The VIRGO Physics Book, Vol.

II

OPTICS and related TOPICS
The Virgo collaboration

1st release : Feb. 2001

March 24, 2005

2

Contents
1 Theory of GW Interferometers
1.1 Shot noise limited interferometry . . . . . . . . . . 1.1.1 Spectral density of power equivalent to SN . 1.1.2 Partially re ecting mirrors . . . . . . . . . . 1.1.3 Elementary Michelson . . . . . . . . . . . . 1.1.4 Frequency stability requirements . . . . . . 1.2 The Fabry-Perot resonant cavity . . . . . . . . . . . 1.2.1 Conventions used throughout this section . . 1.2.2 The Pound-Drever scheme . . . . . . . . . . 1.2.3 The double Fabry-Perot cavity . . . . . . . . 1.3 Optics in a wave Space-Time . . . . . . . . . . . . . 1.3.1 Retarded time in a GW - Simpli ed picture 1.3.2 Retarded time in a GW - General picture . . 1.3.3 The A133 Algebra . . . . . . . . . . . . . . 1.4 Signal to Noise Ratio . . . . . . . . . . . . . . . . . 1.5 Resonant cavities in a GW . . . . . . . . . . . . . . 1.6 Michelson Interferometer involving FP cavities . . . 1.7 Recycling . . . . . . . . . . . . . . . . . . . . . . . 1.7.1 standard power recycling . . . . . . . . . . . 1.7.2 detuned power recycling . . . . . . . . . . . 1.7.3 Synchronous Recycling . . . . . . . . . . . . 1.7.4 Signal recycling . . . . . . . . . . . . . . . . 1.7.5 The signal extraction regime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11
11 11 14 15 19 21 21 30 32 35 35 38 42 46 47 49 54 54 61 63 79 84

2 Beam optics and Interferometers

2.1 introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 2.2 A short theory of di raction . . . . . . . . . . . . . . . . . . . 91 2.2.1 The Helmholtz equation . . . . . . . . . . . . . . . . . 91 3

91

4 2.2.2 2.2.3 2.2.4 2.2.5

CONTENTS
The Kirchho integral . . . . . . . . . . . . . . . . . . 92 Application of the Kirchho equation . . . . . . . . . . 95 Consistency of the Kirchho equation . . . . . . . . . . 101 The Fresnel approximation and the paraxial di raction equation (PDE) . . . . . . . . . . . . . . . . . . . . . . 107 2.2.6 The Fraunhofer approximation . . . . . . . . . . . . . . 111 2.2.7 Representation of optical elements . . . . . . . . . . . . 112 Fundamental TEM mode . . . . . . . . . . . . . . . . . . . . . 115 Discrete bases for free space propagation . . . . . . . . . . . . 118 2.4.1 Hermite-Gauss modes . . . . . . . . . . . . . . . . . . 119 2.4.2 The Laguerre-Gauss modes . . . . . . . . . . . . . . . 124 Fabry-Perot: paraxial approximation . . . . . . . . . . . . . . 128 Hypergaussian modes . . . . . . . . . . . . . . . . . . . . . . . 135 2.6.1 construction . . . . . . . . . . . . . . . . . . . . . . . . 135 2.6.2 Angular aperture and Fourier transform . . . . . . . . 136 2.6.3 Normalization . . . . . . . . . . . . . . . . . . . . . . . 137

2.3 2.4 2.5 2.6

3 Numerical methods

3.1 Numerical propagation using Fourier transforms . . . . . . . . 142 3.1.1 On the discrete Fourier transform . . . . . . . . . . . . 142 3.1.2 FFT-based propagation algorithms . . . . . . . . . . . 147 3.1.3 Finding the eld re ected o a resonant cavity . . . . . 153 3.1.4 The Michelson Interferometer . . . . . . . . . . . . . . 157 3.1.5 The power-recycled Michelson interferometer . . . . . . 160 3.1.6 On the intrinsic limitation to basic DFT-based algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . 170 3.1.7 Propagation with magni cation . . . . . . . . . . . . . 171 3.1.8 O -axis propagation . . . . . . . . . . . . . . . . . . . 176 3.2 Hankel transform methods . . . . . . . . . . . . . . . . . . . . 179 3.2.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 3.2.2 Numerical implementation . . . . . . . . . . . . . . . . 185 3.3 Modal expansion . . . . . . . . . . . . . . . . . . . . . . . . . 194 3.3.1 Return to the HG family of modes . . . . . . . . . . . 194 3.3.2 Tilted mirrors . . . . . . . . . . . . . . . . . . . . . . . 196 3.3.3 Parallel translations of the beam . . . . . . . . . . . . 198 3.3.4 Mismatching . . . . . . . . . . . . . . . . . . . . . . . 200 3.3.5 Clipped mirrors . . . . . . . . . . . . . . . . . . . . . . 203 3.3.6 O set and clipping . . . . . . . . . . . . . . . . . . . . 213

141

CONTENTS
3.3.7 Mismatched beams . . . . . . . . . . . . . . . . 3.3.8 Coupling of astigmatic beams . . . . . . . . . . 3.3.9 Properties of the Displacement polynomials . . 3.3.10 Structural properties of Displacement matrices . 3.3.11 Magnitude of displacement matrix elements . . 3.3.12 Numerical results . . . . . . . . . . . . . . . . . 3.3.13 The A266 Algebra . . . . . . . . . . . . . . . . 3.4 Monte-Carlo methods . . . . . . . . . . . . . . . . . . . 3.4.1 Spatial spectra, plane waves and photons . . . . 3.4.2 Propagation . . . . . . . . . . . . . . . . . . . . 3.4.3 Di raction patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5 . 214 . 215 . 217 . 218 . 220 . 224 . 231 . 236 . 236 . 237 . 240 . 243 . 244 . 246 . 247 . 247 . 247 . 251 . 251 . 254 . 259 . 262

4 Real mirrors

4.1 Multilayer coatings . . . . . . . . . . . 4.1.1 Dioptric matrix . . . . . . . . . 4.1.2 Models of stacks . . . . . . . . 4.1.3 Numerical codes . . . . . . . . . 4.2 Surface maps . . . . . . . . . . . . . . 4.2.1 Collimation and attening . . . 4.2.2 Weighted RMS roughness . . . 4.2.3 2D interpolation techniques . . 4.2.4 Backcoupling due to roughness 4.2.5 Zernike polynomials . . . . . . 4.2.6 Roughness and scattering losses

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

. . . . . . . . . . .

243

5 Scattered light 6 Heating issues

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265 5.2 Scattering mirrors . . . . . . . . . . . . . . . . . . . . . . . . . 266 5.3 The scattering coherence function . . . . . . . . . . . . . . . . 268 . . . . . . .

265

6.1 Heating by dissipation in the coating . . . . . . . . . . . . 6.1.1 The Fourier equation and the boundary conditions 6.1.2 Solution as a Dini expansion . . . . . . . . . . . . . 6.1.3 Thermal lensing . . . . . . . . . . . . . . . . . . . . 6.2 Heating by dissipation in the bulk substrate . . . . . . . . 6.2.1 Temperature eld . . . . . . . . . . . . . . . . . . . 6.2.2 Thermal lensing . . . . . . . . . . . . . . . . . . . .

. 273 . 274 . 276 . 279 . 286 . 286 . 288

273

6 6.3 Distortion from coating absorption . . . . . . . . . . 6.3.1 Thermoelastic solution . . . . . . . . . . . . . 6.3.2 Surface analysis . . . . . . . . . . . . . . . . . 6.4 Distortion caused by bulk absorption . . . . . . . . . 6.4.1 Thermoelastic solution . . . . . . . . . . . . . 6.4.2 Surface analysis . . . . . . . . . . . . . . . . . 6.5 Heating processes . . . . . . . . . . . . . . . . . . . . 6.5.1 Transient temperature elds : general method 6.5.2 Transient thermoelastic deformations . . . . . 6.6 Thermoelastic coupling : Coating absorption . . . . . 6.6.1 Dynamical temperature . . . . . . . . . . . . 6.6.2 Dynamical thermal surface distortions . . . . 6.7 Thermoelastic coupling : Bulk absorption . . . . . . . 6.7.1 Dynamical temperature . . . . . . . . . . . . 6.7.2 Dynamical thermal distortions . . . . . . . . .

CONTENTS
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289 . 290 . 296 . 299 . 300 . 302 . 303 . 303 . 311 . 315 . 316 . 323 . 327 . 327 . 330 . 344 . 346 . 348 . 351 . 351 . 352 . 353 . 353 . 356 . 356 . 358 . 362 . 367 . 370 . 370 . 373 . 375 . 376 . 379

7 Mirrors standard thermal noise
7.1 7.2 7.3 7.4

7.5 7.6

7.7

7.8

Damped harmonic oscillator . . . . . . . . . . . The FD theorem . . . . . . . . . . . . . . . . . The Levin generalized coordinate method . . . . Basic linear elasticity . . . . . . . . . . . . . . . 7.4.1 displacement, strain, stress . . . . . . . . 7.4.2 Elastodynamics equation . . . . . . . . . 7.4.3 Boundary conditions . . . . . . . . . . . Mirror as a half-space . . . . . . . . . . . . . . . Finite mirrors . . . . . . . . . . . . . . . . . . . 7.6.1 A solution to the equilibrium equations . 7.6.2 Boundary conditions . . . . . . . . . . . 7.6.3 Strain Energy . . . . . . . . . . . . . . . 7.6.4 Some numerical results . . . . . . . . . . Non gaussian beams . . . . . . . . . . . . . . . 7.7.1 Half-space approximation . . . . . . . . 7.7.2 Finite test mass approximation . . . . . 7.7.3 Numerical results . . . . . . . . . . . . . 7.7.4 Realistic modes . . . . . . . . . . . . . . Mirror distortions and energy maps . . . . . . .

. . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . .

343

CONTENTS

7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389 . 392 . 393 . 394 . 394 . 396 . 396

8 Thermoelastic noise

8.1 Introduction . . . . . . . 8.2 Case of in nite mirrors . 8.2.1 Gaussian beams . 8.2.2 Flat beams . . . 8.3 Case of nite mirrors . . 8.3.1 Gaussian beams . 8.3.2 Flat modes . . .

389

9 Modulation and Transfer functions

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399 9.2 Elementary perturbations and audio sidebands . . . . . . . . . 400 9.2.1 Perturbation of mirrors by small displacements . . . . 400 9.2.2 Perturbation of a vacuum by a gravitational wave . . . 402 9.2.3 Algebra of rst order perturbations . . . . . . . . . . . 403 9.3 Interferometer operators . . . . . . . . . . . . . . . . . . . . . 404 9.3.1 Cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . 404 9.3.2 Michelson . . . . . . . . . . . . . . . . . . . . . . . . . 406 9.3.3 Recycled interferometer transmittance and re ectance . 408 9.4 Tuning the interferometer . . . . . . . . . . . . . . . . . . . . 409 9.4.1 Tuning long cavities . . . . . . . . . . . . . . . . . . . 410 9.4.2 Tuning at a dark fringe . . . . . . . . . . . . . . . . . . 411 9.4.3 Tuning the recycling cavity . . . . . . . . . . . . . . . 412 9.5 Modulation, Detection, Demodulation and Transfer functions . 412 9.5.1 General case . . . . . . . . . . . . . . . . . . . . . . . . 412 9.5.2 The special case of quantum noise . . . . . . . . . . . . 415 9.5.3 Transfer functions to an equivalent h(f) . . . . . . . . 418 9.6 Interferometer noises . . . . . . . . . . . . . . . . . . . . . . . 419 9.6.1 Proof masses position noise . . . . . . . . . . . . . . . 419 9.6.2 Quantum noise . . . . . . . . . . . . . . . . . . . . . . 422 9.6.3 Sensitivity curve . . . . . . . . . . . . . . . . . . . . . 422 9.7 Upstream noises . . . . . . . . . . . . . . . . . . . . . . . . . . 423 9.7.1 Laser frequency noise . . . . . . . . . . . . . . . . . . . 423 9.7.2 Laser amplitude noise . . . . . . . . . . . . . . . . . . 426 9.7.3 Modulator noise . . . . . . . . . . . . . . . . . . . . . . 428

399

8

CONTENTS

although obviously some general principles are recalled. This is not a compilation of Virgo notes. nor a course of physics. of the materials. Evolution of the technology. The principle of the document is thus to present all theoretical contributions of the Virgo teams in a comprehensive way. The second reason is to prevent loss of memory in the collaboration. 9 . and check wether the hierarchy is still valid.Preface This is one of the volumes of the Virgo Physics Book. although obviously it contains some of them. etc. This is necessary for several reasons. It is signed by "the collaboration". in a comprehensive document. have been carried out on request and with the help of it as a whole. the sum of the theoretical researches carried out during the R&D period (not fully over. It is convenient to be able to nd in one place the state of the art in the various domains of modeling. for instance. considering future improvements). even if the relation with Virgo is not direct. the principles must be available. In this spirit. because all theoretical or numerical works reported in. some "frequently asked questions" are treated. The rst reason is to present outside the collaboration. in order to show what physical e ects have been studied. For doing this. The third reason is to provide a reference document for the young searchers entering the collaboration allowing them to become e ciently acquainted with the principles of the experiment. The aim of this enterprise (VPB) is to keep track of the various theoretical or numerical studies carried out during the development of the Virgo conceipt and its realization... makes necessary to re-estimate from time to time the orders of magnitude of di erent sources of noise.

of interferometers).fr N. distortions).B. but only a virtual presence on the net. Moreover. The various principles of optical modeling are presented (propagation of light. simulation of cavities. about thermal noise issues. It is clear that further contributions will be added from time to time.. this document will be improved by correcting misprints and errors. Please report possible errors or misprints to vinet@obs-nice.. The principles of interferometry are recalled. and this is the reason why it will not have a general paper version. about seismic noise suppression and superattenuator physics and possibly other topics ? and for which future contributors are known. about General Relativity and gravitational wave theory basic background. and also the methods of analyzing mirrors.: From time to time. The modulation-demodulation theory is also described. and the part of thermal noise studies related to the mirror substrates though a special volume should be dedicated to thermal noise issues. theory of the interaction of GW with detectors. other volumes are expected. or adding some material. . The opto-thermal problems are studied (thermal lensing.. This is the present status of this document.10 CONTENTS This Volume II contains some works related to optics and other issues strongly related to optics. then the various con gurations of interferometers.

In fact the light is produced and received as a ux of photons. at the microscopic level as a random stationnary process having a mean in agreement with a classical theory. It is classically shown that the expectation value of a random variable N obeying a Poisson law of parameter m is E N ] = m. but a variance that can be understood only by reference to quantum theory.1.1 Spectral density of power equivalent to SN . and in concrete terms. during the time interval t. the Poisson law is identical to a gaussian law having the same moments. that during a time interval t.1 Shot noise limited interferometry Shot noise is produced by photodetectors currently used in all domains of photonics. This means that the probability of detecting exactly n photons is: n ?m m pn = e n! where m is the only parameter of the Poisson probability distribution. the photocurrent appears. for instance.Chapter 1 Theory of GW Interferometers 1. represents the mean photon ux. and it is shown. the number photons that a photodiode can detect is a random variable N whose probability law is Poissonnian (a general law for all processes consisting in random arrivals). Even with very stable lasers and cooled detectors. and it variance is V N ] = m. the energy deposited on the diode is e = P t = NhP 11 1. if the mean number of photons is larger than about 50. On the other hand. In fact.

and saying that P is a random variable. so that we shall ignore in all the sequel.12 CHAPTER 1. We can write the function x(t) as : x(t) = xk for k t < t < (k + 1) t The spectral density of any stationary centered process has the general de nition : 2 3 1 E 4 Z T e?i tx(t) dt 25 Sx( ) = Tlim T !1 0 . We shall consider in the sequel that given an incoming power P0. Calling P0 the averaged value of P . and its frequency (hP is the Planck constant). In other words. consider P as the power actually detected. the two-sided spectral density of power equivalent to shot noise is 0 SP (f ) = P0hP (1. it is clear that there is an equivalence between saying that N is a random variable.1) The fact that the preceding formula gives actually the two-sided SD can be shown as follows. calling x the statistical variable P ? P0. x(t) de nes a stationary centered stochastic process. then the quantity P0 h appears as a white spectral density. the detected energy (and consequently the averaged power) is a random variable of mean P0 . a quantity very close to 1 in present infra-red detectors. Now. On successive time slices of duration t. we see that E N ] = P0 t hP and consequently (Poisson) : V N ] = P0 t hP It is now possible to consider the variance of P : 2 2 V P ] = V N ] hP t2 = P0hP t The quantity 1= t may be regarded as the ideal bandwidth of the detector. THEORY OF GW INTERFEROMETERS where P is the power of the light beam. so that. is the quantum e ciency of the detector.

so that E x k xm ] = V P ] and km 2Z 23 T ?i t X E 4 e x(t) dt 5 = V P ] t2 sinc( 0 k t=2)2 = n tV P ] tsinc( t=2)2 (with the de nition: sinc(x) sin(x)=x). and the one sided spectral density to be used in practical problems is simply : SP ( ) = 2P0 hP as for a white noise. SHOT NOISE LIMITED INTERFEROMETRY If we choose T an integer multiple of t.1. so that the preceding function is almost at in the audio region. we get easily : ZT n?1 X ?i(k+ ) t e?i t x(t) dt = xk e t sinc( 0 1 2 13 k=0 t=2) so that ZT 0 e?i t x(t) dt = 2 X k. and with T = n t. .m xk xm e?i(k?m) t t2 sinc( t=2)2 The variables xk are uncorrelated. this is nally 0 SP ( ) = P0 hP sinc( t=2)2 One easily sees that the total variance is recovered by R integrating over neg1 ative and positive frequencies (and remembering that ?1 sinc(x)2 dx = ) The single-sided spectral density is thus : SP ( ) = 2P0 hP sinc( t=2)2 The integration time t can be chosen very short.1.

1: Partially re ecting mirror In interferometry. There are two complex numbers zR and zT expressing respectively the relative re ected and transmitted waves. the source of A being switched o . a few 10?6 ). it is mandatory to have a very small p (usually a few ppm. We consider a mirror as a plane surface of vanishing thickness.e. i. it undergoes exactly the same processes with the same coe cients (the mirror is invariant in a space re ection). we have (see Fig. Requirements on the arguments of zR and zT come from the mirror viewed as a 4 ports element.14 CHAPTER 1. Namely. When the two amplitudes are present simultaneously.1 for notation): AR = zRA . we have thus: AR = zRA + zT B 1.2 Partially re ecting mirrors . AT = zT A Conservation of the total power requires that jzRj2 + jzT j2 = 1 ? p where p expresses possible absorption (dissipation) in the mirror.1. when a wave of complex amplitude A reaches the mirror's surface. It is mandatory to take into account the phase jumps caused by re ection or transmission at a mirror surface. THEORY OF GW INTERFEROMETERS B A BR AR AT Figure 1. If a second wave of amplitude B reaches the mirror coming from the opposite direction.1. a light source provides a beam that is often splitted into two or more waves propagating along di erent paths. For our present purposes.

3 Elementary Michelson . The light coming from a laser is split into two distinct paths ended by mirrors. SHOT NOISE LIMITED INTERFEROMETRY 15 BR = zT A + zRB Remark that we call AR the sum of all waves going to the left. We call rs and ts the re ection and transmission coe cients of the splitter. t) are real numbers verifying r 2 + t2 = 1 ? p A simple interferometer design is shown on Fig. in the calculation.1. then re ected and recombined on the splitter where the interference occurs.2.1. that will be kept throughout this document. in terms of arguments: In order to preserve power balance at each interference occuring at the surface of a mirror. we must have jARj2 + jBRj2 = (1 ? p) jAj2 + jB j2 on the other hand. and BR the sum of all waves going to the right. and k the wave 1. take into account this phase jump of =2 between the re ected and the transmitted wave.1. B ) of complex numbers.1. z T = t where (r. we get jARj2 + jBRj2 = jzRj2 + jzT j2 jAj2 + jB j2 + (zRzT + zRzT ) AB + AB we therefore must have (zRzT + zRzT ) AB + AB = 0 for any couple (A. we must. we could as well call BT and AT the same waves. is Arg(zR) ? Arg(zT ) = (2n + 1) 2 (n 2 N) zR = i r . which clearly requires zRzT + zRzT = 0 or. If we consider the power balance. One possible choice. using the preceding equations.

if now = (2n + 1) .16 CHAPTER 1. we see that the outgoing power can be controlled by . We can consider for brevity that the splitter is well balanced and 2 rs = t2 = 1=2. (r1 ? r2)2 P PDC.d = 0 4 . r2 are reasonably near unity. if x(t) = 0. The outgoing power is : s P (t) = PDC + P (t) with 2 2 PDC = 1 P0 r1 + r2 + 2r1r2 cos 4 where = 2k(a0 ? b) is the static tuning of the interferometer. The amplitude of the laser wave is A and the outgoing is B . the length of arm 1 is a = a0 + x(t).b = (r1 + r2) P0 4 which is almost 1 if both r1.2: A simple Michelson experiment number (k 2 = . One has B = rs ts r1e2ika + r2e2ikb so that 2 2 2 BB = rs t2AA r1 + r2 + 2r1r2 cos 2k(a ? b)] s Suppose now that the device aims to measure a very small variation of the length of one arm. we say that the interferometer is tuned at a bright fringe. For instance. being the wavelength. If = 2n . 2 PDC. THEORY OF GW INTERFEROMETERS r 2 b Laser A rs t s B a r 1 Figure 1. where jx(t)j .

One already sees that if the two coe cients are close to 1 . It is easily seen that the optimal value 0 is such that cos 0 = ? r< r> where r< is the smallest of r1.3). the tuning of the interfometer is near a dark fringe. due to shot noise.and r> the largest. SHOT NOISE LIMITED INTERFEROMETRY 17 which can be made as small as wanted by equalizing r1 and r2.1. The answer is given by computing the signal to noise ratio : ( (f ) = S Pff)) S ( P The spectral density SP of power equivalent to shot noise is : 2 2 SP (f ) = 1 P0hP r1 + r2 + 2r1 r2 cos 2 The spectral density of signal is : 2 2 S P (f ) = r1 r2 P02 sin2 k2Sx(f ) where Sx(f ) is the SD of x viewed as a stationnary process. and this determines the contrast of the inteferometer. We say that the interferometer is tuned at a dark fringe. In practice. We have thus 2 2 (f ) = 2r1 r2 hP0 f ( ) k2Sx(f ) P where sin2 f ( ) = r2 + r2 + 2r r cos 1 2 1 2 (see Fig.1. even in the absence of signal. If x is not zero. r2 . we have f ( 0) = r1 2 > . knowing that there is a uctuation of the power. it is not so easy to make r1 = r2.1. there is a time varying component P (t) = r1r2P0kx(t) sin The question is now : What is the minimum variation x that we could detect. When optimally tuned.

4 0.6 f(α) CHAPTER 1.7 0.9 0.5 0. and we have replaced the motion of both mirrors of hL=2 by a unique motion of mirror 1 by hL. THEORY OF GW INTERFEROMETERS 0. we have x(t) = L h(t) where L is the roughly equal arm lengths of the arms.3: Optimization of the SNR so that the optimal SNR is 2 P (f ) = 2r< h 0 k2 Sx(f ) The minimum detectable x can be evaluated by taking = 1.0 0.2 0.0 2. . If further we assume that the small displacement x(t) is caused by a gravitational wave h(t).0 α [Rd] π α0 3.8 0. and this gives 1 h Sx(f )min = 2r2 k2PP 0 < It is more physical to consider the root spectral density : s 1 Sx=2(f ) = 4 2hP P0 where we have set r< 1.3 0.18 1.0 Figure 1.0 2.1 0.5 4.5 3.

The laser source is not in practice a purely monochromatic source. not by upsizing the laser itself. We shall represent the laser optical amplitude as: 1. The shot noise is not the only limitation to laser metrology.1. which means the distance between mirrors. but by creating a resonance surtension on the Michelson increase the arm length. we get 1 Sx=2(f ) 1 Sh=2(f ) 1:064 m of the 1:2 3:8 10?17 m Hz?1=2 10?21 Hz?1=2 With a 3 km arm length. SHOT NOISE LIMITED INTERFEROMETRY The root spectral density of h equivalent to shot noise is nally: s 2hP 1=2(f ) = Sh 4 L P 0 19 With the Virgo laser (P0 20 W) and the wavelength Nd:YAG ampli er. The laser frequency is determined by the optical length of the laser cavity. not by adding kilometers of tunnels. this gives In fact. All these parameters are in general coupled to external sources of mechanical or thermal noise. but by creating a resonance in the 3 km arms Creation and characterics of resonances are thus a very important item we are going to analyze and discuss in details.1. so that the instantaneous frequency of the laser may be viewed as a random process. this means that two orders of magnitude are missing for having some hope to detect gravitational waves. but also the index in the ampli er medium. We shall see that these two orders can be gained by enhancing the laser power.1. according to the theoretical litterature. and the index of the medium in between mirrors and ampli er medium.4 Frequency stability requirements Alaser = A0 e?i! t ei 0 (t) .

if we admit a 1% asymmetry rate. less easy to symmetrize than actual geometrical lengths. The power reaching the photodetector is: h2 2 i P (t) = 1 r1 + r2 + 2r1r2 cos 2k(b ? a) + (t ? 2a=c) ? (t ? 2b=c)] 4 We have thus a spurious phase: 1 (t) = 2 (t ? 2a=c) ? (t ? 2b=c)] = 1 (t ? (a + b)=c + (b ? a)=c) ? (t ? (a + b)=c ? (b ? a)=c)] 2 ' b ? a @@t (t ? (a + b)=c) c assuming the di erence d b ? a small compared to the coherence length of the laser. and (t) a random centered process. which is: s 2hP sn = P we must obtain a sepctral density of frequency noise: s c 2hP (f ) < 2 d P we see the importance of having a good symmetry (a small d) between the two arms. where !0=2 is the nominal frequency of the laser.20 CHAPTER 1. If we take the parameters already used above. because rstly we want a safety margin of at least 1 order of magnitude with respect to the shot noise. We have thus (t) = d 2 (t) c . so that the requirement is rather in the range of 10?6 Hz=Hz1=2. secondly the shot noise will be reduced by 1 order of magnitude by recycling. this results in a requirement of (f ) < 2:10?3 Hz=Hz1=2 The realistic situation is even more demanding. anf nally. the shot noise induced phase was about 10?10 Rd=Hz1=2. the arm lengths will be seen to result from resonance e ects. This implies that if we want to reduce the corresponding phase noise to a level comparable to the shot noise. THEORY OF GW INTERFEROMETERS where (t) is the instantaneous frequency.

it is partially re ected and partially transmitted. the interference of the incoming wave and the returning wave is constructive and a strong intracavity wave builds up. A Fabry-Perot cavity is made of two parallel mirrors. We keep the convention explicited above: Ain being the incoming amplitude. pi (i = 1. x) = A(x) e?i!t A simple propagation step along a path of length L in a vacuum is therefore represented by a phase factor. transmitted and absorbed. and moreover.4. As seen above.1 Conventions used throughout this section A(t. We call ri. and .2. Atrans the transmitted. ti . Atrans = t Ain r. it is partially re ected.2. The transmitted wave is re ected by mirror 2. When light enters the cavity through mirror 1. we have : Aref = ir Ain . accounting for absorption in the coating or scattering into a di erent mode due to mirror geometrical imperfections (p can be as low as a few ppm (10?6 ) for supermirrors as Virgo's). t being respectively the re ection and transmission coe cients of the mirror (real numbers). On Fig. light can be stored. then returns to mirror 1 where it is recombined with the incoming wave and partially transmitted to the exterior.. and the relation between amplitudes will be A(x + L) = eikL A(x) with k = !=c = 2 = .1. when a light ray encounters a mirror. 2).2 The Fabry-Perot resonant cavity We assume a monochromatic light source. If the phase after a round trip in the cavity allows it. and we describe in the present section the (ideal) light beam circulating inside the interferometer as a plane wave. We have the power balance : r 2 + t2 = 1 ? p where p is the loss coe cient. so that the optical eld at any place x of an optical system is of the scalar form 1. c being the velocity of light. Aref the re ected. we have spatially separated the left and right propagating waves for the sake of clarity. the parameters of the mirrors.1. we consider a given component of the electric eld. THE FABRY-PEROT RESONANT CAVITY 21 1.

we have a series of resonant frequencies 1 c n = n+ 2 2L so that The spacing between two successive resonances is called Free spectral Range (FSR). The inverse case will be presented later. so that the integer n is close to 6 109. so that = n + . The length of the cavity is L. whereas the optical frequency (at = 1:06 m) is about 3 1014 Hz. c FSR = 2L For a 3 kilometers cavity (as in VIRGO).4: Fabry-Perot cavity Ain the incoming wave. We rst discuss the case when the length of the cavity is xed. The ratio S = B=Ain is called surtension factor . THEORY OF GW INTERFEROMETERS B A in L A ref M1 M2 Figure 1. the FSR is close to 50 kHz. and noted FSR. For a given L.22 CHAPTER 1. and the frequency of light variable. Its maximum value is Smax = 1 ?t1 r r1 2 The width of the resonance line may be evaluated as follows. We can write the interference at M1 for the intracavity wave as : B = t1 Ain ? r1r2 e2ikLB 1 B = 1 + r tr e2ikL Ain 1 2 Clearly a resonance occurs when e2ikL = ?1. We assume that the frequency is close a resonance.

F contains only a photometric F FSR . 23 We have 2kL = (2n + 1) + 2 FSR The surtension coe cient takes on the form t1 S = 1 ? r1r2 exp 2i FSR Its square modulus gives the ratio between the intensities : jS j2 = This is t2 1 2 + 4r1 r2 sin (1 ? r1r2) FSR 1 i2 1 + 1?r r sin FSR Proximity of the resonance allows to replace the sine by its argument.2) and the Full Width at Half Maximum (FWHM) of the resonance is nally : FWHM for the nesse of the cavity.2.1. The values of such that the surtension is half its maximum are : FSR = 2F = One can note that we have described the cavity by an extra set of parameters F and FSR equivalent to r1r2 and L. so that 1 2 jS j2 = Smax i2 h 1 + 2F FSR 2 jS j2 = Smax h 2pr r 1 2 1 2 with the following de nition pr1r2 F = 1?r r 1 2 (1. THE FABRY-PEROT RESONANT CAVITY with FSR .

5 0.5.9 1.0 δν/∆νFSR Figure 1.8 0.1 0.3) 1 1 2 2 For a cavity operated in the re ection mode.24 10 0 CHAPTER 1.7 0. THEORY OF GW INTERFEROMETERS 10-1 |S/Smax|2 10-2 10-3 10-4 0.1. The wave re ected o the cavity can be computed by Aref = ir1Ain + ir2t1e2ikLB by substituting the value of B. and F = 100 (dotted line). The exact expression for the resonance can be written under the form 2 jS j2=Smax = h 2F 1 i2 1 + sin FSR see Fig.5: Resonance line shape for a nesse of F = 10 (solid line). we get Aref = i R Ain Where R is the re ectance of the cavity. de ned as (1 ? ) 2 e2ikL R = r1 + + r rp1erikL (1.2 0. having a nite re ectivity of the input mirror (M1). Frequency unit is FSR information about mirrors.0 0.6 0. whereas FSR contains a geometrical information about the cavity. a high re ectivity end mirror (M2) and reasonable .4 0.3 0.

If now.7). Note that = 0:5 FWHM correspond to half the maximum absorption and to a dephasing of =2 with respect to resonance. and can be better understood in a simpli ed model.85 and r2=0.6: Absorption line of a cavity for r1=0. This is classical in all oscillators.9996 0.1. THE FABRY-PEROT RESONANT CAVITY 1.99998. it can be seen that the global re ectance is about unity.1.6 and Fig.9992 0. losses (p1.2.9990 -5 -4 -3 -2 -1 0 1 2 3 4 5 1-A/2 1-A δν/δνFWHM Figure 1.9995 0. in terms of displacement. A is the maximum of absorption.9994 0. we have resonant lengths given by Ln = n + 1 2 2 showing that the displacement separating two successive resonances is : LFSR = 2 It is easy to show that the width of the resonance. with a small peak of absorption at resonance. is LFWHM = 2F . which is ideally the case in a GW interferometer.9997 0. p2).9991 0. the frequency of the light source is xed and the length of the cavity variable.9993 0.9999 25 Intensity reflectance 0.9998 0.0000 0.(Finesse F ' 19:3). The phase of the re ected wave undergoes a rapid transition of 2 when crossing the resonance (see Fig.1. instead of resonant frequencies.

(1.36 1. it is possible to compute r1r2 from F : s 2 2 r1r2 = 1 ? F 1 + 4F 2 ? 2F 2 If F is much larger than 1.(1.57 0. A key parameter is indeed the nesse.79 0. de ned by eq.14 2.3). We assume a frequency that is slightly detuned with respect ot resonance by an amount so that : 2kL = 2k L + 2 = +2 1f 0 where the reduced frequency f is the ratio of the o set to linewidth : FSR F f FWHM .50 CHAPTER 1. Conversely.2) and depending only on the parameter r1r2. THEORY OF GW INTERFEROMETERS Phase reflectance 4.99998 We develop now an approximate model of a cavity relying on the fact that the nesse is large compared to unity. It will prove useful for our further discussions of more complex systems involving cavities. and the phase factor 2kL.28 5.26 6.00 -5 3π/2 π π/2 -4 -3 -2 -1 0 1 2 3 4 5 δν/δνFWHM Figure 1.93 3. we can limit the expression at the rst order in 1=F . and take r1r2 = 1 ? F Consider now the re ectance of the cavity Eq.7: Phase re ectance of a cavity for r1=0.71 3.85 and r2=0.

assuming p very small. THE FABRY-PEROT RESONANT CAVITY with 2k0L mod 2 ]. at resonance. where p accounts for all losses in the cavity. or optimal coupling. the incoming eld is increasingly absorbed by the cavity until . we have : 27 ? (1 1) 2 e2i f=F r2R = r1r2 1 ? r? pe2ir2f=F r 1 2 (1. By expanding r2R at rst order in 1=F we get : + r2R = ? 1 ? pF =2if 2if 1? The quantity = pF = is called coupling rate and it is easily seen that 0 < < 2.1. whence ! 0 < r1r2 < 1 ? p = 1 ? p=2 0 < 1 ? F < 1 ? p=2 q ! 0 < pF < 2 We see that the re ectance at resonance is Most of the properties of the FP cavity can be known by only knowing its coupling rate. The re ectance of the cavity can thus be written (by putting r2 ' 1 at this point) : 2 R = ? 1 ? ? +if if 1 2 R(0) = ?(1 ? ) so that = 1 corresponds to total absorption of light.2. For running from 0 to 1 the cavity is overcoupled.4) 2 We set (1 ? p1 )r2 = (1 ? p). We have indeed obviously 2 0 < r1 < 1 ? p 1 2 2 2 ! 0 < r1 r2 < (1 ? p1)r2 = 1 ? p then. this means that.

and viceversa. call the undercoupling constant. The intensity re ection coe cient is : jRj2 = 1 ? 1(2 ?f 2) +4 The re ected phase is : ! ?1 2f Arg R] = + tan 1 ? + tan?1 (2f ) (1. there is no re ected wave) total absorption.8 -5 -4 -3 -2 -1 0 1 2 3 4 5 δν/δνFWHM Figure 1.95 3π/2 σ = 0. THEORY OF GW INTERFEROMETERS σ = 0.05 0 σ = 1. The surtension coe cient is de ned by 2 S = j1 + r tr1 e2ikLj2 1 2 . and the re ected phase becomes more and more unsensitive to frequency detuning (see Fig. then past 1. the re ectivity becomes near unity (because the input mirror becomes the only part of the FP visible from the exterior) . We should better therefore.8).1. Note that the coupling is strong when the coupling constant is weak. the eld is decreasingly absorbed until total re ection.32Ε−03 Phase reflectance π π/2 σ = 1.8: Transition from over to undercoupling. but we keep the present de nition for the sake of brevity.5) Increasing values of progressively decouples the cavity from the incoming eld.28 2π CHAPTER 1. (NB : For =1.

1. when we shall study the conversion of a phase change into an amplitude change. it becomes (2 ? S = p (1 + 4f ) ) 2 showing that the maximum of intracavity power is reached at optimal coupling ( = 1). This will be discussed later..5) is (f ) = + tan?1 1? 2f ! + tan?1(2f ) For f very small. we have the slope d = 2 (2 ? ) 2 dL 1? p . In general. the losses small and the nesse moderate). which is the current case in GW interferometers. THE FABRY-PEROT RESONANT CAVITY 29 In the coupling rate notations. the slope is simply : d = 8F dL In terms of the coupling rate and absolute displacement of the mirrors. so that there is no re ected wave. The in nite slope at optimal coupling is very appealing.e.1. in the Michelson interferometer section. If the coupling rate is small (i. where the slope is a maximum. the slope of ( ) is d = 2 (2 ? ) d 1? p FSR It could seem that the optimum detectivity is near the optimal coupling. this is at rst order : ? ? (f ) = 2(2? ) f = 2(2? ) F 1 1 FSR In terms of the coupling rate.2. detectable by a diode. at resonance we have therefore S = 2F 0 The phase re ectance (see eq.. but unfortunately corresponds to total absorption. and absolute frequency detuning.

In order to act for correction upon the frequency of the source. i. for 1p f = 2 ?1 these two extrema being ! 2 ?1 p? = tan ?1 showing that the phase re ectance becomes at as ! 2. the phase re ectance has two extrema. It is obtained by a modulation technique : the light source is phase modulated at frequency mod.2 The Pound-Drever scheme As a rst example of application of this simple model of a re ection operated cavity. an error signal is needed. the variable to be servoed. the slope would be: d = 4 n dL so that n 2F = note that this is exactly the surtension at resonance : n = S0 In the undercoupling regime (1 < < 2). we can expand at rst order the preceding expression. we consider the so-called Pound-Drever servo scheme. the amplitude entering the cavity is of the form : A(t) = A0 exp i cos(2 mod t)] exp ?2i Lt] where L is the frequency of the source. is the modulation depth. THEORY OF GW INTERFEROMETERS This allows to nd a relation with an equivalent number n of non interfering round trips in a multipass cell of same length : in such a situation. which means that after passing the modulator crystal. and if it is small. yielding A(t) = A0 e?2i Lt + i A0 e?2i ( L+ mod )t + i 2 A0 e?2i ( L? mod )t 2 . 1. in which the goal is to keep a given light source in resonance with a reference cavity.30 CHAPTER 1.2.e.

THE FABRY-PEROT RESONANT CAVITY 31 We can recognize in this sum. and R the re ectance for the two sidebands. is practically 1. the demodulation is said in phase.2. we assume the modulation frequency antiresonant. Now for the sidebands.e. and consequently R . Each of these three waves is di erently re ected by the cavity. so that the DFC becomes explicitly : h i DFC = i 2 A0A0 X ei + X e?i with X = R?R B (t)B (t) = A0A0 RR ? i 2 (RR? ? RR+ )e?2i mod t ? i 2 (RR+ ? RR?)e2i mod t . The demodulation current is : D(t) = ei e2i mod t + e?i e?2i mod t For =0.1. and considering that a low pass lter retains only the DC terms in the result. the carrier and two sidebands added by the modulator. we have : ?2i ( L + mod )t + i R e?2i ( L ? mod )t 2 R+ e 2 ? where R represents the re ectance of the cavity for the carrier. The demodulated signal is the product BB D. i. the re ectance of the cavity. This amplitude is partially directed to a photodiode delivering thus a current proportional to B (t) = A0 R e?2i Lt + i The demodulation consists in mixing the latter current with the modulation current with a variable dephasing . If we call B (t) the re ected amplitude. such that it is shifted by half a FSR from resonance. denoting by f the o set of the source frequency with respect to resonance in linewidth units : 2 R = ? 1 ? ? +if if 1 2 It is assumed that f does not exceed 1. we get for the demodulated ltered current (DFC) : h i DFC = i 2 A0A0 ei (RR+ ? RR? ) + e?i (RR? ? RR+ ) The approximate model presented above allows to compute this expression. and in quadrature for = =2. We have. At antiresonance.

1. because it is the basis of the so-called "power recycling" setup.32 1. spaced by distances l and L.3 The double Fabry-Perot cavity It is interesting to investigate what happens when we install a Fabry-Perot cavity inside a Fabry-Perot cavity.25 -0.9: Pound-Drever error signal for laser stabilization on a reference cavity or nally (2 ) X = 4i1 +?4f 2 f This shows that the demodulation must be in quadrature. The system we are considering is described on Fig.00 0.75 -1. M2.50 -0.2. We see that there exists a range of frequency on which the error signal is practically proportional to the frequency excursion. The error curve has the following appearance (see Fig. M3. We assume L l.00 -4 CHAPTER 1.9).1.1. This o ers us the . used in GW interferometers for enhancing the laser power.50 0. units -3 -2 -1 0 δν/δνFWHM 1 2 3 4 Figure 1. THEORY OF GW INTERFEROMETERS Error signal.75 0. Note that the frequency interval between the two extrema is nothing but the FWHM of the resonance.10: it consists of three mirrors. arb.00 -0. and this is the starting point of the Pound-Drever-Hall technique for servoing cavities on laser light or conversely.25 0. M1. that we shall discuss in details in a foregoing section.

and for the sake of simplicity. We can write the stored amplitude when M3 is removed: 1 B = A 1 + r tr e2ikl 1 2 as already seen. Now the amplitude tranmitted through mirror M2 in absence of mirror M3 de nes the transmittance: T = 1 +t1rt2re e2ikl 1 2 ikl For the compound cavity. and an end mirror M3: T C = 1 + R e2ikL A whereas the re ectance for a wave coming from the right is. t1. Call r1.10: Double Fabry-Perot opportunity to calculate the transmittance of a Fabry-Perot cavity having mirrors M1. THE FABRY-PEROT RESONANT CAVITY A B l C L 33 M1 M2 M3 Figure 1.2. and in the same spirit. r2.1. T . let us neglect the losses. take r3 = 1. M2. according to a preceding study: 2ikl 2 R = 1r++ rr1ee2ikl r 1 2 Now the question is: How to choose the phases 2kl and 2kL in order to maximize the intracavity power jC j2 ? It is worth to compute explicitly the result: t1 ikl C = 1 + r r e2ikl + te22e (r + r e2ikl) A ikL 1 2 2 1 . we can evaluate the amplitude C just as we would do for a cavity having a virtual mirror of parameters R. t2 the corresponding parameters.

If we assume L l. and 2kL mod 2 ]. If we therefore take a frequency excursion small compared to L.34 CHAPTER 1. "2 1. we get t2t2 1 2 2 = 2 jC j 2 (1 ? r2 )2 jAj (1 ? r1) which is clearly the maximum value. We can consequently de nitely neglect the phase change in R. THEORY OF GW INTERFEROMETERS we see that for 2kl 0 mod 2 ]. We can write the result as a global surtension: " 2# + = (1 ? r1)(1 + r2) S0 = jjC jj2 A reso (1 r1)(1 ? r2) We know that a cavity at antiresonance is far more re ective than any of its two simple mirrors: if we assume r1 = 1 ? "1 and r2 = 1 ? "2. we can consider T and R as constants. It means that the short cavity must be antiresonant. Moreover. it will be a fortiori small compared to L. we have Rantireso = R0 = 1r1 + rr2 1 ? "12"2 + r1 2 which shows that the global transmittance is second order with respect to the individual transmittances. with "1. and owing to the preceding remark. but it is multiplied by (1 ? r2). Its is easy to check that the phase of R changes by a negligible amount. and the long one resonant. not only = l is much smaller than = L. We have: 2kl = 2 so that l #) 2 r1(1 ? r2 ) = R0 1 + 2i l (r1 + r2)(1 + r1r2 ) thus. the free spectral range L of the long cavity is much shorter than that l of the short cavity. we know that both the transmittance and the re ectance of an antiresonant cavity are almost independent on the frequency over a large interval in between two successive resonances. We have 2kL +2 l + + R = 1 r2 r rr1 + 22ii rr1r = = l + 12 1 2 l r 1 + 2i r1+r l = R0 1 + 2i r r 1 2 ( " 1+r1 r2 1 2 L .

