that may have risen during the growing process. In this
way the fibre can be made more regular by simplifying the modelization and
analysis phases.
The used chemical solution is composed of nitric acid by 75% (HNO3 at
a 70% concentration), fluoridric acid by 15% (HF at a 59.2% concentration)
and acetic alcohol by 10% (CH3 COOH). Etching occurs via a redox reaction
followed by the dissolution of the oxide by HF that acts as a complexing
agent. The reaction can be schematized as

Si + HNO3 + 6HF → H2 SiF6 + HNO2 + H2 O + H2 . (6.14)

Points on silicon surface randomly become oxidation or reduction sites. These

act like localized electrochemical cells, sustaining corrosion currents of about
100 A/cm2 , which are relatively large currents. For the kind of solution under
consideration, each point spends, on the average, the same amount of time
being an anode or a cathode site; this leads to an isotropic etching, i.e. an
etching whose corrosion velocity of the material is the same in all directions.
The nitric acid gives birth to NO2 , that is reduced at a cathode site which
148 Chapter 6. Research on new materials for suspensions

produces free holes:

2NO2 → 2NO− +
2 + 2h . (6.15)

Simultaneously, silicon is promoted to a higher oxidation state at an anodic

site through positive charges in the form of holes:

Si0 + 2h+ → Si2+ . (6.16)

The Si2+ combines with OH− to form SiO2 :

Si2+ + 2OH− → Si(OH)2 → SiO2 + H2 O (6.17)

and SiO2 is then dissolved by HF to form a water soluble complex of H2 SiF6 :

SiO2 + 6HF → H2 SiF6 + 2H2 O . (6.18)

The nitric acid has a complex behavior and its concentration manages a
chain of reactions that form an autocathalitic cycle for holes production and
HNO2 . The acetic acid is a frequent substitute for water as a diluent; it has
a lower dielectric constant w.r.t. water, thus it produces a minor dissociation
of HNO3 , yielding a higher oxidation power for the etching. Furthermore,
acetic acid is less polar than water and it can help to achieve proper wetting
of slightly hydrophobic silicon.
In figure 6.17 it is shown a triangular diagram [116] that allows to find
the various etching velocity depending on the concentrations of the three
solution components. The little circle represents the working point chosen
for the solution described above.
The etching is performed in a long basin filled with the HNA solution.
The fibre can enter the solution staying in a horizontal position. This is
a much more efficient way of doing the etching w.r.t. a vertical immersion:
in the latter case the reaction products, which are hotter and come to the
solution surface, do create vertical gradients — either gradient of the reagents
6.4 Measurement of the thermoelastic peak 149

0
100

25

)
.2%

W
eig
75

49

ht
F(

%
,H

,H
50

NO
1 mm/min
ht
50
eig

3
(69
W

.5%
100 µ m/min 75

)
25
10 µ m/min
100
0
100 75 50 25 0
Weight %, CH3COOH

Figure 6.17: Diagram of the etching velocity for an HNA etching solution.
Three iso-velocity lines are shown together with the working point corre-
sponding to the solution used for the fibre etching described in the text.

concentration, or gradient of temperature — that can deeply modify the

velocity of the etching itself. In fact, it was experimentally found that the
fibre surface is etched faster at the top than at the bottom. With a horizontal
etching the above mentioned problems are avoided and the reaction velocity
is seen to be constant all along the fibre.
The etching procedure consists of three steps. First of all an etching last-
ing some minutes is performed in order to remove the superficial impurities.
Secondly, after deciding the target diameter of the fibre and calculating the
duration time of the etching, which depends on the particular HNA chosen
solution, a second etching is performed. During this phase it is also possible
to apply nail varnish on the fibre surface in regions that are already thin or
on the part chosen to become the head to be inserted in the clamp. Finally
the nail varnish is removed with acetone and the fibre is subsequently dipped
into a fluoridric acid solution to deeply clean the fibre from the oxide that
may still be present on its surface. An example of a fibre etching is shown in
150 Chapter 6. Research on new materials for suspensions

figure 6.18.

Figure 6.18: The profile of a fibre at a particular angle is shown before (solid
blue line) and after (dashed red line) an etching. It is evident that the
diameter has reduced of about 200 µm.

6.4.3 Modeling the fibre

From the data obtained with the profile measurement apparatus, described
in § 6.3.2, it is possible to reproduce a 3-D model of the fibre to be used
for a finite element analysis (FEA) in the framework of the ANSYS R 8.1 [103]
software.

Profiles alignment

§
The data coming from the profile measurement ( 6.3.2) are 6 vectors of num-
bers representing the longitudinal views of the same fibre every 30◦ of rotation
along the fibre axis. The data acquisition has not a definite and reproducible
starting point of one profile w.r.t. the subsequent rotated one. Therefore, to
make the fibre profiles correspond to one another it is necessary to manually
align them. The procedure is done with the help of a MATLAB R 6.5 [117]
program that acquires the data and shows the different profiles on the same
6.4 Measurement of the thermoelastic peak 151

Figure 6.19: Profiles at various section angles are shown for the part of
the fibre near to the tip. The alignment of the different data sets (one for
each angle) is realized manually over-imposing along the x axis the points
preceding the 400 µm-diameter line crossing indicated on the left by the
orange dashed line.

Figure 6.20: Profiles at various section angles.

152 Chapter 6. Research on new materials for suspensions

plot. In figure 6.19 the beginning of the 6 sets is shown after the alignment,
done by choosing as starting points the ones preceding the crossing of the
400 µm-diameter line indicated in orange. In figure 6.20 an example of the
aligned profiles is shown for a 177.5 mm-long fibre. It is evident that the
fibre is not a perfect cylinder but it is also evident that the various profiles
are very similar to one another. The aligned profiles are stored in a matrix
for further analysis.