The free spectral range is thus L 50 kHz. If we put a simple input mirror with 2 re ectivity r2 = 0:882. this gives: 1 S = S0 1 + 2Fsuper L where Fsuper is the super nesse.3. some arbitrary in the choice of the coordinates allows simpli cations by partially removing this arbitrariness. of same re ectivity r1 = 0:882 and if we tune the short cavity at antiresonance. In 1.1 Retarded time in a GW . we get a re ectance of 0. and a linewidth of 31 Hz.3.998 for the short cavity. assume the length of the long cavity to be L = 3 km.Simpli ed picture . so that the linewidth of the 2 cavity is near 1 kHz.1. de ned by = S0 4R sin2 ( = L ) (1?R)2 2 Fsuper The linewidth is accordingly: L p R 1?R = F L super For instance.3 Optics in a wave Space-Time When studying gravitational waves (GW). giving a super nesse of 1595. the nesse is near 50. This is: 35 S = T2 1 + R2 ? 2R cos 2 1 L 1+ If we replace the sine by its argument. 1. OPTICS IN A WAVE SPACE-TIME and the surtension is 2 jT j2 S = jjC jj2 = A j1 ? Re2i = L j2 where R. Now if we add a second mirror. T have their antiresonant values.

y directions. Assume that the GW is propagating along the z direction. h+ being so small. z only. light follows a null geodesic. and h 1 the gravitational perturbation propagating as a wave. we can write : g dx dx = 0 or in detail. consider the round trip experiment. the only e ect is a phase change during propagation. Now. 1 dx = cdt 1 + 2 h+(t) (1.36 CHAPTER 1. then we have : 00 0 0 01 B C h = B 0 h+ ?h 0 C B0 h h 0C @ A + 0 0 0 0 We shall assume in what follows that the z direction is orthogonal to the plane of the optics laboratory (or of the antenna). which is safely veri ed for known GW sources. dy. In this case. the Space-Time metrics is of the form g = +h where = diag(1. i. in which a light ray is rstly emitted from .e.6) where the sign depends obviously on the propagation direction. 0 = c2dt2 ? dx2 ? dy2 ? dz2 + 2h dx dy + h+ dx2 ? dy2 where h+. ?1. dx. called h+ and h . if dx represents the space-time elementary vector separating two events encountered by the light ray. taking dx = (c dt. dz). THEORY OF GW INTERFEROMETERS the so-called TT-gauge. Let us see this in detail : For a path lying along the x direction we have simply : 0 = c2dt2 ? dx2 + h+ dx2 or as well. We know that in a vacuum. ?1. are functions of t. and consider the propagation of a light ray along the x. It can be shown that there is no change of direction of the light ray during its interaction with the GW as long as the GW frequency is negligible compared to the EM frequency. ?1) is the Minkowski tensor of Special Relativity. This tensor reduces to two independent components.

except that the rst order term must be changed of sign. Assume now that t2 = t is the detection time. such that h(t) = h cos( t). One way of detecting gravitational waves could be to measure the excess time delay between emission and back detection of light.6) with the + sign : 1 c Z t h (u) du L = c(t1 ? t0) + 2 + t then the light ray is re ected back and returns to the origin at time t2. using again (1. for instance by analyzing solar system radar ranging data (in the microwave domain. We have the relation : 2L + 1 Z t h(u) du tr = t ? c 2 tr This is an implicit equation in tr but very easy to solve at rst order in h : Zt L tr = t ? 2c + 1 2 t?2L=c h(u) du If we consider a monochromatic wave of frequency g = =2 . Some experiments have been proposed using this principle. but the principle is the same). we have then. we get 1 Z t h (u) du 2L = t2 ? t0 + 2 c + t In the sequel we shall omit the index + in the GW amplitude and write simply h(t) instead of h+(t). then received at abscissa L > 0 at time t1 : we have using (1.sign : 1 c Z t h (u) du ?L = c(?t2 + t1) ? 2 t + by subtracting the last equation to the preceding. and t0 = tr the unknown time at which the light ray was emitted (retarded time).6) but with the . OPTICS IN A WAVE SPACE-TIME 37 abscissa 0 at time t0. 1 0 2 1 2 0 . We have generally : L tr = t ? 2c + h L sinc( L=c) cos ( (t ? L=c)) c where = 1 along the x axis and = ?1 along the y axis. the result is L tr = t ? 2c + h L sinc( L=c) cos ( (t ? L=c)) c Clearly the result is identical for a round trip along the y axis.1.3.

going to point B (of coordinates ~B ) and returning to A. ~ . as will be done later. ~ . We shall therefore assume now a gravitational signal propagating along a direction ( . The results found are useful in order to determine signal-to-noise ratios. THEORY OF GW INTERFEROMETERS The preceding analysis assumed a normally incident gravitational wave. we get : 1. ~). ~ = B cos C ~ @ A A @ A @ 0 ? sin cos these unit vectors build an orthonormal frame.3. ~ = B cos sin C . having an optimal polarization state. de ned as ~ 0 1 0 1 0 ? sin 1 cos cos sin cos w = B sin sin C . We know that there exist a coordinate system de ned by the basis (w.2 Retarded time in a GW . A quite di erent purpose is to analyze the angular response of an antenna. ~) a b ~. the spatial part of h can be expressed ~ a b as hij = h+(aiaj ? bibj ) + h (aibj + aj bi) In terms of vectors (~. and vectors ~.38 CHAPTER 1. ~) ~ a b (we assume the basis orthonormal).General picture hij = (h+ cos 2 +h sin 2 )( i j ? i j )+(?h+ sin 2 +h cos 2 )( i j + j i) which shows that up to a rotation. ~ . We denote by r r . ). consider now a light ray starting from point A (of coordinates ~A ). ) as hij = h+ ( i j ? i j ) + h ( i j + j i) This being said. ~ ). The transverse vectors (~ . we can express the wave amplitude (with new h+. ~ ) by some rotation of angle : are related to ( ( ~ = cos ~ ? sin ~ a ~ = sin ~ + cos ~ b In terms of the basis vectors (w. in which the perturbation to the metric tensor is 00 0 0 01 B C h = B 0 h+(t) ?h (t)) 0 C B 0 h (t) h (t 0 C @ A + 0 0 0 0 we use the vector w used above.

the position of the photon is now parametrized by ~(t) = ~A ? c(t ? tm) ~ r r n and after a similar calculation. after integration : c Z tm H (1 ? w:~ )t0 ? w:~ =c + w:~ t ] dt0 (1. we get : c Z tm H (1 ? w:~ )t0 ? w:~ =c + w:~ t ] dt0 ? 2L = c(t ? tr) ? 2 ~n ~ rA ~n r tr . Along the path n of a photon from A to B.1. the position ~ of the photon r can be parametrized by ~(t) = ~A + c(t ? tr) ~ r r n so that ~ r n d = c dt 1 + 1 H t ? w: (~A + c(t ? tr ) ~ )] 2 If we denote by tm the time of arrival at B. we have thus : 0 = c2dt2 ? d 2 ? hij ninj d 2 from what we obtain 1 d = c dt 1 + 2 H (t ? w:~=c) ~r where H hij ninj .7) L = c(tm ? tr ) ? 2 ~n ~ rA ~n r tr during the return trip from B to A. we get. in the unperturbed space) distance from A to B. we get c Z t H (1 + w:~ )t0 ? w:~ =c ? w:~ t ] dt0 (1.e. The general expression of the space-time element is ds2 = c2dt2 ? d~2 ? hij dxidxj r for a trip from A to B ~ = ~A + ~ r r n where 0 L and ~ is the unit vector directed along AB. If the trip begins at time tr .8) from (1.8) ?L = ?c(t ? tm) + 2 tm ~n ~ rB ~nm By subtracting (1.7). OPTICS IN A WAVE SPACE-TIME 39 L the ordinary (i.3.

THEORY OF GW INTERFEROMETERS cZt ? 2 t H (1 + w:~ )t0 ? w:~B =c ? w:~ tm] dt0 ~n ~r ~n m At zeroth order in h.1. we have w:~ = w:~M = 0. we have i 1h H (t) = 2 H e?i t + H ei t We can write eq. we have tm = t ? L=c .9 under the form L tt = t ? 2c ? 1 H tr + H tr 4 where Z t?L=c tr = exp f?i! (1 ? w:~ )t0 ? w:~A=c + w:~ (t ? 2L=c)]g dt0 + ~n ~r ~n t?2L=c Zt + exp f?i! (1 + w:~ )t0 ? w:~B =c ? w:~ (t ? L=c)]g ~n ~r ~n after some straightforward algebra.9) Consider now a particular gravitational frequency fg = =2 .40 CHAPTER 1. we nd n ~r ~n tr = L e?i (t?L=c) ei w:~M =c ei L=2csinc (1 ? w:~ ) L=2c] + c o +e?i L=2csinc (1 + w:~ ) L=2c] ~n where ~M = (~A + ~B )=2 epresents the coordinates of the middle of the segr r r ment AB. so that ~n ~r ? = 2L e?i (t?L=c) sinc( L=c) r c t?2L=c . Note that in the case where w is orthogonal to the plane containing ~ the optical path. and assuming this plane to contain the origin of the coordinates. tr = t ? 2L=c So that the expression of the retarded time is : Z t?L=c L tr = t ? 2c ? 1 H (1 ? w:~ )t0 ? w:~A=c + w:~ (t ? 2L=c)] dt0 ? ~n ~r ~n 2 t?2L=c 1 Z t H (1 + w:~ )t0 ? w:~ =c ? w:~ (t ? L=c)] dt0 ? 2 t?L=c ~n ~ rB ~n (1.

.3. having arms directed along the x and y directions respectively.2)2 ? (~ :~ 1.2 = h+ (~:~ 1. if we assume 1 h+.2 = ~0 + ~ 2L=2 r r n r r n so that apart from a common phase factor we can drop out by changing the origin of the time. OPTICS IN A WAVE SPACE-TIME 41 exactly as in the preceding subsection. returning to eq. we have for the middles of the north r and west arms respectively : ~M. we have a unit vector ~ 1.1.2 = 2c 1 1 12 .1 = ~0 + ~ 1L=2 . e?i t + h+. ~M.2) L=2c] ~n The same way. and along the west arm (y). L=2c hei L=2csinc (1 ? w:~ ) L=2c] + ~n ~ n1. (t) = 2 h+. Along the north arm (x).2 = 2 i +e?i L=2c sinc (1 + w:~ 1.2) n n n n And the north and west excesses in round trip dephasing for an optical wave of circular frequency ! is : can be written as : !L H e?i t ei L=c + c:c: 1.9. for the sake of brevity by h ~r (~M . If we n n note ~0 the coordinates of the splitter. we have the north and west functions : 1 ei w:~ . for instance. ) = 1 ei w:~M =c ei L=2csinc (1 ? w:~ ) L=2c] + r ~n 2 i +e?i L=2csinc (1 + w:~ ) L=2c] ~n Now.1. we can write it under the compact form L tr = t ? 2c ? 1 H L (~M . ) e?i (t?L=c) 2 c r 2 c r where the function is de ned. )e?i (t?L=c) ? 1 H L (~M . . a unit vector ~ 2.2)(~ :~ 1. . Now. ei t we can write h i H = h+ (~:~ )2 ? (~ :~ )2 + 2 h (~:~ )(~ :~ ) n n n n Let us now consider a whole interferometer. we have the north and west gravitational amplitudes h i H1.2)2 + 2 h (~:~ 1.2 1.

and A(t) its value at the beginning. and consequently. we have a frequency dependent antenna pattern. and we can take simply 1 = 2 = 1. whatever the various enhancements will be.42 CHAPTER 1. we have the following pattern (see g.1. so that ( . We have B (t) = A(tr) If we note A(t) = Ae?i!t . = =100. ) ' jH1 ? H2j or. For arms as long as 3 km. we have at 1 kHz. the antenna pattern will be given by ( .11). ( . ) = j h+(1 + cos2 ) cos 2 ? 2h cos sin 2 j In the case of purely h+ sources (binaries in a plane perpendicular to the line of sight). ) = jH1 1 ? H2 2j we have explicitly H1 = h+ (cos2 cos2 ? sin2 ) ? h cos sin 2 H2 = h+(cos2 sin2 ? cos2 ) + h cos sin 2 and also ( L=2c) : 1 i sin cos nei sinc (1 ? sin cos ) ] + e?i sinc (1 + sin cos ) ]o 1 = e 2 1 i sin sin nei sinc (1 ? sin sin ) ] + e?i sinc (1 + sin sin ) ]o 2 = e 2 At high frequencies. if we are interested in the directivity pattern of a Michelson. Our light ray is in fact a monochromatic plane wave of frequency = !=2 . 1. so that the dependence of the 's in frequency can be neglected. THEORY OF GW INTERFEROMETERS The Michelson topology is essentially designed for monitoring 1 ? 2. Call B (t) the (complex) amplitude at the end of the round trip.3.3 The A133 Algebra Let us now turn to wave optics. when = L=2c is not negligible.

00 .1. OPTICS IN A WAVE SPACE-TIME 43 -1.36 1.57 0.79 Figure 1. 1.79 0.3.11: Directivity pattern for h+ sources. Angle runs from 0 to .00 0.57 -0.79 0.14 x y 2. angle from ? =2 to =2.57 3.

Let the incoming amplitude be of the form 1 A(t) = A0 + 1 h A1e?i t + 2 h A2ei t e?i!t 2 The scaling factor is h because we assume the GW to be the only cause of generation of sidebands in the whole (unknown) optical system.44 we get CHAPTER 1. After a 1st order expansion of the exponential. This is necessary because in interferometers. of frequencies g from one single frequency . we have used the abbreviations: !L=c and L=c. light undergoes several times the action of the GW in order to enhance the signal production. we get 1 1 B (t) = B0 + 2 h B1e?i t + 2 h B2ei t e?i!t with the following notation : B0 = e2i A0 B1 = e2i( + )A1 ? i sinc( )ei(2 + )A0 . we write B (t) = Ae?i!te2i!L=c + !L i 2 hA c sinc( i !L 2 hA c sinc( L=c) e2i!L=cei L=c e?i(!+ )t + L=c) e2i!L=ce?i L=c e?i(!? )t It clearly appears that the action of the GW was to create two sidebands of very low amplitude. THEORY OF GW INTERFEROMETERS B (t) = Ae?i!tr = Ae?i!(t?2L=c) exp ih !L sinc( L=c) cos ( (t ? L=c)) c Since we are always at rst order in h. We have then 1 1 B (t) = A(tr) = A0 + 2 hA1e?i te2i + 2 hA2ei te?2i e?i!te2i e?i h sinc( )cos( t? ) For shortening the formula. Now let us see what happens if the incoming optical wave is already modulated and exhibits two sidebands.

10) @ A ?i sinc( )ei(2 ? ) 0 e2i( ? ) It is easy to check that the set of all operators having the form 1 0 O00 0 0 O = B O10 O11 0 C A @ O20 0 O22 is stable for any algebraic operation. OPTICS IN A WAVE SPACE-TIME 45 B2 = e2i( ? )A2 ? i sinc( )ei(2 ? )A0 We see that if we de ne \generalized amplitudes" as rank 3 vectors having the carrier amplitude. We call it \A133" for brevity. and even may be given a structure of non-commutative algebra isomorphous to the algebra of rst order expansions. A1. A2) and B = (B0.3. B1. by setting A = (A0. The basic algebraic operations are de ned by The sum : (A + B)ij = Aij + Bij The product : (A B)ii = Aii Bii (A B)i0 = Ai0B00 + Aii Bi0 The inverse : 1 (A?1)ii = A 0 (A?1)i0 = ? A AiA 00 ii ii . B2) the amplitude after a round trip that we have precedently computed may be written in the form : B = XA where X is the linear round trip operator de ned as 0 1 e2i 0 0 X = B ?i sinc( )ei(2 + ) e2i( + ) 0 C (1.1. the upper sideband and the lower sideband respectively as coordinates.

1. well inside the tolerances of the mirror coatings. we know that the output wave is given by : " P (t) = Aout Aout = t # S ( g ) = jS10S00 + S20S00j The DC component of the output is proportional to jS00j2.46 CHAPTER 1. and in this case. The result is that. up to a normalization factor. The diagonal elements Oii represent action of that element on the carrier and the sidebands. lenses) there is no frequency dependence because the gravitational perturbation causes a negligible frequency shift. THEORY OF GW INTERFEROMETERS An A133 operator may be associated to any optical element of a complex optical system. so that our main concern.4 Signal to Noise Ratio We can start with a pure monochromatic wave Ain = (A. the SNR is proportional to : SNR( g ) / jS10 e?i' + S20 ei' j 00 00 . In fact the only non-diagonal operators are those corresponding to propagation of light in a vacuum over long distances. 0) " # h S e?i t + h S ei t e?i!t Aout = A S00 + 2 10 2 20 The corresponding detectable power is. Often (mirrors. 0 . the whole optical system has an associated A133 operator describing its behaviour. and calling Pin the incoming power : = Pin jS00j2 + h S10S00 + S20S00 e?i t + h S20S00 + S10S00 ei 2 2 The signal amplitude at frequency g is thus S being the A133 system operator. the corresponding operator is simply scalar. after some (A133) algebra.

is the A133 operator associated to a Fabry-Perot cavity.10. We take the same notations as in Fig. the spectral density hSN( g ) equivalent to the quantum noise is obtained by taking a unitary SNR : s h jS00j hSN( g ) = 2PP in jS10S00 + S20S00j We see that evaluation of the SNR of any optical GW detector eventually reduces to calculation of the Si0 of the whole system.1. The intracavity (vector) amplitude B obeys . B = t1Ain ? r1r2XB where X is the round trip operator just de ned above (Eq.12) It is possible to compute the components of F : 0 1 F 0 0C F = B G+ F+ 0 A @ G? 0 F? .5. 10 20 00 1.11) hP Inversely. before addressing more complex structures. RESONANT CAVITIES IN A GW 47 where 'ij is the argument of Sij .5 Resonant cavities in a GW The rst element we need. We have as well. We have thus B = 1 + r1r2X]?1 t1 Ain The re ected amplitude is : Aref = i r1 Ain + i t1r2 XB = i r1 + (1 ? p1)r2 X] 1 + r1r2X]?1 Ain so that the re ectance of the cavity is the operator F = r1 + (1 ? p1 )r2 X] 1 + r1r2X]?1 (1. with the correct normalisation : s SNR( g ) = 2Pin jjS10j + jS20j ei(' +' ?2' )j h( g ) (1.1.1.5.

and fg = g = FWHM the reduced gravitational frequency. For the sake of simplicity.12). F the ordinary re ectance of the FP for the upper and lower sidebands respectively. A symmetrical gure can be obtained with jG+ j.12 : 2i ? F = r1 + (1 r rp1e)2ri2e 1+ 1 2 2i( F = r1 + (1 ? rp1e)2ri(2e ) 1 + r1 2 ) (1. We have then. this can be approximated by F = ? 1 ? ? +i 2if f 1 2 2 f F = ? 1 ? ? +i( i(f f f ) g ) 1 2 g (1. . the upper sideband becoming resonant.16) (1. THEORY OF GW INTERFEROMETERS F is the ordinary re ectance of the FP for the carrier. The modulus of G? has also a resonance for f = 0 and for f = fg .17) G = i 2F L (1 ? 2i f ) 2 ? 2i( f f )] 1? g where f = = FWHM is the reduced detuning of the light source from resonance. we see that the modulus of G+ has a resonance for f = 0 (resonance of the carrier) and a second resonance when f = ?fg . When we vary the detuning.13) (1.15) (1.1. we use again the notation : = kL = L=c (recall that =2 is the GW frequency). the lower sideband becoming resonant (see Fig. after direct evaluation of F according to Eq.14) 2 )ei(2 ) G = ?i (1 + rt1rr2e2sinc( + r r e2i( )) 1 2 i ) (1 1 2 In the coupling rate ( ) formalism.1.48 CHAPTER 1.

6 Michelson Interferometer involving FP cavities We take the classical Michelson geometry.9 0.4 f_{g}= 1 fg = 1 0. The expressions of F1 and F2 for perfectly identical but orthogonal cavities lying respectively along .3 0.5 0.12: E ciency of lower sideband generation vs detuning of the source for three reduced GW frequencies. Solid line :fg = 0. MICHELSON INTERFEROMETER INVOLVING FP CAVITIES 49 1. Note that even when optically identical.1.0 -3 -2 -1 0 1 ∆f 2 3 4 5 f_{g}= 2 fg = 2 Figure 1. and consequently we must denote the corresponding operators by di erent notations (see Fig. long dashed line : fg = 2 1.2 0. but replace the end mirrors by two identical Fabry-Perot cavities.1. The transmitted amplitude is Atrans = ?rsts e2ikaF1 + e2ikbF2 Ain whereas the re ected amplitude is Aref = i t2e2ikaF1 ? rs2e2ikbF2 Ain s Note that we neglect phases of the order of 2 g a=c. short dashed line : fg = 1.8 0. small phases of order 2 g a=c.7 0.1 0.6. We neglect in this rst aapproach.13).0 0. F1 and F2.6 f_{g}g= 00 f = |G-| 0. the e ect of a GW on them will be di erent.

THEORY OF GW INTERFEROMETERS F L 2 b A in a A ref F 1 L A trans Figure 1. assuming a perfectly symmetp rical splitter (rs = ts = 1 ? ps =2). We can de ne a transmittance and a re ectance A133 operator an obvious way. by Atrans = TMic Ain Aref = i RMic Ain The elements of these operators are as follows.00 = ?(1 ? ps)eik(a+b) cos k(a ? b)] F TMic.11 = ?(1 ? ps )eik(a+b) cos k(a ? b)] F+ TMic.50 CHAPTER 1. F2 = B ?G+ F+ 0 C @ A @ A G? 0 F ? ?G? 0 F? The opposite signs of the o -diagonal elements re ect the signature of a + polarized gravitational wave having the x. for the transmittance: TMic. are : 0 1 0 1 F 0 0 F 0 0 F1 = B G+ F+ 0 C .22 = ?(1 ? ps )eik(a+b) cos k(a ? b)] F? TMic10 = ?i(1 ? ps )eik(a+b) sin k(a ? b)] G+ TMic20 = ?i(1 ? ps )eik(a+b) sin k(a ? b)] G? . y axes as polarization directions.13: Geometry of a Michelson with FP cavities the x and y directions.

18) If we assume the carrier at a dark fringe.19) g .6.10 = (1 ? ps )eik(a+b) cos k(a ? b)] G+ RMic.11 = i(1 ? ps )eik(a+b) sin k(a ? b)] F+ RMic.00 = i(1 ? ps )eik(a+b) sin k(a ? b)] F RMic. the sidebands are transmitted. neglecting ps at this level : i ?i ? ? +q e SNR(fg ) / 4F L p12+ 4 f 2 1 q e 2 1 + 4( f + fg )2 1 + 4( f ? fg )2 where ! ?1 (2( f + fg )) ? tan?1 2 f + = tan 1? ! ?1 (2( f ? fg )) ? tan?1 2 f ? = tan 1? + After some algebra. we nd the following result : SNR(f ) / 8(1 ? =2)F L 2 4 31=2 (1 ? + 4 f )2 + 4(1 ? )2fg2 5 (1 + 4 f )2 ((1 ? )2 + 4 f )2) 1 + 8( f 2 + fg2) + 16( f 2 ? fg2)2 (1. we get 0 1 0 1 0 00 iF 0 0 C TMic = (1?ps )eik(a+b) B ?iG+ 0 0 C . and conversely.1.22 = i(1 ? ps )eik(a+b) sin k(a ? b)] F? RMic. MICHELSON INTERFEROMETER INVOLVING FP CAVITIES 51 and for the re ectance : RMic. RMic = (1?ps )eik(a+b) B 0 iF+ 0 A @ A @ ?iG? 0 0 0 0 iF? This allows to study the SNR of a simple Michelson having FP cavities as arms.20 = (1 ? ps )eik(a+b) cos k(a ? b)] G? It is evident that when the interferometer is tuned at a dark fringe for the carrier. We have in the coupling rate formalism. The SNR takes the form : F F SNR( g ) / (1 ? ps ) sin k(a ? b)] G+ jF j ? G? jF j (1.

this is is 0. For . Better idea is to increase the laser power. the surtension coe cient is S = 1=p. The optimum value of F occurs theoretically for = 1. For a 10 W laser source. This corresponds to F = =p. On the other hand. and it is worthless to try higher nesses as long as a means of reducing thermal noise has'nt been found .52 CHAPTER 1. For p = 3 10?5 . for the optimal coupling of the cavities. this corresponds to a nesse of 105. we have simply s 8F L q ? =2 1 PL h(f ) SNR(fg ) = g 1 + 4fg2 2hP q we plot hereafter the spectral density of equivalent h for various values of F for a 20W light source at = 1:064 m. The sensitivity at low frequency is a function of F . (see Fig.1.14) is the maximum presently reasonable for a CW monomode. But 20W (as assumed in Fig.e. i. THEORY OF GW INTERFEROMETERS 10-20 h equivalent to shot noise Hz-1/2 10-21 10-22 F=50 F=100 F=200 10 -23 F=400 10 2 10 3 Gravitational frequency Hz 10 4 10 1 Figure 1. stabilized laser. Let us keep however in mind that the improvement due to increasing the nesse occurs only at low frequency.14: Simple Michelson with FP cavities : Spectral density of h equivalent to shot noise if the cavities are at resonance ( f = 0). and this means here a surtension of ' 3 104 .14). the limitation of the sensitivity is due to thermal noise. because the whole curve is then globally lowered.3 MW stored light power.1. But at low frequency. when = 1.

this gives ! 10Hz 2=3 Fopt( g ) = 13782 FSR g But the maximum is very at. and a good approximation of the optimal coupling rate is : ! ! 2 1=3 = p FSR 2=3 p opt = q2 2 g The optimal nesse is therefore : !1=3 !2=3 FSR Fopt( g ) = p g For instance. the length of the cavities being xed. Let us remark that for given g . as explained hereafter. and it is not necessary to require the true optimum. this corresponds to Fopt( g ) = The pseudo-optimal nesse for g = 1 kHz is for instance F = 50. we have : s 4 g PL h( ) SNR( g ) = s g 2 2hP g 1+ 2 g opt (0) (0) FSR g . In terms of this reference frequency. A value of such that q = 2 is quite su cient. MICHELSON INTERFEROMETER INVOLVING FP CAVITIES 53 gaining 1 order of magnitude. = 50 kHz.1. The parameter q is very high even for g = 10 Hz. This is quite feasible. The pseudo-optimal nesse depends of a reference frequency g(0) which is an equivalent parameter. the SNR di ering from its true optimum by only 10%. with p = 3 10?5 .6. but the result can be achieved with a much more elegant and convenient solution. we would have to lock in phase an array of 3 such lasers. the SNR is of the form s 8 L 1 p ? =2) PL h( ) (1 SNR = g p 1 + q2 2 2h P with q 2 g =p FSR and consequently is a maximum for a nite value of .

6) One important point is that.1.16 for notation). the SNR is reduced. This regime of operation. of no bene t in the simple Michelson con guration. This formula is valid except for too small values of g(0). 1.54 CHAPTER 1. working out of resonance.15: Michelson with detuned cavities (F=100) where opt is the optical frequency.(1. as can be red directy on Eq. When the two cavities have a common detuning. the re ectances of the cavities are much higher than in the tuned case.7. 1. At this frequency. and namely from the previous section that when tuned at a dark fringe.19). Drever to build a cavity with one extra mirror (the recycling mirror) and the Michelson as a second mirror (see Fig. It has been proposed a long time ago by R. the transmittance of the Michelson being a minimum. 10kHz].5 {≅σ∆}φ = 0 ∆f = 0 10-23 10 100 1000 Gravitational frequency [Hz] 10000 10-22 Figure 1. the loss due to the frequency o set of the carrier is somewhat compensated by the resonance (see Fig1. THEORY OF GW INTERFEROMETERS 10-19 h equivalent to shot noise [Hz -1/2 ] 10-20 {≅σ∆}φ = 2 ∆f = 2 10 -21 {≅σ∆}φ = 1 ∆f = 1 {≅σ∆}φ = 0. But a resonance occurs when the upper sideband created by the GW becomes resonant (for fg = f ). as will be shown later. For the interval 10 Hz.7 Recycling It is clear from conservation laws in general. it is valid.1 standard power recycling . We see the huge scale factor provided by the cavities. becomes interesting when recycling is applied. its re ectance is a maximum.5 ∆f = 0.

after some algebra. The Michelson operators for reection and transmission being respectively RMic and TMic. Using the preceding results about the Michelson operators.16: Recycled Michelson with FP cavities By controlling the resonance of this recycling cavity .20 = ?i DD TItf 00 tr (1 ? ps )eik(l+a+b) cos F = ? D . and l the length of the recycling cavity. we have for the re ectance and transmittance of the complete interferometer : h ih i RItf = rr + (1 ? pr )e2iklRMic 1 + e2iklrr RMic ?1 (1. the surtension coefcient enhances the power reaching the splitter.21) We are especially interested in the TItf 10.20) h i TItf = eikltr TMic 1 + e2iklrrRMic ?1 (1. giving the SNR. RECYCLING F 2 55 b A in l A ref Recycling mirror a F 1 A out Figure 1. and the SNR is increased. The A133 operator corresponding to this con guration is easily obtained by copying the simple Fabry-Perot operators.1.7. we obtain ( k(a ? b)) : h i tr(1 ? ps)eik(l+a+b) G sin + i rr (1 ? ps)eik(2l+a+b) F TItf 10.20 components.

0. The SNR takes on the simple form s F 4qL (2 ? ) tr(1 ? ps ) PL h( ) SNR = g 1 + 4fg2 1 ? rr (1 ? ps )j1 ? j 2hP Where we see directly how increasing the coupling factor increases the Michelson SNR. It will be even smaller here.e.e. 1) : It is always possible to tune the path di erence between the two arms at a dark fringe ( =2 mod2 ]). and we have seen that the coupling rate in a simple Michelson must be relatively small. multiplied by the surtension factor : r SNR(fg ) = SNRMic(fg ) 1 ? r (1t? p ) jF j (1. THEORY OF GW INTERFEROMETERS Da = 1 + i rr(1 ? ps )eik(2l+a+b) sin Fa with the following de nition (a = ?1. At this point. but decreases the recycling factor. and the length l of the recycling cavity in order to obtain resonance.56 CHAPTER 1. i. that maximizing the recycling surtension factor. the SNR is simply the SNR of a Michelson. we are free to choose the best recycling re ectance rr . Anyway. : D = 1 ? rr (1 ? ps ) jF j where F refers to the (assumed common) re ectance of the cavities. i.22) r s In the so called standard recycling sheme. we assume the FP cavities at resonance ( f = 0). This happens when rr opt = (1 ? pr )(1 ? ps)j1 ? j giving Sr opt s F 4qL (2 ? ) PL h( ) 1 ? pr = (1 ? ps ) 1 ? (1 ? p )(1 ? p )2(1 ? )2 g r s 1 + 4fg2 2hP s The mirror losses will be taken very small (of the order of 10 ppm). because the recycling factor would be .

i. Remark that this value is half the pseudo-optimum for the simple Michelson. and small losses). This sharp maximum makes the SNR very sensitive to the GW frequency at which the SNR is optimized.7. With physically signi cant parameters (frequencies in the detection range 10 Hz. the maximum is sharp (see Fig. RECYCLING 57 destroyed by a large cavity absorption.e. the losses external to FP's : we have 1 ? pITF = (1 ? pr )(1 ? ps )2 ) pITF ' pr + 2ps Fopt = .1. q2 is very large: Consequently.17). with q = 2 g =p FSR : 1 q2 3 + q2 2 + 3 2 ? 1 = 0 2 2 for avoiding an exact but useless and cumbersome resolution of this equation. It is therefore not unrealistic to consider that the total losses are dominated by the cavity resonant absorption.10 kHz]. must be very small. the SNR can be approximated by a simple formula. small (pr + 2ps 2 1). Call pITF the losses encountered in the recycling mirror and the splitter.1. in terms of nesse. we get the following equation. we rather solve it in q: ?3 2 q2 = 1 (1 + =2 2 =2) Now we remark that. and we can take the approximation opt = 1 q FSR (0) g or. But here. even for low GW frequencies (10 Hz). The optimal SNR is then s PL h( ) 4 L 1 r 1=2(2 ? ) SNR( g ) = p p g 2 2hP 2 1+ 2 p g FSR When searching for the optimal value of . and however. 2 (0) Where g is the GW frequency for which the SNR is optimized.

17: SNR vs for three GW frequencies.1 νg = 20 Hz νg = 50 Hz 0. The small diamonds show the approximate optima theoretically derived 10 0 F = 2500 SNR for optimal recycling 10-1 F = 500 F = 125 F = 50 10-2 10-3 10-4 10 0 10 1 10 2 Gravitational frequency [Hz] 10 3 10 4 Figure 1.10 Figure 1.58 CHAPTER 1. THEORY OF GW INTERFEROMETERS 0.00 0. .0 0. The small stars point the GW frequency at which the SNR was optimized.18: SNR vs frequency for four nesses.2 SNR for optimal recycling νg = 10 Hz 0.05 coupling factor σ 0.

then f is small. The e ciency of recycling crucially depends on the quality of the re ectivity. The surtension factor reads : 2 tr Sr = 1 + ir (1 ? p ) eik(2l+a+b) sin F r s in this expression. The amplitude in the recycling cavity has a peak at the recycling resonance. assume = =2 and =2+ k(2l + a + b) . and is of order 1 Hz. the re ectivity decreases. RECYCLING 59 2 The losses internal to FP's are still p = 1?(1?p1)r2 . we see how the re ectivity of the Fabry-Perot cavities play a central role.1.23) g 2 2hP pITF + p g g 1+ opt (0) FSR (0) g (0) the parameter p FSR=2 has the dimension of a frequency. motivated section 3. given by 2 tan?1(2 f ) where we have assumed a small . a high nesse being needed. the second is the SNR of a simple Michelson. This is the reason why at low frequency.7. This strong requirement of very re ecting cavities was the cause of a number of numerical optics studies that in turn. It is interesting to evaluate the width of the resonance line when the frequency of the source varies. and the e ect of recycling becomes negligible. The only frequency dependent quantity (in this approximation) is the phase of the re ectance. and an optimal nesse for a given GW frequency is not signi cantly better that a simple Michelson when that frequency is very low. If the frequency excursion is small compared to the cavity linewidth. The rst term represents the gain due to optimal recycling. We can conclude that a power recycled Michelson. the dominating phase is obviously given by the re ectance F . so that we can write : 2 1 (0) Sr = Sr 1 + (4F f= )2 R . having an optimal recycling rate. we can expect this slope to be reinforced by the recycling nesse. In this subsection and in the next one. Neglecting non essential small terms leads to : s 2g 1 PL h( ) s SNR( g ) = r (1. We can take for the modulus of the re ectance its value jF j = 1 ? at resonance. Since the phase re ected by cavities has already a sharp slope. the coupling rate increases.

numerically obtained.7 0.6 0. 1. THEORY OF GW INTERFEROMETERS Finesse 60 50 40 30 20 10 0 0. the linewidth is 1 kHz.0 0. Sr(0) = 50. It is also clear that . This nesse depends obviously of the tuning of the Michelson. For a 3 km long. = 6:366 10?4 .60 100 90 80 70 CHAPTER 1. Detuning reduces the re ectance of the Michelson.19.2 0. we show the exact line shape for such parameters).1.3 0.5 0. The full width at half maximum of the surtension peak can be therefore estimated by rec = 2FR FWHM (recall that FWHM i the linewidth of the cavity). (hence rr = 0:962.19: Variable nesse by detuning the dark fringe where Sr(0) is the peak height for a given detuning of the dark fringe . 50 nesse cavity.0 Dark f.9 1. of 19. f the reduced frequency excursion. (corresponding to a cavity nesse of 50 ). say ps = 2 10?5. detuning π/2 − δ [Rd] Figure 1. so that 20Hz rec very near the exact value.1 0. For standard values. we nd FR 78.8 0. as can be seen on Fig.4 0. and q r (1 ? p )(1 ? ) sin FR = 1 ? rr (1 ? ps )(1 ? ) sin r s the recycling nesse.64 Hz (on Fig.20.

03 -0.05 detuning of the laser frequency / linewidth Figure 1. We develop the symmetrical case. This helps tuning the interferometer.7.?? the full width at half maximum of the recycling width is plotted.05 -0. and the antisymmetrical detuning. the re ectivity of the cavities is enhanced. On Fig.01 0.02 -0.01 0. and the lower sideband in the second arm. but also increases the recycling linewidth. Owing . the upper sideband is resonant in the two arms. the upper sideband is resonant in the rst arm.02 0. in which one cavity is detuned by f .00 0. The basic idea is to exploit at the same time the resonance (frequency 0) of a cavity for one sideband (such that L g = 0) and the fact that the carrier being out of resonance. so that nally. and consequently the recycling e ciency also.1.7.8 10 0 -0.2 detuned power recycling We consider the case of a power recycled Michelson with detuned cavities.04 -0. in the second case.03 0.04 0. We restrict our attention to two special cases giving the same result for the SNR : The symmetrical detuning.20: Linewidth of the recycling cavity / linewidth of the long cavities. and never the lower sideband. in which the two cavities have the same detuning f . 1. the e ect is identical. RECYCLING 50 61 40 {@sd=p/2} δ=π/2 Surtension factor 30 20 {@sd=} δ=1. A detuning wrt darkfringe increases the recycling width a detuning with respect to the dark fringe ( 6= =2) not only decreases the maximum recycling gain. In the rst case. and the other one by ? f .

power recycled Michelson is : p s PL 8F L(1 ? =2) tr SNR(fg ) = 1 ? r (1 ? p ) S 2h h( g ) r s P where ( f ) is the FP's modulus re ectance.0 0.3 0.4 0.8 0. THEORY OF GW INTERFEROMETERS Recycling width [Hz] 300 200 100 0 0.6 {@sdn}_{min} = 20 Hz δνmin = 20 Hz 0.9 1.5 π/2 − δ [Rd] 0.0 Figure 1.1 0.19.62 400 CHAPTER 1.1.2 0. it simultaneously happens. and (1 ? + 4 f 2)2 + 4(1 ? )2f 2 S = (1 + 4 f 2)((1 ? )2 + 4 f 2)(1 + 8( f 2 + f 2)g+ 16( f 2 + f 2)2) g g recall that The optimal recycling is obtained when s (2 ? ( f ) = 1 ? 1 + 4 f)2 2 rr = (1 ? pr )(1 ? ps)2 The e ciency of recycling essentially depends on the re ectivity of the cavities. for fg = f that one of the sidebands is resonant.7 0.21: Linewidth of the recycling cavity vs dark fringe detuning 2 (a ? b)= to the general Eq. . and the re ectivity of the cavities. When the detuning is not zero. the SNR for a detuned.

For instance. The eigenfrequencies S . RECYCLING 10-21 63 h equivalent to shot noise [Hz -1/2 ] 10-22 10-23 {@sD}f 2 2 ∆f = = {@sD}f 1 1 ∆f = = {@sD}f 0.1. to a given TEM00 mode of frequency 0. The basic idea of synchronous recycling is to have two identical cavities. and a coupling.5 ∆f = = 0.1. This is the reason why it is possible to have a better SNR for fg in the neighborhood of f . A di er from 0 by an amount depending of the coupling. the di erence S ? A may fall in 1. When the coupling tends to zero.22). In such a system a system of supermodes exists. the frequencies S . corresponding to combinations of the individual eigenmodes of one cavity. A tend to the same limit 0. The maximum SNR is 1 SNRmax = 8F L(1 ? =2) q 1 ? (1 ? pr )(1 ? ps ) 2 v u (1 ? )2 + 4(1 ? )(3 ? ) f 2 + 16 f 4 s P u L t 2 )((1 ? )2 + 4 f 2 )(1 + 16 f 2 ) 2hP h( g ) (1 + 4 f The title of the present section could have been \how to make a narrow band optical detector by 6 orders of magnitude better than bar detectors".22: Detuned recycled Michelson (F=100) higher than when the carrier is resonant. corresponds two supermodes. If the coupling is very weak.7.3 Synchronous Recycling . and to degeneracy.5 {@sD}f 0 0 ∆f = = 10-24 10 100 1000 Gravitational frequency [Hz] 10000 Figure 1. a symmetrical (S) and an antisymmetrical (A). (see Fig.7.

and waiting the signal on the S frequency (or vice-versa). We feel that if the GW frequency is exactly this beat note. It has been rst proposed by Ph. The two cavities (of length L) are facing each other. When the central cavity is at resonance. it is possible to tune the central cavity by changing the distance l. roughly speaking.23). When the central cavity is at antiresonance. (or a short one) and we can expect the phase modulation to increase inde nitely. THEORY OF GW INTERFEROMETERS L r l r L F1 F 2 Figure 1. the light will accumulate positive phase shifts during the rst half GW period. Without changing the two FP's. If we call F the re ectances of the (identical) cavities. We assume in the following simple model no losses. so that. and a gravitational perturbation is able to pump energy from one mode in the other. Another way of understanding what happens in coupled cavities is to consider the beat note between these A and S modes. It is worth studying the e ect on a simpli ed model involving only two coupled optical cavities (see Fig. this region is itself a cavity and we call it the central cavity. then will be transferred to the second cavity at the moment when the phase becomes negative in the rst. at a frequency which is the gap S ? A (Think to coupled pendulums). its transmittance is a minimum. The principle of operation is thus to tune the coupling at a minimum. The light can be transmitted through the central region of length l. Picasso 4] to use this e ect in high Q superconducting microwave cavities. the light source on the A mode. its tranmittance is a maximum. we have for a round trip in the central cavity : (iF eikl)2 = 1 . In fact. and of r for the two inner mirrors.64 CHAPTER 1. and the coupling is weak. Bernard and E. a re ectivity of 1 for the two end mirrors. Let us consider the resonance condition for a wave to remain stored in the system. it sees always a long arm. and the coupling is strong.1.23: System of coupled cavities the audio range. and positive in the second. The result is that the stored energy is periodically exchanged between the two cavities.

The round trip phase becomes simply 2kL = + 2 f so that the re ectance reduces to the pure phase factor " # 2 f + 2 tan?1 r cos(2 f=F ) Arg(F ) = + F 1 ? r sin(2 f=F ) For the phase factor corresponding to the central cavity. RECYCLING 65 Two series of solutions can be obtained by taking iF eikl = 1 symmetrical mode iF eikl = ?1 antisymmetrical mode in case of zero losses. f = = FWHM which is simply the ratio of the detuning to the linewidth of the cavity.24) 1 ? r sin(2 f=F ) 2 4 2L f F . We have thus 4 0= mod2 ].7.1. the re ectance of one cavity is of modulus 1 : ?2ikL + 2ikL F = 1r+ ree2ikL = e2ikL 11+ rree2ikL + If we take the resonance as a reference frequency. we can write 2kL = 4 c0L + 4 L c where 0 is the resonance frequency of the (isolated) cavity. The resonance conditions become " # ?1 r cos(2 f=F ) 2 tan 1 ? r sin(2 f=F ) = (2n + 1) ? 2 ? ' ? 2 f ? F lL f F leading to the S-modes equation : " ! #?1 r cos(2 f=F ) = tan ' + + 1 + l (1. and the unknown detuning giving a resonance in the coupled system. we have kl = 2 c 0l + F lL f The constant phase ' = 2 0l=c can be considered as the tuning of the central cavity. and we can work with the reduced detuning already used above.

In order to study the minimum of coupling. The tuning has period . so that we retrieve a similar situation at ' = where the S-frequency is near the preceding A-frequency. and a weak splitting of the resonance lines. and a maximum of line splitting This maximum is half the FSR (the interval between the two white spots on the gure). we turn to our simpli ed model. THEORY OF GW INTERFEROMETERS S-mode Supermode eigenfrequency −∆/4 −∆/2 −3∆/4 −∆ A-mode 0 π/2 π Tuning of the central cavityϕ/2 3π/2 2π Figure 1. The value ' = =2 corresponds to resonance of the central cavity. The numerical solutions are plotted on Fig. and we have for the phases : + 2 tan?1(2 f ) + 2c l ( 0 + ) 0 (S ? modes) 2 . which will be of some use anyway in the sequel. The value ' = 0 corresponds thus to antiresonance. and the frequency gap between the A and S modes at this tuning.1. The round trip phase in the central cavity is equal to 2'.24. For zero losses. then to a minimum of coupling.24: Relative detuning of the A and S supermodes vs tuning of the central cavity The A-modes equation can be obtained a similar way : " ! # r cos(2 f=F ) = ? tan ' + + 1 + l 1 ? r sin(2 f=F ) 2 4 2L f (1.25) These are implicit equations in the unknown detuning f .66 0 CHAPTER 1. the parameter is zero. thus to a maximum of coupling.