Shadow-projection correction

Data have to be corrected for a geometrical effect due to the setup of the pro-
file apparatus. The laser-shadowmeter system systematically overestimates
the diameter at the various acquisition angles: the parts of the fibre project-
ing the shadow are not always the expected ones because the fibre is not a
perfect cylinder. To understand this effect refer to figure 6.21, in which

30o

a) b)

Figure 6.21: Estimation of the fibre section from profile measurements. Pro-
jecting directions of the shadows are indicated by black arrows and estimated
section points are in red (a). Without any correction the “true” section of
the fibre (blue ellipse) is falsely interpreted to be the red polygon (b).

the “true” fibre section is drawn in blue and the direction of the projecting
shadow is indicated by black arrows every 30◦ . Without making any correc-
6.4 Measurement of the thermoelastic peak 153

a) b)

Figure 6.22: Correction of the estimated fibre section from profile measure-
ments. Two consecutive shadow projections identify construction points (the
green squares). The new estimated section points (cyan triangles) are cho-
sen to be the mid-points of the segments delimited by the green squares (a).
A comparison among the “true” section (in blue), the fake estimation (in
red) and the corrected one (in cyan) is given (b). It is evident that the
peanut-shell deformation is corrected.

tions, the shadows are interpreted to come from the red points and not from
the tangential points between the fibre and the black arrows. It is evident
that elliptical sections are distorted toward a peanut-shell shape. The more
the eccentricity, the more the deformation and if the fibre were perfectly
cylindrical this effect would disappear. This is not the case, because most of
the sections are roughly elliptical.
Therefore an algorithm done with MATLAB R software was developed to
correct the introduced bias. It resides on the idea that the real section of
the fibre is contained inside the envelope of all the projection directions of
the shadows. As shown in figure 6.22 a, two consecutive shadow projections
intersect in construction points identified with green squares. The new esti-
mated section points (cyan triangles) are chosen to be the mid-points of the
segments delimited by the green squares. A comparison among the “true”
section (in blue), the fake estimation (in red) and the corrected one (in cyan)
154 Chapter 6. Research on new materials for suspensions

is given in figure 6.22 b. It is evident that the peanut-shell shape, present in

the red curve, is corrected toward a shape much more similar to the real one.

The finite element model

The corrected data are imported in ANSYS R thanks to a routine done with
AWK [104], a pattern-directed scanning and processing language. The routine
converts the data file into an ANSYS R macro file that executes all the com-
mands to create the keypoints. They are intermediate point-like structures
used as frames to construct areas, volumes and, finally, the elements of the
3-D model. These steps of the model creation are done with an automated
ANSYS R macro file that has just to be slightly corrected from fibre to fibre,
depending on the number and the length of the profiles and on the mean
section of the fibre itself.
Firstly the lateral surface of the fibre is created. It is simply a set of
triangles sharing edges. The triangles have two keypoints of one section and
one keypoint of the previous section or the following one. A view of the
lateral surface of a fibre is shown in figure 6.23 for two adjacent rings of
triangles.
Secondly areas associated to each fibre section are created. They have
no physical significance but they are of fundamental importance for creating
intermediate volumes, delimited by 2 of these areas and by 24 lateral trian-
gles. The volumes are glued together to form a unique object and the model
is ready to be meshed.
Usually it suffices to mesh volumes with an automated command provided
by ANSYS R, called “free mesh”. In some particular cases2 , the mesh has to be
guided by an intermediate one, generally done on 2-D surfaces. This is the

2
This usually happens when the model is a 3-D model with one dimension that extends
much more or much less w.r.t. the others.
6.4 Measurement of the thermoelastic peak 155

Figure 6.23: A part of the lateral surface creation of the fibre model is
shown. The surface is constructed by the definition of triangles connecting
two adjacent fibre sections.

case, and a mesh is preliminarily done with PLANE2 elements on the slice
areas. The elements are triangular and their dimensions are taken in order
to have 12÷16 elements for each section. This feature is obtained imposing
single elements to have a size proportional to the section area. In figure 6.24
several fibre sections are shown together with the PLANE2 elements on them.
It is evident that the number of elements per section is constant even if the
section is widely changing dimension.
Thanks to the grid provided by the PLANE2 elements on the sections, it is
straightforward to mesh the whole model. It suffices to force no line division3 ,
in order not to have too many elements, and to use the standard free-mesh
command with 3-D SOLID92 elements. A view of the model elements is
shown in figure 6.25 where a set of them has been removed to permit the
vision of the other underlying elements.
The elements can be assigned the material properties, i.e. Young’s mod-

3
This means that the mesher cannot use parts of the already defined lines in building
the elements of the model.
156 Chapter 6. Research on new materials for suspensions

Figure 6.24: Several sections of a fibre are shown together with the triangular
elements on them. The number of elements per slide is not depending on the
slide dimension.

Figure 6.25: View of elements of a fibre model. Some of them are removed
to allow the vision of the underlying elements.

ulus, Poisson’s ratio and the density. In order to simulate the clamping,
the last section on one end of the fibre is constrained in all the degrees of
freedom.
The model is ready to perform a modal analysis on it. In the modal
6.4 Measurement of the thermoelastic peak 157

analysis the software extracts the frequencies and the shapes of the resonance
modes. It allows also the user to look for stress and strain distribution on
the surface or inside the simulated objects.

6.4.4 Phi measurements

Measurements of the thermoelastic peak are performed following the proce-

§
dure described in 6.4.1.
The measured values for a 308 mm long fibre are shown in figure 6.26; the
free length of this fibre after the clamping is 289 mm. As it will be explained
§
in 6.5, from the analysis of these data the values of Young’s modulus E, of
the thermal expansion coefficient α and of the thermal conductivity κ can
be evaluated. Using the measured value of E, an “effective” diameter can
be defined for each resonance mode. On assuming a cylindrical-shaped fibre
with this diameter it is possible to predict a value for the loss angle for each
resonance frequency. These predictions, with the error coming mainly from
the uncertainty in the diameter measurement, are also shown in figure 6.26.
It is quite evident that there is the presence of an excess loss. This excess
can be due to contaminants present on fibre surface: after the production
of the fibre, the external surface is in fact subject to oxidation; furthermore
the deposition of some impurities on it, during the growing process, is also
possible. To verify this hypothesis a chemical etching process was performed
§
on the fibre as described in 6.4.2, removing the first silicon layers. The part
of the fibre inside the clamp was not etched: in this way the narrowing of the
fibre just outside the clamp can help to further reduce the clamping losses.
In figure 6.27 the measured values of the loss angle for the same fibre
after the etching are shown. The free length of the fibre after the clamping is
278 mm and the diameter reduction shifts the position of the thermoelastic
peak to higher frequencies.
158 Chapter 6. Research on new materials for suspensions

Figure 6.26: Measured loss angle for a 289 mm (free length) fibre, with an
average diameter of 746 µm. The squares represent the thermoelastic contri-
bution as predicted by the model described in the text.

Figure 6.27: Measured loss angle for the same fibre as that shown in fig-
ure 6.26 after the etching process; the average diameter is now 574 µm, while
the free length is 278 mm.
6.4 Measurement of the thermoelastic peak 159

The comparison between these values and the expected thermoelastic

losses, computed as previously described, points out that the excess loss is
not present anymore, thus confirming the surface contamination hypothesis.
A small excess is still found at low frequencies: this is probably due to clamp
losses.
In figure 6.28 the measured values of φ for another fibre — 111.5 mm
long — treated with chemical etching are shown. Since this fibre has a

Figure 6.28: Measured loss angle for a 111.5 mm (free length) fibre. The blue
and the red squares represent the measured loss angle values relative to “a”
and “b” modes respectively.

roughly elliptical section, it is evident that there are two sets of modes,
corresponding to the different length of the two ellipse axes. An unloaded
cylindrical fibre of length L, diameter d and clamped at the upper edge shows
a set of resonance frequencies given by [30]:
s s
1 EI 2 d E 2
fi = ki = k , (6.19)
2πL2 ρS 8πL2 ρ i
160 Chapter 6. Research on new materials for suspensions

where ki = αi /L, αi are the solutions of

cos(α) cosh(α) + 1 = 0 (6.20)

and E, S and I are Young’s modulus of the material, the section and the
cross-section moment of inertia of the fibre. A very good approximation for
αi values is (
1.8755 , for i = 1 ,
αi = 1
(6.21)
(i − 2 π) , for i > 2 .
Therefore, in the case of an elliptical fibre there will be a split of each reso-
nance in doublets. The frequencies will roughly correspond to the ones of the
cylindrical fibres with sections tangent to the ellipse as shown in figure 6.29.

a
b b−mode

a−mode

Figure 6.29: The section of an elliptical fibre is shown together with the
direction of the split modes due to the difference between the 2 semi-axes
“a” and “b” of the ellipse. The modes relative to the shorter axis are at a
lower frequency and usually called a-modes; the ones relative to the longer
axis are at a higher frequency and called b-modes.

6.5 Extraction of silicon thermo-mechanical

properties

The analysis done for extracting the silicon thermo-mechanical properties

from the measurement of the thermoelastic peak resides on the finite element
model realized with ANSYS R (see § 6.4.3). The model is constructed on the
6.5 Extraction of silicon thermo-mechanical properties 161

data coming from profile measurements, as described in § 6.3.2. Therefore

it is important to estimate carefully the errors coming from the profilation
procedure.

6.5.1 Estimating the errors of the profilation

The data coming from the profilation apparatus are pure numbers, taken as
ratios R between the signals coming from the detection photodiode and the
reference one (see figure 6.14 on page 143). Therefore they need a calibration
to be converted in displacements. It is better to do the calibration just before
and after the profiles measurement on a fibre.
A series of wires of well known diameter is measured with the profilation
apparatus bringing a set of data like the ones reported in figure 6.30. The

Figure 6.30: Profile of the series of wires used for calibrating the profile
measurement apparatus.

ratio R and its error are taken to be the mean and the mean square difference
of the oscillations in each of the six steps. The values are fitted with a straight
line R(d):

R = ad + b (6.22)
162 Chapter 6. Research on new materials for suspensions

and the value of the diameter can be extracted:

R−b
d(R) = (6.23)
a

The error σd on d is calculated as

1 2
σd2 = (σ + d2 σa2 + σb2 ) (6.24)
a2 R

and it is around 100 µm, i.e. around 10%, dominated by the σa2 term. To
measure the fibre profile there is also an independent method, more precise
but extremely slow. It resides on the possibility of taking several photos of
the fibre profile with a digital camera set on a microscope. The photos are
merged into a unique image that is processed by a MATLAB R program that
finds the edges of the fibre. The error for this method is a composition of
the error on the pixel/mm conversion factor and the error on the position of
the edge found by the algorithm. In figure 6.31, the data obtained with the
camera profile (green bars) and the standard procedure (red bars) are shown
for a part of a fibre. It is evident that there is a overestimation of the errors for
the standard procedure. That is the case for all the regions inspected along
the fibre. Thanks to the agreement found, it seems reasonable to downscale
the error on the standard profile measurement to the value coming from the
camera profilation, i.e. 3.5%.