For ' = =2. It is in particular easy to compute the minimum line splitting : " # 1 1 + tan( =4) = 1 fS ? fA]min = 2 tan( =4) corresponding. For the current situation (L=3 km and F =100). the validity range of the present model.1. Higher values can be obtained by a di erent tuning of the central cavity : The general result is 1 c g = S? A = cos ' 2F L .7. the frequency gap is g =500 Hz. If we intend to use this device to detect GW by coupling the A and S modes with the gravitational perturbation.1. i. the model consequently fails. The same reason causes the divergence of the S-mode at ? =2. The approximation used is valid only for detunings much smaller than the FSR. If we neglect it. this is the reason of the divergence of the Amode at this point.25) is to be compared with the preceding. we have seen that the detuning of the A-mode is half the FSR. RECYCLING 67 + 2 tan?1(2 f ) + 2c l ( 0 + ) (A ? modes) 2 We can write as well " # 1 tan ' + + l f ?1 fS = 2 2 4 2F L S The term l=2F L is very small for kilometric cavities of nesse ' 100 and a metric central cavity. to c S ? A ]min = FWHM = 2F L In other words. we see that we have to use high nesse and long cavities. we have the very simple results : ?1 fS = 1 tan ' + 4 2 2 fA = ? 1 tan ' + 4 2 2 The following plot (Fig.e. we can see a good agreement with the exact calculation. in terms of frequency. If we restrict our attention to the neighbourhood of ' = 0. the minimum splitting is nothing but the linewidth of the cavity.

The optical path has been split for clarity.25: Approximate model of degeneracy removing by coupling We have now to study the response of a real system involving a light source and a detector. If (as likely) these switching elements induce losses.ϕ π/4 π/2 Figure 1. Drever 3] after a very di erent approach than Ph.1. and it could seem strange to separate between the incident and the re ected wave o a cavity. We rst consider the counterclockwise wave (see Fig.27) and evaluate the A133 re ection operator. The scheme of Fig. THEORY OF GW INTERFEROMETERS Normalized detuning of the supermodes 5 S-mode 0 A-mode -5 -10 −π/2 −π/4 0 Tuning of the central cavity. these waves are recombined on the splitter. The splitter and the square path allow to launch two rotating waves in the ring cavity. We have rstly for the intracavity wave : B = tr Ain + rr rte2iklF2 F1 B or : h i?1 B = tr 1 ? rr rte2iklF2 F1 then Aout = i rr Ain ? i tr rte2iklF2 F1 B . so that the situation is almost that of the gure.68 10 CHAPTER 1.26 was suggested years ago by R. these losses can be localized in the mirror rt. Picasso.1. The coupled cavities are in what we call ring cavity on the gure. Bernard & E. one clockwise and one counterclockwise. It is however possible by using polarization rotators and polarization sensitive re ectors.

RECYCLING 69 F2 ring cavity b r f transfer mirror rt a F1 rr Recycling mirror light source splitter rs r f Figure 1.1.27: Ring cavity and counterclockwise optical path .26: Sketch of the synchronous recycling con guration F2 b A in A out Ring cavity a B F1 Recycling mirror Figure 1.7.

THEORY OF GW INTERFEROMETERS so that the re ectance is : h ih i?1 = rr ? (1 ? pr )rte2iklF2 F1 1 ? rr rte2iklF2 F1 Obviously. F are the same as in section 7 . 0. is the optical path inside the square cavity. we can compute the transmittance of the whole system : 2 2 T = ?i t2rf ei R? + irs2rf ei R+ s Where the two transfer mirrors of the square cavity have been assumed identical. The di erence of with respect to the preceding subsection (two isolated cavities) is due to the fact that we have now two extra mirrors.70 CHAPTER 1. if we return to the splitter. for . and R+ = rr ? t2rte2iklF2 F1 1 ? rr rte2iklF2 F1 ?1 r h i Da = 1 ? rr rt e2iklFa2 (a = ?1. we have the re ectance : h ih i R? = rr ? (1 ? pr )rte2iklF1 F2 1 ? rr rte2iklF1 F2 ?1 Now. for the clockwise optical path. we can write 2 T = ?i 1 (1 ? ps )rf ei (R? ? R+) 2 A direct calculation gives R? ? R+]10 = ?t2rt e2ikl 2G+ (F?F+) r DD+ Where the de nitions of G . 1) The 10] component of T is thus : 4i(2 ? )2(1 ? ps )r2 r ei t2f T10 = F L (1 ? 2i f )2 1 ? 2i( f + f )]2 1 ? rf rt e2iklrFg2] 1 ? r r e2iklF 2 ] g r t r t + One would obtain a similar expression for T20 by changing the sign of fg . The preceding expression exhibits a sharp resonance peak when the resonance condition 2kl + 2Arg(F ) 0 mod2 ] is met. If further we assume a perfectly symmetrical splitter.

This should be kept in mind in any comparison. The long cavities are detuned by a microscopic change in length making their new resonance shifted by an amount f . representing the tuning of the central cavity. In particular. and the antiresonance is 2kl . the resonance condition for the central cavity is now 2kl 0. is p fS ? fA]min = 1 ? p For having a large SNR at the normalized GW frequency fg(0) > 1 ? . RECYCLING 71 recycling (Mr ) and transfer (Mt) each adding a phase of =2.26) 4 and v u u 1 ? =2 !2 (1. we have thus to solve ! ?1 2 f + tan?1 (2 f ) = ?kl tan 1 ? or. For f corresponding to a resonance of the ring cavity. so that kl is a constant. 2(2 ? ) f = ? tan(kl) 1 ? ? 4 f2 this gives two solutions : 2 3 v ! u 1 6 1 ? =2 + u 1 ? =2 2 + 1 ? 7 fS = 2 4 tan(kl) t tan(kl) 5 and Note that 2 3 v !2 u u = fA = 1 6 1 ? kl) ? t 1 ? kl2 + 1 ? 7 4 tan( =2 5 2 tan( ) fA fS = ? 1 ? (1.27) fS ? fA = t tan(kl) + 1 ? We remark that the minimum frequency gap.7. We assume the laser frequency given. which happens for 3 2 1 ? =2 5 kl = tan?1 4 q (0)2 fg ? (1 ? ) .1. we follow the following scheme : Tune the central cavity in such a way that fS ? fA = fg(0).

only one of the two components T10 and T20 can be made resonant. we can therefore neglect T20. Obviously.S 2 1 + 4 fA. If we work on T10. at resonance.28: Gravito-optical pumping fA and fS are now determined.72 CHAPTER 1. a similar process can be carried out starting from the upper frequency fS and going to the lower level fA : this is described by the T20 component of the A133 transmittance (see Fig. At the end of this process. Unfortunately.1.28). We have : v ! u 1 + (1 ? pr )rt2 2 2 ? u 1 + (1 ? pr )rt2 2 2 2 ? (1 ? p ) A S A S t rr ]opt = r ( 2 + 2 ) r r ( 2 + 2) t A S t A S . and vice-versa. A and S are the modulus re ectances : 2 = 1 ? (2 ? ) A. Tune the cavities in such a way that f = fA. and for 2 1 ? rr rte2iklF+ = 1 ? rr rt 2 S because the tuning satis es the modes S equation when fg = fg(0). THEORY OF GW INTERFEROMETERS νL ν A νL νS ν S ν A νg νg Figure 1. when the gravitational frequency is larger than the minimum gap. when fg = fg(0). we see that the resonance condition is met simultaneously. it is cumbersome to derive the optimal rr .S We have then. In the general case. for 1 ? rr rte2iklF 2 = 1 ? rr rt 2 A because the tuning satis es the modes A equation. 16(1 = 2 2 (0) jT10j = F L (1 + 4 f 2 )(1 + 4 ?2)(12) trr frg 2 )(1 ? r r 2 ) fS ? r t A r t S A We can still x the recycling rate by opimizing T10 with respect to rr .

The optimum value of the recycling mirror re ection coe cient is simply rr]opt = (1 ? pr )rt(1 ? ) And the optimal peak value of jT10j is jT10jpeak = 2 pL K ( ) with the form factor p 2 1? K ( ) = 1 ? (1 ? p )(1 ? )2 RC and 1 ? pRC = (1 ? pr )rt2. it is possible to show that A S =1? which is particularly remarkable. a value of K very close to 1 is obtained already for the pseudo. But the shape of the curve is so at. The spectral density of h equivalent to shot noise is : s h(f0) = 2hP SNR 1 P L peak. The peak SNR at resonance is therefore s 2 L PL h( ) SNRpeak.optinum ' 20psr (see g.Max = p 2h g P for a wide range of reference GW frequencies.1. The form factor K has the maximum value 1.29).Max .e when fg(0) = fm = 1 ? .1. however. situation in which we have A = p S = 1 ? . that this value is misleading. i. being independent on the tuning of the cental cavity. After that. For f (0) too small. RECYCLING If we use the de nitions of A 73 and S .7. the sum 2 + 2 will obviously depend on the A S tuning : 2 2 2 + 2 = 1 + (1 ? )2 ? (2 ? ) A S (0)2 2 + 4fg It is however reasonable to optimize the SNR for the lowest possible GW p frequency. whereas the zero frequency limit of the (wideband) power recycling scales as 1=pp. Remark that this peak value scales as 1=p. the SNR falls to zero. obtained for the approximate value = (2pRC )1=3.

fg = fg(0) + f .2 0.6 0. so that fA + fg = fS + f .8 psr=10-5 -5 psr=10 psr=10-3 p_{sr}=10^{-4} p_{sr}=10^{-3} psr=10-4 SNR form factor 0. = 1:064 10?6 m.74 1. l=3 km.e. we assume the laser being locked on the antisymmetric resonance. and the gravitational frequency in the neighbourhood of the gap fg(0).0 10-5 10-4 10-3 Finesse σ 10-2 10-1 10 0 Figure 1. For this purpose. The white dots are the strict optima. the varying terms are : The fast varying phase " # " # ?1 2( fS + f ) +tan?1 2( f + f )]?tan?1 2 fS ?tan?1 2 f ] = tan S S 1? 1? expanded at rst order in f this gives )(1 f2 = A f with A = (14(2 ?2 + 4 ?f 2]+ 4 4 S )f 2] ? ) S 1+ S . the black dots mark the pseudo-optima For cavity losses p = 3 10?5 .29: Form factor K for the SNR vs nesse = pF = .4 0.0 CHAPTER 1. i. we have h(f0) ' 2:3 10?25 Hz?1=2 It is now necessary to study the width of the resonance. Consider the SNR : 2)2 2 g r SNR( f ) / 16F L (1 + 4 f 2 )(1 + 4( f (1 ? )=)(1 t?fr r 2 ) (1 ? r r 2 e2i ) A + fg 2 r t A r t + A In this expression. THEORY OF GW INTERFEROMETERS 0. PL=20W.

7. in order to have a good re ectivity of the cavities. which varies very little. whose variation is like second order.1. 2 + = 2 S The term. It can be thus understood. The study of the shape of the resonance line can thus be carried out on the only term : h i j1 ? rr rt 2 e2i j ' (1 ? rr rt 2 )2 + 4rr rt sin2(A f ) 1=2 + S 2 0 q 1 31=2 2 sin(A f ) 2 2 r rt = (1 ? rr rt 2 ) 61 + @ r ? S r 2 A 7 5 S 4 1 r r t S fg(0) + f 1 + 4( fS + f )2 expression very similar to a cavity resonance. and numerically checked that the variations of 2 around the S resonance + can be neglected. RECYCLING 75 The re ectivity of the cavities for the upper sideband : (2 ? ) 2 = 1? + 1 + 4( fS + f )2 This di ers from unity by a small amount. the second order expansion gives 2 + 8 (2 ? ) 2 fS f + 4 (2 ? )(1 ? 212 fS ) (1 + 4 fS ) (1 + 4 fS )3 We have already seen that the best value of is very small. with the super nesse q 2 SF = 1 ? rrr rrt S2 r t S The linewidth (FWHM) od the SNR is thus : 2 q fg = 3(1 ? rr rt2 s ) A rr rt S p . More speci cally.

We have seen that small values of are pseudooptimal. The SNR (with optimized recycling rate) becomes : SNR(fg ) = 16F L (1 + 4 2 fA)(1 + 4 (1 ? )fg(pRC 2 )(1 ? (1 ? pRC )(1 ? fS PL h( ) g 2hP It is easy to see that the 1st order expressions for the S and A detunings are respectively : = fS = 21 ? z=2 tan( 4) fA = ? (1 ? =2) tan(z=4) 2 s +2 ) ) 2 )j1 ? (1 ? pRC )(1 ? ) 2 e2i j A S .30: E ect of the detuning of the central cavity on the response of synchronous recycling interferometer A simpli ed model will help to get simple estimates of the peak value and the linewidth of the SNR. We can then try a rst order approximation in and a fortiori in the various losses.76 10-22 CHAPTER 1. THEORY OF GW INTERFEROMETERS h equivalent to Shot Noise [Hz-1/2] 10-23 10-24 z={@sp}/2 z=π/2 z=3{@sp}/8 z=3π/8 z={@sp}/4 z=π/4 10-25 40 50 60 Gravitational frequency [Hz] 70 80 Figure 1.

1. we have simply 2 S 2 A 77 and also 1 ? (1 ? pRC )(1 ? ) 1 ? (1 ? pRC )(1 ? ) For the varying phase factor. RECYCLING where z = 2kl is the tuning of the central cavity.7. we have The SNR is : 2 S 2 A ' 1 + 2 sin2(z=4)] ' 1 + 2 cos2(z=4)] = 4(1 ? =2) sin2(z=4) f (1 ? =2)fg(0) sin(z=2) L SNR = 8 p (1 ? ) 1 + 2 cos2(z=4)] 1 + 2 sin2(z=4)] 2 !23?1=2 s PL h( ) 8 sin2(z=4) 41 + 5 g 2 (z=4)] f 2h 1 + 2 sin P and nally : 4 z=2) SNR = 2 L 1 (1 ? =2) 3 +sin(2(z=2) p sin . we nd then 2 2 1 + 4 fS = 1 ? cos (z=4) sin2(z=4) 2 2 = 1 ? sin (z=4) 1 + 4 fA cos2(z=4) whence 4(1 ) 2 2 (1 + 4 fS )(1 + 4 fA ) = sin2(? 2) z= = 1 ? 2 sin2(z=4) = 1 ? 2 cos2(z=4) The gravitational resonance frequency is 1? fg(0) = fS ? fA = sin(z==2 2) If we can neglect the ring cavity losses pRC (a few 10?5 ) with respect to (up to 1%).

max P (z) (1. 1. The general formula is : 1 ? =2 (0) (1.28) SNRpeak. it is possible to adjust the resonance for a GW frequency equal to or larger than the linewidth of the cavities.28). and P (z) is a form factor.FWHM. THEORY OF GW INTERFEROMETERS 2 !23?1=2 s 8 sin2(z=4) PL h( ) 41 + 5 g 2 (z=4)] f 2hP 1 + 2 sin from where we conclude that the overall peak value.29) FWHM g = sin(z=2) The best response of the interferometer is obtained for the lowest GW frequency. the minimum resonance frequency is given by the linewidth of the cavity. when the central cavity is exactly antiresonant (z = =2). and the minimum gravitational linewidth (FWHM) : fFWHM.min = p 3 In terms of gravitational frequencies. taking the value 1 for z = =2 : 4 z=2) P (z) = 3 +sin(2(z=2) (1. the value of the SNR resonance peak is: SNRpeak = SNRpeak. we nd the relation with the cavity linewidth : p g. corresponding to z = =2 is 1 (1.CHAPTER 1.31) sin .max = 2 L p 78 This peak corresponds to the resonance frequency fg(0) = 1 ? =2 in other words.30) where the maximum peak value has been expressed above (Eq.min = 3 FWHM Let us summarize the results for small and z not far =2 : By varying the tuning of the central cavity.

We have seen other methods giving a comparable result.32). Here. the linewidth has to be very thin.d. The parameters of the mirrors are labeled by r.32) g. For the optimal operation point (z = =2).7. and the nesse very high. The gravitational frequencies creating a sideband for which the signal cavity is antiresonant are enhanced.1. The A133 operator corresponding to the whole setup may be constructed by successive shells. We rst consider the Michelson (mic) as a black box having two inputs. The sketch of the setup and the notation are shown on Fig.4 Signal recycling . one more bene t is to enhance the constrast of the interferometer by the spatial ltering e ect of the extra Fabry-Perot installed at the output (But this is out of the scope of the present chapter).s. and the GW linewidth increases. and the length of the dual recycling cavity is z.FWHM FWHM 4 4) These approximations remain true as long as does'nt exceed a few %. This allows to modify the sensitivity curve and have a gain factor at a given frequency range of special interest. The ratio g = g. Meers 2].1. we have 1 Q ' p = p 3pF 3 Signal recycling was proposed some years ago by B. The general formula for the GW linewidth is : p 2 sin2( = 3 1 +sin2(z=z=4) (1.7. It has also a re ectance RS and a mic 1. The dark fringe port plus the signal recycling mirror form a resonant cavity whose re ectivity can be tuned. It has therefore an A133 re ectance RW and mic W a transmittance Tm ic (see Fig. the sensitivity decreases. In fact we already know from the preceding study that the SNR tends to zero when the resonance peak tends to zero. cannot more be kept small and the approximation fails.1. For very low gravitational frequencies. The idea is to add one more mirror after the output port of the interferometer in order to store the sidebands generated by the GW. RECYCLING 79 When the central cavity is progressively detuned from antiresonance. the GW resonance frequency increases.FWHM gives an idea of the equivalent Q of the resonator. the length of the power recycling cavity is l. West (as in the preceding sections) and South (because the dual recycling reinjects from the South).31. The lengths of the short arms are a and b.

THEORY OF GW INTERFEROMETERS Michelson F2 Power recycling Mirror b Ms l Splitter z a F 1 West Mr Dual recycling mirror South Md Figure 1.32: West input .80 CHAPTER 1.31: sketch of the Dual recycling con guration 1 Mic i RW mic Tmic W Figure 1.

The corresponding operators are : h i?1 Tpritf = treiklTmic 1 + rr e2iklRmic . F and G having the de nitions set in section 8.7. and a re ectance Rpritf for a south input.34 and Fig. we have simply.1.35) has a transmittance Tpritf for a west input. RECYCLING S Tmic 81 Mic i RS mic 1 Figure 1. F2 = B ?G+ F+ 0 C @ A @ A G? 0 F? ?G? 0 F? F. and the Michelson tuned at a black fringe.1.33: South input S transmittance Tmic for the South input port : The corresponding operators are easy to compute : 2 RW = t2e2ikaF1 ? rs e2ikbF2 mic s W Tmic = ?rsts e2ikaF1 + e2ikbF2 2 RS = ?rs e2ikaF1 + t2e2ikbF2 mic s S Tmic = ?rsts e2ikaF1 + e2ikbF2 If we assume the splitter to be strictly symmetrical.1. after setting m = (a + b)=2. Now the power recycled interferometer (pritf) (see Fig. i RW = ?RS = 2 (1 ? ps)e2ikm(F1 + F2) = Rmic mic mic i W S Tmic = Tmic = ? 2 (1 ? ps )e2ikm(F1 ? F2) = Tmic with 0 1 0 1 F 0 0 F 0 0 F1 = B G+ F+ 0 C .

34: West input on a power recycled Michelson l Michelson rr 1 i Rpritf Figure 1. THEORY OF GW INTERFEROMETERS 1 l Michelson rr Tpritf Figure 1.35: South input on a power recycled Michelson .82 CHAPTER 1.

multiplied by an extra surtension factor : d SD = 1 ? i(1 ? p )tr e2ik(m+z) F s d + Remark the opposite signs in the two factors of the denominators : The best e ciency is obtained for resonance in the recycling cavity.7.10 = ?i 1 ? i(1 ? p )rtret2dik(m?zpF)e] 1 + i(1 ? + )r e2ik(l+m)F ] (1. the dual recycling setup (see Fig. so that Arg(F ) = . the two sidebands cannot be both antiresonant (except at zero gravitational frequency). It is possible to tune the long Fabry-Perot's at resonance.33) +) + ps r s d (1 s ik(l+2m+z) G TD. if we choose for instance. as already derived in a previous section. and antiresonance in the signal cavity.1. we nd : (1 s ik(l+2m+z) G TD.34) +) ? ps r s d It is easy ro recognize in these formulas the SNR for a power recycled Michelson.36: West input on a power and dual recycled Michelson h i?1 Rpritf = ?Rmic + rr e2iklTmic 1 + rr e2iklRmic Tmic Finally.1. RECYCLING 1 pritf 83 z rd TD Figure 1. Obviously.20 = ?i 1 ? i(1 ? p )rtret2dik(m?zpF)e] 1 + i(1 ? ? )r e2ik(l+m)F ] (1.36) has a transmittance for a west input TD. . to make the (10)-component resonant. h i?1 TD = tdeikz 1 + rde2ikz Rpritf Tpritf After some elementary algebra.

it is always possible to tune the dual recycling cavity to meet resonance. Mizuno 5].1. f being the gravitational reduced frequency.84 CHAPTER 1. which corresponds to f0 = 1. This regime. for the same nesse) was called 'Signal extraction' by J. 6]. increases sharply to a high and the signal surtension factor may be written as SD = 1 ? r (1 ? p ) etd =2+2k(m+z)+Arg(F )] d s + i where. and thus losing in maximum sensitivity. THEORY OF GW INTERFEROMETERS to tune the power recycling cavity so as to obtain resonance.7. the sensitivity is almost at (there is a knee at a higher frequency). we conclude that the detuning giving the sensitivity peak at given f0 is ?1 2 ? 2tan (2f0) where we have set 4 (m + z)= .1. exhibiting a broadband response (broader than the standard recycling. by taking 2k(l + m) + 2 clearly.37) If is small.5 The signal extraction regime .38) We remark that for = ? =2. (see Fig. The explanation is that the at curve is the result of a con ict between the low-pass response of the Michelson (1=(1 + 4f 2 ) and the signalrecycling gain factor which starting from a low value at f = 0 (the SNR is out of resonance though the FP's are resonant). (2 ? ) + = 1? 1 + 4f 2 " # ?1 2f + tan?1 2f ] Arg(F+) = + tan 1 ? + 1. with the condition : # " ?1 2f + tan?1 2f ] 2k(m + z) + tan 1 ? 2 The sharpness of the dual resonance is a function of rd (see Fig.

95 d R 0.99 R d = 0.80 R d = 0. RECYCLING 85 10 1 SNR (arb. units) Rd = 0.7.80 10 0 R 0.95 R R = 0.1.90 0.37: SNR of dual recycling con guration for various recycling rates (values of rd ) .99 10-1 0 100 200 frequency [Hz] 300 400 Figure 1.00 R R 0.90 R d = 0.

57 10 -1 da tan rd pow ec er r ing ycl 10 1 10 2 10 3 Gravitational frequency [Hz] 10 4 Figure 1.81 10 0 xxxxxxxx δ = 1.57 S xxxxxxxx δ = 1.18 xxxxxxxx δ = -1.86 CHAPTER 1. THEORY OF GW INTERFEROMETERS 10 2 Spectral sensitivity (arb.38: Spectral sensitivity of dual recycling con guration for various detunings of the signal recycling cavity 4 (m + z)= mod 2 ] . units) 10 1 xxxxxxxx δ = -1.37 xxxxxxxx δ = -0.00 xxxxxxxx δ = 0.

the ratio of the gravitational frequency to the linewidth of the cavity (fg = g = ). and that the re ectance modulus is much higher at antiresonance than at resonance ). This result can be understood by looking at the expression of the SNR (1. RECYCLING 87 constant value when the FP's arrive to anti-resonance.e. i. especially at high nesses.10j = 2F L(2 ? ) Gr 1 ? 2if + (1 ? ptd)r 1 ? + 2if ] = g s d g 2F L(2 ? ) G td r 1 + (1 ? ps )(1 ? )rd ? 2ifg 1 ? (1 ? ps )rd] Which makes clear that the bandwidth is now g = 1 ? (1 ? p )r s d . Recall that.7.1. when the cavities are at resonance.10 = 2F L(2 ? ) 1 ?12if Gr g 1 + (1 ? ps)rd 1??+2ifg 1 2ifg where Gr is the resonant power recycling gain (unsensitive to GW frequency) : Gr = 1 ? (1 ? ptr)r (1 ? ) s r This yields jTD.33). (recall that there is a phase ip when a FP transits from resonance to anti. when power recycling is resonant and signal recycling antiresonant : td TD. the upper sideband generated by the GW in one cavity is G+ = i 2F L(2 ? ) 1 ?12if g and the re ectance of the cavity for that upper sideband is 1 ? + 2ifg + = ? 1 ? 2ifg where fg is the normalized gravitational frequency. that the antiresonance frequency range is much larger than the resonance. The SNR takes thus the form (up to a phase factor and neglecting the length of the SR cavity). and the coupling coe cient.

5 s/L = 0. Assume a power recycling interferometer having nesse 100 long FP cavities.1 SNR (arb. units) 10 0 Signal rec. due to the e ect mentionned above. by increasing the recycling rate rd. It is even possible to play with the length of the signal cavity. :δ = − π/2 10-1 s/L = 4. and nally about 500 kW in the FP cavities. The optimum power recycling rate corresponds to a surtension 800.e-03 10-2 10 1 10 2 10 3 Gravitational frequency [Hz] 10 4 Figure 1. It is interesting to note that it is possible. THEORY OF GW INTERFEROMETERS Standard rec. this gives 18 kW on the splitter. This creates local resonance e ects. i. Starting from a 20 W laser.e. assuming lengths much longer than the recycling cavity (see Fig.39: SNR in the regime of "signal extraction". L is the length of the arms so that. for several ratios s=L of the signal cavity.88 10 1 CHAPTER 1. it is possible to keep constant the product F 1 ? (1 ? ps )rd] and thus the bandwidth of the detector. to have almost exactly the same SNR spectral pro le with standard power recycling.1. and with power recycling + signal extraction. The following extreme example will help to understand it.39). ? =2 mod 2 ]. . s/L = 0. even if the nesse is very high.

which is 1. radiation pressure.9. The nesses of the long FP cavities are 1000.. RECYCLING 10-22 89 Shot noise limited spectral sensitivity [Hz-1/2 10-23 10-24 10 1 10 2 GW frequency [Hz] 10 3 10 4 Figure 1.4. the power surtension is only 80. F = 1000.69. a higher value would give a at response but with a loss of sensitivity.6 kW on the splitter. . Dotted line : rd = 0.69.40: Solid curve : Standard power recyling. we can compare the SNR in the two situations (see Fig.1.1.7. The power stored in the FP's is still about 500 kW. The re ection coe cient of the signal recycling mirror 2 is rd = 0. A smaller value would give a standard power recyling type response peaked at f = 0.. the coincidence is caused by the particular choice of rd. Assume now a dual recycling interferometer in the signal extraction regime. Short dashed curve : Dual recycling-Signal extraction. and under optimal power recycling. 2 2 2 optimal recycling.40). with F = 100 and optimal recycling.) are identical in the FP cavities in both . Long dashed curve : rd = 0. rd = 0. thermal distortions. We see that the drawbacks caused by high powers (thermal lensing.

The ultimate logics of the signal extraction regime is reached when the cavities are optimally coupled (all the light power is absorbed in the FP's). THEORY OF GW INTERFEROMETERS cases. but very di erent in the power recycling cavity. the power recycling rate being zero.90 CHAPTER 1. and nevertheless. . the bandwidth large. This is of some importance when power dissipation is taken into account (see further chapters).

1 The Helmholtz equation . The basis of the SDT is the Kirchho equation. combined with the ability to achieve a dark fringe have been an actual worry at the beginning of interferometer projects. and have triggered a lot of optical simulations of FP cavities and interferometers. It was essential to have theoretical models for light propagation. We have seen that a good re ectivity of these cavities is a key condition for e ciency in recycling. in order to see clearly what means the paraxial approximation which is in fact more widely used. z) in a homogeneous dielectric medium of refractive index n.1 introduction In interferometers. y.2.2 A short theory of di raction A component of the the real Electric eld. The theory used up to now for this purpose is the Scalar Di raction Theory (SDT) (this seemed su cient. obeys the wave equation (c = 2:997925 91 2. owing to the very weak departure of the optical elements from an ideal shape).Chapter 2 Beam optics and Interferometers 2. 2. This re ectivity. say E (x. it seemed therefore useful to recall it and its derivation. we need to store light in long cavities in which light propagates back and forth.

4) V S 2. real lasers have a nite linewidth. y. Recall that for two arbitrary elds A and B . In fact.2 The Kirchho integral ~ represents the inward normal to the surface S surrounding the volume n V (see Fig. The following notation has been used: @ n ~ @n ~ r . So far as the dimensions of an optical system are small compared to the coherence range. For a monochromatic wave of frequency = !=2 . z) = 0 (2.1) 108 km:s?1 being the speed of light in a vacuum) " # n2 @ 2 E (t. we can set E (x.2. y. z) ei!t (2. The role and the status of the Kirchho theory is therefore seldom known: Is it \exact" or \approximate". and a nite coherence range. z) = 1 E (x. BEAM OPTICS AND INTERFEROMETERS (2. in a naive representation as a pure monochromatic wave. This is usually the most delicate part in optics books. Even kilometric optical systems may be treated assuming purely monochromatic waves. The lasers used in gravitational wave interferometers are highly stabilised in frequency and have huge coherence ranges. we obtain the Helmholtz equation h i + k2 E (x. x. z) e?i!t + E (x.1). and often skipped by stressed readers.2. y.3) where k n!=c. then with respect to what ? We try to address these issues and give answers at the end of the present section. z) = 0 ? c2 @t2 The light coming from a laser can be viewed. known as Green's theorem: # Z I " @B @A ds (A B ? B A) d~ = ? r A @n ? B @n (2.92 CHAPTER 2. the monochromatic approximation remains valid. we have a relation between a volume integral and an integral on the surface bounding the same volume.2) 2 and for the amplitude E of the electric eld. y. y.

we get: ~ ~ E (r0) 0 G(~ ? r0) ? G(~ ? r0) 0 E (r0) = ? (~ ? r0) E (r0) r ~ r ~ r ~ ~ . with (2. z) = E (~) of the Helmholtz equation (2. satisfying r h i + k2 G(~) = ? (~) r r (2. y. and G(~) a Green function.5) We have as well.6) ~ by multiplying both sides by E (r0) . A SHORT THEORY OF DIFFRACTION S n 93 n V n Figure 2.e. h 0 involving 2d order derivatives. i.2.2.3) r . we can write h i ~ E (r0) 0 + k2 G(~ ? r0) = ? (~ ? r0) E (r0) r ~ r ~ ~ and obviously.3): G(~ ? r0) r ~ h 0 i ~ + k 2 E (r 0 ) = 0 by subtracting these two equations. 0 + k 2 i G(~ ? r0 ) = ? (~ ? r0 ) r ~ r ~ (2.1: Consider now a solution E (x.

On the surface at in nity.7) r ~ ~ ~ ~ whereas for ~ in the left half space. The result is that at any point ~ in the right half space we have: r E (~) = r I h S i ~ ~ ~ r ~ E (r0) n0 r0G(~ ? r0) ? G(~ ? r0)n0 r0E (r0) ds0 (2. we have: r I h i ~ ~ ~ r ~ E (r0) n0 r0G(~ ? r0) ? G(~ ? r0)n0 r0E (r0) ds0 = 0 r ~ ~ ~ ~ S (2.(2. we get: E (~) = ? r Z h V ~ E (r 0 ) 0 G(~ ? r0) ? G(~ ? r0) r ~ r ~ 0 i ~ ~ E (r0) dr0 by using Green's theorem. it can be shown that the surface integral vanishes. If the point is outside. we have a Green function. no source at in nity). and this property will be exploited below. Let us de ne r0 = r 0.94 CHAPTER 2. z 0 ]. r the integral vanishes. due to the radiation condition on the eld (outgoing waves.5) is the simple spherical wave: so that by taking eikr G(~) = 4 r r (2. There are thus two half spaces that we refer to as the left half space. Now.9) ik 0 ~0) = e 0 G1(~ ? r r 4 ~ ~ with 0 = j~ ? r0j.3) holds. bounded by a closed surface S . By integrating the preceding equation over the volume. Assume now that the surface S extends to in nity. y 0. if we consider x . and the right half space respectively.(2.8) Now it is well known that a solution of Eq. this becomes: E (~) = r I h S i ~ ~ ~ r ~ E (r0) n0 r0G(~ ? r0) ? G(~ ? r0)n0 r0E (r0) ds0 r ~ ~ ~ ~ provided the point ~ is inside the closed surface S . BEAM OPTICS AND INTERFEROMETERS Consider now a volume V in which Eq.

7.12) is especially interesting. It can however be widely exploited. y0. Assume that the surface z = 0 contains a hole.2. within the hole. ?z0]. We can assume that at the immediate right of the surface z = 0. and a zero contribution by the G2 integral.3 Application of the Kirchho equation .11) r~ ~ ~ ~ where G = G1 + z=0 G2 with arbitrary. we note that it is the symmetrical r ~ of the preceding with respect to the plane z = 0.2. We may thus add any multiple of G2 to G1 without changing the result: ZZ h i ~ ~ ~ r~ E (~) = r E (r0) n0 r0G(~. r0)n0 r0E (r0) dx0 dy0 (2. and identical to the eld coming from the left. This means that on the right side of the plane. r0) = G1(~ ? r0) ? G2(~ ? r00) r~ r ~ r ~ (2. which is assumed to be given throughout the aperture r 2. by slightly changing its meaning. A SHORT THEORY OF DIFFRACTION 95 eik 00 G2(~ ? r00) = 4 00 r ~ (2. If the surface S is this plane. because it gives a Green function that is zero on the surface z = 0. transmitted through the hole.13) z=0 this is the Kirchho equation The preceding equation. but taken in the strict sense. establishing a relation between the eld inside a volume and the eld at the boundary is exact. r0) dx0 dy0 (2. the eld is zero outside the hole. because the correct way to impose boundary values is out of this theory.2. in the following situation. the eld is simply the eld at the left.10) ~ with 00 = j~ ?r00j and r00 = x0. We can change the sense of the Kirchho equation (2. we rst need to know E . and that a primary electromagnetic wave is coming from the left (see Fig.2. any point in the right half space. of almost no practical interest: It could seem that in order to compute E . one is the coming one E1(~).8) being exchanged.2). We obtain simply ZZ ~ ~ ~ r~ E (~) = r E (r0) n0 rG(~. which greatly simpli es the equation. The special choice G(~.2. will give a non-zero contribution by the G1 integral. the equations (2. r0) ? G(~.13) by introducing two elds.

r0) r~ n0 rG(~.14) D ~ ~ r~ The function K (~. and the second one E2(~). q 2 = z + 2 + z2 ? z = z + p 2 2 z + +z so that " # ik 2 ik = eikz p e exp (2. r r~ 1+ k q (x ? x0)2 + (y ? y0)2 + z2. we obtain: ! i Z Z E (r0) eik 1 + i z dx0 dy0 E2(~) = ? r (2.(2. which is to be computed from the preceding r one by using a Kirchho -like formula: ZZ ~ ~ ~ r~ E2(~) = r E1(r0) n0 rG(~. it is necessary to extract the rapidly oscillating term in exp(ik ) by writing.15) 1 ~ k D For a numerical implementation of Eq. with 2 (x?x0)2 +(y ?y0)2.15). BEAM OPTICS AND INTERFEROMETERS E2 (r) r D E1 (r) z=0 Figure 2. r0) is called the Di raction Kernel.16) z + z2 + 2 .96 CHAPTER 2. r0) dx0 dy0 (2. where With this explicit formula. It can be explicitly computed: ! ik i z 0) = ? i e K (~.2: D.

A SHORT THEORY OF DIFFRACTION 97 The rst exponential. 1 cm]. As a rst example. whereas the second exponential is slowly oscillating. we compute the di raction pattern of a rectangular aperture illuminated by a constant amplitude.2. The far eld theory foresees a central spot surrounded by rings.2.b=1cm] directed along y.e. The window containing the source was -1 cm. numerically. The distance of the observation plane is z=1 km. which is much more convenient. 1 cm] -1 cm.a=0.e = 3:83171 ) rdark.and the discretization grid was 200 200 points. and goes out of the integral. the dark rings correspond to solutions rdark of J1 2 za r = 0 the rst zeros of the Bessel functions J1 are 1 i. and the discretization grid used for numerical integration was 200 200 points. The well known far eld theory 7] gives a central spot of rectangular shape.3) As a second example we compute using Eq. The radius of the aperture is a = 1 cm. and its width -a=-0. The aperture has its length -b=-1cm. with its longer dimension along x.15) the di raction pattern of a circular aperture illuminated by a constant amplitude.2. the distance of the observator is z = 1 km.5cm. The computational window containing the source was -1 cm. the wavelength is = 1 m. and its shorter along y. the wavelength is = 1 m.5cm] along x. xn = n2az = n 10:6 cm and " # sin kby = 0 z ym = m bz = m 5:3 cm 2 The plot is logarithmic with respect of the light intensity (Fig.1 6:5 cm . The dark lines correspond to solutions of " # sin kax = 0 z i.(2. rapidly oscillating represents pure propagation. 1 cm] -1 cm. 1 cm].

5cm.05 0.10 0. BEAM OPTICS AND INTERFEROMETERS 0. b=0.20 0.15 -0.15 -0.20 -0.00 0. at z=1 km for the Nd:YAG wavelength: Distribution of log(I) .15 0.05 0.15 0.3: Di raction pattern of a uniform rectangular aperture a =1 cm.20 -0.00 -0.05 0.10 -0.98 CHAPTER 2.20 Figure 2.10 -0.10 0.05 -0.

3 11:9 cm 17:2 cm The plot (Fig.15 0.4) is logarithmic with respect to the light intensity. along the propagation direction z) cut of the intensity distribution (Fig.20 99 0. give access to the near eld.2. Numerical exploitation of the Kirchho formula. In this case.2.05 0. The near eld theory of this case is analytically di cult.10 -0.00 0. See for instance a longitudinal (i.2.00 -0.20 -0.05 0.15 -0. The two precedent di raction .2.20 -0.10 -0.05 0. The transmitted eld is computed starting from z = 0. A SHORT THEORY OF DIFFRACTION 0.e.5).15 -0.4: Di raction pattern of a uniform circular aperture a =1 cm at z=1 km for the Nd:YAG wavelength: Distribution of log(I). it can be seen that the dark rings coming out of the numerical calculation are in agreement with the far eld theory.20 Figure 2.15 0.1mm radius. the circular aperture had 0.10 0.10 0. longitudinal cut 2 3 = 7:01559 ) rdark.05 -0.01 mm.2 = 10:17347 ) rdark.

1 mm for the Nd:YAG wavelength: Distribution of log(I) .750 0.002 0.250 0.001 0.500 0.002 -0.5: Di raction pattern of a uniform circular aperture a = 0.100 CHAPTER 2.003 0.000 0.001 0.004 -0.004 Figure 2. BEAM OPTICS AND INTERFEROMETERS 1.003 -0.000 -0.000 0.

A SHORT THEORY OF DIFFRACTION 101 patterns are roughly known after their far. this is a case where use of the Kirchho formula is necessary. that the di raction kernel tends to a delta function for z ! 0. the amplitude extends to in nity.e. But let us consider a rather exotic window shape. i. whereas the computing window is nite.17) .2.7). In the preceding case. Even the far eld is rather intricate to estimate.9). and spurious rings are generated.2. z) = with Z1 ?1 dx0 Z1 ?1 dy0 " !# eik 0 ? eik 00 @z0 4 0 4 00 E1(x0.2. Use of the Kirchho integral gives however the result at any distance: (see Fig. the question of the agreement of the approximation done in the preceding subsection with reality could be raisen: Let us try to discuss this issue.6). this is the amplitude of the light taken at the input mirrors of the Virgo cavities.2. It seems to di ract without any lobe. g. even if it becomes negligible for radial distances larger than w0. Because we have changed the meaning of the surface integral.8.2. y.15). It is necessary to take a window much larger than the gaussian radius of the beam. y. Finally let we consider the gaussian beam (this will be studied later in detail). that the secondary eld E2 will reduce to the input eld E1 when z ! 0. 0) z0 =0 (2. of amplitude " 2 2# + E (x. as can be seen on Fig. faint lobes can be observed on a logarithmic plot (see Fig.4 Consistency of the Kirchho equation E2(x. Too narrow windows are understood like a diaphragm.2. as for instance a regular 5 folded star (seee. nor even by the numerical (to be presented farther) Fourier transforms methods because of its sharp edges. because it cannot be treated analytically due to the complexity of the aperture. even with a computing window as wide as 30 cm. In fact. We give this example. y0. It is not obvious by only looking at the Kirchho formula (2. The Kirchho equation can be written under the form: 2.2.eld approximation (see below). For w0=2 cm. 0) / exp ? x w2 y 0 in the plane z = 0.

6: A starred pupil: a regular polygon in a circle of radius 1 mm 0.001 -0.000 -0.000 0.001 x y 0.102 CHAPTER 2.001 0.001 -0.000 -0. BEAM OPTICS AND INTERFEROMETERS -0.000 0.001 0.001 Figure 2.001 .001 -0.

004 -0.004 Figure 2.003 -0.7: di raction pattern at 1 m from the starred source.000 -0.002 0.002 -0.000 0.001 0.004 -0.001 -0. .003 0. A SHORT THEORY OF DIFFRACTION 103 0.2.003 0.2.002 -0.002 0.001 0.001 0. The thin circle indicates the size of the initial starred window.004 0.003 -0.

2500. 1500.8: Di raction of a gaussian wave from 100 m to 3 km .10 Figure 2.08 -0.06 -0.02 0. 2000.06 0. 1000.00 0. -0. 500.104 CHAPTER 2.04 -0. 0.02 0.10 -0. BEAM OPTICS AND INTERFEROMETERS 3000.04 0.08 0.

2500.05 0.2.10 -0.9: Di raction of a gaussian wave from 100 m to 3 km. 500.15 0.logarithmic plot .00 0. -0.05 0. A SHORT THEORY OF DIFFRACTION 105 3000. 1000.2. 1500.20 -0.20 Figure 2.10 0. 0.15 -0. 2000.