6.5.2 Young’s modulus

From (6.19) it follows that for a cylindrical fibre the value fi /ki2 is constant.
For an elliptical one two preferred values are expected. That is what happens,
for example, with the 111 mm long fibre as it can be seen in figure 6.32. The
data, indicated by blue squares, follow two different trends reflecting the
ellipticity of the fibre section; the values are not constant since the fibre
geometry is irregular.
6.5 Extraction of silicon thermo-mechanical properties 163

Figure 6.31: Comparison between profile measurements on a part of a fibre,

obtained with the digital camera (little green bars) and with the standard
procedure (big red bars).

Figure 6.32: Plot of fi /ki2 versus the mode number for the 111 mm long fibre.

With the FEA on the modeled fibre it is possible to predict the values
fi /ki2 and find the best Young’s modulus E best to use in order to obtain a set
164 Chapter 6. Research on new materials for suspensions

of frequencies fimodel that minimizes the quantity

X  f exp − f model 2
i i
. (6.25)
i
ki2

As for this fibre, E best is 150 GPa with an error of 11 GPa coming mainly
from the diameter uncertainty. In figure 6.32 there are also shown, with
magenta squares, the values predicted by the FEA using for Young’s modulus
the obtained value E best . The agreement between data and prediction is
very good: since the model considers a fibre whose properties are constant
along the length, it is possible to conclude that the geometry of the fibre
is taken into account properly and that any variation of the mean4 Young’s
modulus is small along the fibre. Using FEA it is also possible to identify
some modes which are not purely transversal; in these modes the plane of
oscillation changes a lot along the fibre axis (see figure 6.33 a) or the violin-
string oscillations are coupled with “breathing” modes (see figure 6.33 b).
These modes are not used in thermoelastic analysis since there is not a simple

a) b)

Figure 6.33: Two non-purely-transversal modes of the 111 mm long fibre are
shown. In the mode at 6104 Hz (a) the plane of oscillation is changing a lot
along the fibre axis. In the mode at 8602 Hz (b) the transversal oscillation is
coupled with an “explosion” mode.
4
It can happen that Young’s modulus has big changes on the scale of some millimeters
or less; in these cases the modes only “see” an average value of Young’s modulus, and it
is this quantity that has to be considered.
6.5 Extraction of silicon thermo-mechanical properties 165

model of the loss angle for these situations.

An analogous analysis on the 308 mm long fibre leads to a value of
(174±12) GPa for Young’s modulus.

6.5.3 The coefficient of linear thermal expansion

The experimental value of the thermoelastic peak together with its theoretical
prediction — equation (3.54) — allows to estimate the coefficient of linear
thermal expansion α. By using the known specific heat cV = 707 J/(kgK)
and density ρ = 2330 kg/m3 of silicon [111], for the 308 mm long fibre the
extracted peak value is φ0 = (10.12±0.12)×10−5 .
The error is due to the fact that the thermoelastic peak is sampled only
at the resonance frequencies and the maximum amplitude is estimated us-
ing the φ trend around the peak. The maximum for the peak value was
taken equal to the intersection point of the two steepest straight lines that
can be constructed with the couples of experimental points just before and
after the peak. The minimum for the peak estimate was taken equal to the
maximum measured values. An example of the process on this fibre is given
in figure 6.34. Assuming for T the measured value of 293 K it follows that
α = (2.56±0.11)×10−6 K−1 . For the 111 mm long fibre the extracted value
for the peak is φ0 = (17.26±0.24)×10−5 and α = (2.54±0.13)×10−6 K−1 .

6.5.4 The coefficient of thermal conduction

The τth parameter in the thermoelastic curve, discussed in § 3.2.2, allows

a determination of the thermal conductivity κ of the fibre. The heat flux
characteristic time is
cV d2
τth = F (6.26)
κ
where d is an effective distance for the heat flow and F is a geometrical
constant keeping into account the shape of the fibre section. For the measured
166 Chapter 6. Research on new materials for suspensions

Figure 6.34: An enlargement of loss angle measurements around the ther-

moelastic peak is shown for the 308 mm long fibre. There are also indicated,
as “min” and “max”, the limits of the error interval taken for the peak value.

roughly-elliptical fibres an estimate of d can be deduced for each mode by

assuming a cylindrically shaped fibre oscillating at that frequency. In this
way the elliptical section is approximated with two circular sections, one
with the larger axis as a diameter and the other with the dimension of the
shorter axis, for the two main oscillating directions (refer to figure 6.29).
Nevertheless the larger curvature radius assumed for the external circular
section leads to an underestimation of the heat gradient and consequently to
an overestimation of τth ; the opposite happens assuming the internal circular
section. A better first order estimation can be obtained by modifying (6.26)
with the introduction of a parameter c correcting the form factor known for
a cylindrical geometry

1 cV d2
τth = (1 ± c) , (6.27)
2.16 κ
6.6 Future work 167

where the sign + in front of c is used when the inscribed circular cross section
is taken into account. The values found with a fit by using this expression for
the 111 mm long fibre are κ = (146±13) W/(mK) and c = 0.18±0.002, only
considering the pure transverse resonance modes. The error was evaluated
varying the model parameters inside their errors. For the 308 mm fibre,
performing the same fit, the obtained values are κ = (138 ± 11) W/(mK)
and c = 0.001±0.001. In table 6.1 the physical parameters measured for the
111 mm long fibre and the 308 mm long one are summarized.

L (mm) E (GPa) α (K−1 ) κ (W/(mK))

111.5±0.5 150±11 (2.54±0.13)10−6 146±13
308.0±0.5 174±12 (2.56±0.11)10−6 138±11

Table 6.1: Measured parameter for two different silicon fibres.

6.6 Future work

The activity described in this chapter has been done in the framework of
a larger research program on new materials for interferometric detectors of
the 3rd generation. The group that I belong to is constructing a cryogenic
facility to measure, on fibres, the coefficient of thermal conductivity κ, of
thermal expansion α and the loss angle φ as a function of temperature. In
fact, it is of crucial importance to have the possibility of measuring these
properties easily, in a robust way and on fibre shaped samples, in order to
characterize silicon and new silicon-based materials, whose properties have
not been measured yet.

Measuring κ(T )

Inside a cryostat, already installed and working at liquid nitrogen temper-

ature, there will be a metallic copper box — whose temperature is tunable
168 Chapter 6. Research on new materials for suspensions

thanks to coil heaters — inside which a fiber is clamped. The specific thermal
conductivity of the fiber will be calculated by measuring fiber geometry and
its absolute conductivity (through thermometers at the end of an induced
temperature gradient along the fibre itself).

Measuring α(T )

The same cryostat will be used for the absolute measurement of the linear
coefficient of thermal expansion.
Because of the low value of thermal expansion expected for silicon, which
is of the order of 10−8 K−1 (see figure 6.1), the displacement detector should
have a stability of 1 nm over a period of 10÷15 minutes. The choice of a
suitable optical layout fell on Fabry-Perot cavities.
The final design of the full facility is sketched in figure 6.35. A 20 cm

Figure 6.35: Layout of the thermal expansion measurement facility.

a) 200 mW Yag laser; b) Faraday isolator; c) EO modulator; d) mixer;
e) InGaAs photodiode; f) Zerodur cavity; g) flat-concave optical cavities;
h) piezo; i) vacuum tank; j) copper shield; k) cryostat; l) quarter wave plates.
6.6 Future work 169

long Zerodur reference cavity controls the frequency of a 200 mW YAG laser
through a piezoelectric actuator, which has its negative expansion coefficient
partially matched by a washer of aluminum with a suitable thickness. The
laser beam coming out of the reference cavity tracks the changes of length
in the measuring cavity. The information on the fibre length is contained in
the signal fed into the piezo.

Measuring φ(T )

The facility realized for measuring fibers loss angles can be used at cryogenic
temperatures with minor changes.
Particular care should be paid in order to realize a clamp that does not
become too tight or too loose on the fibre going towards low temperatures.
The realization of a composite clamp, made of two different materials —
whose thermal expansion coefficients add up to be near the silicon thermal
expansion — is being investigated: particular clamps like the one shown in
figure 6.36 have been designed and realized. In the future they will be tested

Figure 6.36: Sketch of a realized clamp made of brass and aluminum.

on silicon fibres.
170
Chapter 7

Conclusions

This work has moved from the idea of using finite element analysis (FEA) as
a general tool for thermal noise studies relating to improve the performances
of interferometric gravitational wave antennae.
On the one hand FEA has been used to simulate some thermal noise con-
tributions for the Virgo mirrors. For the first time ever, a complete model of
a Virgo mirror has been used to perform Brownian thermal noise calculations
through the numerical dynamic approach, which relies on Levin’s method.
The analysis is a harmonic FEA, of the same kind as other analyses already
made, but presents a series of new important elements:

ˆ the analysis is performed including in one model only all the ancillary
components on the mirror: the coating, the markers, the spacers and
the magnets (in former analyses all those contributions were investi-
gated separately as worsening inputs);

ˆ the mesh of the coating is driven in a new way, with a double advantage:
– it allows to apply in a more efficient and precise way the pressure
profile on the mirror;

– it allows the creation of relatively simple models on which para-

metric analyses can be run;

171
172 Chapter 7. Conclusions

ˆ the coating is simulated using SHELL99 elements, which allow to define

up to 250 layers with different properties.

The results are in agreement with the ones by Levin [46] and by Numata [18].
Besides,

ˆ the effects of the mirror resonance modes on the read-out have been
calculated on the actual model of a Virgo mirror through a modal FEA
and

ˆ a parametric analysis has been done moving the position of the markers
and showing, for the first time, the heuristic relation between the dis-
tance of the markers from the laser beam axis and the induced silicate
bonding Brownian thermal noise.

On the other hand, in the framework of research of new suspension ma-

terials for future interferometric antennae, a facility for measuring fibre loss
angle φ has been realized. For the first time, measurements on crystalline
silicon fibres have been performed with the ring-down technique.
FEA enters that experiment in a novel way: an apparatus to measure the
fibre profiles — which, for production problems, is not perfectly cylindrical
as aimed to be — has been made and the profile data are used to create
finite element models. The creation of the 3-D models of such thin objects
resides on the particular mesh-driving process used. A modal analysis allows
to find the resonances of the fibers helping the Q-measurement procedure
and a parametric analysis w.r.t. Young’s modulus E allows to fit E itself
from the modeled frequencies and the theoretical wave numbers.
The values obtained for the measured loss angles are consistent with the
predicted thermoelastic peaks and are encouraging thanks to the possibility
of clamping the fibre without evident clamping losses.
Chapter 7. Conclusions 173

In the future the efforts will be focused on performing:

ˆ a simulation of thermoelastic noise in the mirror;