19) g(x. z) = eiz pk ?p ?q 2 2 2 f E1(p. q. q. q. q) A useful result can be found in 9]. the Kirchho equation becomes.18) and the reciprocal transform by Z Z f (x. and therefore may be transformed into a simple algebraic product by a 2D Fourier tranform in the variables (x0. q. y0). y) f (2. y. The function R R (2. q) = dx dy eipxeiqy f (x. y) = 41 2 R dp R dq e?ipxe?iqy f~(p. q. Recall that the 2D Fourier Transform of any function f (x. BEAM OPTICS AND INTERFEROMETERS 0 00 q = (x ? x0)2 + (y ? y0)2 + (z ? z0)2 q = (x ? x0)2 + (y ? y0)2 + (z + z0)2 It is clear that the integral may be viewed as a 2D convolution product. z) = has the following FT: 2 eik x2+y2 +z2 px2 + y2 + z2 p eiz k ?p ?q g(p. z) = 4@z0 @ i ? i 2 k 2 ? p2 ? q 2 2 k 2 ? p2 ? q 2 z0 =0 2 2 2 2 2 2 f E1(p. y) of integrable square modulus is de ned by: Z Z ~(p. z) = i pk2 ? p2 ? q2 ~ After a Fourier transform.106 CHAPTER 2.20) . 0) and eventually reduces to f E2(p. 0) (2. using this result: 2 2 2 p p 13 2 0 i(z?z0 )pk ?p ?q (z eip+z0 ) k ?p ?q A5 ep f E2(p.

despite a serious change of meaning with respect to the Green theorem.2. which becomes. and we can write: The Fresnel approximation . and as will be seen later on. in resonant cavities. z) = G( . We can conclude that. the nal data of slice #n being the initial data for slice #(n+1). after a FT: ~ @z2 + k2 ? p2 ? q2 E (p. 0) ) E2(x.(2.15) becomes negligible. We can add the following remark: If we interpret the 2D Fourier transform in the transverse plane as a continuous expansion on plane waves of various directions. q.2. y. it is possible to split space into successive slices along the propagation direction. de ned as the Fourier Transform of the di raction kernel has the very simple form ~ (2. the propagator reduces to 1. and the di raction to r ~ 2 2 2 p f f E2(p. q. showing that the di raction kernel reduces to (~ ? r0).5 The Fresnel approximation and the paraxial di raction equation (PDE) As soon as the distance z separating the input aperture from the observation plane is much larger than the wavelength. 0) = E1(x. the Kirchho formula is strictly equivalent to the wave equation. q. we see that the propagator is nothing but the phase change along the z axis of this special plane wave: ~ ~ G(p. z) = 0 Anyway. ) denote that direction. q = k sin sin where ( .2. . provided that re ections at each cut do not exist or are ignored. at least in the case where initial data are given on a plane screen. This scheme can be used in compound systems with interfaces. for z ! 0. z) = eikz cos 2. It follows that if convenient. q. A SHORT THEORY OF DIFFRACTION 107 which shows that the propagator. z) = eiz k ?p ?q We see that this is perfectly consistent with the Helmholtz equation.21) G(p. 0) = E1(p. q. the 1=k term in Eq. 0) as could be expected. y. by identifying p = k sin cos .

22 is small. and we can for instance say that the elementary amplitude created at ~ by the small elementary source r ds(x0. the amplitude is the sum of all these wavelets. For this purpose. y) = E1(x0. we are in the paraxial regime. we have thus p 2 3 ikz eiz k ?p ?q 5 eikz = 42i pk2 ? p2 ? q2 = 2i ke 2 2 2 It is therefore necessary that = ?i= . we can require that the propagation of an inde nite plane wave is the same plane wave.108 CHAPTER 2. y0)dx0 dy0 Z Z eik dE2 (x. This leads to the paraxial di raction integral: p=q=0 . This means that Z Z eik ikz = e dx0 dy0 R2 is: The integral is easy to compute. y0) is seen from the observation point (x. y). y0) = E1(x0. and the distance Z long enough. as said above. y0) dx0 dy0 D where is some coe cient to be determined. we can neglect the quantity q (x ? x0)2 + (y ? y0)2 with respect to z.22 is often referred to as the \Huyghens-Fresnel" equation. BEAM OPTICS AND INTERFEROMETERS I ik ~ E2(~) = ? i E1(r0) e cos( ) ds0 r D (2. If in Eq. This was known long before Kirchho 's theory. being the value at p = q = 0 of the Fourier transform of eik = that is known. which is the mathematical justi cation to the Huyghens principle and to the Fresnel formula. It can be (and was) derived heuristically by considering all points of the aperture as elementary sources of spherical waves: At any point of the right hand side half space. Eq. up to a phase factor. except in the phase factor. If the observation point is near the optical axis.22) where is the angle under which the element of aperture centered at (x0.2.2.

y) = e?Z(x +y ) where Z is any complex number of positive real part. q are restricted to small values due to the behavior of the function to be propagated ("small" means p.24) (2. it can be shown that 2 2 p q e G(p. y 0. because the Fourier transform of KP is easy to compute. z) = exp ?i z(p2 + q2) KP 2k 2+ 2 4 (2. If is small. q.23) All consequences of this formula are said having been obtained within the Fresnel approximation. z) = eiz k ?p ?q by assuming that the values of p. A SHORT THEORY OF DIFFRACTION E2(x. q k). y. 0) exp ik (x ? x ) + (y ? y ) dx0 dy 0 E1(x 2z D 109 (2. Remark that this equation is the convolution product of the eld E (z = 0) with the simpli ed (paraxial) di raction kernel " # i exp ik x2 + y2 KP (x. q.25) Obviously. we have ' sin = pp2 + q2=pk2 ? p2 ? q2.2. p.2. k2 ? p2 ? q2 may be thought of as the coordinates of the wave vector of an elementary plane wave. the di raction is "adiabatic" along z (if z is regarded as an evolution parameter). so that p the angles of the rays with respect to the axis are small. q. y. For a function of the form G(x. z) = ? iz exp(ikz) " ZZ 02 0 2# 0 . we could have deduced it from the "exact" propagator ~ Gexact(p. Then . we can thus write: 2 2 2 p . z) = ? z 2z Use of the Fourier transform is especially convenient here. q) = Z e? Z in particular the propagator is " # g (p. This is one more version of the paraxial approximation. being the direction of that elementary wave.

z) g E (p. 0) ( 2 + 2) 2 This is a very convenient way. It is clearly equivalent to the Fresnel integral.27) If the eld is expected to propagate mainly in the z direction. y. z) = eikz E = 0 (2. More speci cally.28) 2 @x + @y2 is the transverse Laplace operator. we intend to use the approximation @E kE @z for neglecting second order derivatives of E . z) in which the envelope E (x.110 CHAPTER 2.e. y. q. Consider the Helmholtz equation: h i + k2 E = 0 (2. q. z) = eikz e?i k An alternative way of computing E2 is therefore: g f E (x. y. y. The Paraxial Di raction Equation E (x. for by taking the Fourier transform of Eq. z) = K (p. as will be shown later. so that the Helmholtz equation becomes: where 2ik @z + T T] E (x.28) with respect to x. we obtain: . with a slow expansion in the transverse plane.(2. z) is assumed to depend slowly on z. q. the rapidly oscillating factor having been extracted.26) One can derive from the Helmholtz equation an approximate equation called The paraxial di raction equation (PDE) which is equivalent to the paraxial di raction integral.This is the PDE. y. 2 P 1 (2. BEAM OPTICS AND INTERFEROMETERS zp q ~ G(p. we can use the slowly varying envelope approximation scheme. i.

25). the order of magnitude of the argument of the quadratic term in the complex exponential is 2 < az = NF NF is called Fresnel number. 2 y .6 The Fraunhofer approximation " . q. 0 ~ E (x. If the observation distance is so large that NF may be neglected. y0.2. z) = ? z z z z Z The ultimate approximation for a di racted wave holds when the very far eld is considered. y. z) = 0 the solution of which is of the form " # 2 2 ~ (p. y. y. A SHORT THEORY OF DIFFRACTION i~ 2ik @z ? (p2 + q2) E (p. 111 h # " # " # x02 + y02 exp ?i xx0 exp ?i yy0 E (x0. 0) dx0 dy0 exp ?2i z z R2 which is nothing but the Fourier transform of the incoming amplitude: " # i exp i x2 + y2 E 2 x . z + z) = E (p. we can write simply " # i exp i x2 + y2 E (x. z) = ? z z " # " # Z xx0 exp ?2i yy0 E (x0.2. q.2. z) = ? z z 2. 0) dx0 dy0 exp i z z z R2 If we assume the transverse extension of the initial amplitude bounded by a radius a. z) exp ? i (p + q ) z ~ E 2k in which we recover the propagator (2. q. y0. The Fresnel-Huyghens integral can be written as: " # i exp i x2 + y2 E (x.

2. For a uniform circular aperture.20): p f f E1(p. and E2(x. a] ?b.112 CHAPTER 2. when r=z is su ciently small. y. y.10). BEAM OPTICS AND INTERFEROMETERS This is the Fraunho er approximation. so that we have the ratio P1 = S 2 P z 0 The action of thin optical elements. Strictly speaking.7 Representation of optical elements . for a rectangular and uniform aperture ?a. y. the eld arriving on the mirror's surface should be computed fromn the eld in the plane by Kirchho 's equation. one nds immediately 2 b2 2 jE (x. b]. r < a.4. Anyway. and allows to compute quickly the properties of the di racted eld for z very large. we have S 2 I (0) I0 z where I(0) is the on axis intensity in the far eld. z) the eld on the mirror's surface. z)j2 = a J1 2 z r explaining the pattern of Fig.2. Consider for instance the re ection o a curved mirror of curvature radius Rc and diameter D. we have as seen in (2.2. For instance. q. in the very far eld. 0) the eld in the plane z = 0 . Assume the mirror to close the aperture in the plane z = 0 (see Fig. z) = eiz k ?p ?q 2 2 2 2. 0) E2(p. we nd ar 2 jE (r. and I0 the initial intensity. If we consider the total power P0 passing through the aperture . It is more convenient to discuss in the Fourier space. q.2. we get S I (0) 2 z 2 P0 The total power received by an equal area in the far eld is P1 = S I (0). Calling E1(x. like thin lenses or nearly at mirrors on the optical amplitudes can be modelled without using a di raction integral.3. z)j2 = 16az2 sinc 2 zx sinc 2 zy 2 explaining the pattern of Fig.

this aperture is of the order of tens of microradians.2.2.10: re ection o a curved mirror The argument of the imaginary exponential can be written as: q q z k2 ? p2 ? q2 = kz + z k2 ? p2 ? q2 ? k or q 2 2 q 2 p +q z k2 ? p2 ? q2 = kz ? z(p 2+ q ) k 1+ 1? k It is necessary to estimate the di erent orders of magnitude of these terms. 2 2 2 The quantity p2 + q2 is determined by the spatial behavior of the input wave.y) z=0 Figure 2. A SHORT THEORY OF DIFFRACTION D/2 113 Rc z z=f(x. For long baseline interferometers. . If the spatial frequencies are of the order of magnitude of w0. for instance for a TEM mode of Fourier transform 2 2 2 ! ~(p. q) = exp ? w0 (p + q ) 4 we have !2 p2 + q2 4 2 g 2 2 w2 k k 0 w0 g being nothing but the divergence angle of the beam.

0) .45 km) this is: z(p2 + q2) 3:8 10?9 2k The conclusion is that we can write with a good accuracy f f E2(p. z) = eikz E1(p. We z(p2 + q2) D 2 2k 8 Rc Taking again Virgo gures (D=35 cm. on the other hand. q. q. g = 1:7 10?5 Rd. whereas z is of the order of tens of micrometers. It is clear that we can neglect this term. Rc=3. the Rayleigh parameter is related to the curvature radius (in a at/parabolic cavity of length L) by q zR = L(Rc ? L) = 1 Rc where = q R=L R=L ? 1 the factor is of the order of the unity (For Virgo.114 CHAPTER 2. More precisely. BEAM OPTICS AND INTERFEROMETERS for Virgo. The argument of the imaginary exponential therefore reduces to but q 2 2 z k2 ? p2 ? q2 = kz ? z(p 2+ q ) k z(p2 + q2) z z 2 2k w0 zR where zR is the Rayleigh parameter of the beam (see below). we have for a parabolic mirror zmax = D2 =8Rc . about 1km for GW interferometers. have thus: ' 2:97).

z) = eA(z) eikr =2q(z) 2 . for a parabolic mirror. z) E1(x0. FUNDAMENTAL TEM MODE which by inverse Fourier transform gives simply 115 E2(x. well adapted to gaussian beams in the paraxial approximation. y) = e?ikf (x. we have " # 2 (x2 + y2) R = exp i (2. 0) In other words. y) = e2ikf (x. y) and the re ection operator is simply the phase factor .y) E1(x. we reverse the point of view. y) For the re ected wave E3(x. so that if we consider the amplitude on a surface of equation z = f (x. y).2. If we would compute E2 knowing E3 we would nd (the propagation direction being reverse) Therefore E2(x. y. In particular.29) Let us recall that this only holds for "thin" optical elements.28 under the axially symmetrical form depending on two unknown functions of z: (r. R = e2ikf (x. z).y) E1(x. y.3 Fundamental TEM mode It is possible to nd a special solution of 2. y0. in the Fresnel equation Z E2(x. y ? y0. 0) dx0 dy0 we have shown that the kernel Kp is a 2-D delta function for small z. z) = eikz E1(x.30) Rc 2. y. y) = eikf (x. y. z) = eikz Kp (x ? x0. y) E3(x.3.y) (2. we can write simply E2(x. in the above discussed sense.y) E3(x.

and then " # 1 A(z) = ln 1 + iz=b or as well 3 2 1 5 ? i arctan(z=b) A(z) = ln 4 q 2 =b2 1+z on the other hand we have. so that 1 A(z) = ln z ? i b + C The arbitrary integration constant C may be chosen in order to have A(0) = 0. i. the wave 2 is a real gaussian function of parameter w0.28 provides two coupled di erential equations: dq = 1 and dA = ? 1 dz dz q . (i. w(z) and R(z): ik = ? 1 + ik 2q(z) w2(z) 2R(z) The de nitions of w(z) and R(z) are consequently: q w(z) = w0 1 + z2=b2 from where we get rstly substituting this expression in 2.e. of the form exp(?r2=w0 )).116 CHAPTER 2. We have then q(z) = z ? i b. separating the real from the imaginary part of 1=q: ik = 1 + i 2q(z) z + b2=z b + z2=b de ning two new real functions . BEAM OPTICS AND INTERFEROMETERS q(z) = q0 + z It is convenient to choose the constant q0 in such a way that at z = 0.e. This clearly happens if 2 q0 = ?i kw0 = ?i b 2 2 The parameter b = kw0 =2 is called Rayleigh range. C = ? ln(?1=ib).

3. z) is called TEM(0. and R(z) is the curvature radius at the same point. z) is a very special one. The extra phase arctan(z=b) appearing during propagation with respect to a plane wave is called Gouy phase.y 117 z Figure 2. but contain the same information as the complex function q(z) often called complex curvature radius We have nally the complete solution for the envelope: (r. The solution (r. One can nd other solutions by considering the product of (r. z) = q 1 2 2 e?r =w(z) eikr =2R(z) e?i arctan(z=b) 1 + z =b 2 2 2 The factor exp(ikz) may be added for representing the rapidly varying part. FUNDAMENTAL TEM MODE x. z) by polynomials in the variables (x=w . The solution (r.0) propagation mode. It is the fundamental mode of two families of modes discussed below. y=w). These two real functions have concrete optical meanings.11: Di raction of a gaussian wave: equal intensity and equal phase surfaces R(z) = z(1 + b2=z2) w(z) is the beam half-width at abscissa z. .2.

and the LaguerreGauss modes LG(m.0)(x.118 CHAPTER 2.n) when polar coordinates are convenient. by introducing the scalar product: Z h f . z) is. y. Obviously. The more often employed bases for studying cavities and laser beams. with r2 x2 + y2. The fundamental mode has been de ned above as: TEM(0. whose corresponding vectors are called Transverse Electromagnetic Modes.0)(x. y)g(x. z) = w(z)2 2 2 2 where w(z) gives the radius of the beam. z) = LG(0. g i = 2 dx dy f (x. s 2 eikz e?i arctan(z=b)e?r =w(z) eikr =2R(z) (2.n). f i = 2 dx dy jf (x.32) R is nothing but the light power of the beam crossing the transverse plane at z: restriction to L2 is therefore not too demanding. BEAM OPTICS AND INTERFEROMETERS 2. y. are the Hermite-Gauss modes HG(m. The form of w(z) suggests a widening of the beam (see Fig.0)(x. y) of integrable square modulus may be given the structure of a Hilbert vector space.4 Discrete bases for free space propagation The set L2 of all complex functions f (x.31) R If we think to these functions in terms of optical amplitudes at a given point of the path of a light beam having the preferred propagation direction z.11) during propagation. y)j2 (2. y) (2.0)(x. R(z) the curvature radius of the phase surface. y. z) = HG(0. It is possible to nd discrete bases. and are labelled by two indices: TEM(m. an angle aperture can be evaluated by lim w(z) z!1 z This gives the gaussian aperture angle g = w 0 . a number of bases can be constructed for this Hilbert space. b the Rayleigh range.n) when rectangular coordinates are convenient.33) TEM(0. y. we see that Z kf k2 = h f .2.

whereas u is a real function of z. and secondly the clear necessity to include a variable scaling factor in the transverse plane accounting for the extension of the wavefront. Now we furthermore require that separately: ! @ 2P + 2ikx @u + u @P + 0P = 0 u2 @X 2 (2. and P. i. and for the curvature radius of the wavefront. DISCRETE BASES FOR FREE SPACE PROPAGATION 119 2. as seen in the fundamental mode.e. The reason for these choices are rstly a separation of the variables x and y.34) @Y 2 where we used the notations X u(z)x and Y u(z)y. After straightforward calculations. Now we require the function q(z) to be the same as in the fundamental solution. z) = eA(z)eikr =2q(z)P u(z)x] Q u(z)y] where A and q are complex functions of z alone.4.35) @z q @X 2 Extended solution ! @ 2Q + 2iky @u + u @Q + 00P = 0 (2. Owing to the fact that u must be real. we expect the unknown function u(z) to be inversely proportional to w(z). 00 are real arbitrary constants. Let us look for solutions of the form: (x.36) @Y 2 @z q @Y where 0.4. Q real functions. it is necessary that @u + u @z q u2 and . y.1 Hermite-Gauss modes The fundamental solution found above can be extended in the following scheme. as also P and Q. @q ? 1 = 0 @z in order to keep the same dependence for the width of the beam. the paraxial di raction equation becomes: ! ! ! ! @A + 1 + k2r2 @q ? 1 +2ik @u + u x @P Q(Y ) + y @Q P (X ) 2ik @z q q2 @z @z q @X @Y ! 2 2 2 @ P Q(Y ) + @ Q P (X ) = 0 +u @X 2 (2.2. In this spirit.

2. in this case.2. an arbitrary constant.35 becomes: 2 @ 2P ? 4 X @P + 0P = 0 w2 @X 2 w2 @X or @ 2P ? 2X @P + 0w2 P = 0 (2.37 exist. P (X ) Hn (X ) u(z) = p 2 2 b +z where . we have: @u + u = ibu @z q z 2 + b2 p 2 2 = i kw3 so that eq. if 0 w2 2 = 2n where n is any integer.37 de nes the Hermite polynomial of order n. This is nally p 2 u(z) = w(z) .120 CHAPTER 2.2. But now. BEAM OPTICS AND INTERFEROMETERS u < @u + z ? ib @z be purely imaginary.37) @X 2 @X 2 We know that polynomial solutions of eq. may be chosen in such a way that u(0) = p 2=w0. This is ( or ) = 0 1 @u = ? z u @z z 2 + b2 which gives the obvious solution w(z) being the function de ned above in the fundamental solution. eq.

The Hermite polynomials are de ned by: Hn(x) = ex2 ! d n e?x ? dx 2 (2. y.39) . Several properties of these functions are very convenient. eq.34 reduces to: Q(Y ) Hm (Y ) ! @A + 1 ? (m + n) 4 = 0 2ik @z q w2 @A + 1 + i(m + n) = 0 @z z ? ib b(1 + z2=b2) or so that: and 1 A(z) = ln z ? ib ? i(m + n) arctan z b eA(z) = (1 + z2=b2)?1=2 e?i(m+n+1) arctan(z=b) The HG basis It has been shown that the PDE has Hermite-Gauss solutions of the form ! p y ! ikz H p2 x HG(m.n e m w(z) Hn 2 w(z) 2 2 2 e?i(m+n+1) arctan(z=b) e?r =w(z) eikr =2R(z) (2. and with 00w2 121 2 = 2m we nd Now.n a normalization constant to be de ned later.36.2.38) where the functions Hn (X ) are the Hermite polynomials and cm. and we recall them herafter without any proof.4. the same discussion holds for eq.n)(x. z) = cm.2.2. DISCRETE BASES FOR FREE SPACE PROPAGATION Obviously.

n = w2 2m+n1 ! n! (2.43) m They obey as well a closure relation: 1 X p 2p1p! Hp (x)Hp(x0) e?(x +x0 )=2 = (x ? x0) p 2 2 (2.42) mn The normalization constants for the HG modes are therefore: 1=2 2 cm.41) (2.40) (the bracket means the integer part) They obey the following di erential equation: 00 Hn (x) ? 2xH 0n(x) + 2nHn (x) = 0 Their derivatives are given by: 0 Hn (x) = 2n Hn?1 (x) They obey a recurrence relation: Hn+1 (x) = 2x Hn (x) ? 2n Hn?1 (x) They obey an orthogonality relation Z1 p 2mm! Hm(x) Hn (x) e?x dx = ?1 2 (2.45) (it can be shown using the recursion formula) . BEAM OPTICS AND INTERFEROMETERS The explicit expression is: (?1)s (n ?n2!s)!s! (2x)n?2s s=0 n=2] X Hn (x) = (2.122 CHAPTER 2.44) There is a translation formula: Hn (x + =2) = n X k=0 k Cn Hn?k (x) k (2.

4.n)(Z.2.46) Hn?2k (x) There is a reduction formula: min(m. p. p.n)(1. DISCRETE BASES FOR FREE SPACE PROPAGATION There is a scaling formula: n=2] X n! Hn ( x) = k=0 k !(n ? 2k )! n?2k 2?1 k 123 (2. The HermiteGauss functions correspond to Z = 1. We even give a more general formula under the following form. .n) X m! n! 2s Hm (x) Hn (x) = (m ? s)!(n ? s)!s! Hm+n?2s (x) s=0 (2. rather with the FT of the product of two modes.49) note that this formula has nothing to do with the Fourier transform of a TEM mode. we see that the HG modes are eigenvectors of the Fourier transform.n) (Z. Let ! p x p y x2 + y 2 2 w Hn 2 w exp ?Z w2 (m. q) = p p 2 2 2 8 2 2 2 2 There is a useful Fourier transform: 1 Z p e?x Hn (x) eipx dx = (ip)n e?p =4 (2.n)(2. p.47) It is possible to give the general expression of the Fourier Transform of any mode .48) For Z = 1 (HG functions) this is simply: ! " ! # pw H p exp ? w2(p2 + q2) qw 2H e (m. q) = w m p 4 2 n 2 In a certain sense. y ) = Hm where Z is any complex number of positive real part. The special case Z = 2 gives ! ! " # w2 ipw m iqw n exp ? w2(p2 + q2) e (m. q) = w i 2Z ? Z 2 Z Z ! ! " 2 2 2# + qw pw Hn p p exp ? w (p4Z q ) Hm p p 2 2Z ? Z 2 2 2Z ? Z 2 (2. The Fourier Transform is: m+n 2 (m+n)=2 e (m. x.

14. 2. .. 2. The intensity pattern of some HG functions is shown on the gures 2. 2.2 The Laguerre-Gauss modes Using polar coordinates (r.13. of the form !n ikz p2 r (n) 2 2 LGm.4.12. y) in the transverse plane.50) n The functions L(m )(X ) are the generalized Laguerre polynomials. a new class of solutions to the PDE can be found.. They are de ned by ! ex d m xn+me?x (n) (x) = Lm m! xn dx . ) instead of (x.124 CHAPTER 2.n e w(z) Lm (2r =w(z) ) 2 2 2 e?i(2m+n+1) arctan(z=b) e?r =w(z) eikr =2R(z) cos(n ) (2. z) = cm.n(r. BEAM OPTICS AND INTERFEROMETERS A consequence of the preceding integral (or an application of the generating function as well) is the expansion of a plane wave in terms of Hermite polynomials: X )n eipx = e?p =4 (2ipn! Hn (x) n n 0 2 The rst Hermite polynomials are explicitly: H0(x) = 1 H1(x) = 2x H2(x) = 4x2 ? 2 H3(x) = 8x3 ? 12x H4(x) = 16x4 ? 48x2 + 12 H5(x) = 32x5 ? 160x3 + 120x H6(x) = 64x6 ? 480x4 + 720x2 ? 120 etc.

01 -0.04 -0.12: Intensity pattern of a HG11 mode for w0 = 0.03 0.03 0.02 m .03 -0.1e+00 1.01 0.02 0. DISCRETE BASES FOR FREE SPACE PROPAGATION 125 0.01 0.03 -0.2.04 -0.02 -0.00 0.4.04 0.04 5.02 -0.5e-06 5.01 0.4e-01 1.6e+00 2.2e+00 Figure 2.02 0.00 -0.

9e+00 5.03 -0.3e+00 2.2e+00 Figure 2.03 0.126 CHAPTER 2.00 -0.4e-06 1.6e+00 3.04 -0.02 0.01 0.04 -0. BEAM OPTICS AND INTERFEROMETERS 0.04 3.01 0.00 0.02 0.02 -0.01 0.03 0.03 -0.04 0.13: Intensity pattern of a HG20 mode for w0 = 0.02 -0.02 m .01 -0.

02 -0.2e+02 1.02 -0.01 0.4.03 0.00 0.02 0.5e+02 Figure 2.7e+01 1.04 -0.03 -0.04 1.00 -0.01 0.02 0.03 -0.2.03 0.9e+01 7.01 -0.01 0.2e-04 3.04 0.14: Intensity pattern of a HG32 mode for w0 = 0.02 m .04 -0. DISCRETE BASES FOR FREE SPACE PROPAGATION 127 0.

. in a certain sense eigenmodes called TEMm.n modes.128 CHAPTER 2.16. there is already a storage in the transverse plane They have a parabolic equiphase surface . They have the signi cant two following properties: They are of nite transverse extension.2. BEAM OPTICS AND INTERFEROMETERS They obey the recursion relation: n) ) n (m + 1)L(m+1 (x) = (2m + n + 1 ? x)L(m )(x) ? (m + n)L(n?1 (x) m The rst ones are as follows: L(0n)(x) = 1 L(1n)(x) = n + 1 ? x 2 L(2n)(x) = (n + 1)(n + 2) ? (n + 2) x + x 2 2 3 L(3n)(x) = (n + 1)(n + 2)(n + 3) ? (n + 2)(n + 7) x + n + 3 x2 ? x 6 6 2 9 The normalization relation for the Laguerre polynomials comes from 11]: Z1 + n L(m )(x)2 xn e?x dx = (mm!n)! 0 so that the normalization constants cmn are: s 2 cmn = w (1 + m!(m + n)! ) n0 The intensity pattern of some LG modes is given in the maps 2. 2.17.15.5 Fabry-Perot: paraxial approximation It has been seen that the free space propagation operator has.2.

02 -0.02 -0.04 7.00 0.5e-01 5.01 0.04 0.04 -0.9e-01 Figure 2.02 m .02 0.01 -0. FABRY-PEROT: PARAXIAL APPROXIMATION 129 0.00 -0.03 -0.01 0.0e-01 7.03 0.2.04 -0.02 0.01 0.03 -0.8e-05 2.5.03 0.15: Intensity pattern of a LG20 mode for w0 = 0.5e-01 9.

02 m .01 0.01 0.1e-01 7.16: Intensity pattern of a LG21 mode for w0 = 0.0e+00 Figure 2.01 -0.00 0.02 0.5e-01 5.03 -0.04 0.04 -0.03 -0.03 0.00 -0.7e-07 2.04 -0.02 0.130 CHAPTER 2.04 1.6e-01 1.03 0.02 -0. BEAM OPTICS AND INTERFEROMETERS 0.02 -0.01 0.

02 -0.5.00 0.2e-01 1.04 5.2e+00 2.03 0.04 -0.4e+00 2.01 0.02 0.03 0. FABRY-PEROT: PARAXIAL APPROXIMATION 131 0.01 0.2.00 -0.04 0.03 -0.8e-37 7.01 0.02 0.03 -0.9e+00 Figure 2.02 -0.01 -0.02 m .04 -0.17: Intensity pattern of a LG22 mode for w0 = 0.

18: Any Fabry-Perot cavity with curved mirrors . BEAM OPTICS AND INTERFEROMETERS M2 M1 z=0 w0 waist θg Figure 2.132 CHAPTER 2.

a class of perturbations of the mirrors having the right symmetry will pump power from one mode to the other due to the nite linewidths (see below). re ecting the mode on itself (see Fig. In particular. so that we can write ! b2 Rc = L 1 + L2 b (Rayleigh parameter) having the de nition previously encountered. 0) are well separated from the reference mode. in order to be matched to M1.2. This gives q b = L(Rc ? L) Clearly.n. provided it is near resonance.2. assuming a =2 dephasing at each re ection. Let us discuss this issue now. this is possible only if Rc > L.n mode. at a distance L. If two modes have by chance close eigenfrequencies.p = c b .18) A mode matching two parabolic mirrors not always exists. n. FABRY-PEROT: PARAXIAL APPROXIMATION 133 The second property allows to make "matched" mirrors. Consider for instance a planespherical cavity with a plane input mirror M1. exactly as the modes of a closed box. and a spherical mirror M2 of curvature radius Rc . of shape adapted to the equiphase surface. the stored wave must have a phase curvature radius of Rc at z = L. is : ? 2(m + n + 1) tan?1 L 2p b the eigenmodes of the cavity are thus labeled by 3 integers.0 modes are especially well coupled with the TEM0. depending on the curvature radii of the mirrors and on the cavity length. The size of the waist is q w0 = b= The resonance condition. Then.pL ? 2(m + n + 1) tan?1 L + = 2p m. We have: 4 m. it is better to choose the geometrical parameters in such a way that the nearest transverse modes (m. n) 6= (0. The TEM0.0 in case of misalignment of mirrors. This is a stability condition for that type of cavity.n. If this condition is ful lled. if the cavity is operated on its fundamental mode. the cavity is able to store any TEMm.p the total dephasing of the m. We call m. the stored wave must be at its waist at z = 0 on the input plane.5.n.1 and TEM1. The frequency spacing between modes is a very important feature in a cavity. p mode over a round trip in the cavity.

n.p ? 0.134 CHAPTER 2.0. is FSR = c=2L. The frequency o sets of the 15 nearest tranverse modes are given in the following table.p0 . . the length of the arms is L = 3km and the curvature radius of the end mirror Rc = 3.45 km.n.n. we see that the frequencies of the modes are given by m.51 has no zero solutions for (m.38238. BEAM OPTICS AND INTERFEROMETERS We see that the frequency gap between two successive longitudinal resonances. ( p = 1). the curvature radius Rc can be chosen under the following constraints: it must be larger than L it must not cause a too large magni cation factor between the input mirror and the end mirror it must give a value of such that equation 2.0.51) The distribution of resonances being periodic.p = FSR p ? 1 + (m + n + 1) 2 0. In the case of Virgo. it is su cient to study it over a FSR. so that ' 0. or Free Spectral Range (FSR). n) small. Given the length L of the cavity. Assume the operation mode has frequency the distances of the other modes are: m.p = m.p0 = (p ? p0 + (m + n) ) FSR (2. with = tan?1(L=b)= .

0) = 1 2 dx0 dy0 (x ? x0.6. In the case of planespherical cavities. On the assumed at input mirror.79 2 38211. we try a similar method.78 6 14704. y ? y0) b 2.1 construction . it appears than gaussian modes are not the best choice. Large spots are better than sharp. y.32 14 17654. The nearest are the family (m+n=8) which are not easy to couple to (0.64 7 33810. With this respect. 2.0).48 4 26458.62 3 7352.49 Remark that the (0.02 12 29408.72 10 41162. was dealing with a symmetrical cavity. The idea of constructing more homogeneous modes has been proposed long time ago by laser scientists in order to better exploit ampli er media: such modes are called hypergaussian.18 15 36760.86 9 22056.34 5 45563.94 1 19105. A way of constructing almost at modes has been explored by D'Ambrosio ( 16]). In this work.0) modes are well separated from the (0.16 11 10302.6 Hypergaussian modes It will be shown in a foregoing section that the thermal noise (random motion of the mirror's surface) depends on the area of the light spot on the mirror. HYPERGAUSSIAN MODES 135 Mode order (m+n) Frequency o set (Hz) 8 2950.88 13 48514. we consider the eld as a superposition of gaussian modes according to : Z (x.1) and (1. and that there is no coincidence for orders lower than 15.0) by a simple perturbation. D'A.6.2.

20. uniformly distributed on the disk of radius b. and where s " # 2 exp ? x2 + y2 (x.21 2. BEAM OPTICS AND INTERFEROMETERS is the disk of radius b. The aperture angle of the beam is obviously much smaller than the gaussian's. 0) = 2 2 w0 exp ?w0 2=4 J1( b) b . The beam may be viewed as the convolution product of a gaussian of waist w0 with a uniform distribution on the disk r < b. i. On the at mirror. is the superposition of such modes with various o sets. A detailed calculation gives thus.2.19 The intensity pro le after 3 km propagation is plotted on Fig. L) = 2 w2 w2 0 0 0 0 b 0 where r=w.21.2.2 Angular aperture and Fourier transform It is remarkable that the mode is practically unchanged along the propagation. y = 0).136 where CHAPTER 2. we have taken the following values: s w = L ' 3:2 cm 0 2 0 0 b = 4w0 ' 12:8 cm We use a simple Simpson numerical integration technique. The wavefront is shown on Fig. centered at (x = 0. In other words.2. The di cult point is to make a mirror having the pro le shown on Fig. Z 1+ iL=zR (zR being the Rayleigh parameter). The initial intensity pro le is as shown on Fig.6. q.e. y) = 2 2 w0 w0 is a classical TEM00 mode. w is the beam width at the distance L. q 2 w = w0 1 + L2=zR Following E. It is straightforward to express the eld propagated at a distance L.2. The Fourier transform of the beam amplitude is therefore the simple product of the Fourier transform of the elementary gaussian beam with that of the disk. y. and where I0(z) denotes the 1st kind modi ed Bessel function.d'A. propagation of each elementary gaussian mode being known: s 2 2 Z b=w e?Z( ? ) e?2Z I (2Z ) d (x. q 2 2 ~ (p.

For w0 = 2 cm and b = 10 cm.00 0. and the aperture angle is practically determined by b (see Fig.50 Intensity [arb.10 radial coordinate [m] 0. By identifying = k .6.00 1.00 137 2. Owing to the Parseval-Plancherel theorem. HYPERGAUSSIAN MODES 3. It is di cult to compute directly the power carried by such a mode by simply integrating the intensity in the plane (x.22).19: Intensity pro le on the at input mirror where 2 p2 + q2. 0)j dp dq 4 2. Instead.50 1. we get " #2 2 = / exp ?2 2 = 2 2J1 ( = b ) j ~ (p.00 0.50 0. and b = =2 b is the Bessel aperture angle. 2. b happens to be 10 times smaller than g . y)j (2. This is consistent with the fact that the width of the beam is practically constant along the di raction length.3 Normalization . y).00 0. units] 2.05 0.20 Figure 2.15 0. 0)j g = b where g = = w0 is the gaussian aperture angle.2. we can write for the norm P: Z Z 2 dx dy = 1 2 ~ P = j (x. we do it in the Fourier space. q.52) 2 j (p.6. q.

00 1. using a precedent result: Z1 2 2 2 P = 2 w20 2 exp(?w0x2=2b2) J1(xx) dx b 0 The integral can be carried out.50 Intensity [arb.15 0.53) (2.55) 8b2 128b4 3072b6 .00 0.20: Intensity pro le at 3 km so that.00 0.20 Figure 2. yielding: 2 P (w0. so that b=w0 is large. using the asymptotic values of the Bessel functions we get: " ( #) 2 4 6 2w0 1 ? w0 ? 3w0 ? 45w0 ? ::: F 1?p b (2.138 3.00 0.50 1.50 0.10 radial coordinate [m] 0.54) with h i 2 2 2 F 1 ? exp(?b2=w0 ) I0(b2=w0 ) + I1(b2=w0 ) where I0(z) and I1(z) are the modi ed Bessel functions of the 1st kind.00 CHAPTER 2.05 0. When w0 is small. Units] 2. b) = 2 w20 F b (2. BEAM OPTICS AND INTERFEROMETERS 2.

50 1.00 139 Phase shift [Rd] 2. 0) = 0 0 0 0 2 b w0 F and nally.50 2. y ? y ) (2.6. L) = 0 w0b F 0 (2.10 radial coordinate [m] 0. w Z 1 + iL=zR. at any distance L: 2p Z b=w exp ?Z (r=w ? )] exp(?2Zr =w) J (2Zr =w) d w (x. zR w0 w0 Z:Z .00 0.57) with the same notation as above: q 2= .20 Figure 2.00 0.50 3. HYPERGAUSSIAN MODES 4.00 1. y.05 0.00 3.2.56) (x.00 0. y.50 0.21: Wavefront at 3 km The normalized at mode at its waist is: 1p Z dx dy (x ? x .15 0.

e-05 3. dashed line: circular aperture. BEAM OPTICS AND INTERFEROMETERS 1. Solid line: gaussian distribution.7 P(θ).2 0.e-05 angle [Rd] 2.5 0.6 0. units] 0.22: Angular distribution of the at beam.1 0. 1.e-05 Figure 2. [arb.9 0.4 0.140 CHAPTER 2. Red dashed line: resulting angular distribution .3 0.0 0.0 0.8 0.

Anyway. the polar representation will be convenient. instance: by sampling its complex amplitude on a rectangular grid in the x. and thick lenses both. thin lenses or almost at mirrors pure refraction or re ection. In case of very small misalignments of mirrors in a resonant cavity. These simple functions will be plane waves in the case of Fourier (or Hankel) transform based methods TEMm. For studying any property of the instrument. In the case of axial symmetry. each of the corresponding algorithm belong to the class of "spectral methods" expanding the optical amplitudes on a basis of simple functions for which the di raction/refraction problem is already solved. ) plane by expanding it on a discrete basis of modes to each of these decisions correspond a special way of computing the eld di racted at a distance z. and of space between them.n modes (HG or LG) in the modal methods 141 . There are a lot of ways of representing numerically a light beam. y plane by mapping it on a polar mesh in the (r. The method to choose depends obviously of the type of e ects we want to analyze.Chapter 3 Numerical methods An optical instrument is generally composed of optical elements like lenses. we have to represent the action of each of these elements on a light beam. mirrors. Everyone understands that space produces a di raction of the beam. the modal expansion will do the job. for.

The smallest frequency interval we can consider is obviously f = 1=T . . The sampling will thus be: 1 fm = m T and the samples of the Fourier Transform are: N ?1 e m e (m=T ) = X e2i mj=N j (3.1 Numerical propagation using Fourier transforms When the complex amplitude is sampled on a rectangular grid with equally spaced sampling points. If we consider the vector e m . because the longest time interval on which the function can be studied is T .142 CHAPTER 3. it is therefore su cient to compute f e m . m = 0. it is possible to use a discrete 2D Fourier Transform to propagate the eld.T ].(3. and write approximately: N ?1 ~ (f ) = T X e2i fjT=N (jT=N ) N j=0 Now it is possible to sample the function in the frequency domain too.1) with the notation j = (jT=N ). The Discrete Fourier Transform (DFT) comes from the crude approach of the numerical Fourier Transform of any function (t) which is zero outside the interval 0. The Fourier Transform then reduces to: Z ~ (f ) = T e2i ft (t) dt 0 For a numerical integration. : : :. N ? 1g. NUMERICAL METHODS 3. we can cut the interval in N slices of width t = T=n. several remarks arise It is easily seen that j =0 3.1 On the discrete Fourier transform e m+N = e m showing that the DFT has period N with respect to m.1.1) expresses the DFT. Eq.

1: Assign frequencies to the DFT samples Clearly m = 0 corresponds to the mean of the function (t).3. the second half of the vector f e mg contains the negative frequencies (see Fig.(3.∆ f Figure 3.1)) The maximum frequency is thus 1 fmax = N T 2 Consider a Fourier transform followed by the reciprocal: N ?1 e m = 1 X e2i mj=N N j=0 j (as already seen. it is easily seen that e N ?m = e ?m and as a result. the time element is T=N . and thus to the value at f = 0 of its Fourier transform. so that the time frequency element is 1=N ) then XX e n = 1 N ?1 N ?1 e2i m(j?n)=N N m=0 j=0 but ?N + 1 N ?1 X m=0 j j?n e2i m(j?n)=N ( e2i (j?n) ? 1 = 0 if j 6= n = e2i (j?n)=N ? 1 N if j = n N ? 1. . and the frequency element is 1=T . Now. so that and consequently. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS143 0 1 2 3 N-2 N-1 f=0 f= ∆ f f=2 ∆ f f=-2 ∆ f f=.1.

m 2 N j=0 j 0? 1 NX1 e2i m0j0=N 0 ei j=N 0 ? 2 j 0 =0 (3. this is the rst property to check.2m0 N 0 ?1 g = 1 T 0 X e2i m0 j=N 0 ei j=N 0 N. Practically.m 2 N j=0 j X 1 T 2N 0?1 e2i m0j=N 0 + 2 N0 j =N 0 0? 1 T NX1 e2i m0j0 =N 0 + 2 N0 j 0 =0 j and by renaming j = j 0 + N 0 in the second sum. Assume that N = 2N 0 even.3) j 0 +m . we have: g 2N 0.144 CHAPTER 3. NUMERICAL METHODS en = n (3. when implementing any DFT algorithm.m0 N0 N 2 In the case where m = 2m0 + 1 is odd. We can write N ?1 g = T X e2i mj=N N.m 2 N j=0 j j 0 +m if we note (1) and (2) the two halves (of lengths N 0 = N=2) of the input vector (of length N ). and m = 2m0 too. Let us denote here by g N. we have the following property g g = 1 (1).m the N -points DFT of . N 0?1 g = 1 T 0 X e2i m0j=N 0 N.2) which shows that the "approximate" of the inverse FT is the exact inverse of the "approximate" FT.m0 + (2)0.m N j=0 j X T 2N 0?1 e2i m0 j=N 0 = 2N 0 j =0 j By splitting the sum into two segments we get N 0 ?1 g = 1 T 0 X e2i m0 j=N 0 N.

1)). there is an optimal density of samples. It must be clear that the result of a DFT is not a sampling of the result of the continuous transformation. This is the fundamental remark that led to FFT algorithms. We consider the function ! t2 F (t) = exp ? 2 2 g 2N 0 . Its Fourier transform is: p ~ F (f ) = 2 exp ? 1 (2 f )2 2 and we compare a N -sample of the continous FT of F to the DFT of a N -sample of F (see table (3. The gain is for a 2 dimensional FT: " # N 2 Log2N for N = 128 this is a gain larger than 300 ! But this is at the price of a restriction of the validity of the method (very small di raction angles). The linear algebra involving vectors of size N and rank N DFT's is perfectly closed due to eq. The size of the window must be chosen such that the function takes vanishing values near the ends of the window. Moreover.(3.2m0 in our experiment. The interval over which the function is sampled is called window. but it represents a world di erent from reality. FFT routines allow to compute rank N DFT's with N Log2N algorithms instead of N 2. we see that the (1) (2) = 1 g 0+ g 0 (3.2).m 2 and the conclusion is that the N -ranks Discrete Fourier Transform essentially reduces to two N=2-rank partial transformations of the two halves of the input vector.3.1. which implies that when the size of the sample is changed.m N 0 . we take the time constant =1s.4) N 0 .1. the size of the window giving the optimal agreement changes too. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS145 If we introduce the new function preceding equation reads: j e2i j=N j. The result is a tremendous increase of the computational speed of Fresnel di raction (FFT) compared to the general Kirchho integral. We give for instance the result of a basic experiment. . The nite step integration which was at the starting point of the algorithm can only converge towards the true FT as N increases.

For instance. if a 1024 1024 2D sample is needed. q. Moreover.2 10?17 In this case. q. FFT provides a factor of roughly 104 gain in CPU time. reducing in principle a N -DFT to two N=2-DFT's. z)g E1(p. since at N =64. grows as N log2(N ).5) f we need the discretization of K (p. We have seen that the DFT has the dichotomic property. All other properties are exactly those described for the DFT. if we intend to use a DFT for computing the di raction integral according to the scheme g f E2(x.146 CHAPTER 3. z) = exp ? i K z F2 F2 x y is: .8 10?12 64 17s 7. N has obviously to be an integer power of 2. z) = ? z 2 z ! z(p2 + q2) f K (p. In this elementary scheme (due to J. q. q. n. where Fx is the x side of the 2D spatial window. it can be seen that increasing the size of the sample is useless.9 10?7 32 14s 1. the ultimate precision of the computer is reached. which is a tremendous improvement with respect to the naive DFT scheme. Tukey 10]). The basic of the FFT is to recursively compute any N -DFT from a series of initial 2-DFT's. y. We have also q = 2 =Fy . it is clear that the number of recursions is log2(N ). q. z) with respect to p. the increment will be p = 2 =Fx. z + z) = KP (p. The Fourier Transform of the paraxial di raction kernel ! i exp ?i k(x2 + y2) K (x. z) = exp ?i 2k Now. y. z) (3. The discretization is therefore of the form: " !# m2 + n2 f(m. so that the number of operations. Cooley and J. in a 2D Fourier Transform. NUMERICAL METHODS Sample size Optimal window rms error 16 10s 6. We remember that the frequency increment in the DFT is f = 1=T where T is the time window. In terms of spatial circular frequencies.