ˆ a complete simulation of thermoelastic noise in fibres;
ˆ a modelization of the clamping dissipative processes;
ˆ cryogenic measurements of thermal conduction coefficient, thermal ex-
pansion coefficient and loss angle on silicon fibres.
174
Bibliography

[1] A. Einstein, Sitz.-Ber. Preuß. Akad. Wiss. (1916) 514

[2] J.H. Taylor and J.M. Wiesberg, Astrophys. J. 345 (1989) 434

[3] T. Nakamura, M. Sasaki, T. Tanaka and K.S. Thorne,

Astrophys. J. 487 (1997) L139

[4] T. Nakamura, Proc. 2nd TAMA Workshop on Grav. Wave Detection,

Univ. Acad. Press (2000) 123

[5] G.A. Prodi et al., Proc. 2nd Edoardo Amaldi Conf. on Grav. Waves,
CERN, July (1997) 148; http://www.auriga.lnl.infn.it

[6] P. Astone et al., Phys. Rev. D 47 (1993) 362;

http://www.lnf.infn.it/esperimenti/rog

[7] P. Astone et al. Astrophys. J. 7 (1997) 231;

http://www.lnf.infn.it/esperimenti/rog

[8] E. Mauceli et al., Phys. Rev. D 54 (1996) 1264; http://sam.phys.lsu.edu

[9] A. Abramovici et al., Science 256 (1992) 325;

http://www.ligo.caltech.edu

[10] A. Giazotto, Phys. Rep. C 182 (1989) 365; C. Bradaschia et al., Nucl.
Instr. Methods A 289 (1990) 518; http://www.virgo.infn.it

175
176 BIBLIOGRAPHY

[11] K. Danzmann et al., GEO600: Proposal for a 600 m Laser Interferomet-

ric Gravitational Wave Antenna, Max-Planck Institut für Quantenoptik
Report 190, Garching, Germany (1990);
http://www.geo600.uni-hannover.de

[12] K. Tsubono et al., Proc. Intern. Conf. on Grav. Waves, Sources and De-
tectors, World Scientific, Singapore (1997); http://tamago.mtk.nao.ac.jp

[13] A. Buonanno, Y.B. Chen, Phys. Rev. D 69 (2004) 102004

[14] A. Brillet et al., Advanced Techniques, in “The Detection of Gravita-

tional Waves”, D. Blair ed., Cambridge University Press, Cambridge,
UK (1999) 369-405

[15] P.R. Saulson, Fundamentals of Interferometric Gravitational Wave De-

tectors, World Scientific, Singapore (1994).

[16] A. Nakagawa, B.A. Auld, E. Gustafson and M.M. Fejer, Rev. Sci. In-
strum. 68 (1997) 3553

[17] F. Bondu, Dark Fringe Shot Noise Sensitivity, Virgo note,

VIRNOTOCA1390243 (2003)

[18] K. Numata, Direct Measurement of Mirror Thermal Noise, thesis, Uni-

versity of Tokyo (2002)

[19] P.R. Saulson, Phys. Rev. D 30 (1984) 732

[20] M. Beccaria et al., Class. Quant. Grav. 15 (1998) 3339

[21] J.-Y. Vinet et al, Phys. Rev. D 56 (1997) 6085

[22] G. Losurdo, Ultra-Low Frequency Inverted Pendulum for the Virgo Test
Mass Suspension, Ph.D. thesis, Scuola normale superiore di Pisa (1998)
BIBLIOGRAPHY 177

[23] M. Punturo, The Virgo Sensitivity Curve, Virgo note, VIR-NOT-PER-

1390-51 (2001)

[24] A. Einstein, Annalen der Physik 17 (1905) 549

[25] H.B. Callen and T.A. Welton, Phys. Rev. 83 (1951) 34

[26] R.F. Greene and H.B. Callen, Phys. Rev. 83 (1951) 1231

[27] H.B. Callen and R.F. Greene, Phys. Rev. 86 (1952) 702

[28] R.F. Greene and H.B. Callen, Phys. Rev. 88 (1952) 1387

[29] H. Nyquist, Phys. Rev. 32 (1928) 110

[30] L.D. Landau and E. M. Lifshitz, Theory of Elasticity, Addison-

Wesley (1959)

[31] F. Bondu, Étude de Bruit Termique et Stabilisation en Fréquence du

Laser du Détecteur Interférométrique d’Ondes Gravitationelles Virgo,
Ph.D. thesis, Univesité de Paris-Sud, Centre d’Orsay (1996)

[32] J.M. Ziman, Principles of the Theory of Solids, 2nd ed., Cambridge Uni-
versity Press (1979)

[33] J.M. Ziman, Electrons and Phonons. The Theory of Transport Phenom-
ena in Solids, Oxford University Press (2001)

[34] C. Zener, Phys. Rev. 52 (1937) 230

[35] C. Zener, Phys. Rev. 53 (1938) 90

[36] J. Ferreirinho, Internal Friction in High Q Materials, in The Detection

of Gravitational Waves, D.G. Blair ed., Cambridge University press, first
reprint (1991) 116-151
178 BIBLIOGRAPHY

[37] A.M. Gretarsson and G.M Harry, Rev. Sci. Instrum. 70 (1999) 4081

[38] G.M. Harry et al., Class. Quant. Grav. 19 (2002) 897

[39] P.R. Saulson, Phys. Rev. D 42 (1990) 2437

[40] V.B. Braginsky, V.P. Mitrofanov, V.I. Panov, System with Small Dissi-
pation, University of Chicago Press (1985)

[41] A.V. Granato and K. Lücke, Appl. Phys. J. 27 (1956) 583

[42] G.I. Gonzáles and P.R. Saulson, Phys. Lett. A 201 (1995) 12

[43] M. Kajima, N. Kusumi, S. Moriawaki and N. Mio, Phys.