1.3. z).le. If we have spatially squeezed beams.mil) then ind1=i-1 else ind1=i-1-n endif do j=1.j) the complex array (of size n n) representing the amplitude at z = 0. j.sin(phase)) enddo enddo 3.le. but rather write the propagator according to the same convention. Call a(i.2.mil) then ind2=j-1 else ind2=j-1-n endif square=ind1**2+ind2**2 phase=-pi*lambda*z*square/window**2 ktilde(i. Clearly. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS147 If we remember that the DFT has a speci c way of sorting the frequencies. we could think that we have to correctly arrange the FFT of the input eld. For a square computation window of size window the FORTRAN sequence f calculating K (i.n. it is more e cient not to correct the FFT's.iflag) represents any . and propagation steps can be linked into series corresponding to the various interfaces of an optical system.1.3. The propagation step is for instance c------------------------------------------------------------c the subroutine named cfft2d(m.2 FFT-based propagation algorithms A step z of propagation will be carried out following the scheme showed on Fig. one direction can be larger and more sampled than the other). could be: mil=n/2+1 do i=1.j)=cmplx(cos(phase).n if (j. ~ before multiplying by K .ar. The optical amplitudes are sampled on a rectangular grid (it is not necessary to use a square grid.n if (i.

and represented by complex arrays: c------------------------------------------------------------subroutine defmirr(n.window.n a(i.reflect.n : size of the array to be transformed c ar : the array to be transformed. and on return.n.radius.j)=a(i.148 Direct space CHAPTER 3.j)*ktilde(i.mir) c c returns an array of samples of the phase equivalent c of the mirror for given parameters c c n : rank of the arrays c window : computation window c reflect : photometric amplitude reflectivity .1) do i=1.2: propagation step c procedure carrying out the 2D-FT of a complex array c c m.-1) c------------------------------------------------------------- after what a(i.n.j) represents the propagated amplitude.j) enddo enddo call cfft2d(n.a. NUMERICAL METHODS ∆z -1 FFT FFT Multiply by propagator Fourier space Figure 3. Mirrors are sampled on the same grid.curvature.n do j=1.a. the c transformed array c iflag = 1 : direct transform c iflag = -1: inverse transform c-----------------------------------------------------------call cfft2d(n.

n) c data pi/3.j real*8 reflect.pi.j)=ain(i.1.i.j)=0 else phase=2*pi*rp2*curvature/lambda mir(i.n x=(i-1)*dx-window/2 do j=1.3.mir(n.j)*mir(i.rp2 real*8 window.x.curvature. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS149 c c c c c integer n.j) enddo .064d-6/ c dx=window/n do i=1.phase complex*16 ci.141592653589793d0/ data ci/(0.1.lambda.gt.radius*radius) then mir(i.n do j=1.radius.d0)/ data lambda/1.d0.-dsin(phase)) endif enddo enddo return end radius : radius (half size) of the mirror curvature : inverse of curvature radius mir : returned array representing the mirror implicit none A re ection will then be carried out by a simple term to term product of the amplitude by the mirror : do i=1.n aref(i.n y=(j-1)*dx-window/2 rp2=x*x+y*y if (rp2.y.j)=ci*reflect*dcmplx(dcos(phase).dx.

then propagate over the distance L =3 km. with w0 '2 cm.150 enddo CHAPTER 3.3) these dependences. NUMERICAL METHODS The main example we shall study is the case of a Fabry-Perot cavity. We can carry out this experiment by numerical propagation.3. For measuring the distance between two complex amplitude arrays. we use the Hilbert Space metrics: 2 n?1 31=2 s X je ? e j25 d(e1. the re ected wave exactly coincide. we can also propagate a TEM00 wave having such a waist (w0 2 cm) that the curvature radius of its wavefront matches a 3.ij 1.j =0 . e2) = 4 n 2.3 10?15 Now. i. It is remarkable that propagation looks just like a lens (see below) in the Fourier space. We see on (Fig. In the paraxial theory. As an example we can make the following numerical experiment.ij which maybe interpreted as the square root of the total power in the interference of the two waves. with the original. so that di raction and refraction processes are exchanged by the FT. starting from the waist (plane wave). The shown intensity distribution represents the computational noise.3. round trips in the cavity will be implicitly replaced by a direct propagation through a series of thin lenses.4). second by using one step of the FFT scheme. and the cross pattern re ects the square grid used for the discrete sampling in the window. The result depends of the size of the window. Sample size Optimal window rms error 32 34 cm 4. with a sampling of n n over a grid of size s. and compare the image after the round trip with the original. In fact. and on the sampling rate. after the return trip.9 10?5 64 49 cm 1.6 10?8 128 70 cm 4. Start from the TEM00 amplitude at its waist (this situation exists at the corner mirrors of the Virgo cavities. Then make numerically the two wave interfere. rst by using the analytical formula giving the continuous paraxial result. The following table gives an idea of the convergence of the discrete world towards continuous. The di erence between the theoretical propagated mode and the numerically propagated one can be visualized as an interference (see Fig.9 10?15 256 81 cm 8.45 km curvature radius mirror.

40 0. n=50 (squares) n=100 (circles).50 0.60 0.3. n=150 (triangles) .1.70 0. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS151 10-1 10-2 10-3 10-4 10-5 10-6 10-7 10-8 10-9 10-10 10-11 10-12 10-13 10-14 10-15 0.20 0.30 0.80 Figure 3.3: Round trip error vs window size (m).

10 -0.05 0.00 0.20 -0.15 0.00 -0.20 Figure 3. NUMERICAL METHODS 0.15 -0.15 -0.10 0.152 CHAPTER 3.10 0.05 0.4: Interference fringe between exact and numerically propagated TEM00 .20 -0.05 -0.15 0.20 0.05 0.10 -0.

which. x3. to solve eq. for instance given by (2. the linear operator PL is a n2 n2 rank operator. of the mirrors. y) is entering a Fabry-Perot cavity. as can be seen on g.2 are the re ection operators on the two mirrors respectively. By increasing the order of the calculation (the . the initial guess could be the ideal TEM00 mode 00(x. in order to simulate the ne tuning of the cavity.6) (see Fig. y) = < Ein .1. A more accurate study shows that the resonances are slightly di erent from their theoretical values. This is a consequence of the discretization of the eld.(3.1.5 for notation). The inital tuning is assumed to correspond to a TEM00. For instance. y2 . we have to take an input eld not strictly orthogonal to the TEM00. It is possible to bring into evidence the various eignemodes of a parabolic Fabry-perot cavity. The intracavity eld B (x. the re ected amplitude is obtained by Eout = M1 Ein + PL M2 PLB .1. Once the intracavity eld B (x. for large n would lead to invert huge matrices.3. For instance. 00 > 1 ?t1r r 00(x. we see only the resonances of the fundamental (Fig3. In the following numerical experiment we try to scan the di erent resonances of a VIRGO type cavity by adding a varying phase 2 0.6) by successive iterations. y) obeys the following equation: B = t1 Ein ? r1r2 M1 PL M2 PL B (3. this is an implicit linear equation.6.5). M1. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS153 3. It is therefore much more convenient if possible. adding terms in x. If the input eld is a pure TEM00. if we study small geometrical defects of the mirrors surfaces. provided some inital guess of the intracavity eld. and it could be in principle solved by matrix inversion. of the propagator.3. But for a n n sampling grid. Mathematically. y). y) is found. The discrete world has di erent rules.30). The speed of the convergence to the solution depends on the nesse of the cavity. other modes being orthogonal to the input eld are never excited. In order to excite higher order modes. : : : to the phase of the input wave allows resonances of TEMmn up to m + n = 4.3 Finding the eld re ected o a resonant cavity Assume an amplitude Ein (x. with the correct surtension coe cient: Bguess (x. 2 ] to the propagator. y) 1 2 But other choices are possible.

e+00 1.00 1.e-02 0.e-01 1.05 2.5: Fabry-Perot cavity : resonances of the fundamental mode .14 4.28 Figure 3.19 5. NUMERICAL METHODS 1.e+01 1.e+02 1.09 3.24 6.154 CHAPTER 3.

09 2. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS155 3 1.14 Figure 3.n modes.52 1.e+00 1.e+02 1 4 2 5 1.6: Fabry-Perot cavity : excitation of the TEMm.62 3. .e-01 1.1. and are labeled by m + n.05 1.57 2.e+01 1.e-02 0.00 0.3. Dashed lines correspond to the rst theoretical resonances.

it is necessary to achieve a ne tuning of the propagator. p = 1. At the lower order we get . 2. C > 0 0 making clear that the value of which corresponds to resonance is = ?Arg < 0 . this small discrepancy vanishes. 1g (with a unique index) the basis of TEM modes. we get 2 3 X b0 = t1a + ei 4b0 < 0 . and search for the maximum. C 0 > + bp < 0 .7) where A is the input eld.156 CHAPTER 3. It is better to use the following scheme.7 with the fundamental. The brute force method would consist in computing the power in the intracavity eld for various values of the varying phase precedingly introduced. In order to drive the cavity exactly at resonance. If the input eld is the fundamental mode TEM00 with an amplitude a. and if for the sake of simplicity we note f p . the coe cients fbp . C p >5 p>0 p>0 t b0 = 1 ? ei <1 a . NUMERICAL METHODS preceding was carried out on a 64 64 grid). C 0 >] The one way propagator must therefore be corrected by the phase factor ei =2 The phase discrepancy of discrete vs continuous eigenmodes of the cavity is given in the following table. Let us denote by C the cavity operator: C = M 1 PL M2 PL and an arbitrary phase representing the ne tuning of the cavity. The intracavity eld B obeys the implicit equation B = t1 A + ei C B (3. but this would result in a very time costly code. we can write: X B = b0 0 + bp p For a small perturbation. By taking the scalar product of eq. p > 0g are rst order quantities. : : : .3.

Another example corresponds to a misalignment of a mirror. E ei = <jE1j:.e.5 10?15 Rd 128 70 cm -6. the total output eld is Etot = E1 + ei E2 where represents the di erential optical path between the two arms. For instance a corner mirror of one cavity has a pointing error of 10?8 Rd. The structure of the fringe (see Fig. 32 34 cm -7. let < E1 . In the ideal case.1. and in which one spherical mirror has a wrong curvature radius (1% error): After computation. .1 10?16 Rd 3.3. i. we consider a Michelson having the same parameters as Virgo.1.8) is analogous to the intensity pattern of a TEM01 mode. we take a perfect reference cavity and recombine the two re ected elds.jE2 j> 1 E2 As an example. The interference between the two slightly di erently curved wavefronts gives a series of rings of which one is visible in the non zero zone of the globally gaussian intensity. we read that the relative power on the dark fringe is 1:6 10?3 . Denoting by E1 . It is necessary to adjust this phase to obtain the darkest eld. q 2 q q Ptot = P1 ? P2 + 4 P1 P2 sin2 ( ? )=2] so that it is clear that we must take = .4 The Michelson Interferometer It is easy to model a Michelson interferometer having two cavities as arms. the intensity eld having the structure shown on Fig.7. E2 the amplitides re ected by the two cavities. The relative power on the dark fringe is 4:810?8 .3. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS157 Sample size Optimal window phase corr. E2 > = jE1j:jE2j ei it is easy to obtain the following equation for the power .3. the phase is and the resulting eld is zero. for instance is order to study the eld re ected o a cavity having imperfect mirrors. In the general case.4 10?8 Rd 64 49 cm -4.

040 -0.040 x .020 0.020 Figure 3.000 -0.040 -0.158 CHAPTER 3.000 0.7: Michelson interferometer: Dark fringe pattern for 1% curvature radius error on a far mirror -0.020 0. NUMERICAL METHODS 0.020 0.040 y 0.

020 Figure 3.3.8: Michelson interferometer: Dark fringe pattern for 10?8 Rd pointing error on a corner mirror -0.040 -0.000 0.020 0.000 -0.040 y 0.020 0.1.040 x . NUMERICAL PROPAGATION USING FOURIER TRANSFORMS159 0.020 0.040 -0.

It is much more convenient to use one loop.9 for notation) We denote by Ri. We begin by a rst estimate of the recycled eld E and of the two intracavity elds F1 and F2 (see Fig.160 CHAPTER 3. and P2 a propagation from South to West by re ection on the splitter.12 cm 2. as well for the curvature radius than for the roughness pattern. M22 for the West cavity.1. Matched Waist RMS roughness C01077 3584 m 2.5 The power-recycled Michelson interferometer It is possible to model a power recycled interferometer by an external loop.8 nm C02017 3624 m 2.e. the eld in the dark fringe is B given by B = RS P 01R11P1 + TS P 02R21P2] E + RS P 01T11C1 F1 + TS P 02T21C2 F2 For instance. The following table summarizes the main parameters: Mirror # Curv.6 nm The roughness maps are shown on Fig. Then the process is iterated until the hilbertian distance between two successive estimates is small enough.10 and Fig. NUMERICAL METHODS 3. propagation/re ection/propagation).3.15 cm 3. we have taken the maps of two recently produced end mirrors (C01077 and C02017 respectively).11 respectively . and the corresponding M21. Rad. F2) of the internal elds corresponding to the easily computed ideal situation (perfect mirrors).3. and used the preceding algorithm for checking the best mutual attitude of both when installed in a powerrecycled interferometer. starting on an estimate of the recycled eld E then involving two subloops for describing the two FP cavities. Ti respectively the operators associated to the re ection and the transmission of mirror Mi . The 6 mirrors involved are: the recycling mirror MR. At the end. P1 a propagation along the North short arm through the splitter. F1. the corner mirror M11 and the far mirror M12 of the North cavity.3. We start from three estimates (E. The splitter is MS . A such code would be however very time consuming without necessity. then new estimates can be computed according to the following scheme: E new = TRA+RR P1R11P1 + P2R21P2] E old+RRP1T11C1 F1old+RRP2T21C2 F2old F1new = T11P1 E old + R11C1 F1old F2new = T21P2 E old + R21C2 F2old where Ci denotes a round trip in cavity #i (i. The two mirrors are not perfectly identical.

9: Sketch of a power recycled Michelson .3. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS161 M22 F2 M 21 A E F1 D Mr B M 11 M 12 Figure 3.1.

040 Figure 3. NUMERICAL METHODS x 0.10: Roughness in the axial zone of C01077 -0.000 -0.020 -0.060 y 0.040 0.020 -0.162 CHAPTER 3.060 0.060 .060 0.020 0.020 0.040 -0.000 -0.040 0.

11: Roughness in the axial zone of C02017 -0.060 y 0.020 -0. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS163 x 0.060 .020 -0.040 Figure 3.040 -0.000 -0.020 0.040 0.1.020 0.040 0.060 0.3.060 0.000 -0.

The green crosses are obtained by the inverse calculation: map #1 is left unchanged. The results are summarized in the following plot (Fig.164 0.15 0.20 CHAPTER 3.12): The various types of dots correspond to consistence tests. y) plane by angles 1 and 2 respectively. Cyan triangles are obtained by rotating the two mirrors by indentical angles. The two surface maps may be rotated in the (x.05 0.10 0. The fact that the values found for the same angular di erence but di erent o sets are only almost equal is due to the necessary interpolation that causes some uctuations of the mirrors surfaces. and map #2 rotated by opposite angles.00 -180 -150 -120 -90 -60 -30 0 30 60 90 120 150 180 Figure 3.3. and the dark fringe computed as above. A merit gure proportional to the SNR can be evaluated according to the formula s M = Pic1 Pic2 Prec . The red squares are obtained by setting the rotation angle of map #1 to zero and varying the roation angle of map #2. NUMERICAL METHODS 0.12: Power on dark fringe for various mutual angles (see text for comments) It is instructive to run the code with di erent mutual angles.

so that: s F = Pic1 2Pic2 Prec also proportional to the square root of the power stored in the recycling cavity. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS165 where Pic1 (resp. The maximum merit factor is therefore p Mmax = 795:8= 50 = 112:54 W1=2 The quantity plotted on Fig.3. we take the geometric average. for our two cavities: F1.14. It is known ( ?]) that the signal to noise ratio of a recycled Michelson with FP cavities is proportional to the nesse of the two cavities. i.14 we have varied the input waist and computed the corresponding merit factor.2 Prec The two cavities having in general di erent nesses. so that. . The nesse of a cavity can be estimated by P F = 2 Pic in where Pin is the incoming power.8 W.3. Pic2) is the power in cavity #1 (resp. so that the power entering the cavities is Pin = 25 W. 3604 m. this is Prec = 50 W. It seems reasonable to choose a waist such that the curvature radius of the wavefront matches the averaged curvature radius of the two mirrors. In Fig. and the intracavity power Pic = 25W 2 50= = 795.3.3. Its optimum value corresponds to a recycling surtension of 50 and cavity nesses of 50.2 = Pic1. # 2). In Fig.e. the question of the mode to be injected in the interferometer could be raised. This explains the structure of the merit factor. We have here Pin Prec=2. For 1 W laser power. and Prec the power in the recycling cavity.1.13 (red squares) is M=Mmax SNR=SNRmax Because the mirrors have not exactly the same curvature radii.

80 0.75 -180 -90 0 mutual angle [degrees] 90 0.15 0.25 0.10 0. NUMERICAL METHODS 1.166 CHAPTER 3.13: SNR vs mutual angle of end mirrors (red squares).20 SNR/SNRmax 0.90 0. relative power on the dark fringe (green triangles) .85 0.00 0.00 180 Figure 3.05 0.95 0.

1.135 0.00 SNR/SNRmax 0.14: SNR vs waist of input mode .3 Figure 3.9 2.88 1.96 0.2 2.1 waist of the input TEM00 [cm] 2. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS167 1.3.0 2.92 2.

The green disks correspond to the same situation (di erent curvature radii) but with zero residual roughness. the dark fringe has the following pattern (see Fig. mirrors have a nite size (35 cm diameter). The blue crosses correspond to two identical mirrors without roughness. In all cases.15): . The computation grid was a 1m side square.3. 3624m curvature radius. 3604m. giving 256 256 samples. NUMERICAL METHODS the red sqares correspond to the actual SMA mirrors in the nominal reciprocal attitude (marks up). Finally. in the best situation.168 CHAPTER 3. The three dotted vertical lines correspond to values of w0 such that the wavefront has respectively 3854m. of same curvature radius 3604 m.

18 -0.06 0.8e+00 -2.00 0.1.5e+00 -1.06 -0.18 0.12 -0.06 0.12 0.1e+00 1.3.15: darkest fringe: intensity pattern (logarithmic plot) .12 -0.18 -0.12 0.8e-01 1.06 0.5e+00 Figure 3.00 -0. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS169 0.18 -3.

for the rank N and the window F is pmax = N F This is a limitation on the structure of the admissible optical amplitudes: if their variations are of scale shorter than min 2 =pmax = 2F=N . and the algorithm fails.170 CHAPTER 3. Owing to the condition 0 < max. where w0 is the waist (gaussian radius of the spot on the mirror). For a gaussian beam of amplitude h i 2 f (r) = exp ? r2=w0 having a Fourier transform given by h i 2 f~( ) = w exp ? 2w0=4 = p2 + q2. 2w0 and 3w0. We know that the maximum spatial frequency.1. special care must be taken of the angular limitation induced by the nite sampling of optical amplitudes. we can set pmax k max where represents the propagation direction with respect to the main optical axis. We can thus write: N = max = 2F 2D D being the size of the sampling interval. This means that larger divergences are forbidden. we get the condition D < 2w0 with 2 0 3. corresponding to the Shannon frequency. by substituting = k .6 On the intrinsic limitation to basic DFT-based algorithms . aliasing will follow. we get h i 2 f~( ) = w exp ? 2 = g 0 where g = w0 is the gaussian divergence angle of the beam. The Fourier variable p being interpreted as a transverse component of an oblique wave vector. NUMERICAL METHODS The three black circles have diameters respectively w0 = 2. The spectrum becomes negligible for 0 = g where of the order of 3 or 4.135 cm. When chosing the spatial window and the rank of the transform.

the window must be signi cantly larger than w0. Now.3. the intensity distribution were out of. so that we get the new limitation w D > N0 The compatibility of these two conditions requires that 2 N > 2 which is easily met. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS171 On the other hand. Moreover. If a beam is foreseen as very divergent. For intance assume we want to propagate a TEM00 mode of waist w0 = 5 mm. we see that is must be much larger than w1.1. over a distance of L = 3 km (the wavelength is about 1 m). in the initial window. say F > w0. However. we would have serious troubles if after di raction. the wavefront is certainly undersampled. say F 10 w1 2m. The beam width after di raction on the distance L is w1 203 mm. it may be di cult to choose a window and a sampling rate adapted to the situation. If we decide to take a common square computation window at both ends of the path. and the preceding . This implies that the maximum spatial frequency is pmax = 2 N F 2 where N is the sampling rate. q) = 2 w0 exp h?(p2 + q2)w0 =4i so that the maximum frequency can be estimated at about 3. the Fourier transform of the input beam is q 2 2 ~(p.1. we get N 640 which is very demanding in terms of memory and cpu time. or larger than the computation window.7 Propagation with magni cation pmax = 5=w0 = 103 m?1 by comparing with th epreceding expression of pmax.

y.16). After an idea proposed by Sziklas and Siegman ( 13] and ?]). Consider a freely di racting wave which has a beam width w0 at z = 0.3. NUMERICAL METHODS rough estimation based on a pure TEM is still optimistic for a distorted wavefront. and T @x + @y2. in the case of strong focusing or defocusing to remove the convergent or divergent part of the eld.8) x0 = zx y0 = zy (3.172 CHAPTER 3. and a beam width w1 at z = L (see Fig.9) 1 z0 = 2 z1 ? z (3. z) F obeys the following partial di erential equation: " # 2ik (x@ + y@ ) F = 0 2ik@z + T + z x y If now we introduce the new coordinates: (3. z) de ned by h i 1 (x. It is therefore sometimes mandatory to use a modi ed paraxial algorithm based on a function and coordinates transform.10) 0 where and z0 are arbitrary constants. namely take = z0 and z0 + L = w1 z0 w0 . z) is the unknown wave function. z) = z exp ikr2=2z F (x. the paraxial di raction equation is invariant under the combined transformation of function and coordinates. We can exploit this fact. y. y. we may choose the constants and z0 in such a way that the change of coordinates follows the transverse extension of the eld. it is easily seen that the di raction equation becomes (2ik@z0 + 0T ) F = 0 in other words. y. Let us return to the paraxial di raction equation: (2ik@z + T) = 0 2 where (x. consider a new wave function F (x.

y. where w w 0 z1 = w0 L. The procedure for numerical propagation is therefore the following: The initial wave function 0(x. x01 = w0 y w1 1 1 Consequently. One changes of wave function by the formula h i F0 = exp ?ikr2=2z0 0 where z0 L=(w1=w0 ? 1). NUMERICAL PROPAGATION USING FOURIER TRANSFORMS173 this determines z0: the new coordinates are now L z0 = w =w ? 1 1 0 x0 = zz0 x (3. in other words. On computes the magni cation factor w1=w0 for the propagation step (this can be estimated by analogy with a gaussian beam). z) is given. .11) y0 = zz0 y (3.3.12) 2 z0 = z0 ? zz0 (3. y0 = y and the nal plane at z = z0 + L. the initial data may be given in the initial coordinates. and propagates the eld using the propagator # " (p2 + q2) z0 PL = exp ?i 2k where z0 w0L=w1.1. x01 = w0 x. and the propagation step L is xed. the coordinate change is smooth on the initial plane. x00 = x. one chooses the window appropriate for F0.13) so that the initial plane is located at z = z0 where 0 0 z0 = 0.

q) = 1 + ib=z 4(1 + ib=z ) 0 0 we obtain the propagated wave in the Fourier space by f f F1 = P ( z0) F0 where ( z0 Lz0=(z0 + L)): # 2 w0 (p2 + q2) z0 P ( z0) = exp ? i 4b we nd. The Fourier transform is: " 2 # b exp ? w0 (p2 + q2) f F0(p. The initial wave function is: ?r =w 0 (x. NUMERICAL METHODS get the propagated wavefunction F1(x0. y 0) = z0 exp ikr2 =2(z + L) = w0 exp ik r (z0 + L) F (x0. we obtained the corrected wave function " 2 !# ?r =w e?ikr =2z = exp ? r 1 + i b F0(x. y ) = 0(1 + ib=z0)=b z0 + L 1 + i z " . y 0) 1 (x 0 1 2 z0 + L w1 2z0 be aware that the transverse coordinates are now rescaled according to the magni cation factor. y) = e w2 z 2 2 0 2 2 0 2 0 0 0 where b 2 w0 = .174 CHAPTER 3. y ) = e z0 being computed. and return to the true wave function by " 02 # h i 0. after a reciprocal Fourier transform: " 0 # 1 r 2 b(1 + ib=z0) 0. It may be very instructive to examine step by step what happens to a pure TEM00 (an analytic calculation is possible) when treated this way. y0). y 0) = F1(x 2 1 + i z0(1 + ib=z0)=b exp ? w0 b + i z0(1 + ib=z0) the propagated function is thus z0 1 0 0 1 (x .

y0) with R gaussian optics. q) = exp ? i 4 1+ 2 by taking the Fourier transform of F0.3. Let us examine the nature of the corrected waves F0 and F1 used as intermediary data. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS175 ( 0 " #) r2 b(1 + ib=z0) ? i b(z0 + L) exp ? w2 b + i z0(1 + ib=z ) 2 z0 0 0 or as well 1 (x0. multiplying by P and applying a reciprocal Fourier transform. gives the propagated corrected wave function as: ( 02 " p1 + 2 ? 1 #) 0 . consequently.1. It is thus checked that the preceding method gives the same result as the direct calculation giving directly 1 from 0. but the detailed calculation allows to understand how the new algorithm maps a diverging beam onto a collimated one. q z0 = L=( 1 + 2 ? 1) the corrected function is then: ( 2 " #) p r 1+i 1+ 2?1 F0 = exp ? w2 0 the propagator may be written as: " # 2 w0 (p2 + q2) p P (p. This result is not very interesting by itself. which is the classical result (see a preceding section) of L(1+b " 0 # r 2 w2 w2 1 1 = 1 + iL=b exp ? w2 1 0 1 + iL=b 0 " # w0 exp ? i atan(L=b) ? r2 + ik r2 = w w2 2R 0 . y 0) = exp ? r F1(x w2 1 ? i 2 2 0 1 1 2=L2 ). We have rstly ?r =w 0=e we know from the theory of gaussian beams that q q w1 = w0 1 + L2=b2 = w0 1 + 2 ( L=b).

NUMERICAL METHODS where we see that the new wave has the same width w0 and an opposite radius of curvature. ) and set p0 k cos . y + Lq =k) 0 = exp ikL 1 ? 2k2 2 0 0 2 2 0 0 0 2 0 2 0 0 . The diverging behavior of the wave is recovered through the homothetic transformation of coordinates at the end. Let us de ne a corrected eld by 3. Assume the incidence angles are ( . y). The initial wave has been transformed into a collimated wave propagating without magni cation between z0 and z0 + L. it is likely that its Fourier trnasform is peaked at (p0. q0 k sin . Now. we can propagate the corrected eld over a distance L.176 CHAPTER 3. q ? q0) Z = eikLe?ip x?iq y 41 2 dp dq e?iL (p+p ) +(q+q ) ]=2k ~ 1(p. q) " 2 !# p2 + q0 e?ip x?iq y e?ikL (x + Lp =k. y) = e 4 2 dp dq e?iL(p +q )=2k ~ 1(p ? p0. q) = exp ikL ? i 2k (p2 + q2) ~ 1(p ? p0. q ? q0) showing that the FT has been translated in the Fourier plane to reach a central position. although the behavior of the eld is quite reasonable. It is possible to suppress this e ect simply by translating the Fourier transform. q) = ~ 1(p ? p0. y ) e?ip x e?iq y 0 0 the desired e ect follows immediately: ~ F1(p. If the propagated eld is 2 and the corrected propagated eld F2. y) = 1 (x. When the incidence angle is not zero.8 O -axis propagation F1(x. q ? q0) and by taking the reciprocal transform: Z ikL 1 F2(x.1. If the incoming amplitude is denoted by 1(x. the Fourier transform of the incoming amplitude may exceed the limits of the Fourier window. we can write: L ~ F2(p. q0) and possibly out of the Fourier window.

3. we get h i F2(x. the angular direction of the beam is preserved the pure propagation phase is kL(1 ? 2=2) instead of kL this accounts for the removed obliquity that introduces a factor of cos which can be easily corrected if necessary.3.1. so that the eld is expressed in the new cordinates x0 = x + L cos . y + L sin ) 2 where we have put into evidence the following facts: the propagated wave function has been translated. The situation is summarized on Fig. y) = eikL(1? =2) e?ikx cos e?iky sin e?ikL 2(x + L cos . y0 = y + L sin so as to follow the angular direction of the beam. NUMERICAL PROPAGATION USING FOURIER TRANSFORMS177 the factor of exp(?ikL) comes from the fact that our de nition of 2 (the propagated eld) implicitly includes a pure propagation phase of exp(ikL). by returning to the incidence angles. and keep the amplitude map at the center of the window.17 .

17: Corrected o -axis propagation .178 CHAPTER 3.cos θ θ L φ Figure 3. NUMERICAL METHODS w w 1 0 L z 0 z1 Figure 3.16: Beam divergence corrected position φ true position L.

2 Hankel transform methods The preceding numerical methods were designed to study situations having no symmetry.e. involving a Bessel function. we can take bene t of the symmetry and reduce the computational demands by specifying explicitly the symmetry in the calculations.3.1 Theory The 2D Fourier Transform of a function f (x. this results in the Hankel transform. we may assume the optical elements and the optical eld itself having the axial symmetry. owing to what. the integration can be carried out. In the case of the Fourier transform approach. ) f 0 0 If the initial function f is axially symmetrical. In certain cases. we de ne (r. independent of . 3. i. The inverse Fourier transform reads Z1 f~(r) = 21 J0( r) f~( ) d 0 0 . HANKEL TRANSFORM METHODS 179 3. y = r sin p = cos . and we obtain: Z ~( ) = 2 1 J0( r) f (r) r dr f Integrals of this type. are called Hankel transforms. ) and ( .2. y) dx dy R In order to have polar coordinates both in the direct and in the Fourier space. ) = 2 d 1 r dr ei r cos( ? ) f (r. q) = eipxeiqy f (x. This happens namely in the case of thermally induced distortions caused by the beam.2. almost all distortions will keep axial symmetry. y) is Z f~(p. and assume a pure TEM00 mode as the source of heat. ) by: x = r cos . q = sin the transform is now: Z Z ~( . then its transform f~ is also axially symmetrical in the Fourier space. We can neglect the beam's imperfections. In such cases.

i. 1. The orthogonality relation is: Za 2 ' (r) ' (r) r dr = a J02( ) (3. the 2D Fourier transform of f (r) is simply the Hankel transform: Z1 f~( ) = J0( r) f (r) r dr (3.e.16) We are dealing with special physical solutions of the wave equation. amplitudes of nite spatial extension (or almost). There exists a circle of radius a outside of which the amplitude is negligible. = 1. 1 where the .e. such as gaussian waves.14) 2 ?1 In this case. : : : . NUMERICAL METHODS It is cumbersome to keep these factors of 2 throughout all foregoing calculations.15) and its inverse is 0 f (r) = Z1 0 J0( r) f~( ) d (3. 2 which are practically zero for x2 + y2 > 10w0 . 2. the functions ' (r) = J0( r=a). if Db f < bg is the disk in the Fourier space. 1 are the zeros of J1(z). i. it admits a corresponding family of functions: ( ) = J0 ( =b). = 0. and of nite extension in the Fourier space (again gaussian waves). 1 . It is much more convenient to use here a di erent convention in the de nition of the Fourier Transform: Z1 ~(p) = p1 f eipxf (x) dx (3.17) 2 0 Let us note 2 p = a J02( ) 2 Obviously. and there exists a circle of radius b in the Fourier plane outside of which the Fourier transform of the amplitudes is negligible.180 CHAPTER 3. : : : . 2. It is well known that there exists a family of orthogonal functions on the disk Da fr < ag. . = 1.

and Bb the set of functions of negligible outside Db . Owing to the orthogonality relation gives: f0 = f q by substituting in the expansion of f~ we get: =0 0 .3. Such an expansion is called Dini expansion: 1 1 X X f~( ) = f 0 ( ) = f 0 J0( =b) by substituting in (3. HANKEL TRANSFORM METHODS with the notation 2 q = b2 J02( ) we have the orthogonality relation: Zb () () d = q 0 181 (3. that f~ 2 Bb. and expand it on the basis.18) We call Ba the set of all functions of r negligible outside Da.16) we get: 1 X 0 Z1 f (r) = f J0( r) J0( =b) d We can sample the values of f (r) by choosing an elementary distance in the plane. It is convenient to take r = 1=b as the distance element and sample the radii according to =0 0 =0 =0 r = =b so that there is a strong link between the coe cients f 0 introduced in the Dini expansion of f~ and the samples f (r ): 1 X Zb f f (r ) = f 0 J0( =b) J0( =b) d Remark that the transform integral stops to b instead of 1 because we know that the function f~( ) is zero outside Db.15.3.16).2. We can assume in the formulas of the Hankel Transform (3.

The preceding expression. and the spectrum samples will be evaluated at = =a. after sampling becomes: 1 X 1 f J0( =ab) f~ f~( ) = =0 q This is the discrete expression of the Hankel Transform. so that we have the discrete version: 1 X Za f~ f~( ) = f~0 0 J0( r=a)J0( r=a) r dr from the orthogonality relation we obtain: =0 =0 0 =0 =0 .15) of the continuous direct Hankel Transform becomes: Z 1 ~( ) = X f 0 a J0( r) J0( r=a) r dr ~ f Now.19) 0 ) Obviously.182 CHAPTER 3. in the Fourier plane. a linear relation between the vector f~ and the vector f : 1 ~ = X H (+) f f The direct transform is thus represented by the matrix 0( H (+) = 2Jb2J 2( =ab) (3. a similar treatment can be carried out for the inverse transform. The previously introduced function f 2 Da admits a Dini expansion of the form 1 X f (r) = f~0 J0( r=a) so that the expression (3. NUMERICAL METHODS 1 X f J0( =b) f~( ) = q =0 ~ We can now also sample the values of f ( ) by chosing the elementary frequency as = 1=a. we can sample the conjugated variable according to = =a.

20) H (?) = 2Ja(J 2( =ab) 2 0 ) The fact that the studied function f is in the set Ba implies that it takes negligible values for r > a. and we have: 1 X 1~ f (r) = f J0( r=a) =0 p Sampling of the values of f according to r = =b leads to the inverse Hankel Transform: 1 X (?) ~ f = H f with =0 0 (3. HANKEL TRANSFORM METHODS 183 ~ f~0 = f p the Dini expansion of f (r) is now determined. we have = 0.3. The sampling r can therefore stop at r = a. Then we can still freely decide the maximum number N of zeros we shall take into account in the in nite sums encountered in the expressions of the Discrete Hankel Transformn (DHT). then the sampling of must stop at b: =a b ) < ab We can freely decide the size of the computation window a. If the transformed function f~ takes negligible values for > b. : : : . N ab = N )b= N =a . This being done. 2. 1.2. and we have: =b a ) < ab The same result is obviously obtained by considering the Fourier space.

g > = < f~ . g 2 Da . g > = f g =q ~ 0 . f. g(r) is de ned in the direct Hilbert space by Z1 <f . NUMERICAL METHODS r = a =N and the spatial frequency sampling: = N X f~ = H (+) f .21) N J0 ) 0 H (?) = 2Ja(J 2( = )N ) (3. g > ~ it is easily seen that X < f~ . X < f . using the orthogonality. X~ ~ < f . g > = f~ g =p ~ The Hermitian scalar product is invariant by a Fourier Transform.22) 2 0 The Hermitian scalar product of two functions f (r) . then the integral can be stopped at r = a and f . g>= f (r) g(r) r dr if moreover. so that we have as well: < f .184 The radial sampling is thus: CHAPTER 3. g > = f 0 g0 p or. so that. 2. 1. : : : . g may be replaced by their Dini expansions on the ' . N The expressions for the Transforms (FDHT) are nite rank matrix algebra: with H ( ) having the following de nitions: 2 0 H (+) = 2a J2 ( 2( = N ) (3. =0 =a = 0.

all the terms of the recursion can be divided by a common arbitrary renormalization constant. leading to a possible over ow. f > and we have a distance in the Hilbert space. This can be done using a well known algorithm based on the recursion formula for Bessel functions. g) = P (f ? g) If e cient FFT routines are available in all mathematical computer libraries. namely + Jn(z) = 2(n z 1) Jn+1 ? Jn+2 The recursion begins by taking arbitrarily JM = 0 and JM ?1 = 1. M must be chosen su ciently large depending on the argument. Now. An initial guess of being given. and then descending to J1 and nally J0 by the preceding formula. All preceding formulas deal with the two numerical tables of the zeros of J1 (including 0).2. using the well known relation 1 J10 (z) = J0(z) ? z J1(z) this is " # J1( old )=J0( old) new = old 1 ? ? J ( )=J ( ) old 1 old 0 old 3. If during the recursion the Jn 's become to large. the ratio of the two last terms gives the value of J1(z)=J0(z). and J0( ). The power carried by a given amplitude f is now P (f ) = < f . g by q d(f. The calculation . de ned for two functions f.3. a better estimate is found by the Newton formula: J1( old) new = old ? 0 J1( old) which can be iterated until a given accuracy is met. Finally. this is not the same for DHT.2.2 Numerical implementation The problem reduces to the calculation of J1(z)=J0(z). It is easy to obtain the table by the following scheme. HANKEL TRANSFORM METHODS 185 these formula provide the way of computing the scalar product either in the spatial or in the frequency space. this is the reason why we give here the basic ideas for building speci c libraries.

One can easily imagine routines providing at the same time the two families and J0( ). = 0. and H (+) H (?) a projector on Bb. 1. or REAL*8 . by suitably chosing the window a. it is possible to reach the limit accuracy of the computer (' 10?15 ) in double . Practically. except that the last term of the recursion gives J0(z) only after normalization. If we consider a gaussian wave at its waist (w0 = 2 cm) and compare it with its double DHT. the numerical accuracy is the same. but practically. 2 = 7:02 (values already encountered in the di raction problem for a uniform circular aperture) as initial guesses for initializing the Newton re nement process. in which the product DFT?1 DFT is exactly the unity operator. we have N N X X (?) (+) H H f f ' =0 =0 and the corresponding formula in the Fourier space. we obtain the following results: . We conclude that the situation is theoretically di erent from the 2D DFT. because it is not manifest that N X =0 H ( )H ( ) = (?) In fact this is not true. and for N large enough. The question "is the inverse HT actually the algebraic inverse of the direct HT ?" must be considered. with an accuracy depending on the window size a and the rank N . What is true is that the linear operator H (?) H (+) is a projector on Ba . the determination of all f . the initial guess for is ?1 + .186 CHAPTER 3. this means that for a given function f negligible outside Da. N g is done by taking the rst ones from any mathematical handbook: 0 n=1 = 0. : : : . Then for all higher indices. To be more speci c. NUMERICAL METHODS of J0(z) is identical. 1 = 3:83 . the window being correctly chosen. for N in nite. This normalization is done using the well known relation 1 X 1 = J0(z) + 2 J2n (z) Finally. regardless of the rank of the transform or the window size.

8 10?13 50 17 cm 1. q. as seen previously. there is no signi cant numerical discrepancy between the DFT and the DHT The correspondance between an initial eld distribution e0(r) and the eld ez (r) di racted at a distance z is represented by a matrix that can be computed explicitly. z) = exp ? i 2zk p2 + q2 with 2 = p2 + q2. = N X =0 All this can be summarized by the simple linear operation e2. ~ The Fourier Transform of the nal eld is e e2. is expressed in the Fourier variables p . = Where the matrix P is: N X =0 P e1.2 10?7 20 11 cm 1. e1.23) . The paraxial propagator. q. by e G(p.2. and using the sampling = =a. HANKEL TRANSFORM METHODS Sample size Optimal window rms error 10 8 cm 5. ~ H (?) e2. = G ~ And the nal eld itself is e2. (3. we have " # z 2 e G = exp ? i 4 a2 The Fourier tranform of the initial eld is: e1. with N about 50 .3. = ~ N X =0 H (+) e1.0 10?15 187 Thus.

Sample size Optimal window rms error 10 13 cm 6. we do the same numerical experiment. The sampling has been represented by small spots. and we propagate a normalised TEM00 from its waist (w0 = 2cm) over a distance z = 3km. the blue to the di racted one.5 10?13 100 40 cm 1. We compare in the following table the numerically propagated wave with the theoretical.3. The red pro le corresponds to the initial gaussian wave. NUMERICAL METHODS P = N X =0 e H (?) G H (+) (3.18).188 CHAPTER 3. .3 10?3 20 18 cm 3.1 10?5 50 29 cm 9.8 10?15 The intensity of the eld can be represented on a radial plot (see Fig.24) In order to compare with the 2D DFT.

dots: HT samples.10 0.30 Figure 3. window = 30 cm. HANKEL TRANSFORM METHODS 189 4 2 0 log[Intensity] -2 -4 -6 -8 -10 0. N =50 samples.05 0.15 Radius [m] 0.00 0. solid line: di raction theory .25 0.20 0.18: Di raction of a gaussian wave.2.3.

190 CHAPTER 3. NUMERICAL METHODS The di erence between a numerically propagated mode and its exact value is represented on Fig. the same experiment as reported on the previous Fig.19.3.4: .3.

39 Figure 3. HANKEL TRANSFORM METHODS 191 10-24 10-25 10-26 10-27 10-28 10-29 0.2.3.19: Interference fringe between a numerically propagated TEM00 and its exact value .00 0.

we see from the preceding subsection that C is an explicitly known matrix. Recall that if the input amplitude is A. the intracavity eld B obeys the implicit equation B = t1 A + r1r2 ei M1 P M2 P ] B where the phase determines the tuning of the cavity. The function f (r) represents possible geometrical defects of the mirror. Instead of a series of iterations.192 CHAPTER 3. but the matrix inversion may become problematic too. The solution of the preceding equation can therefore be found as h i?1 B = I ? r1r2ei C t1A where I is the identity matrix. and we check the relative accuracy on the solution of the implicit cavity equation. It follows that to any optical system involving distances and mirrors (or thin lenses) can be associated a single matrix which represents explicitly the optical transfer function. representing the e ect of a round trip. If the nesse is high. = M Ain. Finding the eld stored in a resonant cavity is especially convenient in the DHT scheme because it can be done by one matrix inversion. Calling C = M1 P M2 P the cavity operator.e: = k B ? t1A + r1r2ei C k . A mirror will be de ned by the diagonal operator " # a 2 + i k f (a = ) M = i rP exp i k 2R 2 N c N Aref . this is a huge bene t. due to the very small diagonal elements. the iteration scheme in the FFT method converges very slowly. the mirror surfaces are sampled along a radius. and P is the di raction operator described above. as in the DFT scheme. i. tuned on the fundamental mode. An idea of the accuracy of the algorithm can be drawn from the following experiment: We consider a resonant cavity of nesse 50. and Rc the curvature radius. Action on an optical amplitude. we may now solve the problem by a single N N matrix inversion. NUMERICAL METHODS where rP is the photometric re ection coe cient. giving Aref from Ain is: The optical system being assumed axially symmetrical. An example is treated below. In case of moderate nesse (r1r2 not too close to 1).

0 10?15 193 For the re ected eld.3.7 10?12 100 39 cm 3.2. HANKEL TRANSFORM METHODS Sample size Optimal window rms error 20 17 cm 4. we have Aref = M1y A + t1 B so that the re ection o the cavity reduces to a matrix product: h i?1 Aref = M1y + t1 I ? r1r2ei C t1 A .8 10?5 50 29 cm 1.

we take the normalized HGm.n functions as the basis. y). : : : . This suggests that in case of very small perturbations of the system. y) = Emn HGz. as will be seen.m.194 CHAPTER 3. 1. The ideal eld of application of modal expansion is the simulation of the small motions of mirrors in any degree of freedom.3 Modal expansion The principle of a modal expansion is to expand the optical amplitudes on a discrete basis of functions having a known behavior in di racting. there is a basis of functions z.3. In fact. will be kept for other perturbations having the axial symmetry.m. the actual amplitudes could be described with a small number of HG or LG functions. even if this increases the complexity. the propagation operator has a diagonal matrix representation. in a perfect system where all mirrors are matched. A perfect beam in a perfect cavity would be precisely a HG00 or a LG00 as well. NUMERICAL METHODS 3. the propagation problem completely decouples in independent scalar equations. saving much computational power. y) m.n This kind of representation has the key advantage that. 1g is a complete set. Examples of such functions have been already presented as the Hermite-Gauss (HG) functions and the Laguerre-Gauss (LG) functions respectively. and consequently. linear coupling of these modes are caused by perturbations of the optical elements. In all what follows.n (x. one for each mode. including rotations. n = 0. of the mirrors.1 Return to the HG family of modes The set of Hermite-Gauss functions fHGmn (x. 3. Thus any optical amplitude admits a unique expansion of the type X E (x. the HG functions are highly recommended. dealing with its eigenfunctions. m. This means that at all optical element. A general numerical approach of small perturbations must involve small displacements. In this case. y ) = cmn Hm 2x=w(z) Hn 2y=w(z) " 2 2# " 2 2 # y exp ? xw(+)y exp i (xR+ ) ) 2 z (z p p .n (x.

curved mirrors can be considered as well as spherical or parabolic. MODAL EXPANSION 195 having a w(z) parameter equal to what di raction imposes. and all these Hilbert spaces are connected by the coordinate z.3. y)) dx dy Hk 2 w Hl 2 w exp ?2 w2 The total curvature phase factor vanishes. For instance.n (x.m. y) represents consequently only the departure of the surface with respect to the ideal paraboloid. because if we represent the incoming wave's by ei .l where r is the photometric coe cient. and eventually. y ).in 0 1 k. we have the basis w . and where the coe cients Rmn. the matched mirror is e?2i .n(x.m.kl Ekl. at a distance z = L. the e ect of such a mirror on an incoming wave Emn. and at q end mirror.out = ir Rmn. the basis the w . disappears. In general.3. The function f (x. Then we take the scalar product of this wave with the inversely directed wave which contains e?i . we have one Hilbert space per element.kl are given by Z p y p x Rmn. at the waist. where w1 = w0 1 + L2 =b2 The optical elements we shall consider as examples are weak curvature matched mirrors having defects or wrong location. and the re ected wave e?i . We take the opportunity to remark that in kilometric FP cavities.kl = cmnckl 2 Hm 2 w Hn 2 w R ! p x p y x2 + y2 exp (2ikf (x. in will be represented by a matrix operation X Emn. The normalisation constant is: s cmn = w2z)2 2m+n1m! n! ( In other words. The apex equation of a sphere of radius Rc osculating the plane z = 0 is: q z = Rc ? R2 ? x2 ? y2 c The expansion for large Rc gives x2 + y2 ? (x2 + y2)2 + O(a6=R5 ) z = 2R2 c 8R3 c c . y).

25) n k l p p exp(?X 2) exp(?Y 2) exp 2ikf (wX= 2. Z frame is rotated by an azimutal angle and a colatitude angle from the reference frame x. NUMERICAL METHODS (a being the radius of the mirror). so that there is no signi cant di erence between the paraboloid and the sphere.45 km. in the X. is. The rst term is the parabolic approximation. a = 17. Obviously we consider small rotation angles. Y. whose solution is 2 2 u2 2 z = 2x + y ? u tan + 2R sin + O(1=R2 ) c Rc cos c cos where we have set u = x cos + y sin 3.2 Tilted mirrors . and the second is bounded by 4 zmax = 8a 3 Rc For typical values. The general apex equation of a parabolic mirror. z.5 cm.3. We need the apex equation in that reference frame. y. we get a second order equation in z. wY= 2) dx dy The rst case we shall examine is the rotation of a mirror. we have zmax ' 3: 10?15 m. as already said. Z coordinates 2 2 Z = X 2+ Y Rc Suppose that the X. The scalar product can be written w2 c c Z H (X ) H (Y ) H (X ) H (Y ) Rmn. and Rc = 3. Y.kl = 2 mn kl R2 m (3.196 CHAPTER 3. We have X = x cos cos + y cos sin ? z sin Y = ? x sin + y cos Z = x sin cos + y sin sin + z cos By substituting in the apex equation.

It is easy to compute the integral 3.3. the third kept term is less than 10?14 m. It means that the function f (x. We can write indeed: Z1 Imk (p) = e?p =4 ?1 e?(X ?ip=2) Hm (X ) Hk (X ) dX 2 2 p ?1 p By considering a closed loop in the complex plane and using the Cauchy theorem. MODAL EXPANSION 197 Consider now orders of magnitude. Now.45 km curvature we see that the neglected terms were less than 10?21 m. y) introduced above is simply f (x. In other words. If the waves interfere. If not. it means that the rotation angles allow some light o a mirror to reach the opposite one. with a very good accuracy. we have to calculate integrals of the form Z1 Imk (p) = Hm(X ) Hk (X ) e?X eipX dx (3. x and y being at most of order w. it can be immediately seen that Z1 Imk (p) = e?p =4 ?1 e?X Hm (X + ip=2) Hk (X + ip=2) dX 2 2 now. the colatitude angle is less than max = a=L where a is the radius of the mirror. It is therefore possible to take. This is less than 10?4 Rd. with the translation formula this is Imk (p) = e?p =4 2 m k XX s=0 t=0 s Cm Ckt (ip)m?s+k?t Z1 ?1 e?X Hs(X ) Ht (X ) dX 2 . A sophisticated numerical model is useful if the mirrors are aligned enough to allow some interference in the system. With 3.26) 2 with either p = ? 2kw tan cos or p = ? 2kw tan sin . y) = ? tan (x cos + y sin ) In order to compute the rotation matrix. geometrical optics models are quite su cient to describe what happens to the light.45. neglecting lengths of order w2 2=Rc : 2 2 z = x 2+ y ? tan (x cos + y sin ) Rc as the rotated-mirror apex equation.26 by using the translation formula 2.3.

29) so that our result can be expressed as: Imk (p) = p im+k e?p =4 Q (p) mk 2 The rotation matrix takes on the form Qmk ? 2kw tan cos Qnl ? 2kw tan sin e?k w tan =2 Some more details on displacement polynomials will be discussed in a following section.28) (3. nally X p im+k e?p =4 min(m. Remark that kw = w 2 w 2 2 2 p m+n+k+l Rmn.3.kl ( .k)(?2)s m! k! m+n?2k Imk (p) = s! (m ? s)! (k ? s)! p 2 s=0 the same result can be found by using formula 2. NUMERICAL METHODS and with the orthogonality relation of the Hermite polynomials.k) X s=0 (?2)s s! (m ?m!)!k(!k ? s)! xm+k?2s s (3.3 Parallel translations of the beam mn Consider now a parallel displacement of the beam in a region of null curvature.47 and 2. this gives m k X X s t m?s+k?t p s 2 s! st Imk (p) = e?p =4 Cm Ck (ip) 2 s=0 t=0 or.198 CHAPTER 3. ) = p m+in+k+l 2 m!n!k!l! p where g = = w0 is the divergence of the gaussian beam. It is thus natural to introduce the displacement polynomials Qmk (x) = min(m.49.27) (3. g 0 3. y) . The incoming beam was X Ein = Emn mn(x.

MODAL EXPANSION and the translation operator T acts as 199 T:Ein = Ein(x + x.30) .kl = 2 mn kl mk nl 4 then k+l p m+(?k1)l = n+ + Tmn.3. A change of variables leads to " # w2 c c Z 1 dX Z 1 dY exp ? (X + X=2)2 + (Y + Y=2)2 = 2 mn kl 2 ?1 ?1 2 2 exp ? (X ? X=2) + (Y ? Y=2) 2 p " # Hm (X ? X=2) Hn (Y ? Y=2) Hk (X + X=2)Hl (Y + Y=2) By using again the translation formula. we get successively " # w2 c c (?1)k+l Q ( X ) Q ( Y ) exp ? X 2 + Y 2 Tmn.kl 2 m!n!k!l! " # x2 + y 2 exp ? 2w2 Qmk ( 2 x=w) Qnl( 2 y=w) p p (3. y + y) The matrix elements of the operator are: 2 Tmn.kl 2 x=w.3.kl = w cmn ckl 2 Z1 ?1 dX Z1 1 # " # (X + X )2 + (Y + Y )2 exp ? X 2 + Y 2 dY exp ? 2 2 Hm (X ) Hn (Y ) Hk (X + X )Hl (Y + Y ) " with X Tmn.

NUMERICAL METHODS When the wavefront incoming on a mirror of curvature radius Rc has curvature radius R. we get Z1 Mmnkl = cmn ckl e?2r =w eikr =Re?ikr =Rc 2 2 2 2 Hm ( 2x=w) Hn ( 2y=w) Hk ( 2x=w) Hl ( 2y=w) dx dy or where we have set 2 Mmnkl = w cmn ckl Im.4 Mismatching 'mn (x.k( ) = p m k 2 ?1 First of all.l( ) 2 p p ?1 p p w2 1 ? 1 R0 R and introduced the following integrals: 1 Z 1 e?(1+i )x H (x) H (x) dx Im. it is clear that the Im.47). By using the reduction formula (2. Assume for instance the reference beam is a TEMmn HG-mode: 3.200 CHAPTER 3. y) = cmn eikr =2R e?r =w Hm ( 2x=w) Hn ( 2y=w) 2 2 2 p p where the cmn are the normalization constants. M'mni where 'kl is the phase-conjugate of 'kl. y) = e?ikr =Rc 2 so that the mirror's matrix elements being de ned as Mmnkl = h'kl . we get for instance H2m (x)H2k (x) = minX 2k] 2m. if R 6= Rc there is mismatching.k ( ) In. This may happen for instance if the mirror has been displaced along the optical axis. s=0 2m!2k!2s (2m ? s)!(2k ? s)!s! H2m+2k?2s (x) .k are non zero only if m and k have the same parity. The mirror operator is: M (x.3.

3.k (x) + 2(2k ? 1) m?1.46) and the Cauchy theorem on a trivial path in the complex plane that.k (x) + 4k (x + 1) m. for n 0: 2n! xn 0. s=0 2m!2k!2s(2m + 2k ? 2s)! xm+k?s (2m ? s)!(2k ? s)!(m + k ? s)!s! mk (1=Z The same calculation can be done for the odd-odd integral: I2m?1.k (x) = 4(m ? 1)x m?1. and mk (x) mk (x) mk (1=Z ? 1) is a even-matching polynomial de ned as: = minX 2k] 2m. It is better to know the rst orders and nd higher orders by recurrence. The well-known recurrence formula for Hermite polynomials induces the following recurrence scheme: m.2k = Z ?1=2 where Z 1 + i . one can show from the scaling formula (2.n (x) = n! .k (x) = 2(2m ? 1)x 2 m?1. but are not optimal for an e cient numerical computation.2k?1] (2m ? 1)!(2k ? 1)!2s (2m + 2k ? 2s ? 2)! xm+k?s?1 mk (x) = 1+x (2m ? 1 ? s)!(2k ? 1 ? s)!(m + k ? s ? 1)!s! s=0 the factor 1=(1 + x) comes from the fact that the preceding sum happens to be divisible by 1 + x.3.k (x) m. MODAL EXPANSION 201 moreover. for any complex number Z of positive real part: Z n e?Zx H2n(x) dx = 2n!! Z ?1=2(1=Z ? 1)n R we have thus the following result: 2 I2m. These two explicit formulas allow in principle to compute any matrix element.2k?1 = Z ?3=2 ? 1) which de nes the odd-matching polynomial X 1 min 2m?1.k?1 This crossed recurrence scheme can be initiated for instance from the following explicitly known matching polynomials.

2k.2k?1. and M2m?1.n (x) = 2x n! h n?3 2n! (2n + 3)(2n + 1)x4 + 8(n ? 1)(2n + 1)x3 + 4(n ? 1)(6n ? 7)x2 + 3.n (x) = 4x n! +16n(n ? 1)x + 4n(n ? 1)] and for n 1: 2n! xn?1 1.2n.2n?1.2k?1.202 the symmetry breaking between the two indices is only apparent.2n.2k.2n?1.2l?1( ) = (1 +1i )2 q 2m+2n+2k+2l?2 (z) nl(z) 2 2m!(2n ? 1)!2k!(2l ? 1)! M2m?1. The matrix elements have the following expressions: z 1 M2m.01( ) = (1 +1i )2 1 M02.02( ) = 1 + i (1 + 2z + 3z2=2) .n = 4x n! +16(n ? 1)(n ? 2)x + 4(n ? 1)(n ? 2)] mk M2m. NUMERICAL METHODS h i n?1 2n! (2n + 1)x2 + 4nx + 2n 1.2l?1( ) = (1 +1i )3 q 2m+2n+2k+2l?4 mk (z) nl (z) 2 (2m ? 1)!(2n ? 1)!(2k ? 1)!(2l ? 1)! Some examples for the rst orders of self-coupling: 1 M00.n (x) = 2x n! 2n! h(2n + 3)(2n + 1)x4 + 8n(2n + 1)x3 + 12n(2n ? 1)x2+ n?2 2.n (x) = n! h i n?2 2n! (2n + 1)x2 + 4(n ? 1)x + 2(n ? 1) 2.2l( ) = (1 +1i )2 q 2m+2n+2k+2l?mk (z) nl (z) 2 (2m ? 1)!2n!(2k ? 1)!2l! 2 CHAPTER 3.00( ) = 1 + i M01.2l( ) = 1 + i p 2m+2nmkk(+2)l nl (z) +2 2 2m!2n!2k!2l! where z ?i =(1 + i ).

MODAL EXPANSION M11.5 times the half-width of the beam can be considered almost inde nite. When a ! 1.l i M being the mirror operator.n.n.5 Clipped mirrors Real mirrors are of nite size.n .3.2m.2n( ) = 1 + i 2m2mm!n!! zm+n +n n for coupling of 01 to higher orders (n 1): !2 p M01. The nite sizing of an otherwise perfectly matched mirror can be considered as a perturbation.3.l = h m. the matrix elements have the form ?m. in practice. and we address the question of computing the matrix elements of that perturbation in the Hermite-Gauss basis. strictly speaking.l = cmnckl d r dr e?2r =w Hm ( 2r cos =w) Hn ( 2r sin =w) 2 2 0 Hk ( 2r cos =w) Hl ( 2r sin =w) where a is the nite radius of the mirror. but it is so sharply peaked that.2n?1( ) = (1 +1i )2 2m2mm!n!! 2n zm+n?1 +n n 203 p p 3.k. due to the orthogonality of the basis. After a change of variables.11( ) = (1 +1i )3 for coupling of 00 to higher orders: 1 !2 M00. we expect the bracket to vanish unless m = k.k. y) the elements of the basis. The ideal TEM00 beam is. of inde nite transversal extension.3. which exactly means that "one can neglect the di raction losses".2m. the integral becomes Z ?m. Due to the perfect matching of the mirror curvature radius. M k.k. this scalar product reduces to Z2 Za p p ?m.l = c0mn c0kl 2 e?R R dR 2 p 0 p 0 . If we denote by m. n = l.n(x.n. mirrors having radii about 2. It is interesting to quantify these losses in terms of power coupled in higher order modes.

2l+1. ?2m+1.2k. l]( ) = 2 e?R R dR Hm (R cos ) Hn (R sin ) Hk (R cos ) Hl(R sin ) 2 204 CHAPTER 3.2n. n. NUMERICAL METHODS 0 for the radial integral.2n+1.2k+1.2l.2k+1.2n+1.1 Z 2 d H (R cos ) H (R sin ) H (R cos ) H (R sin ) m n k l 2 0 p where 2a=w.32) It is however clear that averaging over angles will eliminate a number of terms.2k.2n. k. k.l = hmnjM jkli = c0mn c0kl ( m. we take the abbreviation 1 Z 2 d f( ) (f ) 2 0 so that the matrix element is ?m.31) Hm (R sin ) = m=2] X s=0 (3. 2m! ! (cos2n sin2m ) = 22m+2nm! n2nm + n)! !( so that we see immediately that the only nonzero elements are: ?2m. ?2m. ?2m+1.k.2l+1 (3.n. n. For the angular average. l]) From the de nition of the Hermite polynomials we get immediately: Hm (R cos ) = and obviously: m=2] X s=0 (?1)s s! (mm! 2s)! (2R)m?2s (cos )m?2s ? (?1)s s! (mm! 2s)! (2R)m?2s (sin )m?2s ? (3. we have in particular: (cos2n+1 ) = (sin2n+1 ) = 0 moreover.33) . The new normalization constants are c0mn = p m+1 2 n m! n! It is convenient to take a shorthand notation Z m.2l.

it is possible to show that.31. for k + l 6= 0 (the special case k + l = 0 is known): ! l ( 0. k. 0. n ? 1. k.3. n. the following: m. MODAL EXPANSION 205 It is straightforward to derive from the recurrence relation obeyed by the Hermite polynomials. k. The point is that a whole family of elements can be computed explicitly. n. 0. l] = m ? 1. We rstly have: " Z n 2s # ?R R2n+1 dR = n! 1 ? e? X 2 e 0 s=0 s! 2 2 so that in particular. l] = m.3.36) then. n. l +1]+2l m. n.35) allowing to compute any matrix element. n. 0. l] (3. k. once some initial elements are known. k. k. k +1.32). 0]) = 1 ? e? 2 (3. we have simply: ( 0.33) and some algebra. l] ? 2(m ? 1) m ? 2. l] (3. 2k.37) where the Cm (x) are the clipping polynomials de ned by m X + 1)! s Cm(x) = (?1)s Cm ((m+ 1)! xs s s=0 The rst clipping polynomials are as follows: (3. n ? 2. l]+2k m ? 1. l ? 1] ? 2(n ? 1) m. formula (3.38) C0(x) = 1 C1(x) = 2 ? x C2(x) = 6 ? 6x + x2 C3(x) = 24 ? 36x + 12x2 ? x3 C4(x) = 120 ? 240x + 120x2 ? 20x3 + x4 .3. 2l]) = (?1)k+l k! l2kk2+! l)! 2e? Ck+l?1 ( 2) !( 2 (3. by using the de nitions (3. k ? 1. n ? 1.34) m.

l + m?1.n?2.0.k. 1.2k. 2k.34): CHAPTER 3.0.n.l ? m m m m?2. 2l?1]) = (?1)k+l k! l2kk2+! l)! 2e? Ck+l?1 ( 2) ? (k + l) Ck+l?2( 2) !( (3.l s s s k+1? k? m?1 ? ?m.0.k. 2k. allow to recursively compute any matrix element. 1.n?1.l 2 The clipping polynomials are orthogonal in the following sense: Z1 e?xCm(x) Cn(x) x dx = mnn! (n + 1)! 0 Result (3.k.l = m.n.n.n?1.k.0. 2k. 2l]) (k + l 6= 0) or p (?1)k+l 2k! 2l! 2 e? C ( 2) (k + l 6= 0) ?0.2l ( 0.k.l?1 ? n n n m. 0. added to the recurrence relations and to the symmetry in the pairs (m.n.2l = 2k+l k! l!(k + l)! k+l?1 2 2 .2k.0 = 1 ? e? ?0. We have namely: s s s l+1? l? n?1? ?m. l).k?1.39) (1) One can introduce the family Cm (x) Cm(x) ? (m +1)Cm?1(x). k) and (n.k.k+1.l+1 + m. 2l ? 1]) = (?1)k+l k! l2kk2+! l)! 2e? Ck+l?1 ( 2) !( These two results. with (3. The complete expression of the coupling coe cient is (c000 = 1): ?0.2l = c02k.l = m?1.37) allows already to study the coupling of the TEM00 with higher order modes. NUMERICAL METHODS 2 h i ! l ( 0. and show that x d (1) Cm (x) = m dx Cm(x) (m > 1) or m?1 (m + 1)! s (1)(x) = ?x X (?1)s C s Cm m?1 (s + 2)! x s=0 so that ! l (1) ( 0.n.206 Moreover.

0.0. We see that for small values of a.4.6 = p5 2 e? (1 ? 3 2=2 + 4=2 ? 6=24) 4 2 ?0.0.0.1.0. we need the coupling factors for the TEM01 mode (k > 0.6 = ? 16 2 e? (1 ? 2 2 + 4 ? 6=6 + 8=120) In order to initiate any recursive scheme.20.8 = p 2 e? (1 ? 3 2=2 + 4=2 ? 6=24) 8 2 ?0. due to orthogonality of the modes. MODAL EXPANSION 207 (recall that 2 2a2=w2). there is a huge loss of power.2.2. then it increases and nally returns to zero when a is large.2.4 = 3 2 e? (1 ? 3 2=2 + 4=2 ? 6=24) p8 30 ?0.2 = 1 ? e? (1 + 2=2 ? 3 4=2) ?1.0. one can see the dependence on a of the coupling factor (power).0.4. and thus the coupling is weak.1 = 1 ? e? (1 + 2) ?0.0. In Fig.0. we get for instance: ?0.3.0.3.0.2k.1 = 1 ? e? (1 + 2 + 4=2) 2 2 2 2 2 2 2 2 2 2 2 2 2 .4 = p3 2 e? (1 ? 2=2) 2 p 2 ?0.3.1.2 = ? 2 2 e? p 2 ?0.2l?1 = 2k+l?1 k! (l ? 1)!(k + l)! e Ck+l?1( 2) ? (k + l)Ck+l?2( 2) And for autocoupling of higher order modes.2 = 1 2 e? (1 ? 2=2) p2 ?0.1.0. l > 1): q i (?1)k+l 2k! (2l ? 1)! 2 ? h ?0.0.4 = ? 43 2 e? (1 ? 2 + 4=6) p ?0.2. The explicit values for the rst matrix elements expressing the coupling of the TEM00 are: p ?0.1.0.6 = ? 45 2 e? (1 ? 2 + 4=6) p 35 ?0.0.

NUMERICAL METHODS 0.01 0.03 0.25 0.2) 0.50 0.00 (2.00 0.02 (2. radius of the diaphragm .05 (2.4) (4.75 2.20: Relative power coupled from the TEM00 into the rst higher order modes vs.208 CHAPTER 3.07 0.0) 0.50 1.75 1.25 1.00 Figure 3.00 a/w 1.04 0.4) 0.06 Coupling rate of the TEM00 0.

4) (2.3.3.0) 10-12 10-13 10-14 2.3 2.4 2.9 3. radius of the diaphragm For realistic values of a=w.2) (2.5 a/w 2. MODAL EXPANSION 10-4 10-5 209 Coupling rate of the TEM00 10-6 10-7 10-8 10-9 10-10 10 -11 (4.0 Figure 3.0 2.2 2.21: Relative power coupled from the TEM00 into the rst higher order modes vs. a logarithmic scale is preferrable (Fig.21): .7 2.1 2.3.8 2.4) (2.6 2.

0. n s XX 2 P (n) = ?0. we show the case of a small diaphragm aperture: the convergence is very slow. we have the cases of realistic values of w=a. because the intense perturbation spread the incident power among almost more all the modes. In Fig. and we have plotted 1 ? P (n).2s?2k. NUMERICAL METHODS It may be also interesting to see how the incident power is ditributed in the coupled modes by seeing the cumulated coupling coe cients. .210 CHAPTER 3.2k s=0 k=0 In (Fig.3.3.23.22).

00 Relative cumulated power in (2k.92 0.94 0.82 0. maximum order . MODAL EXPANSION 211 1.90 0.80 0 5 10 (k+l)max 15 20 25 a/w = 1 a/w =0.96 0.2 0.3.86 0.98 0.2l) modes a/w = 1.84 0.9 Figure 3.88 0.22: Cumulated relative power coupled from the TEM00 into the rst modes vs.3.

minimum order . NUMERICAL METHODS 10-3 Relative power in higher order modes 10-4 a/w = 2 10-5 10-6 a/w = 2.23: Residual relative power coupled from the TEM00 into the higher order modes vs.5 10-7 10-8 a/w = 3 10-9 0 5 10 (k+l)min 15 20 25 Figure 3.212 CHAPTER 3.

2e?2 2 =w2 2 2 2 2 2 2 2 2 2 2 2 Z 0 p 0 2a=w p 2 2 = 1 ? e?2a =w ? 2a2 2 2 e?2a =w + O( 4=w4) w w If the o set is a random process (t). the recycling mirror having a radius of order 5 cm. one may worry about a possible angular displacement of the beam making it interacting with the edges. For instance. The preceding integral can be numerically computed for arbitrary . it is then clipped by a mirror. MODAL EXPANSION 213 We consider the case where the beam being translated by from the optical axis. We assume without loss of generality a displacement along the x direction. we have a=w 2:5.3. The power transmission factor is: " # 2 Z exp ?2 (x ? )2 + y2 dx dy ? = w2 w2 C (a) where C (a) is the area delimited by the circle of radius a representing the mirror's edge.6 O set and clipping or ? = d e? I0(2 2 =w) 0 where I0 is the modi ed Bessel function of the 1st kind. This is p 1 e?2 =w Z 2 d Z 2a=w d e? +2 cos ? = 2 2 2 3.3. we see that it induces a power noise: P (t) = 2 (t)2 P0 w2 with the scaling factor 2 = 2a2 e?2a =w w For a recycling mirror of radius 5 cm. but if we assume a displacement small compared to w we can replace I0 by its Taylor expansion up to 2d order: I0(z) 1 + z2=4 so that "Z p2a=w # 2 Z p2a=w ? + 2 3 d e? ?2 =w ? = 2e d e w2 0 0 " # 2 2 1 ? (1 + 2a2=w2) e?2a =w = e?2 =w 1 ? e?2a =w + w2 .3. so that 5 10?5 2 2 2 2 2 2 .

214 CHAPTER 3. where the m. as m. in the special case p = q = 0. we get s s s 2 2 1 ?2m. We assume the waist precisely on the input at mirror.48 giving a Fourier transform.n = 2 2 2m+n m! n! w1 w2 " 2 Z 2 !# r 1 + w2 H (p2 x=w ) H (p2 y=w ) dx dy exp ? w2 2 n 2 m w2 2 1 " 2 w2 ?1 2 =w2 (1 + w2=w2 )2 1 + w2 1 2 1 so that.n i are the eigen HG functions: s s " 2 2# p 2 1 x + y H ( 2 x=w ) H (p2 y=w ) m 2 n 2 m.2n = 2m2n m!n! w2 + w2 w2 + w2 1 2 1 2 The arithmetic factor (under the square root) has a very low decreasing rate.n (x. y ) = 2 2m+n m! n! exp ? w2 w 2 2 we are in the case of formula 2. which yields a non zero result only for even orders both in m and in n. 1 #m+n 2 m n 4 4 (1 ? w2 =w1 )m+n (?1)m 2m!! (?1)n 2n!! " 2 2# 2m! 2n! 2w1w2 w1 ? w2 m+n ?2m.2n = 2 2 2m+n m! n! w w The scalar product is thus reducible to s s s 2 2 1 ?m. This is done by computing the matrix elements: ?m. For even orders. the coupling coe cients are very small and very slowly decreasing with the order (as could p .7 Mismatched beams The rst case of mismatching occurs when for instance a beam of the TEM00 type of waist w1 enters a cavity having a system of eigenmodes of waist w2. We have thus to expand the incoming beam on the basis of the HG functions of parameter w2. NUMERICAL METHODS 3. at the end. so that if w1 and w2 are very di erent.n m. n grow.n = h .3.

2nj2 = 1 Now.3. In fact: X (1 ? x2)?1=2 = 222kk!!2 x2k (jxj < 1) k k so that. p 3.0 = 1. R2) is simply: !2 ?2m. if the input amplitude and the TEMmn basis are not taken at the waist.n 2 1 2 m 1 2 X and thus X m. for the arithmetic factors can be recognized as those in the Taylor expansion of (1 ? x2)?1=2.3.8 Coupling of astigmatic beams 1 2 1 Consider an astigmatic normalized optical amplitude of the following type: s " # 2 exp ? x2 ? y2 A(x.3. After di raction over a distance L.n j?2m. y) = W W W2 W2 R R 1 2 1 2 1 2 . y) = w w w2 w2 2 The wavefront is assumed at at z = 0. Note that w1 = w2 yields ?2m. " 2 2 #2m+2n 2X " 2 2 # 32 2 2 2m! 2n! w1 ? w2 2m! w1 ? w2 2m5 = (w1 + w2 )2 = 4 22m m!2 w2 + w2 2 2 2m+2n m!2 n!2 w2 + w2 4w1 w2 m. up to uniform phases (propagation+Gouy): s " # 2 exp ? x2 ? y2 exp ?i x2 ? i y2 B (x. R1) with the TEMmn mode of parameters (w2.2n = 2m22mm!n!! 2 2 k 2w1w2 n n w + w ? i w2 w2 (1=R1 ? 1=R2 ) 1 2 2 1 2 " 2 2 k 2 2 # w1 ? w2 + i 2 w1 w2 (1=R1 ? 1=R2 ) m+n 2 2 2 2 w1 + w2 ? i k w1 w2 (1=R1 ? 1=R2 ) 2 (with k 2 = ). and ?0. It is easy to check that the total power is conserved.2n = 0 for m. the amplitude becomes. MODAL EXPANSION 215 be foreseen). the formula for coupling the TEM00 mode of parameters (w1. n 6= 0.

Bi The result is: where W1. and the wavefront two di erent curvatures along x and y. This expresses the rate of incoming power one can couple in a perfect TEM00 mode when the incoming amplitude is astigmatic in such a way that the intensity has two di erent widths along x and y.216 CHAPTER 3. . are derived from w1. W2 on one hand. y) = w2 2 w0 0 Let us compute the scalar product ? = h . b2 are the two Rayleigh parameters corresponding to the two astigmatism directions: 2 2 b1 w1 = . w2 and from the distance L through the ordinary gaussian formulas: s 2 W1 = w1 1 + L2 b 1 j?j2 = r 2 4w0 W1W2 r W W 2 2 (w0 + W12)2 + b R (w0 + W22)2 + b R 2 2 1 4 1 2 2 2 4 2 2 where b w0 = is the usual Rayleigh parameter for the TEM00 mode. R2 on the other. et R1. b2 w2= Suppose now a TEM00 mode: s " # 2 exp ? x2 + y2 (x. NUMERICAL METHODS s 2 W2 = w2 1 + L2 b2 " # b2 1 R1 = L 1 + L2 " # b2 2 R 2 = L 1 + L2 b1.

n(x) = x Qmn (x) ? 2n Qm.n?1 (x) or as well Qm. All can be obtained after some elementary algebra.n (x) 2 Orthogonality: X 1 m x =2 k k ! Qmk (x) Qkn (x) = 2 m! e k 02 mn Addition law: X (?1)k ?xy=2 Qmn (x + y ) k k ! Qmk (x) Qkn (y ) = e k 0 2 Derivative: (3.n (x) + nQm.n+1 (x) = x Qmn (x) ? 2m Qm?1.41) (3. MODAL EXPANSION 217 The fQmng polynomials have useful properties.43) (3. using the recurrence relations of the Hermite polynomials. Value at x = 0: Qmn(0) = 0 (m 6= n) Qmm(0) = (?2)m m! Recurrence relation: (3.n) X (?2)k m!n! m+n?2k Qmn(x) = (3.9 Properties of the Displacement polynomials Qm+1.3.3.44) (3. De nition: min(m.3. that we summarize below without proof.n?1 (x) dx (3.42) (3.46) dQmn (x) = mQ m?1.47) Some of the rst polynomials: .45) 3.40) k! (m ? k)! (n ? k)! x k=0 making clear the symmetry with respect to m and n.

NUMERICAL METHODS 0 1 2 3 1 x x2 x3 x x2 ? 2 x3 ? 4x x4 ? 6x2 2 x3 ? 4x 4 ? 8x2 + 8 5 ? 12x3 + 24x x x x x3 x4 ? 6x2 x5 ? 12x3 + 24x x6 ? 18x4 + 72x2 ? 48 x4 x5 ? 8x3 x6 ? 16x4 + 48x2 x7 ? 24x5 + 144x3 ? 192x x5 x6 ? 10x4 x7 ? 20x5 + 80x3 x8 ? 30x6 + 240x4 ? 480x2 The rst lines are simple: Q0n(x) = xn (3.3. We have seen that rotations and translations can be represented by operators of the form ! u 1 m+n+k+l pm+n+k+l Q (p) Q (q) e?(p +q )=4 Umn.48) Q1n(x) = xn+1 ? 2n xn?1 (3. n) mode requires the following relation: X 1 = jUmn. q) = p nl 2 m!n!k!l! mk where u is a unitary complex number. q)j2 2 2 Energy conservation The energy coming under the form of a given mode k. Conservation of the energy brought by any (m.kl(p.kl(p. q) a couple of parameters representing the two degrees of freedom of the displacement. and (p. n) is in general spread over all others after any non perfect optical element.50) Q3n(x) = xn+3 ? 6n xn+1 + 12n(n ? 1) xn?1 ? 8n(n ? 1)(n ? 2) xn?3 (3. (m.l .51) Miscellaneous: X Qmn(x) = (x ? 2)m ex n! n 0 3.218 0 1 2 3 4 5 CHAPTER 3.10 Structural properties of Displacement matrices Practical use of the displacement operators raises several questions.49) Q2n(x) = xn+2 ? 4n xn + 4n(n ? 1) xn?2 (3.

We do it for de niteness in the case of rotations.kl(p. we have X 1 p e(x +x0 2)=2 (x ? x0) 0 k k ! Hk (x) Hk (x ) = k 2 from what we get Z X 1 1 2 = p ep =2 dx e?x H (x)2 m k jQmk (p)j R k 2 k! and due to the normalization relation 2. we have " #" # X 1 2 = ?(p +q )=2 X 1 jQ (p)j2 X 1 jQ (q )j2 = 1 jUmn. but the method is quite general. 3. We rst take the formula 3.3.45 for m = n). 2 2 2 2 2 2 X 1 1 ep =2 2 k k ! jQmk (p)j = k 2 Z X dx dx0 e?(x +x0 2) eip(x?x0) Hm (x) Hm(x0) 2k1k! Hk (x) Hk (x0) R2 2 2 k .44. this yields 1 Q (p) = (?i)m+k p ep =4 I (p) mk 2 mk then the formula 3. we nd X 1 2 m p =2 (3. q)j 2m+n m! n! e mk nl k l k.26 giving the de nition of the Imk integrals: Z Imk (p) = dx e?x eipx Hm(x)Hk (x) 2 So that Z jQmk (p)j2 = 1 ep =2 2 dx dx0 e?(x +x02) eip(x?x0) Hm(x) Hm (x0) Hk (x) Hk (x0) 2 2 R We have consequently R Due to the closure relation 2. Now.52. if we return to the energy balance.52) 2k k! jQmk (p)j = 2 m! e k (This is an indirect proof of eq.l k 2 k! l 2 l! owing to eq.43.3.3. MODAL EXPANSION 219 This leads to compute jQmk j2.28 de ning the Q polynomials.

?q) = ms lt 3. setting R0mk (x) we have i p m+k 2 m+k m!k! Qmk (x) e?x =4 2 Rmnkl ( . and we have X This was a priori expected. q1 + q2). the result of the sequence is a displacement of parameters (p1 + p2 . NUMERICAL METHODS respectively.3. q1) and (p2.st (?p.st (p2.kl (p. ?q). we can study the coupling of the TEM00 mode with higher order ones (see Fig. represented by matrix Dmn. q 2 w sin( ) p so that.11 Magnitude of displacement matrix elements Consider for instance the rotation matrix: Rmnkl ( . ) = R0mk ( . for Dmn. the inverse displacement (?p. q) Dkl. ) R0nl ( .st (p1 + p2.220 CHAPTER 3. q2) This is a direct consequence of the addition law 3. q1 + q2) kl 0 Transitivity Given any two displacements of parameters (p1. 0) = mk nl kl 0 Dmn.kl (?p. q).kl (0.46. For instance. q). ?q) is represented by Dmn. q1) Dkl.3. but it is instructive to see that it is one more consequence of the addition law 3.kl (p1 .46. we see that the coupling e ciency is very small .kl (p. ) = R0mk (p) R0nl (q) It is interesting to check the numerical values of these matrix coe cients. q2) = Dmn. Inversion Given any displacement of parameters (p.24). It also can be deduced from transitivity. X Dmn. ) We can set as previously p 2 p 2 2 w cos( ) .

MODAL EXPANSION 221 0.n(p) 0.2 0.24: Power coupled from the TEM00 mode into higher order modes (TEM0n) through rotation of a mirror .0 0 1 2 3 p 4 5 6 Figure 3.3.3.3 R0.1 n=1=1 n n=2=2 n=3=3 n n n=5 n=5 n=4 n=4 0.4 0.

3.1 n=4=4 n 0. it makes sense to assume a very weak rate of modes having orders larger that 2. and the correspondance between p and the rotation angle is: = p 2:18 10?6 Rd so that p = 1 corresponds to about 13% of the gaussian aperture g = = w0. the width on that far mirror is about 5.2 n=2=2 n n=3=3 n 0.222 0.25 and 3.0 0 1 2 3 p 4 5 6 Figure 3. couple to higher orders (see Figs. It is also interesting to see how the TEM10 and TEM20 modes. when is small.45 km of the far mirror.n(p) 0. The gaussian aperture corresponds to p 7:8. This makes consistent an approximate model (see next subsection) involving only rst orders modes. as well as indirectly through (10).5 cm.26): This shows that a light initially (00) is weakly directly coupled into the (20) mode. If we assume (Virgo parameters) a waist of 2 cm on the cavity input mirror.4 CHAPTER 3. In this angular region. NUMERICAL METHODS 0. Finally.3 R1. a curvature radius of 3. .25: Power coupled from the TEM10 mode into higher order modes (TEM1n ) through rotation of a mirror for orders larger than 2 and values of p smaller than 1. for instance.

4 0.3.26: Power coupled from the TEM20 mode into higher order modes through rotation of a mirror .3 n=3=3 n R2. MODAL EXPANSION 223 0.n(p) 0.0 0 1 2 3 p 4 5 6 Figure 3.3.1 n=5=5 n 0.2 n=4=4 n 0.

Fig. When the misalignment angle increases.5 0. 3.28).and (02). q) plane has the following pattern (the example of R0055(p.224 1. the stored poser decreases: see Fig. This means that we restrict the expansion to the 6 modes (00). (10).9 0. We can also study the eld re ected by the cavity.7 CHAPTER 3.0 0.24. we consider a at/spherical cavity where the spherical mirror is rotated by an angle .12 Numerical results We check here the results we can obtain from a modal expansion limited to orders up to 2.30.27: Power coupled from the TEMmm mode into itself we show how the rst orders are coupled into themselves (see Fig.3. NUMERICAL METHODS Rm.3. We classically expect a transversal displacement of x = Rc : see Fig. and we study rstly the displacement of the intracavity mode. The four maxima correspond to the maxima already shown on Fig.8 0.0 0 m=0 m=0 m=2 m=2 1 2 3 4 5 p 6 m=6 m=6 7 8 9 10 Figure 3.3 0.3.3. q) is shown on Fig.31.6 0. (11).3.3.1 0.4 0. (01). Namely.32 shows the evolution of the .2 0.27): The complete map of the rotation matrix squared modulus in the (p.3.m(p) 0. (20).

3.50 -5. MODAL EXPANSION 225 10.00 7. q) plane.50 0.00 -2.50 0.00 -7.50 -10.76e+01 -9. logarithmic scale) .00 2.57e+00 -1.57e+01 -1.00 -3.3.50 5.00 -2.00 2.51e+00 Figure 3.50 10.00 -7.28: Power coupled from the TEM00 mode into TEM55 after a rotation.37e+01 -2.50 -5.00 7.00 -10.50 5. ((p.

47e+00 -4. q) plane.226 CHAPTER 3. logarithmic scale) .98e+00 -2.29: Power coupled from the TEM88 mode into itself ((p.95e+00 -7.57e-02 Figure 3.50e+00 -1. NUMERICAL METHODS 10 7 5 2 0 -2 -5 -7 -10 -10 -7 -5 -2 0 2 5 7 10 -9.

3.3. MODAL EXPANSION

227

0.05

0.04

0.03

0.02

0.01

0.00

-0.01

-0.02

-0.03

-0.04

-0.05 -0.05

-0.04

-0.03

-0.02

-0.01

0.00

0.01

0.02

0.03

0.04

0.05

3.57e-07

1.20e+03

2.39e+03

3.59e+03

4.78e+03

Figure 3.30: Map of the intracavity mode for = 2.9 Rd misaligment of the far miror (A rather extreme case). The circle shows the theoretical location of the mode.

228

CHAPTER 3. NUMERICAL METHODS

100 90 80 70 60 50 40 30 20 10 0 0. 1.e-06 2.e-06 3.e-06 4.e-06

Figure 3.31: Intracavity power vs misalignment of the far mirror. Finesse=150 (solid line), Finesse=50 (dashed line). The red curves are obtained from a full FFT simulation over 512 512 samples

3.3. MODAL EXPANSION
1.000

229

0.999 F=50

Cavity reflectance

0.998 F=100 0.997 F=150 0.996 F=200 0.995 0.0 0.5 1.0 1.5 2.0 θ [µRd] 2.5 3.0 3.5 4.0

Figure 3.32: Power re ectance of a cavity for increasing tilt angle and for several nesses re ected amplitude from a cavity having a tilted far mirror. We see that tilting the mirror has a negligible e ect for small angles, and for larger values is equivalent to a detuning, so that the re ectance increases. It is worth to emphasize that this is not due to the longitudinal displacement of the mirror, which has been corrected as if a servo loop were present. The rotation angle being , the apex equation of the mirror (of curvature radius Rc ) is: 2 2 Rc 2 2 z = x 2+ y + x = (x + 2R) + y + 1 2Rc Rc 2 c a corrective phase of = ? Rc 2 = is therefore introduced in the propagator. For higher tilt angles, the reectance of the cavity progressively reaches a constant value which is nothing but the bare re ectance of the input mirror: the far mirror can not more

230
1.00

CHAPTER 3. NUMERICAL METHODS

0.98

Cavity reflectance

0.96

F=150

0.94

F=100

0.92

0.90

0.88 0 2 4 6 8 10 θ [µRd] 12 14 16

F=50 18

20

Figure 3.33: Power re ectance of a cavity for increasing tilt angle and for several nesses. The horizontal lines represent the power re ectance of the input mirror corresponding to the di erent nesses produce interferences in the cavity. This happens when the tilt angle reaches values comparable with the beam aperture: see Fig.3.33. Remark that if the tilt angle is equal to the beam gaussian aperture, the transverse displacement of the intracavity beam is:

x =
so that

g Rc

= w Rc 0

x=w0 = Rc=b where b is the Rayleigh parameter. For the Virgo parameters, this is x=w0 3, so that the input beam is mostly out the intracavity beam. Even if the input eld is a pure TEM00, a small misaligment introduces a second resonance

3.3. MODAL EXPANSION
10 2

231

10 1

Cavity surtension

10 0

10-1

10-2 0.00 0.39 0.79 1.18 1.57 ϕ [Rd] 1.96 2.36 2.75 3.14

Figure 3.34: Resonances of a Fabry Perot cavity with a small misalignment. The red line is the theoretical position of the TEM10 mode at the frequency (or cavity length corresponding to higher order modes, especially the TEM10 or the TEM01 depending on the direction of the tilt. On Fig.3.34, we have scanned the free spectral range and compared the numerical peak with the theoretical position of the (10) resonance. Finally, it is possible to build a Michelson having a reference arm (ideal cavity) and a second arm having a tilted far mirror. The dark nge pattern exhibits a characteristic TEM10 signature. (Fig.3.35).