Lett. A 264 (1999) 251

[44] N. Ohishi, S. Otsuka, K. Kawabe and K. Tsubono, Phys.

Lett. A 266 (2000) 228

[45] K. Yamamoto, Study of the Thermal Noise Caused by Inhomogeneusly

Distributed Loss, Ph.D. thesis, University of Tokyo (2000)

[46] Y. Levin, Phys. Rev. D 57 (1998) 659

[47] J.E. Logan, N.A. Robertson, J. Hough and P.J. Veitch, Phys.
Lett. A 161 (1991) 101

[48] A.D. Gillespie, Thermal Noise in the Initial LIGO Interferometers,

Ph.D. thesis, California Institute of Technology (1995)

[49] E. Majorana and Y. Ogawa, Phys. Lett. A 233 (1997) 162

[50] F. Bondu, J.-Y. Vinet, Phys. Lett. A 198 (1995) 74

[51] A.E.H. Love, A Treatise on the Mathematical Theory of Elasticity, Dover

Publ., New York (1944)
BIBLIOGRAPHY 179

[52] J.R. Hutchinson, J. Appl. Mech. 47 (1980) 901

[53] V.B. Braginsky, M.L. Gorodetsky, S.P. Vyatchanin, Phys.

Lett. A 264 (1999) 1

[54] F. Bondu, P. Hello, J.-Y. Vinet, Phys. Lett. A 246 (1998) 227

[55] A. Nakagawa, E. Gustafson, P. Beyersdorf and M.M. Fejer, Phys.

Rev. D 65 (2002) 082002

[56] Y.T. Liu and K.S. Thorne, Phys. Rev. D 62 (2000) 122002

[57] A. Nakagawa, A.M. Gretarsson, E.K. Gustafson and M.M. Fejer, Phys.
Rev. D 65 (2002) 102001

[58] M. Cerdonio, L. Conti, A. Heidmann and M. Pinard, Phys.

Rev. D 63 (2001) 082003

[59] G.M. Harry, Incorporating Coating Anisotropy into Coating Thermal

Noise, LIGO technical note, LIGO-T0040029-00-R (2004)

[60] V.B. Braginsky, S.P. Vyatchanin, Phys. Lett. A 312 (2003) 244

[61] M.M. Fejer, S. Rowan, G. Cagnoli, D.R.M. Crooks, A.M. Gretarsson,

G.M. Harry, J. Hough, S.D. Penn, P. Sneddon, S.P. Vyatchanin, Phys.
Rev. D 70 (2004) 082003

[62] S.R. Rao, Mirror Themal Noise in Interferometric Gravitational Wave

Detectors, Ph.D. thesis, California Institute of Technology, Pasadena,
CA (2003)

[63] V.B. Braginsky, M.L. Gorodetsky and S.P. Vyatchanin, Phys.

Lett. A 271 (2000) 303
180 BIBLIOGRAPHY

[64] M. Punturo, Advanced Virgo Sensitivity Curve: a Possible Scenario,

Virgo note, VIR-NOT-PER-1390-283 (2004)

[65] J.E. Logan, J. Hough and N.A. Robertson, Phys. Lett. A 183 (1993)
145

[66] J. Kovalik and P.R. Saulson, Rev. Sci. Instrum. 64 (1993) 2942

[67] G.I. González and P.R. Saulson, J. Acoust. Soc. Am. 96 (1994) 207

[68] S. Rowan, G. Cagnoli, P.H. Sneddon, J. Hough, R. Route,

E.K. Gustafson, M.M. Fejer and V. Mitrofanov, Phys. Lett. A 265 (2000)
5

[69] K. Numata, Phys. Lett. A 276 (2000) 37

[70] K. Numata, Class. Quant. Grav. 19 (2002) 1697

[71] K. Numata, Phys. Lett. A 327 (2004) 263

[72] P. Willems and D. Busby, Report to the April 25, 2003 Core Optics
Downselect Commettee Meeting, LIGO technical note, LIGO-T030087-
00-R (2003)

[73] J.R. Smith, G. Cagnoli, D.R.M. Crooks, M.M. Fejer, S.J. Goßler,
H. Lück, S. Rowan, J. Hough and K. Danzmann, Class. Quant.
Grav. 21 (2004) S1091

[74] S.D. Penn, A. Ageev, D. Busby, G.M. Harry, A.M. Gretarsson, K. Nu-
mata and P. Willems, Frequency and Surface Dependence of the Me-
chanical Loss in Fused Silica, arxiv.org/abs/gr-qc/0507097 (2005)

[75] J. Wiedersich, S.V. Adichtchev and E. Rössler, Phys. Rev.

Lett. 84 (2000) 2718
BIBLIOGRAPHY 181

[76] S.D. Penn, G.M. Harry, A.M. Gretarsson, S.E. Kittelberger, P.R. Saul-
son, J.J. Shiller, J.R. Smith, S.O. Sword, Rev. Sci. Instrum. 72 (2001)
3670

[77] A. Ageev, B.C. Palmer, A. De Felice, S.D. Penn, P.R. Saulson, Class.
Quant. Grav. 21 (2004) 3887

[78] J.-Y. Vinet, Noises Produced by Opto-Thermal Couplings in Mirrors,

Virgo note, VIR-NOT-OCA-1390-166 (2001)

[79] K. Yamamoto, S. Otsuka, M. Ando, K. Kawabe and K. Tsubono, Phys.