3.3.13 The A266 Algebra
We often need a fast simulation code for a dynamical model of misaligned and detuned interferometer, in order for instance to study the global control of the system. If the angles are small compared to the divergence of the beam, it is possible to limit at 2d order the modal expansion for each mirror.

232

CHAPTER 3. NUMERICAL METHODS

0.05

0.04

0.03

0.02

0.01

0.00

-0.01

-0.02

-0.03

-0.04

-0.05 -0.05

-0.04

-0.03

-0.02

-0.01

0.00

0.01

0.02

0.03

0.04

0.05

1.66e-10

7.95e-02

1.59e-01

2.38e-01

3.18e-01

Figure 3.35: Dark fringe on the output port of a Michelson for .1 Rd tilted far mirror on one arm

3.3. MODAL EXPANSION

233

Moreover, we see fom the preceding table that the modes of order m; n are coupled to the TEM00 at the m + n order. A consistent 2d order expansion will thus involve the only rst 6 modes, and we shall see that it is possible to carry out all calculations using a 2d order, rank 6 6 matrix algebra that we call A266. If we limit the expansion to the second order, the expression of the mirror operator, using the notation p p p = 2kw cos ; q = 2kw sin has the expression Rmnkl (p; q) = e?(p +q )=4 Rmnkl (p; q) where Rmnkl (p; q) is the following table: 00 10 01 20 11 02 p q p pq 00 1 i p2 i p2 ? 2p2 ?2 ? 2qp2 p 10 i p2 1 ? p2 ? pq ip i pq2 0 2 q 01 i p2 ? pq 1 ? q2 0 i pp2 iq 2 p pq 2 20 ? 2p2 ip 0 1?p ? p2 0 pq q p pq pq p2 p2 p2 1 ? p2 ? q 2 ? p2 11 ? 2 i i ? pq 02 ? 2qp2 0 iq 0 ? p2 1 ? q2 There is an apparent inconsistency in keeping the exponential not expanded, but this is not necessary for numerical computations, and gives much better accuracy when the expansion parameter (for instance = g is not in nitesimal. The free propagation along the optical axis is represented by the diagonal operator Pmn;pq = exp ?i(m + n) arctan L mp nq b where L is the propagation distance and b the Rayleigh Range. It is therefore clear that all operators involved in A266 are of the form M = M 0 + M1 + M2 where the partial operators Mi (i = 1; 2; 3) contain respectively the zeroth order in the perturbation strength ( = g or x=w), the rst order and the second. Moreover, each partial operator has a special structure. We note O3, O12 and O18 the sets of operators having these structures. Namely, O3 is the set of 6 6 operators of the form
2 2 2 2 2 2 2 2

234

CHAPTER 3. NUMERICAL METHODS

to which obviously belongs the propagation operator, and the zeroth order of any operator, O12 is the set of 6 6 operators of the form

1 7 8 9 10

2 3 4 5 6

11 12

to which belongs the rst order part of the operators. O18 is the set of 6 6 operators of the form

then AB 2 O3. The separation in three partial operators is therefore stable. What is more remarkable is the following property which is the basis of A266: if A. it is obvious that if A. B 2 O3 . (A + B )2 = A2 + B2 The product of two operators is de ned by (AB )0 = A0B0. The inverse of an operator is de ned recursively by (A?1)0 = A?1 0 which is a trivial operation. This kind of storage requires 3 + 12 + 18 = 33 places instead of 36 in the general case: There is no waste of memory. (AB )1 = A0B1+A1B0 (AB )2 = A0B2+A1B1+A2B0 note that the structure allows algorithms faster than the standard matrix product. The global structure is stable by the elementary algebraic operations.3. The sum of two operators is trivially de ned by (A + B )0 = A0 + B0. then (A?1)1 = ?(A?1)0A1(A?1)0 . B 2 O12 then AB 2 O18. More speci cally. and any algebraic operation reduces to trivial sums and products. We give below the most necessary. MODAL EXPANSION 1 5 7 9 12 16 6 8 10 13 11 14 17 15 18 2 3 4 235 to which belongs the second order part of the operators. (A + B )1 = A1 + B1.3. if A 2 O3 and B 2 O12 resp O18 then AB 2 O12 resp O18. A0 being diagonal.

ij 0. We can interpret the normalized square modulus of the amplitude as a probability density for a photon to be launched: dP 0 (x. NUMERICAL METHODS h i (A?1)2 = ?(A?1)0 A1(A?1)1 + A2(A?1)0 3. The square root X of an operator A is de ned recursively by the following scheme: q X0 = A 0 which is a trivial operation.ij = A2.ii + 0. then 1 X1. and if the information carried by the phase is not essential. y)j2 ds This doesn't tell us the direction of the photon.1 Spatial spectra. What is interesting is that the di raction phenomena can be represented up to a certain extent by this particle description. y).ii X0. though the quantum nature of light is completely ignored in this approach. We call "photons" these particles for brevity. This is approximately the Newton theory of light. We consider on an initial plane. We know that the angular information on the angles can be extracted from the Fourier transform of 3. A0 being diagonal.jj X2 X2.ij = X A+.g. leading to realistic models of light propagation. it is possible to represent light by particles following straight trajectories between re ections or di usion processes. stray light studies).4 Monte-Carlo methods If the system in which light propagates has a complex geometry. plane waves and photons . By launching randomly a large number of such photons. a given complex amplitude of light A(x.4.236 Remark that there is no need for a matrix inversion algorithm: In fact this is the main reason for the e ciency of A266. statistics can tell us where the light goes and how we can forbid certain areas to it (e. y) = jA(x.jj CHAPTER 3.ij ? (X 1 )ij X0.

Y. ) = 1 jA( . q)j2 = 4 2 2 R On the other hand. )j2 and consequently: 1 Z 2 d Z d jA( . q)j2 = k2 2 Z2 0 d Z 0 ~ d jA( . q = k sin sin we obtain Z R ~ dp dq jA(p. due to the Parseval-Plancherel theorem: Z ~ dp dq jA(p.3.4. the conjugated variables (p. y) will thus be described by the two densities of probabilities: The departure point of a photon will obey the statistics corresponding to dP 0=ds. and it is possible to compute the point (X. is dP 0 (X. y) at z = 0 to hit the small target of area dXdY . x. y)dXdY = dP 00 ( . )d dS d . 3. by using the substitution p = k sin cos . The conditional probability for a photon starting from (x.36).3. q). we have. If we consider the square modulus of the k latter. It is even possible to compute the probability density dP 0=dS of the arrival point. )j2 = 1 ~ 2 0 0 which shows that we can obtain a probability density for the angular distribution by taking dP 00 ( . ). its trajectory is de ned. the statistics de ned by dP 00=d . MONTE-CARLO METHODS 237 the amplitude.2 Propagation After having chosen the departure point (x. y) and the direction of the photon ( . )j2 ~ 2 d Propagation of the light described by the complex amplitude A(x. The 2D Fourier transform of the amplitude is nothing but an expansion in terms of plane waves having for transverse components of ~ .4. and its direction. Y ) at which it hits the plane z = d (see Fig.

NUMERICAL METHODS Y y θ.238 CHAPTER 3. (x. we can replace by d d.ϕ dY dX d x X Figure 3. X. Take a TEM00 normalized mode de ned by the amplitude: s ! 2 exp ? x2 + y2 A(x. X. Y )) dP (x. y) d ds R2 This is the integral expressing the transfer of the probability density from z = 0 to z =q . Y ). X. (x. If we adopt the paraxial approximation. X. Y ) = (x ? X )2 + (y ? Y )2 + d2 For the full density. Y ) dP ( (x. X. X. (x. y. y. y. Y ) being de ned by q (x. Y ) = dS Z 00 q 0 1 (x ? X )2 + (y ? Y )2=d. y. so that: dP 0 (X. y) dx dy dP d R2 d ds We can show on a very simple example how it works. y. and by (x ? X )2 + (y ? Y )2=d. we get thus: dP 0 (X. y. Y ) = dS Z 00 0 1 dx dy 2(x.36: propagation from plane to plane where d = dXdY= 2 is the elementary solid angle corresponding to the target seen from the initial point. y) = w2 w2 . Y ) dP (x.

3. MONTE-CARLO METHODS The Fourier transform is: We have thus: 2 2 2 ~(p.53) ds w2 w2 and ! dP 00 ( . y) at z = 0.37 for an example based on the Virgo parameters: The initial waist is w0 =2 cm. Y ) = 2 exp ?2 X 2 + Y 2 dS w02 w0 2 with v u 2 !2 0 = wu1 + w t w d exactly as in paraxial wave optics. ) = 2 w2 exp ? k2w2 2 (3. and the hit point at z = d is calculated as X =x+d .53). Y ) = d ! ! Z 2(x2 + y2) exp ?2 2w2 (x ? X )2 + (y ? Y )2] 4 2 d2 R2 dx dy exp ? 2 d2 w2 This can be calculated either directly or by Fourier transform (being a convolution product) and the result is ! dP 0 (X. of parameter w0. the photons are assumed emitted at a point given by the 2D gaussian random variable (x. q) = 2 w2 exp ? w (p + q ) A 4 239 p ! ! dP 0 (x.54) 2 d 2 And the transfer equation is: dP 0 (X.4. ) of parameter g = = w0.54).3. Y =y+d We see the agreement between the wave optics theory (lines) and the MonteCarlo result (histograms) . the joint probability density being given by (3. the joint probability density being given by (3. y) = 2 exp ? 2(x2 + y2) (3. See Fig. then the direction of ight is given by a new 2D gaussian random variable ( .

The relation px=p = tan allows then to compute the standard deviation corresponding to the uncertainty px : ! = arctan 4 x . the accuracy on its momentumis px = h=4 x (h is Planck's constant). Solid curve: Di raction theory 3. Histogram: Monte-Carlo simulation.3 Di raction patterns It is not always possible to know the Fourier transform of the incoming amplitude.050 radius [m] 0. the standard deviation being inversely proportional to the distance at the edge. As they pass near the edge of any aperture. photons randomly emitted propagating as in a billiard. This is generally possible at the initial plane.37: Di raction spot at 3 km of a 2 cm waist initial TEM mode. A very interesting procedure has been proposed in 19]. but after propagation in a complex system.075 0. The heuristic argument is borrowed from quantum mechanics: a particle being at the distance x of the edge of a screen may be seen as having its location determined with accuracy x.000 0. the momentum of a photon is known to be p = hk=2 .100 Figure 3.025 0. Now. where a clean source is assumed installed.240 1600 CHAPTER 3. NUMERICAL METHODS photon density [hits m-2] 1200 800 400 0 0. the photons are scattered at random angles. and reaching a given plane. do not allow to reconstruct the complex amplitude necessary to determine dP 00=d .4. Consequently.

The statistics is reported on Fig. if x > 0.55) 5 24 2 2 Where C (x) and S (x) are the Fresnel functions (see 20]).0000 0. x being randomly chosen.0075 -0. 0. Solid curve: Di raction theory We can test the procedure on the well-known problem of di raction by a half plane.0050 -0. Number of photons launched: 106 . The screen is at d = 5m. its direction is drawn as a random deviate knowing its standard deviation (a gaussian deviate works). the line represents the wave optics theory with I0 = 1=2xM (Eq. and the averaged behavior in the fringes region. then. 50. the intensity of the light on it is given 7] by 20 0s 112 0 0s 1123 I (x) = I0 1 6@ 1 + C @ 2d xAA + @ 1 + S @ 2d xAA 7 (3. If we consider a white screen at z = d behind.3. Histogram: Monte-Carlo simulation.3.0025 0.0050 0. the wavelength is = 1:06 m. Remark the excellent agreement in the shadow region.0100 Distance from screen [m] Figure 3. if x < 0 the process stops and a new photon is launched. . 241 density of hits [m-2] 100.3.0025 0. We assume a wide and uniform light beam centered at x = y = 0 falling on a blade masking half of the plane (x < 0). We assume a uniform random law for launching photons at x o the interval ?xM .38 (histogram). due to the loss of information about the phase. MONTE-CARLO METHODS 150.55).0075 0. The hit point on a screen at distance d is X = x + d. xM ].38: Di raction by a half-plane. -0.0100 -0.4.

NUMERICAL METHODS .242 CHAPTER 3.

A slice of dielectric material surrounded by other dielectric materials with di erent indices may thus be expected to behave like a FabryPerot cavity. like metallic layers. Superposing more and more alternatively high and low index layers produces a cascade of Fabry-Perot's. This is fortunate. The fact that the light source has a very narrow linewidth around the nominal wavelength allows using mirrors having a selective re ectance at the same wavelength. we present a more concrete representation of mirrors generally involved in laser optical systems. it is possible to enhance the re ectance of the slice. 243 . and the global re ectivity increases towards unity. By adjusting the round-trip phase inside. This selective re ectance can be achieved by superposing thin dielectric material layers as a coating on a transparent block of dielectric material (substrate).1 Multilayer coatings It is well known that light arriving at an interface separating two dielectric media of di erent refraction indices gives rise to both a re ected and an transmitted wave. in realistic interferometers).Chapter 4 Real mirrors In this section. have irreducible losses due to nite conductivity (using superconducting metallic mirrors is still a dream. or more exactly a nightmare. 4. The global quality of such a mirror results from the quality of the substrate. and from the quality of the coating. because all wavelengths mirrors.

the Maxwell equations impose the form of H. According to the sign of k.1 Dioptric matrix We naively represent light in a dielectric medium of index n as a pair of monochromatic plane waves. separating a medium of index n1 (the left half-space) and a medium of index n2 (the right half-space) (see Fig. and a left-propagating wave F are crossing each other. REAL MIRRORS n2 E2 F2 d2 F 1 µ Figure 4. x. where a right-propagating wave E .244 E1 n1 λ ν σ CHAPTER 4. one magnetic H.1). propagating along the z direction. z) = @ 0 A 0 00 1 H(t. the wave is left or right propagating. of the form 0 ?i!t inkz 1 Be e C E (t. Consider namely a plane z = 0. y. z) = B n jkj e?i!t einkz C @ k A 0 where jkj !=c. x. y.4. one electric E .1. The electric and magnetic elds are (we forget the e?i!t time dependence) : 0 in kz 1 e + e?in kz C E(z) = B 0 @ A 0 1 1 .1: waves at a plane boundary 4. Recall that the Maxwell equations also impose the continuity of the tangential components of both E and H at a boundary separating two dielectric media. Once given E .

we nd the outgoing E2 to the right and F1 to the left. namely E1 from the left and F2 from the right. the relevant coe cients (Q21) to apply when the two media are exchanged are easily deduced from the preceding by simply exchanging the 2 2 1 1 1 2 2 2 2 1 2 2 2 2 1 1 1 1 2 2 1 . Given the incoming elds. we get under the matrix form : ! ! ! E2 = t12 r22 E1 (4. and 0 1 1 245 0 in kz 1 e + e?in kz C E(z) = B 0 @ A 0 0 1 0 H(z) = B n2ein kz ? n2e?in kz C @ A 0 for the right half-space. . . with '2 kn2d2. we have ! n t12 = n 2+n ei' r22 = n ?n e2i' n +n Q12 = t = 2n ei' r = n ?n 21 n +n 11 n +n Obviously. and are given and we want to determine and .1) F1 r11 t21 F2 We call Q12 the matrix operator. assuming that the elds are expressed. Continuity of Ex and Hy brings the two equations : ( + = + n1( ? ) = n2( ? ) from where we get n ?n n = n2 + n1 + n 2+1n 2 1 2 1 n n ?n = n 2+2n + n1 + n2 2 1 1 2 This allows a convenient quadrupole representation of the interface. Owing to the preceding calculation. .4.1. in the medium 2 at a distance d2 from the interface (d2 will represent the layer thickness). MULTILAYER COATINGS 0 1 0 H(z) = B n1 ein kz ? n1e?in kz C @ A in the left half-space. are constant amplitudes.

We thus have the two quadrupole operators respectively attached to a low index and a high index layer. we can write ! ! E2 = Q E1 F1 F2 for the uncomplete stack. has also a quadrupole operator Qstack. But the composition law of Q-like operators is more complicated than ordinary linear algebra. By solving the system with respect to (E1.2 Models of stacks The stack of N layers taken as a whole.1.3) n ?n 2n i' n +n n +n e 1 2 2 1 2 2 1 2 1 2 1 2 1 2 2 2 1 2 1 1 2 2 1 2 2 2 1 1 1 1 1 2 2 1 4. It can be obtained from qlow and qhigh. F3). assuming n2 > n1 : 2n ei' n ?n e2i' ! qhigh = n +n?n n 2+nn ei' (4. REAL MIRRORS subscripts 1 and 2.2) n n +n n +n 2n i' n ?n 2i' ! n +n e n +n e qlow = (4.246 CHAPTER 4. Let q be the layer operator : ! t q = r Introducing intermediate elds. either low or high index. Assume for instance that the operator Q associated to a stack of n ? 1 layers is of the form ! TP Q = R and we want to add one more layer at the right. we get ! ! E3 = Q q E1 F1 F3 . and ! ! E3 = q E2 F2 F3 for the extra layer.

SURFACE MAPS where the operator Q q is given by 247 +P (t ?r ) ! tT 1?Pr 1?Pr R+r(T ?RP 1?Pr 1?Pr Q q = (4. measurements of the surface height are performed by interferometric means. 4. It is now easy to construct the operator corresponding to a given stack. It is . 4. Then the complete stack operator is obtained as Qn = (((Q0 qhigh) qlow ) qhigh) qlow : : : The tuning of the elementary Fabry-Perot's is determined by the thickness of the deposit.1.e. when for instance the rst layer is low index : assume Q0 is the operator corresponding to this layer of low index with vacuum in the left half-space (Q0 is a special case of qlow with n2 replaced by 1). The re ectivity for the nominal wavelength would be 1 ? 4 10?9 . The following plot 4. containing samples measured a the nodes fxij .1 Collimation and attening In general the method of measurement introduces a wedge.4) This is the Q-product of two operators.2. We see that the re ectivity remains high even for up to 20% variation of the wavelength. but the algorithm is very well adapted to numerical methods.2 Surface maps In order to test mirrors before and after coating.4.4.3) : one easily sees a tilt of the mirror axis. This is for instance the result of a measurement of a 7 cm diameter mirror (see Fig. a non zero angle between the symmetry axis of the surface and the optical axis. 4.3 Numerical codes An explicit analytical calculation is obviously untractable and even useless when n is larger than 2 or 3. The result is a 2D data set fzij g. The best re ectivity is obtained with '2 = '1 = =2 (quarter wave) and '0 = for the initial layer. i.2 shows the re ectance of a 30 layers stack : the indices were n1 = 1:4783 and n2 = 2:10225. yij g of a grid.2.

15 0. chosen according to the parameter w of the beam : 2 wij = exp ?2(x2 + yij )=w2 ij . The mean plane is de ned by the linear equation z = ax + by + c where (a.30 -0.2: Variations of re ectance vs wavelength thus necessary to re-orient the surface.2 0.3 0.8 Intensity reflectance CHAPTER 4.25 -0.30 Figure 4.5) > a < xy< x + b+<by<> > c < y=><=z < yz > : a > y +c > where the average symbol < : : : > has the following de nition for any quantity X de ned on the grid : P w X ij < X > = P ij ij (4. REAL MIRRORS 0.7 0.1 0.0 0.20 -0.6 0.5 0.25 0.4 0.05 0.6) ij wij The wij are weights.05 0. We can de ne the surface axis by the normal to the mean plane.9 0. c) are parameters to be de ned by a leat-square criterion.15 -0. giving the normal equations 8 > < a < x2 > + b < xy > + c < x > = < xz > 2 > + (4.00 ∆λ/λ0 0.248 1. b.10 -0.20 0.0 -0.10 0.

025 0.)y)2 (x.035 .4.035 0. z) V () y.015 0. SURFACE MAPS 249 -0.)y)2 (y. y Cov b = V (x)Cov((x)Vz)y) ? Cov(x. Cov(x. z) V ( c = < z > ?a < x > ?b < y > where the variances-covariances have the usual de nition The collimation is obtained by the correction V (x) =< x2 > ? < x >2 .025 -0. ? Cov(x.3: Susbstrate of mirror : surface as measured The solution is x.015 -0.035 -0.005 0.005 0. V (y) =< y2 > ? < y >2 .015 0.035 x y 0. Cov(x. y Cov a = V (y)Cov((x)Vz)y? ? Cov(x.2.005 -0. y) =< xy > ? < x >< y > zij ! zij ? a xij ? b yij ? c 0.025 Figure 4.015 -0.005 -0.025 -0.

035 -0.005 -0.250 CHAPTER 4.025 Figure 4.4.005 0. The next step is to extract the curvature radius of the surface.015 -0. But if the mirror is involved in a cavity.025 -0.035 .035 0.4). This can be done by tting a model of the type z = m r2 + p (r2 x2 + y2) The normal equations are satis ed by 2 ? < r2 m = < r<zr> > ? < r>< 2z > 4 2> p = < z > ?m < r2 > c is related to the curvature radius Rc of the mirror by Rc = 1=2m The determination of the curvature was done by a procedure in which the width w of the light beam is known.005 0. REAL MIRRORS -0.015 0.025 -0.4: Collimated surface one sees the same mirror map after collimation (see Fig.005 -0.035 x y 0.025 0.015 0.015 -0. this 0.

i. yj g. SURFACE MAPS 251 width is precisely determined by the curvature. which allows a better estimate of the curvature. we call it the weighted RMS roughness.2. This is done by interpolation. y). knowing its values on a grid fxi. j +1) where the estimation point falls. j ) ? (i +1.4. then we compute the corresponding w. this is a reason for giving here the basic ideas. p). it can be solved by iterations : one begins with an initial guess of the curvature (it is not likely that we have no idea a priori of the curvature). and it seems that it becomes an implicit problem. Interpolation in 1D data series is straightforward. y) knowing .3 2D interpolation techniques The mirror map has its own sampling grid. but in a 2D data array. As customary.2. and the numerical propagation program has also its (di erent) own. and so on. a reasonable initial guess gives a good corrected value within one cycle only.2 Weighted RMS roughness Once found the coe cients (m. and applied to the collimated and attened surface.4. In fact. one can make the correction 2 zij ! zij ? m rij ? p This is the attening operation. 4. for the sake of simplicity that the sampling grid is equally spaced.5). xi+1 ? xi = x yj+1 ? yj = y so that knowing x and y determines easily the cell (i. gives information on the roughness. The problem reduces eventually to nd an estimation of a function f (x. The statistics p = < z 2 > ? < z >2 with the already precised meaning of < : : : >.2.e. The result of the curvature correction is the residual departure of the mirror from the nearest paraboloid (see Fig. The only point is thus to estimate f (x. It is the relevant parameter for scattering losses estimation (see below). It is therefore in general necessary to convert a map from an initial grid to another. We assume. it is more di cult. 4.

013 0.013 -0.021 Figure 4.5: Oriented and attened surface : An example of a residual roughness the surrounding data : fij . These two triangles determine two di erent planes. fi+1 j+1. y.013 -0. x. yj ).004 0. A linear interpolation formula is of the type z =a +b +c where (x ? xi)= x and (y ? yj )= y.021 -0. We have the choice between two solutions (see Fig. and consequently two estimations of f ( .030 x y 0. it depends on three parameters. REAL MIRRORS -0.021 0.021 .030 0. We could 0. fi+1 j .4. ).013 0. as either in the ABC or in the BCD triangle. where we have set fij f (xi.004 0. and three data are necessary to determine them.030 -0. the point M (x.252 CHAPTER 4.6).004 -0. fi j+1. for instance.004 -0. y) may be viewed.

e. and use the corresponding plane to estimate fM . i. and reduce to ordinary (1D) linear interpolation on the edges.2. f0 = (fij + fi+1 j + fi j+1 + fi+1 j+1)=4. It is easy to see to which of them M belongs. ) = fi+1 j+1 + fi+12j ? fi j+1 ? fij + (fi+1 j+1 ? fi+1 j ) + fi+1 j + fi j+1 ? fi+1 j+1 + fij 2 and obviously di erent formulas. SURFACE MAPS x fi j+1 D C fi+1 j+1 253 M ∆y O f0 η A fi j ∆x B ξ y f i+1 j Figure 4. . The quadratic interpolation introduces a curvature of the interpolating surface that may give spurious e ects. In fact this is equivalent to the following procedure : call O the center of the rectangular cell. A quadratic interpolation is of the form z =a +b +c +d and the four coe cients are completely determined by the four corners of the cell . For instance in the case of Fig. Assign to O the estimate f0 taken as the average of the surrounding values. ) = fij + (fi+1 j ? fij ) + (fij+1 ? fij ) + + (fi+1 j+1 + fij ? fi+1 j ? fi j+1 ) The two methods give the exact values on the nodes.6. BOC. We have now four triangles (AOB. COD. we would have f ( .6: 2D interpolation problem take the average of them.4. DOA) with known node values. the result is f ( . depending on the relevant triangle.4.

Only one propagation direction is in principle allowed in the ring. resulting in interferences on the photodiode used to lock the system.4.4 Backcoupling due to roughness . y axes are orthogonal and within the plane tangent to the mirror. The coordinates are such that the z axis is the spherical mirror axis. y) an incoming gaussian beam. smoothing the surface. We denote by '0(x. matched to the mirror.7: sketch of the modecleaner Higher order polynomials can even allow to have continuous derivatives at the edges. We have s h i h i 2 '0(x. the x. just before entering the interferometer. The reason for such a ring cavity is to avoid spurious re ection of the laser beam o the input mirror. But the incidence angle on the curved mirror is so sharp (about 3 10?4 Rd) that a fraction of the light scattered by the surface may be sent in the counterclockwise mode.7 are at and nearly othogonal. whereas the length of one of the long sides is about 140 m. An example of direct application of the preceding methods has been found in the issue caused by the so-called mode-cleaner installed on the beam.254 CHAPTER 4.2. y) = w2 exp ?(x2 + y2)=w2 exp ?ik(x2 + y2)=2R exp ?ik (x cos + y sin )] 4. The two mirrors forming the basis (see Fig. and the far mirror is spherical with a curvature radius about 180 m. The mode-cleaner consists in a threemirrors ring cavity having the shape of a long equilateral triangle having thus a very sharp angle. ). say clockwise. REAL MIRRORS 2a~9cm L~140 m Figure 4. and incident with angles ( . We can study and evaluate the e ect on the roughness map of the spherical mirror as follows. and eventually causing instabilities. The length of the basis is approximately 9 cm.

Note 2 2 that the function f (x. ) between the re ected and the counterpropagating beams is given by the hermitian scalar product: ?( . y) exp ?2ik (x cos + y sin )] dx dy 2 I (x. y) refers to the wavefront map of the mirror. It represents the natural overlap of the re ected beam with the phase conjugate beam. y) '0(x. expressing the perfect matching of the beam. y) The counter propagating beam 'cis the phase conjugate of '0 : 'c = '0 the coupling coe cient ?( .8). the rst term still remains. ) = h'c . we can write " 2 2 2# Z ?( . y) f (x. 'Ri or. The re ected beam is thus 'R(x. y) = M (x. If = 0.4. this overlap is simply unity.2. due to gaussian divergence. y) is the normalized intensity distribution in the beam.4. including the mean paraboloid plus the residual roughness. ) = exp ? 2 w +2ik 2 I (x. y) ? x 2+ y is nothing but the residual R roughness of the mirror (see Fig. y) ? 2R where I (x. y) f (x. We can express the natural overlap as : h i 2 ?0( ) = exp ?2 2= g R2 ?2k2 Z R . y) f (x. y) = exp 2ikf (x. SURFACE MAPS The mirror operator is 255 M (x. This residue being small compared to a wavelength. y)2 exp ?2ik (x cos + y sin )] dx dy When the roughness is zero. in detail : " !# Z x2 + y2 exp ?2ik (x cos + y sin )] dx dy ?( . ) = R2 I (x. y)] where f (x. y) exp 2ik f (x.

010 0.010 Figure 4. y) f (x. y)].000 2 j?( .015 0. For the Virgo mode-cleaner parameters.015 0. it is straightforward to use the exact formula +ik Z -0.015 y 0. y) exp ?2ik (x cos + y sin )] dx dy+ R2 I (x.010 -0. by the intensity distribution of the beam. as usual. weighted. y) I (x. it is thus possible to reduce the expression of the coupling factor to the accurate approximation Z 2 = 4k 2 j?( . We give on Fig. For values of comparable to the mode-cleaner sharp angle ( MC 3: 10?4 Rd). For numerical computation.005 -0.005 0. we have g 3:15 10?5 Rd. )j2 = J~(2k cos . y)2 exp ?2ik (x cos + y sin )] dx dy 2 R showing that the result reduces eventually to the Fourier transform of the roughness. 2k sin ) 2 where J (x. we see that j?0j2 is extremely small and de nitely negligible.005 0.256 CHAPTER 4. The outer thin circle corresponds to coupling with the counterpropagating beam for all azimuth angles of the -0.000 -0. y) f (x.005 -0. where g = w.8: Residual roughness in the central zone of the MC spherical mirror. Within the angular region corresponding to backscattering. )j I (x. y) exp 2ik f (x.4.9 a map of the backcoupling in angular coordinates. units are m.010 0. REAL MIRRORS x 0.015 .

e-04 -2.e-04 -6. The star shows the optimal orientation of the mirror .e-04 4.2.e-04 -4. SURFACE MAPS 257 8.4.e-04 -4.7e+00 -6.e-04 0.0e+00 -7.e-04 6.e-04 8. -2. 2.e-04 -8.e-04 0. ) (logarithmic scale).e-04 6.3e+00 -5.e-04 4.e-04 -8.e-04 -6.0e+00 Figure 4.9: Coupling coe cient as a function of ( .e-04 2.e-04 -1.0e+01 -9.

18 1. the incidence angle being xed. or as well for all rotations of the mirror around its axis.e. We see that backcoupling depend sharply on this azimuth angle .36 2. then taking the average value of j?j2 over all locations .00 2 0.10 represents the variations of j?j2 along the outer circle. If we consider a gaussian distribution of parameter of these incidence locations. for all possible orientation of the mirror.p dithering 0 = w2 + 2. The inner thin circle surrounds a non sigi cant region where the natural overlap dominates. and has been limited.10: Coupling coe cient as a function of for backscattering incident beam. Fig.4. REAL MIRRORS 10-8 |Γ (f )| 10-9 10-10 0. has this pattern any physical reality ? The answer can be obtained by varying via a small o set the incidence location on the mirror. we call dithering this operation.258 10-7 CHAPTER 4.79 1. is strictly equivalent to taking an analyzing beam of width w For plausible values of (analogous to an error in the centering). and that these preferred orientations require a rather accurate positioning. i. One could rise the question of the sensitivity of this pattern with respect to the centering of the analyzing beam : in other words. The angle of 0.96 2. It also shows that there are preferred orientations.57 φ [Rd] 1. The map being rather unsensitive to small variations of w.37 Rd found on the curve corresponds to a hole in the speckle as seen .75 3. we conclude that it has an actual physical meaning.39 0.14 Figure 4.

5 Zernike polynomials . n ? 2.4. : : : (it ends either at 0 or 1 depending on the parity of n): (n?m)=2 X (?)p(n ? p)! m( ) = c n?2p Rn (4. It is necessary to x the radius a of the mirror. The Rm polynomials obey n Z1 Rm ( ) Rm0 ( ) d = 2(n 1 1) nn0 n + 0 n and the circular functions obey : Z2 sin m# sin m0# d# = mm0 0 Z2 cos m# cos m0# d# = (1 + m0) mm0 0 sine et cosine being obviously orthogonal. there is a need for analyzing and representing the departure of a given mirror surface with respect to the ideal shape. : : : and m = n. A systematic surface analysis consists rstly in nding a family of orthogonal defects over which a real surface can be expanded. then r=a. ( sin m# m( ) Znm ( . and in fact. The special behaviour of cos(0 #) 1 forces us to have the following normalization constant : v u 2(n + 1) u cmn = t (1 + ) m0 4.4. The required orthogonality is obtained with a family of functions Znm ( . SURFACE MAPS on Fig.9 and marked by a star. The variables ( . Zernike for n = 0. #) = Rn cos m# where the Rm( ) are a family of orthogonal polynomials rst introduced by n F. 259 In traditional instrumental optics. #) are separate.2.7) nm n+m ? p ! n?m ? p ! p=0 p! 2 2 The cnm are normalization constants. #) in polar coordinates.2. The orthogonality makes possible to treat separately the various defects. 1. and for instance to subtract any one of them without changing the expansion of the remaining surface.

#) = fnm Znm ( .s. #) d d# 0 0 The generating code for calculation of Rm ( ) is very short : n m fn Z 1Z 2 n=0 m=0or1 c=========================================================== real function zerpol(n. REAL MIRRORS The expansion of a surface of equation z = f (x.m.m.n) then print*.gt.rho real alphaz.rapp.'ERROR ! : m should be =< n !' stop endif c d=(n-m)/2 s=(n+m)/2 ro2=rho*rho rapp=1 do i=0.i real facm.d-1 alpha=-alpha*(s-i)*(d-i)/float((n-i)*(i+1)) rnm=ro2*rnm+alpha . #) with = f ( .d facm=facm*i enddo alphaz=rapp/facm alpha=alphaz rnm=alpha do i=0. y) on the Zernike basis is as follows : 1 n X X f ( .ro2 c if (m. #) Znm ( .rho) implicit none c integer n.rnm.d.alpha.260 CHAPTER 4.n-s-1 rapp=rapp*(n-i) enddo facm=1 do i=2.

eq. after V.1) (2.0 Figure 4.4.8 (2.5 ρ 0.N.6 0.3 0.1) (7.11: Some of the rst radial Zernike polynomials enddo if (m.9 1.Mahajan 21] : .0) then zerpol=sqrt(float(n+1))*rnm else zerpol=sqrt(float(2*(n+1)))*rnm*rho**m endif return end c=========================================================== A plot of the rst Zernike radial polynomials is shown on Fig.0) (3.4.11. The rst Zernike functions have the following interpretation. SURFACE MAPS 4 3 2 1 0 -1 -2 -3 0.2) 261 (1.4 0.2 0.0 0.1 0.7 0.1) 0.2.

2. y) = e2ikf (x. y) Let A(x.y) A(x.6 Roughness and scattering losses B (x.y) dx dy If we assume the real surface near the ideal one. y) be the amplitude of a TEM00 gaussian mode falling on a mirror the surface of which is de ned by the apex equation z = f (x. y) dx dy by A we mean the TEM00 eld that would be re ected by the ideally matched mirror. The re ected wave is (forgetting the scalar or photometric re ection coe cient) Its coupling with the TEM00 mode is given by the hermitian scalar product < A. y). y) ? r2=2R) ? 2k2(f (x. REAL MIRRORS name Piston x-tilt y-tilt Defocus Astigmatism Astigmatism Primary x coma Primary y coma Triangular astigmatism Triangular astigmatism Primary spherical Secondary astigmatism Secondary astigmatism 4.y) A(x.262 n 0 1 1 2 2 2 3 3 3 3 4 4 4 m m Zn 0 1 1 2 cos # 1 p 2 sin # 2 0 p3(22 ? 1) 2 p6 2 cos 2# 2 p 6 sin 2# 1 p8(3 3 ? 2 ) cos # 8(3 3 ? 2 ) sin # 1 p 3 3 p8 3 cos 3# 3 p 8 sin 3# 0 p 5(6 4 ? 6 2 + 1) 2 p10(4 4 ? 3 2) cos 2# 2 10(4 4 ? 3 2 ) sin 2# CHAPTER 4. y) e2ikf (x. and write 2 Z e?2r =w h1 + 2ik(f (x. so that 2 Z = w2 e?2r =w e?ikr =Re2ikf (x. y) ? r2=2R)2 + : : :i dx dy = w2 2 2 2 2 2 . B >. so that the coupling of A into itself is given by Z = A (x. then we can expand the phase at second order.

this is j j2 = 1 ? 4k2 q and the losses due to the mirror's imperfections are simply p = 4k2 q . SURFACE MAPS 263 f (x. y) ? r2 =2R)2 dx dy q = w2 is what we called "weighted RMS roughness". y) ? r2 =2R) dx dy = 0 2 2 Then the function f (x. y) is de ned up to a piston (additive constant) which can always be chosen such that Z e?2r =w (f (x. and the quantity 2 Z e?2r =w (f (x. y) ? r2=2R is nothing but what we called " attened" surface in a preceding section. so that 2 2 = 1 ? 2k2 q in terms of power.4.2.

264 CHAPTER 4. REAL MIRRORS .

so that any source of noise. But the design and the nature of these ba es must be such that the remedy makes nothing worse than the disease. This is why models of scattering are useful. the modulation will contaminate the stored TEM mode. mirrors not only re ect light and dissipate a part of it into heat. must be assessed. Due to the seismically driven motion of the vacuum pipe. but GW interferometers are designed to measure better than 10?11 Rd:Hz?1=2 phase changes. seismic noise is re-entering the readout beam. unless the mirror surface is polluted.1 Introduction Due to the imperfect nature of the surface of the re ecting coating. The symmetrical process is possible: Di use light coming from any direction may be partially coupled into a TEM mode. of the order of a few ppm. re ection of the di use light on the metallic walls of the vacuum pipe. In supermirrors as those used in gravitational wave interferometers. even very weak. and we have a by-pass of the seismic isolation system. Scattering is a process in which a perfect TEM mode is coupled to partial waves of any direction. The most evident scenario is scattering of light o a mirror. so that the amplitudes of scattered light. it has been soon seen that a system of ba es for trapping scattered light was necessary. It is clear that this double scattering process is extremely weak. If that di use light is phase modulated for any reason. the total losses (thermal dissipation + scattering) is very low. but also scatter light in all directions.Chapter 5 Scattered light 5. In order to avoid such a catastrophe. depending on the size of the defects of the re ecting surface. 265 . is extremely small. then inverse scattering on the emitter or any other mirror.

thus very small compared to the usual wavelength (1 m). we can expand the exponential and write: Z px ~ ~(~)j2 = ei~(_~ ?x0) n1 + 2ik f (~ ) ? f (x0)] ? 2k2 f (~ )2 + f (x0)2 ? 2f (~ ):f (x0)o ~ ~ j p x x x ~ . we have: Z ~ px ~ x ~(~)j2 = ei~(_~?x0 )e2ik f (~)?f (x0)] 0(~ ) 0(x0)d~ dx0 j p x ~ x ~ Owing to the hypothesis that f . Mirrors installed in GW interferometers as Virgo have roughnesses of rms value of a few nm. for studying scattering.2 Scattering mirrors The scattered light we are faced with.1) Here the stationarity implies that the autocorrelation function does not depend on the location in the plane. Suppose now that a light beam described by the amplitude (~) is imr pinging normally to the re ecting surface. and let us call (~ ) the re ected x beam's amplitude. SCATTERED LIGHT 5. where ~ represents the coordinates in the plane where x x we project the surface. The departure of the surface of a mirror from its ideal geometrical shape can be represented by a two-dimensional random process f (~ ). but only on the separation vector. the roughness is isotropic. is the autocorrelation function: C (~ ? x0) = hf (~ ):f (x0)i= 2 x ~ x ~ (5. We can assume without loss of generality that it is a centered process: hf i = 0 we also assume the process stationary: hf 2 i = 2 But the relevant statistics of the process. It will be further assumed that the autocorrelation function depends only on the length of the separation vector: C (~ ? x0) = C (k ~ ? x0 k) x ~ x ~ in words. We have: x x (~ ) = e2ikf (~) (~ ) x By taking the Fourier transform. is generated by re ection of light beams on mirrors with weak roughnesses.266 CHAPTER 5.

that we call specularly re ected beam. taking signi cant values only in the neighborood of ~ = ~ . We can assume that the Fourier transform of the autocorrelation p 0 function (i. SCATTERING MIRRORS x ~ x ~ 0 (~ ) 0(x0 )d~ dx0 267 By taking the expectation value.2. and scattering. the angular distribution is sharply peaked. one having the same angular distribution as the incoming beam. We identify with scattered light this contribution.5. and one having an angular distribution given by the properties of the surface. namely the power spectral density of f . and we have: ~p hj ~(~)j2i = (1 ? 4k2 2)j (~)j2 + 4k2 2C (~) p p (5. We have: Z p p Pspec=Pin = 41 2 (1 ? 4k2 2)j (~)j2 d~ = 1 ? 4k2 2 which shows that the scattering losses are given by: = 4k2 2 and we have: 1 Z 4k2 2C (~) d~ = 4k2 2 = ~p p Pscatt=Pin = 4 2 We can express the distribution of scattered light as: 1 dPscatt = ~p Pin d~ p 4 2 C (~) 1 Z C (~):j ~(~ ? ~)j2 d~ (5. we get: Z px ~ hj ~(~)j2i = (1 ? 4k2 2)j (~)j2 +4k2 2 ei~(_~?x0 )C (~ ? x0) 0(~ ) 0(x0)d~ dx0 p p x ~ x ~ x ~ or as well: hj ~(~)j2i = (1 ? 4k2 2)j (~)j2 + 4k2 p p 2 For gaussian beams.2) ~q p q q 4 2 . It can moreover be seen that the incoming power is shared between specularly re ected light. it is clear that the re ected light is the sum of two contributions.e. the power spectral density) does not appreciably vary on angles of the order of the angular width of the beam. and even more for hypergaussian beams.3) Under this form. In the preceding integral. the beam function can therefore be treated as a Dirac function.

is the coherence function. For larger angles.3 The scattering coherence function The central concept for a wave optics treatment of light scattered from a beam (gaussian or at). We have seen in the preceding section that the light scattered o a mirror of roughness f (~ ) can x be viewed as emitted by the source s(~ ) = 2kf (~ ) (~ ) x x x . a direct measurement of the surface by using a pro lometer can be carried out. 2 and by indentifying the Fourier coordinates to angles according to p ~ (k sin cos .4) Pscatt d 2 where Z p( ) sin d = 1 0 and nally. For very small angles. Now. and we can set dPscatt ( ) = p( ) (5. SCATTERED LIGHT where we have explicitly taken into acount the isotropy of the autocorrelation. k sin sin ). dPscatt ( ) Pscatt d is a normalized function. corresponding to long correlation distance defects.268 CHAPTER 5. by comparison between the two last equations: ~ C (k sin ) = 2 p( ) (5. we can write: d = d2 d 2 sin d d = 4 2 d d~ p 4 so that 1 dPscatt = 1 C (k sin ) ~ 2 P d in 5.5) Information on the normalized angular density of scattered power (ADSP) can be obtained by di erent ways depending on the angular range. a direct measurement of the ADSP is possible.

5. the expectation value C (d. We can consider the r wave generated by this elementary source after di raction along the distance d. we get k2 2 Z ~ p p x ~ hsd(~ ):sd(y0i = 44 2 C (~) e?i~(~?x0) Kd (~ ?~ )Kd (y0?x0) (~ ) (x0) d~ dx0 d~ y ~ y x ~ ~ x ~ x ~ p Now. This means that the integrand takes non negligible values y only in the small domain where the neigborhood of ?k~ =d intersects that y so that .3.6) This can be computed as follows. y0) = hsd(~ ):sd(y0)i y~ y ~ (5. it may be noted that if the distance d is larger than a few m. THE SCATTERING COHERENCE FUNCTION 269 where (~) is. ~. Firstly we have Z sd (~):sd(y0) = 4k2 Kd (~ ? ~ )Kd (y0 ? x0) f (~ ) f (x0) (~ ) (x0) d~ dx0 y ~ y x ~ ~ x ~ x ~ x ~ By taking the expectation value. Now. we have y Z sd(~ ) = Kd(~ ? ~ ):s(~ ) d~ y y x x x 2 We shall call coherence function of the scattering process. the incoming optical amplitude. It can be computed using the di raction kernel: x Kd (~ ) = ? id eik~ =2d x so that. the coordinate ~ +d~=k falls y p outside the actual beam for values of p slightly di erent from the maximum ~ p ~0 = ?k~=d. if we denote by sd(~ ) the propagated wave. it can be checked that Z p py y Kd (~ ? ~ )e?i~:~ (~ ) d~ = e?id~ =2k e?i~:~ d(~ + d~=k) y x px x x p 2 Z ~p py ~0) = 2 C (~) e?i~:(~?y~0 ) d(~ + d~=k) d (y0 + d~=k) d~ ~ p p C (d. this becomes: Z x ~ y x ~ ~ x ~ x ~ hsd(~ ):sd(y0i = 4k2 2 C (~ ? x0) Kd (~ ? ~ )Kd (y0 ? x0) (~ ) (x0) d~ dx0 y ~ after replacing C by its Fourier integral. as above. y y y p 4 where d is the beam amplitude di racted at a distance d. ~.