Lett. A 280 (2001) 289

[80] K. Yamamoto, M. Ando, K. Kawabe and K. Tsubono, Phys.

Lett. A 305 (2002) 18

[81] K. Yamamoto, S. Otsuka, M. Ando, K. Kawabe and K. Tsubono, Class.

Quant. Grav. 19 (2002) 1689

[82] K. Yamamoto, S. Otsuka, Y. Nanjo, M. Ando and K. Tsubono, Phys.

Lett. A 321 (2004) 79

[83] A.M. Gretarsson, G.M. Harry, S.D. Penn, P.R. Saulson, J.J. Shiller,
W.J. Startin, Proc. 3rd Edoardo Amaldi Conf. on Grav. Waves, July
12-16 (1999)

[84] D.R.M. Crooks et al., Class. Quant. Grav. 19 (2002) 883

[85] S.D. Penn et al., Class. Quant. Grav. 20 (2003) 2917

[86] M.N. Inci, Simultaneous Measurements of Thermal Optical and Linear

Thermal Expansion Coefficients of T a2 O5 Films, ICO 19, Firenze, Italy,
August 25-30 (2002)
182 BIBLIOGRAPHY

[87] C.-L. Tien, C.-C. Jaing, C.-C. Lee, and K.-P. Chuang, J. Mod.
Opt. 47 (2000) 1681

[88] V.B. Braginsky and A.A. Samoilenko, Phys. Lett. A 315 (2003) 175

[89] S.P. Baker, C.R. Ottermann, M. Laube, F. Rauch, and K. Bange, Proc.
Mat. Res. Soc. Symp., Pittsburgh, PA, 436 (1997) 71

[90] A, Gillespie and F. Raab, Phys. Lett. A 178 (1993) 357

[91] A, Gillespie and F. Raab, Phys. Lett. A 190 (1994) 213

[92] G. Cagnoli, L. Gammaitoni, J. Kovalik, F. Marchesoni, M. Pun-

turo, S. Braccini, R. De Salvo, F. Fidecaro and G. Losurdo, Phys.
Lett. A 237 (1997) 21

[93] S. Traeger, B. Willke and K. Danzmann, Phys. Lett. A 225 (1997) 39

[94] S. Rowan, S.M. Twyford, J. Hough, D.-H. Gwo and R. Route, Phys.
Lett. A 246 (1998) 471

[95] A.Y. Ageev, I.A. Bilenko, V.B. Braginsky and S.P. Vyatchanin, Phys.
Lett. A 227 (1997) 159

[96] A.Y. Ageev, I.A. Bilenko and V.B. Braginsky, Phys. Lett. A 246 (1998)
479

[97] A.M. Gretarsson and P.R. Saulson, Monitoring the Thermal and Non-
Thermal Excitation of Fibers,
http://www.ligo.org/pdf public/saulson.pdf

[98] G. Cagnoli, L. Gammaitoni, J. Kovalik, F. Marchesoni and M. Punturo,

Phys. Lett. A 255 (1999) 230
BIBLIOGRAPHY 183

[99] V.B. Braginsky, V.P. Mitrofanov and K.V. Tokmamov, Phys.

Lett. A 218 (1996) 164

[100] G. Cagnoli, L. Gammaitoni, J. Hough, J. Kovalik, S. McIntosh,

M. Punturo and S. Rowan, Phys. Rev. Lett. 85 (2000) 2442

[101] S. Rowan, R. Hutchins, A. McLaren, N.A. Robertson, S.M. Twyford

and J. Hough, Phys. Lett. A 227 (1997) 153

[102] A.M. Gretarsson, G.M. Harry, S.D. Penn, P.R. Saulson, W.J. Startin,
S. Rowan, G. Cagnoli and J. Hough, Phys. Lett. A 270 (2000) 108

[103] Ansys version 8.1, ANSYS Inc., Canonsburg, PA, USA (2004)
http://www.ansys.com

[104] A.V. Aho, B.W. Kernighan, P.J. Weinberger, The AWK Programming
Language, Addison-Wesley (1988)

[105] M. Punturo, F. Travasso, Evaluation of the bulk loss angle of the Virgo
Mirrors from C2 data, Virgo note, VIR-NOT-PER-1390-263 (2004)

[106] D.R.M. Crooks et al., Class. Quant. Grav. 21 (2004) S1059

[107] P.H. Sneddon, S. Bull, G. Cagnoli, D.R.M. Crooks, E.J. Elliffe,

J.E. Faller, M.M. Fejer, J. Hough and S. Rowan, Class. Quant. Grav. 20
(2003) 5025

[108] S.M. Hu, J. Appl. Phys. 53 (1982) 3576

[109] D.F. McGuigan et al., J. Low Temp. Phys. 30 (1978) 621

[110] C.C. Lam and D.H. Douglas, Phys. Lett. A 85 (1981) 41

[111] Material Properies Database (MPDB), http:/www.jahm.com

184 BIBLIOGRAPHY

[112] G. Cagnoli, L. Gammaitoni, J. Kovalik, F. Marchesoni and M. Punturo,

Rev. Sci. Instrum. 71 (2000) 2206

[113] D.H. Yoon and T. Fukuda, J. Cryst. Gr. 144 (1994) 201

[114] V. Chani, A. Yoshikawa, Y. Kuwano, K. Hasagawa and T. Fukuda,

J. Cryst. Gr. 204 (1999) 155

[115] http://www.ni.com/labview

[116] S.K. Ghandi, VLSI Fabrication Principles, Wiley, New York,

2nd ed. (1994)

[117] Matlab version 6.5, The MathWorks Inc., Natick, MA, USA (2002)
http://www.mathworks.com