~. By substituting the Fourier transforms of the amplitudes. ~. ~ and y0 are so close together that y~ y we can write equally ~ y ~ ~ ~ C (?k~=d) = C (?ky0=d) = C (k ) = 2 p( ) where is the angle locating the direction of the small domain around ~ and y ~0.270 CHAPTER 5. y0) takes signi cant values. and that it di ers by a very small amount from ~p ~p ~~ the value ~0 = ?k~=d. We therefore replace C (~) by C (~0) C (p00) in the p y integral. Over this small domain. y0) = 8 3 eik(y ?y0 2)=2d y~ 1 Z d~ d~ dp0 e?i~:Y e?i~(d~=k+Y =2) ~ (~) eip0(d~=k?Y =2) ~ (p0 ) q p ~ q~ p q ~ d p ~ q ~ d ~ 4 16 but the Fourier transforms of the progagated amplitudes are equal to the Fourier transforms of the initial amplitudes. giving Z p y ~p ~0) = 2 C (~0) e?i~:(~?y~0 ) d(~ + d~=k) d (y0 + d~=k) d~ ~ p p C (d. ~. times the propagator 2 2 2 e?idp =2k 2 so that: 1 Z d~ d~ dp0 e?i~:(Y +(~?p0)d=k e?i(~+p0):Y =2 e?i(p ?p0 )d=2k ~(~) ~ (p0) p ~ ~ q p ~ q~ p ~ p ~ 4 16 2 2 C (d. It is possible to give a shorter version of the preceding integral. ~. y y y p 4 ~ When C (d. y y~ d 8 3e kq 2 k 2 ~ y ~ with Y ~ ? y0. it can be assumed that the function C has very small variations. this is 2 C (d. by the y change of variables y ~ p = ~ ? 2kd (~ + y0) ~ q we obtain ! ! Z 2 d~ + 1Y d~ ? 1Y ik(y ?y0 2)=2d d~ e?i~:Y q~ 0) = ~ d q ~ q C (d. y0) = = 8 3 eik(y ?y0 2)=2d y~ 2 2 . ~. SCATTERED LIGHT ~ ~ of ?ky0=d.

y0) = 8 3d2 eik(y ?y0 2)=2d d~ ~(~ ? kY =2d) ~ (~ + kY =2d) (5. y C (d.8) . We have q 2 ~(~) = 2 w0 e?p w =4 p 2 2 2 2 2 2 0 C (d. the explicit calculation is straightforward.3.5.7) y~ p p ~ p ~ In the case of a fundamental gaussian beam at its waist w0. y0) = 8 3 eik(y ?y0 2)=2d y~ 2 2 and consequently Z d~ ~(~ ? kY =2d) ~ (~ + kY =2d) = 4 2e?k w Y p p ~ p ~ 2 2 0 2 =8d2 so that nally where g = w0 is the gaussian angular aperture of the initial beam. ~. ~. ~. This shows the memory e ect of the initial beam even after di usion and di raction. THE SCATTERING COHERENCE FUNCTION the ~ integration gives a Dirac function. y0) = 2 d2 p( ) eik(y ?y0 2)=2d e?(~?y~0 ) =2d y~ 2 2 2 2 g (5. so that q 271 1 k2 Z d~ dp0 (p0 ? p ? kY =d) e?i(~+p0):Y =2 e?i(p ?p0 )d=2k ~(~) ~ (p0) p ~ ~ ~ ~ ~ ~ p ~ 4 2 d2 p Z ik(y ?y0 2)=2d d~ ~(~) ~ (~ + k Y =d) = 8 3d2 e p p p ~ which yields the symmetrical expression for the coherence function: Z C (d.

272 CHAPTER 5. SCATTERED LIGHT .

1 Heating by dissipation in the coating Consider a cylindrical mirror receiving the beam of a laser : A fraction of the light power is dissipated in the coating. It induces secondly distortions of the mirror's surface called thermal aberrations. the heat generation is practically localized around the optical axis. Methods of simulation for these processes need rstly a study of the steady state. These e ects initiate non linear processes : the rate of heating depends on the stored light power. 6. The mirror being suspended in a vacuum by very thin wires.Chapter 6 Heating issues The laser beam circulating through or re ected o the mirrors carries high light power. These mirrors dissipate a small but non zero rate of this power into heat. so that there is also a source in the bulk. especially in the resonant cavities. and the resulting temperature eld presents gradients. according to 273 . The only way for restoring thermal equilibrium with the surrounding walls is to radiate excess energy under the form of infrared radiation. a fraction of the power is absorbed per unit of length of the path inside. The pro le of the beam being sharp (gaussian). the stored light power depends in turn on the cavities tuning. If the substrate is crossed by the beam. it cannot appreciably lose heat by conduction nor by convection. The non uniform temperature eld induces rstly an index eld known as a thermal lens. then the transient case will be addressed. so that there is a source of heat over one face. increasing their internal temperature. which in turn depends on the thermal lensing and aberrations.

1). To be speci c.1) where K is the thermal conductivity (sorry. and p(r) the density of power deposited in the material . and the values used in numerical applications : . C its speci c heat . Let z be the coordinate along the symmetry axis (the optical axis).6. We address the case of surface heating (bulk heating will be addressed in a coming section). with the notation used throughout this chapter. and r the radial coordinate (see Fig. let us give these parameters in the case of silica. 6. the density of the material . Simple analytical solutions can be found in the case of axial symmetry.1. a material frequently used for making mirrors. HEATING ISSUES r=a laser beam z=-h/2 z=h/2 Figure 6. don't confuse with the unit K (Kelvin) of absolute temperature !) .1: Cylindrical mirror heated by laser beam the blackbody law.1 The Fourier equation and the boundary conditions In a general time dependent situation the heat eld obeys the Fourier equation : C @t ? K ] T (r. z) = p(r) (6.274 CHAPTER 6.

h=2) = 0(T 4 ? T04) @z .1) : On the face z = h=2.4 10?7 K?1 Y Young modulus 7. 10?5 m?1 Thermal expansion coe . 0 = 0:8 SB is plausible for SiO2.17 dimensionless If we assume the stationary state. The latter expression is non linear.6. but we hope. We assume that in case of thermal radiation. Let us detail the boundary conditions in the case of a cylindrical body (cf Fig.6. that the temperature excess with respect to room temperature n F + K rT ]surf = 0 where F (W:m?2) is the escaping ux.3 1010 N m?2 Poisson ratio 0. namely the balance of heat uxes on the limiting faces. by a correction (emissivity correction) taking into account the nature of the material (Please do not confuse the SB constant SB with the Poisson ratio ).1. 5.87 10?5 K?1 Dissipation rate (coating) 10?6 dimensionless Linear absorption coe . according to where 0 is related to the Stefan-Boltzmann constant SB 5:67 10?8 W m?2 K?4 that holds for the true blackbody radiation. z) = 0 We must add to this equation the boundary conditions. this reduces to the Laplace equation : T (r.38 W m?1 K?1 dn=dT Thermal index coe . ?K @T (r. in case of low absorption. the heat eld obeys the static homogeneous Fourier equation. and if there is no internal heat sources. the escaping ux is h i n F]surf = 0 T 4 ? T04 surf . -0. in which the radiation losses exactly balance the incoming power. ?K rT the internal ux at boundary and n the normal. HEATING BY DISSIPATION IN THE COATING 275 parameter name value units density 2202 kg m?3 C Speci c heat 745 J kg?1 K?1 K Thermal conductivity 1. T0 is the temperature of the surrounding wall.

z) = 4 0T03 T (a. (still assuming axial symmetry). Assume T ? T0 = T .2) @z On the face z = ?h=2. On the edge of the cylinder. we nd only radiation losses : (6. z) @r 2 2 6.276 CHAPTER 6. so that it was reasonable to linearize with respect to T0. Of course.1.4) ?K @T (a. h=2) (6. so that the incoming power ux is (P being the beam power) : P I (r) = 2w2 e?2r =w Note the change in the sign for the radiative part. this means that T 4 ? T04 4T03 T It will be understood in what follows. z) = 0 r . one must check at the end of the calculation that : T ? T 0 T0 was a correct assumption. represented as a boundary layer. h=2) = 4 0T03 T (r. so that the boundary condition becomes ?K @T (r. that T is the excess of temperature caused by the laser beam with respect to T0.3) @z where is the loss rate due to dissipation of light power into heat (a few ppm). h=2) + I (r) (6. HEATING ISSUES will be small. the Fourier (or Laplace) equation is @r2 + 1 @r + @z2 T (r.2 Solution as a Dini expansion In cylindrical coordinates. and the presence of an extra surface heat ux generated by absorption in the coating. ?h=2) = ?4 0T03 T (r. we have a balance of three heat uxes : ?K @T (r. We assume that the incoming beam is a TEM00 wave of half width w.

a]. : : :g the solutions of J1 ( ) ? J 0 ( ) = 0 (6. n = 1. expressed by Eq. HEATING BY DISSIPATION IN THE COATING 277 A solution of this equation is called harmonic. Call f n . The temperature eld can nally be written as an expansion of the type X T (r.6.5) The values of k are the kn = n =a. A. using a reduced radiation constant = 4 0T03 a=K : kaJ1(ka) ? J0(ka) = 0 An equation like the preceding one has an in nite discrete number of solutions de ning the possible values of k.1. z) = Anekn z + Bne?kn z J0(kn r) n It is well known from the Sturm-Liouville theorem that the functions fJ0( n r=a) . and the fJn(z) . n = 1. In cylindrical coordinates. there exist harmonic functions of the form T (r. B are arbitrary constants. 2. : : :g form a complete orthogonal basis for functions de ned in the interval 0.4 reads thus : K k J1(ka) = 4 0T03 J0(ka) or. n 2 Zg the Bessel functions. having normalization constants cn 20] such that Za J0( n r=a)J0( n0 r=a) r dr = c1 nn0 0 n with 2 cn = a2( 2 +2 2n)J ( )2 n 0 n . z) = J0(kr) A ekz + B e?kz where k. 2. It is probably worth to recall at least that @xJ0(x) = ?J1(x) 1 @x + x J1(x) = J0(x) The last boundary condition.6.

This gives the constants An. The precision improves very fastly with N .6. z) = ( n + )2 ? ( n ? )2e?2 nh=a J0( n r=a) n K (6. The error is quite negligible for N > 30 (Fig. the radius is large enough that the di raction losses are negligible. HEATING ISSUES (6. Finally the temperature eld is shown on Fig.3 reduce to a linear system ( ( n ? )?2 An ? ( n + )Bn = ? pn a?n =K n ( n + )An ? ( n ? )?2 Bn = 0 n where for the sake of brevity. This is equivalent to say that in the preceding integral. one nds : Za pn = cn I (r) J0( n r=a) r dr 0 and substituting the expression of I (r) yields Z a 2P 2 2n pn = a2( 2 + 2)J ( )2 0 w2 J0( n r=a) e?2r =w r dr n 0 n In the cylinders used as mirror substrates.8) The reconstruction of I (r) by expansion on the J0( n r=a) allows to determine the maximum number N of terms to consider for convergence of the above series.278 CHAPTER 6.6. n 0 n . ?n = exp(? n h=2a).2 and 6. the result is 20] : 2 2! P w2 n n pn = a2 ( 2 + 2)J ( )2 exp ? 8 a2 (6.2 one can compare the exact gaussian intensity pro le with formula 6.6.6) In particular.7) 2 2 Then the boundary conditions 6.6 with only 10 terms. On Fig. Bn : n An = pna e?3 nh=2a ( + )2 ? ( ?? )2e?2 nh=a K n n n Bn = pna e? nh=2a ( + )2 ? ( +? )2e?2 nh=a K n n The temperature eld is now fully determined : X pn a ? nh=2a ( n ? )e? n(h?z)=a + ( n + )e? nz=a e T (r.3). the intensity pro le can be expanded on this basis : X I (r) = pn J0( n r=a) n by inverting the latter relation.4. the bound a may be replaced by 1 without changing appreciably the result. In this case.

z) where dn=dT is the index temperature coe cient of the material. z) dz Z (r) = dT ?h=2 With the preceding expression of the temperature eld.2: dashed line : Beam intensity pro le.5 shows the shape of a plane wavefront after passing through a disk like the Virgo mirrors.020 0. z) = dT T (r.040 0.3 Thermal lensing .010 0. or excess optical thickness Z (r) : dn Z h=2 T (r. solid line : reconstruction with only 10 Dini terms The rst consequence of a temperature eld being installed in the bulk material is to create an index eld according to : dn n(r. The e ect 6.030 0. HEATING BY DISSIPATION IN THE COATING 1600 1400 1200 N = 10 279 Intensity [W m-2] 1000 800 600 400 200 0 0. For Silica we have dn=dT ?0:87 10?5 K?1.1.1.000 0. The e ect of the index eld is to change the wavefront of a passing optical wave by an extra path. we nd dn X pn a2 1 ? e? nh=a Z (r) = dT (6.6.6.050 radial coordinate [m] Figure 6. We call thermal lensing this kind of distortion.9) ? h=a J0( n r=a) n K n n + ? ( n ? )e n Fig.

5 1. We can estimate the curvature taken by a plane wave after crossing the disk by calculating the nearest paraboloid.03 0.04 0.0 1. HEATING ISSUES Absolute error on intensity [W m-2] N = 25 N = 20 0.5 -1. leads to a normal system having the solution 2 r ? < r2 c = < r Z (<)r> > ? < r>< 2Z (r) > 4 2> p = < Z > ?c < r2 > where the weighted average < f > of any function has the de nition Za <f > W (r) f (r) r dr 2 2 0 .5 0.02 0. and we want to minimize Za 2 Q(c.5 -2.01 0.0 0. like a real lens.0 -1.3: Error in the reconstructed intensity pro le and order of the Dini expansion of thermal lensing is as a rst approximation to change the curvature of the wavefront. p) vanish.0 0.280 2. exactly as in the treatment of imperfect mirrors (see preceding chapter). The requirement that the partial derivatives of Q(c.0 -0. The apex equation of such a paraboloid is z = c r2 + p. p) = W (r) Z (r) ? c r2 ? p r dr where 0 4 W (r) = w2 e?2r =w is the gaussian weighting function.05 radial coordinate [m] Figure 6.00 CHAPTER 6.

040 -0.030 -0.050 x .000 0.010 -0.030 0.040 -0.000 -0.1.010 0. HEATING BY DISSIPATION IN THE COATING 281 0.9 X Z (r) = znJ0( nr=a) where the coe cients zn are known by the preceding theory.020 -0.6.010 0.030 -0.040 0.020 0.030 0. Hot point : 13.020 0. The equivalent displacement Z (r) being known under the form 6.3K With the weight W (r) we nd (assuming w a) : 2 < r2 > = w 2 4 < r4 > = w 2 so that 4 < r4 > ? < r2 >2 = w 4 Let us derive a useful rule for the computation of the parameters c and p.040 0.050 Figure 6.020 -0. we nd rstly < J0( n r=a) >= e? nw =8a 2 2 2 -0.050 y 0.050 n -0.4: Temperature eld in a Virgo mirror for 1W dissipated in the coating.010 0.

5: Thermal phase lens in a a = 17.06 -0.02 0. the curvature c is : X 2 c = ? 41 2 zn n e? nw =8a a n 2 2 2 2 2 2 The mean optical thickness < Z > is X < Z > = zn e? nw =8a 2 2 2 n . HEATING ISSUES Thermal dephasing [Rd] 14 12 10 8 6 4 2 0 -0.04 -0.06 Figure 6.282 20 18 16 CHAPTER 6. h = 10 cm silica mirror for 1W dissipated in the coating (absolute value) and so that < r2J0( n r=a) >= ! w2 1 ? 2w2 e? nw =8a 2 8a2 2 2 2 2 < r2J0( n r=a) > ? < r2 >< J0( n r=a) > = ? n e? nw =8a < r 4 > ? < r 2 >2 4a2 and nally.04 0.5 cm.00 r [m] 0.02 0.

ignore it. the focal length of the lens : Rc = f = 1=2c 425:5 m. These formulas will be exploited also in foregoing calculations with other coming de nitions of zn .0e-06 -0. 283 -5. For the Virgo parameters. We could therefore in principle.e.06 Figure 6.6. we nd a curvature radius of the wavefront. We can compare the parabolic t to the original thermal lens on Fig. HEATING BY DISSIPATION IN THE COATING 0.1.W (Note that f is inversely proportional to the dissipated power). The dissipation rate could easily be di erent by . then 2 2 replace J0( nr=a) by ? n exp(? nw2=8a2)=4a2 and you get the curvature of the wavefront.0e-07 Thermal lensing [m/W] -1.02 0.6 .0e-06 -1. In fact the curvature is proportional to the absorbed power.5e-06 -3.6. A perfect parabolic lensing could be compensated by a suitable matching of the beam. which depends in turn of the dissipation rate in the coating. i.6: Thermal lens and its weighted parabolic approximation for 1 W absorbed in the coating and the piston : p = X n 2 zn (1 + n w2=8a2) e? nw =8a 2 2 2 The recipe is thus : take the formula giving the thermal lens Z (r) .04 0.00 r [m] 0.04 -0.0e-06 -2.5e-06 -2.06 -0. we have espressed it in m and restablished the sign.02 0.

e-06 -0.02 0. but 2 2 2 1 2 2 2 2 .7. resulting in extra losses for the beam.02 0. we can compute the coupling e ciency of the thermal lens between to matched waves.04 -0.e-06 CHAPTER 6.e-07 -1. and it is therefore di cult to compensate for the two lenses simultaneously.284 1.6.04 Figure 6.00 r [m] 0. HEATING ISSUES Thermal lensing anharmonicity [m/W] 5.7: Thermal lens : anharmonicity a factor of the order of 2 between two mirrors. -5. Concerning the losses. Consider for instance the incoming wave : 2 e?r =w eikr =2R 1 = w and an outgoing wave (after passing the lens) 2 e?r =w eikr =2R 2 = w The two waves have equal w for the lens cannot magnify the beam. the di erence between the actual lens and a paraboloid is plot. the lensing is not exactly parabolic.e-07 0. Moreover. and this means that it couples the nominal TEM00 mode with higher order modes. we call it anharmonicity by analogy with the potential theory. On Fig.

the square modulus of gives the e ciency : = 1 ? k2 < 2 > so that the coupling losses are simply L = 4 2< 2>= 2 . h = 0. we have thus Z ikp 1 W (r) 1 + ik (r) ? k 2 2 (r) r dr = e 0 the mean < > is zero. in which the defects have double weight. This is L 0:14 W?2 Assume = 1 ppm. 2 2 2 2 1 2 . p0) 4:15 10?15 m2=W2.6. Pintracavity = 10 kW. For our a = 0. so that = 1 ? k2 < 2 > =2 = 1 ? k2Q(c.175 m. The e ciency of the coupling depends on the scalar product = < 2. one nds L 5 ppm. Both are normalized according to Z1 (r) (r) r dr = 1 0 The lens can be expressed as r2 + p + (r) (r) = 2f where (r) is the anharmonic residue. we nd (numerically) Q(m0. p)=2 where c and p are the optimal values we just found. Note that the losses due to mirror roughness were computed exactly the same way. 1ei > this gives 4 Z 1 e?2r =w eik r (1=R +1=f ?1=R ) eikp eik (r) = w2 0 The matching condition is precisely 1 = 1 +1 R2 R1 f and if is much smaller than the wavelength. except that an extra factor of 4 appeared. HEATING BY DISSIPATION IN THE COATING 285 di erent curvature radii.1 m mirror.1. due to that special case of re ection.

z) = I0(r) e? z where I0 is the lossless solution of Maxwell's equations. as will be seen later). z) = 2w2 e?2r =w (6. the intensity obeys : 6. and 2 pn = P 2 ( 2 + 2n J ( )2 e? nw =8a a n )0 n 2 2 2 .286 CHAPTER 6. so that we can consider a heat source distributed in the bulk material. of the form P p(r) = 2w2 e?2r =w The heat equation now reads : P ?K T (r.2 Heating by dissipation in the bulk substrate A slightly di erent model must be used when we consider the heating process caused by dissipation of light power by its propagation through an absorbing medium.5 (in fact it is the same.2.1 Temperature eld I (r. z) = ? dI I0(r) dz diss We shall neglect di raction e ects inside the medium for the Rayleigh range of the beam ( 1 km) is much larger than the medium thickness ( 10 cm). HEATING ISSUES 6. The power dissipated in the medium per volume unit is therefore at rst order in : " # p(r. Transparent solids as silica have a small but nite linear absorption rate . so that as a function of z (the optical axis).10) We know from the coating study that 2P e?2r =w = X p J ( r=a) n 0 n w2 n 2 2 2 2 2 2 where the n are the discrete family of zeros of an equation similar to 6.

It will be satis ed if the n are the same as in the coating study. the face z = ?h=2 gives the same condition.10 admits a special solution Tspec(r) given by X Tspec(r) = tn J0( n r=a) n P tn = K ( 2 + 21)J ( )2 e? nw =8a 0 n n A general solution of 6.3. z) = 4 0T03 T (a. The particular choice of cosh rather than a combination of exp(? nz=a) and exp( nz=a) is justi ed by the symmetry of the problem : the heat source is independent of z. h=2) = 4 0T03 T (r. eq. On the face z = h=2 .6. On the edge ?K @T (a. and the temperature eld must therefore be symmetrical with respect to the meridian plane.6. Anyway. The global temperature eld is now : X T (r. z) = (An cosh( nz=a) + tn) J0( nr=a) n Now the boundary conditions reduce to radiation losses on the faces and on the edge. z) @r and this is the same condition as 6. that can be taken of the form X Tgen eh = An cosh( nz=a) J0( n r=a) 2 2 2 with n where the coe cients An are arbitrary. it is easy to guess that the n will be exactly the same as in the coating study. we have ?K @T (r. HEATING BY DISSIPATION IN THE BULK SUBSTRATE 287 In fact.10 requires still a general solution of the homogeneous heat equation. as could be foreseen. The last boundary condition determines the An : An = ?tn sinh( h=2a) + cosh( h=2a) n n n .2. h=2) @z Owing to the symmetry.

03 0.11) 2 ( n + 2)J0( n )2 0 n The pro le of the temperature eld is given on Fig.00 -0.2.288 CHAPTER 6.6.06 0.02 0.03 0.06 -0. 6.12 0.03 -0.03 -0.2 Thermal lensing We now know how to compute the thermal lens : dn Z h=2 T (r.00 0.01 0.09 -0. z) = K n sinh( n h=2a) + cosh( n h=2a) n 2 exp ? nw2=8a2) J ( r=a) (6.15 y 0.2 K for 1 W dissipated and the temperature eld is now determined : " # P X 1? cosh( n z=a) T (r.04 0. HEATING ISSUES 0.09 0.04 -0.12 .15 Figure 6.8: Absorption in the bulk : temperature eld. z) dz Z (r) = dT ?h=2 this gives : " # dn Ph X 1 ? (2 a= n h) sinh( nh=2a) Z (r) = dT K n sinh( n h=2a) + cosh( n h=2a) n -0.05 x -0.8.02 -0.05 -0. Hot point : 3.01 -0.

02 0.06 Figure 6.04 0. if the temperature is not uniform (we have seen that this is the .9: Thermal lens and its parabolic approximation (1 W dissipated)Bulk absorption 2 exp(? nw2=8a2) J ( r=a) 2 ( n + 2)J0( n)2 0 n One sees on Fig. 6.02 0. according to methods developped above.e-06 -0. Moreover.3 Distortion from coating absorption An other e ect of temperature changes in a solid is its thermal expansion.6.9 a plot of the lens pro le and of its parabolic best t. DISTORTION FROM COATING ABSORPTION 0. The focal length is f 412 m:W and the losses L = 0:15 W?2 Note that these results are very similar to those obtained in the case of coating heating. 289 Thermal lens (bulk) [m/W] -1.6.06 -0.3.00 r [m] 0.e-06 -3.04 -0.e-06 -2.

1 Thermoelastic solution For isotropic solids (e. We rst recall the linear thermoelastic equations. and it is related to the thermal expansion coe cient by : = (3 + 2 ) The equilibrium equation is : @j ij = 0 In the case of axial symmetry. the nonzero strain components are Err = @r ur . the mirror re ecting surface is distorted. using cylindrical coordinates. Moreover.3. is a linear relation between the stress tensor ij (~) r and the strain tensor via a constant rank 4 tensor : X Cijkl Ekl ij = k. The partial derivatives of ~ de ne a rank 2 tensor Eij (~) called strain : u r 1 Eij (~) = 2 @iuj (~) + @j ui(~)] r r r A generalization of Hooke's law linking applied force to displacement in the distortion of a spring. In the case of coating absorption. The atoms of the distorted solid are displaced with respect to the reference solid by a displacement vector ur ~ (~).290 CHAPTER 6. if a temperature eld T is present.g. the fact that the temperature eld is harmonic greatly simpli es the solution. In particular. and we have to estimate the e ect both in the case of the coating and bulk heating process. the elastic tensor reduces to only two independent components : ij = ij E + 2 Eij and are known as the Lame coe cients.l 6. HEATING ISSUES case when the heat source is the laser beam) stresses are developed inside causing distortions of the solid. fused silica) . an extra stress arises and this becomes 29] ij = ij ( E ? T ) + 2 Eij is the stress temperature modulus.

e. z) . the displacement of the surface z = ?h=2 holding the re ective coating. z) r0 dr0 E (r. z) = 2( + ) T (r.6. z ?h=2 provided a suitable determination of the unknown function (r). Remark that (r) is exactly our target. i. z) = 2( + ) r2 0 Ezz (r. a] and z 2 ?h=2. DISTORTION FROM COATING ABSORPTION Erz = 1 (@r uz + @z ur) 2 E = ur r Ezz = @z uz the stress/strain relations are 8 > rr = ? T + E + 2 Err > < = ? T + E +2 E > zz = ? T + E + 2 Ezz > : rz = 2 Erz and the equilibrium equation are ( @r rr + ( rr ? )=r + @z rz = 0 (@r + 1=r) rz + @z zz = 0 291 (6. We rst nd the strain tensor : 1 Z r T (r0. z) = 2( + ) T (r. All the following derivations aim to eventually nd (r). z) ? r2 0 1 Z r T (r0.12) (6. and that the coating is located on the z = ?h=2 face.13) Recall that the mirror is a cylinder of radius a. z) r0 dr0 Err (r. z) r0 dr0 ur = 2( + ) r 0 "Z z # 0) dz 0 + (r) uz = 2( + ) T (r.3. of thickness h. h=2]. One can check that the equilibrium equation is satis ed by a displacement vector of the form 1 Z r T (r0. that the coordinates are chosen in such a way that r 2 0.

z) = 2( + ) @z and we have to choose in such a way that the preceding expression vanishes.14 is identically satis ed : @z rz = 0 . z ) = 2( + ) r 0 @z ?h=2 @r The equilibrium equations reduce then to : @z rz (r. we have for the same reason (T being harmonic) : " # 00(r) + 0 (r)=r + @T (r.14 we get " # @ rz (r. ! @ 2T = ? 1 @ r @T @z2 r r @r so that eq.15) By substituting the expression of rz into eq. z) r0 dr0 (r. z 0) dz 0 + 1 r @T (r0 . z ) r0 dr0 rz (r. z) ? r2 0 zz (r.6.292 1 Z r T (r0. z ) = ? The stress tensor is in turn : CHAPTER 6. z) = 0 (6.14) (@r + 1=r) rz (r. HEATING ISSUES " # Z Z 0 (r) + z @T (r. z ) r0 dr0 Erz (r. z ) = 0 " # Z Z 0 (r) + z @T (r. z) = ? + T (r. z 0) dz 0 + r @T (r0 . z) = 0 (6. ?h=2) (@r + 1=r) rz (r. z) r0 dr0 (6. ?h=2) r @r @z rr (r.16) @z 2( + ) @r r 0 @z2 but there is no heat source inside the material. z) + 1 Z r @ 2T (r0. so that T obeys the homogeneous Fourier equation T = 0 i. z) = 4( + ) ?h=2 @r 0 @z .e.6. that is ! 1 @r r @ (r) = ? @T (r. z) r0 dr0 + r2 0 1 Z r T (r0. z) = @T (r. Now.

It is easy to compute rr (a. ?h=2) r00 dr00 + C 0 ln(r) + C 0 r 0 @z where C and C 0 are arbitrary constants. h=2) = 0 all are identically ful lled except the rst one. z) = e ( n + )2 ? ( n ? )2e?2 nh=a J0( n r=a) n K We have Za 2 Z n 2 2 J ( r=a) r dr = a J (x) x dx = a J ( ) = a J ( ) 0 0 n where we have used the de nition 6. z). z ) = ? + aK 2 n 0 0 2 n 1 n n 2 n 0 n . Now the stress component rz is explicitly known : the axis requires C Z z @T (r.5 of n . z) ? @T (r0. and since it has been shown that @z rz = 0. ?h=2) = 0. z ) = 0 rz (a. This gives P rr (a. We recall the expression found in the preceding section for the temperature eld 6. h=2) = 0 zz (r. z0) dz0 + rz (r. z ) = 0 rz (r. the regularity of ~ on u 0 = 0. Obviously.6. These conditions are here : rr (a.8 X pn a ? nh=2a ( n ? )e? n(h?z)=a + ( n + )e? nz=a T (r. z ) = 0 The boundary conditions express the balance of applied forces and torques at the limiting surfaces.3. DISTORTION FROM COATING ABSORPTION 293 the solution of which is Z r 0 Z r0 @T (r) = ? dr0 (r00. z ) = 2( + ) ?h=2 @r " # ! 1 Z r @T (r0. we have in fact simply rz (r. ?h=2) r0 dr0 r 0 @z @z This last form makes it clear that rz (r.

z) = A + B z. This is done by the new displacement vector 2 ur(r. -0.6. and by adding a stress of the form rr (a.05 0. Then the Saint-Venant principle tells us that the resulting solution is almost everywhere near the exact solution. and adding the correction ~ to ~ . CHAPTER 6. z) (see g.294 -10000. z) = ? (3 + 2 ) (A z + B z2=2) ? 4 (3++ 2 ) B r2 some calculation shows that rz as well as zz are identically zero.05 Figure 6. it will be possible to almost exactly cancel the edge stress. By suitably chosing the arbitrary constants A and B . we can remove the global resultant radial u u force exerted on the edge and the resultant torque. z) = 2 (3++ 2 ) (A r + B rz) 2 uz (r. But we are 2 2 2 X . z ) = A + B z .00 z [m] 0. and rr (a. z). HEATING ISSUES Θrr [N m-2] -15000.10) shows that the dependence on z is quasi linear. But we have to nd a new solution of the elastic equations satisfying the boundary conditions and giving a linear rr (a.10: The radial stress rr on the edge is a quasi-linear function of z e?w n=8a ( n ? )e? n(h?z)=a + ( n + )e? nz=a 2 2 ( n + )2 ? ( n ? )2e?2 nh=a n ( n + )J0( n ) A plot of rr (a. -20000. except maybe in a small neighborhood of the edge.

z) dz B = ? h3 rr ?h=2 By substituting the expression of rr (a.6. so that the approximation should work quite well. and the total displacement ~ vector eld is U = ~ + ~u. On the following gure 6.12 one sees the error rr(a. we get P X e?w n=8a 1 ? e? nh=a (6. one can see the linear function ?A ? B z superimposed to the function rr(a.11. z). The displacement is de ned up to a constant u 2 2 2 2 2 2 one nds .17) A = YKh 2 2 ? h=a n ( n + ) n J0 ( n ) n + ? ( n ? )e n and ? n h=a ? 1 + e? n h=a nh 12 Y P a X e?w n=8a 2a 1 ? e B = ? Kh3 2 2 2 n + + ( n ? )e? nh=a n ( n + ) n J0 ( n ) (6. The displacement is now fully determined. DISTORTION FROM COATING ABSORPTION 295 interested in the region "seen" by the light beam.6. The relation is = (1 + Y ? 2 ) )(1 = 2(1Y ) + so that 2( + ) = (1 + ) and + = Y On g. z) + A + B z)2 dz ?h=2 1 Z h=2 (a.3. z) dz A = ?h rr ?h=2 and 12 Z h=2 z (a. z). By minimizing Z h=2 Q = ( rr (a. z)+ A+Bz.18) It has been found more convenient to use the Young modulus Y and the Poisson ratio instead of the Lame coe cients.

as can be seen on Fig6. ?h=2) = 12? B r2 Y But the Saint-Venant correction appears very small in the region of optical interest.3.19) ( n + )2 ? ( n ? )2e?2 nh=a J0( nr=a) ? 1] and uz (r.05 0. CHAPTER 6.11: Edge stress function (solid line) and its linear t (dashed line) vector. We arbitrarily choose a displacement zero for r = 0 and z = ?h=2. ?h=2) = (1 +K ) P ( 2n+ 2)J 2( ) 0 n n n n + ? ( n ? )e?2 nh=a (6.296 -10000. We have thus the special result X e? nw =8a uz (r. we wish to estimate the departure of the distorted face from an ideally parabolic surface. -20000.13.05 Figure 6.2 Surface analysis As in the case of thermal lensing. -0. 2 2 2 6. HEATING ISSUES Θrr [N m-2] -15000.00 z [m] 0. The apex equation of the paraboloid being ^ Z (r) = c r2 + p .

6.21) ( n + )2 ? ( n ? )2e?2 nh=a for the Virgo corner mirrors (a = 0. h = 0. ?h=2) ? c r2 ? p where W (r) is the normalized intensity of the light spot.02 m). p) = W (r) uz (r.12: Residual edge stress we nd the parameters c (curvature) and p (piston) by minimizing Za h i2 Q(c. DISTORTION FROM COATING ABSORPTION 200 297 ∆Θrr [N m-2] -100 -400 -0. with the notation xn = n w2=8a2 : ?xn h i X p = ? (1 +K ) P ( 2 + n e2)J 2( ) 1 ? (1 + xn)e?xn 0 n n n n + ? ( n ? )e?2 nh=a (6.00 z [m] 0.1 m and w = 0.20) 4 a ? n? ) 0 n n n n 2 and.175 m.05 0.3. we nd for instance a curvature c ?8:6 10?5 m?1W?1 2 2 2 0 . This yields ?2 nh=a ) X 3e? nw =4a + c = ? (1 + 2K P ( 2n+ 2)J 2( ) ( n+ )2? ((n ? )e2e?2 nh=a (6.05 Figure 6.

5e-07 -2.0e-07 -0. For the Virgo corner mirrors.0e-08 uz(r) [m] -1. The beam half-width was w = 0.15 Figure 6. CHAPTER 6. we show the distorted surface in the optically interesting region.10 -0.6.14.05 0. de ned above.00 r [m] 0. One can evaluate these coupling losses as customary by L = 16 2 Q(c.0e-07 -1. Without S-V corection (dashed line) and with S-V correction (solid line). p)= 2 where c and p have their optimal values.02 m and for the curvature radius (Rc = 1=2c): Rc ?5818 m:W On Fig. and the nearest paraboloid. .13: Distortion of the re ecting surface. HEATING ISSUES -5.10 0.298 0.05 0. the losses are quadratic. one nds the loss rate L 3 10?3 =W2 Recall that the displacement being linear with respect to the absorbed power.15 -0. This distorted surface couples the TEM00 mode with higher order modes causing coupling losses.

299 -5. and = n =a.06 -0.06 Figure 6.0e-08 uz(r) [m] -1.02 0.04 -0.5e-07 -0. We are bound to solve the full system of thermo-elastic equations.04 0. z) = tn(z) J0(kn r) where kn namely.14: Distortion Z (r) of the re ecting surface and the nearest (intensity weighted) paraboloid 6.6.4 Distortion caused by bulk absorption In the case of bulk absorption. The functions tn(z) are of the form tn(z) = pn 1 ? n cosh(knz)] n P ?2 282 pn = K exp(+ n2wJ =( a )) ( 2 n ) 02 n n n sinh( nh=2a) + cosh( n h=2a) .0e-07 -1.00 r [m] 0. and the preceding short method cannot be employed.02 0.4. Let us recall that the temperature eld is given by : X T (r. the temperature eld is not any more harmonic. DISTORTION CAUSED BY BULK ABSORPTION 0.

we nd the equilibrium equations : ( 00 2 0 An ? knAn] ? kn ( + ) (Bn + knAn) ? tn ] = 0 00 2 0 Bn ? kn Bn ] + @z ( + )(Bn + kn An) ? tn] = 0 The rst consequence is that h 2 2i 0 @z ? kn (An + kn Bn) = 0 of which the odd solution is 0 A0n + kn Bn = knCn sinh(kn z) (6. The source of heat being independent on z and the temperature. z) = PP (kn An(z) + Bn (z)) ? tn (z)] J0(kn r) + 2 Pn Bn (z) J0(kn r) zz n 0 rz (r. z) = n (kn An(z) + Bn n kn An (z ) J1 (kn r)=kn r 0 0 (r. z) = (kn An(z) + Bn(z)) J0(kn r) n The stress tensor is in turn : P P 0 0 rr (r.1 Thermoelastic solution We search for a solution of the form ( ur (r. we expect An .22) 0 where Cn is an arbitrary constant. z) = PBn (z) J0(kn r) n Erz (r. HEATING ISSUES 6. z) = 1 n(A0n(z) ? kn Bn(z)) J1(kn r) 2 so that the trace is X 0 E (r. consequently. z) = Pn An(z) J1(kn r) uz (r. z ) = Pn (kn An (z ) + Bn (z )) ? tn (z )] J0(kn r) + 2 Pn kn An (z ) J1 (kn r) 0 (z )) ? tn (z )] J0 (kn r) + 2 (r. z) = Pn knAn(z) J10 (kn r) P E (r. The strain tensor components are : Err (r.4. z ) = n (An (z ) ? kn Bn (z )) J1 (kn r) after some algebra. an even function of z. z) = P n kn An(z) J1(kn r)=kn r 0 Ezz (r.300 CHAPTER 6. z) = Pn Bn (z) J0(knr) where An(z) and Bn (z) are unknown functions to be determined according to the equilibrium equations and the boundary conditions.

inserting in the rst. This result (Eq. h=2) = 0 gives # " + 0 00 + 2 n cosh n ? + 2 sinh n Cn + 2 sinh n Cn = and the condition zz (r. using 6. h=2) = 0 yields " # + 0 00 cosh n ? + 2 n sinh n Cn ? 2 cosh n Cn = 2 pn + pn n kn ( + 2 ) kn( + 2 ) n sinh n ? k ( pn n ) (sinh n + n cosh n) +2 n (6.22 : + 0 00 0 Bn(z) = (Cn ?Cn ) sinh(kn z) ? 2( + 2 ) Cn sinh(knz) + kn z cosh(kn z)] ? ? 2k (pn+n2 ) sinh(kn z) + knz cosh(kn z)] n The arbitrary constants are determined by the boundary conditions on the surfaces z = h=2. We have seen indeed on the preceding case that the needed correction to the displacement is practically negligible on the central area of the mirror. where light is actually interacting with the surface. The condition rz (r. The conditions on the edge r = a are ignored.6.6.4. yields h i 2 2 0 ( + 2 ) @z2 ? kn An = ( + )kn Cn cosh(kn z) ? kn pn 1 ? n cosh(kn z)] 00 of which the even solution including one more arbitrary constant Cn is : 00 0 An(z) = Cn cosh(kn z) + 2( + 2 ) Cn knz sinh(kn z) + + + k ( pn 2 ) 1 + n knz sinh(kn z)=2] n + then.23) .22) allows to express Bn as a function of An and to insert it in any of the two equilibrium equations. DISTORTION CAUSED BY BULK ABSORPTION 301 being an even function and Bn an odd one. For instance.

is # X pn sinh n " sinh n n ? 2 J0(kn r) uz (r. due to the nearly parabolic pro le. it is possible to compute Bn( h=2) : " # pn sinh sinh n n Bn(h=2) = k ( + ) ?2 n n n + sinh n cosh n so that the displacement at z = h=2 (symmetrical to the displacement at z = ?h=2.6. with the nearest paraboloid. this is in detail 2 X ( n 2= 2 Z (r) = (1 + K) Pa (exp+?2) w J 8(a )) 2 2 n 0 n n n # " 2 sinh n ? sinh + cosh J0( nr=a) n + sinh n cosh n n n n where we have replaced the Lame coe cients ( . The curvature radius is rc = ?16388 m:W and the losses L = 9:3 10?5 W ?2 These gures are signi cantly di erent from the case of coating heating. is is n n h=2a.15 the shape of the distorted surface in the center region. h=2) = + kn n + sinh n cosh n n Calling Z (r) the apex of the distorted surface. HEATING ISSUES This is a linear system in (C 0. by a rough factor of 3 for the focal length. C 00) the solution of which " # pn 2 sinh n 0 Cn = k ( + ) ? n n n + sinh n cosh n " # pn 2 ( + ) n cosh n ? sinh n + 00 Cn = ? 2k ( + ) + 2 n n n + sinh n cosh n Now.4. and even by two orders of magnitude for the losses. computed according to the method already experimented in the previous problems.2 Surface analysis One can see on Fig. CHAPTER 6. and ) by the Poisson ratio and the linear thermal expansion coe cient . .302 where. 6.

r. then we switch on the laser. r) z=?h=2 6.1 Transient temperature elds : general method .02 0.00 r [m] 0.06 -0. r. z) refers to the internal source of heat. z) C @t (6. HEATING PROCESSES 5.8e-07 303 5.2e-07 5.5. r.06 Figure 6. The boundary conditions remain " # @ ?K @z T = ?4 0T03T (t.04 0.0e-07 4.5 Heating processes Up to now we have treated the steady state solution supposed to be reached at thermal equilibrium between the mirror and the world around the vacuum vessel. Now we turn to the question of temperature evolution from a given state to a new one. In the time dependent case the heat equation is (we keep axial symmetry) : " # @ ? K T (t.04 -0.4e-07 5.5.02 0. and the temperature begins to increase until the steady state.6e-07 Distortion (bulk) [m/W] 5.8e-07 -0. z) = S (t.15: Distortion of the re ecting surface and the nearest (intensity weighted) paraboloid 6. h=2) + S2(t.6. r. For instance the mirror is at a uniform temperature T0.24) 1 where S1(t.

z) = T1(r. z) + Ttr(t. 00 .m 304 " CHAPTER 6. r) is the surface source of heat localized on the coating (at z = ?h=2). z) where Ttr(t. z) ?K @r T r=a where S2(t.24 satisfying the boundary conditions is known since the preceding sections. the sources are time independent. a.e. In the special case of a static source (a constant power laser beam). The functions nm(t) and m 00 nm (t) must satisfy 0 @ nm + K (k2 + 0 2 ) 0 = 0 @t C n m nm 00 @ nm + K (k2 + 002) 00 = 0 @t C n m nm whose non exploding solutions are " # 0 (t) = 0 exp ? K (k 2 + 0 2 )t nm nm C n m " # 00 (t) = 00 exp ? K (k 2 + 00 2)t nm nm C n m where the nm are extra arbitrary constants.24 is therefore of the form T (t. The general solution of 6. r.# @T ?K @z = 4 0T03T (t. r. r. HEATING ISSUES 0 where 0m. satisfying the homogeneous heat equation and the homogeneous boundary conditions (i. z) = nm (t) cos( m z ) + nm (t) sin( m z )] J0 (kn r) n. It is convenient to de ne the following time constants : C 0 nm = K ( 0 2 + k 2 ) m n . h=2) z=h=2 " # @ = 4 0T03T (t. z) is the transient part. and kn are arbitrary constants. The transient temperature can be searched under the separated form X 0 0 00 00 Ttr(t. r. z) of 6. reduced to outgoing radiation). r. and a special static solution T1(r.

that can be easily computed. whereas for satisfying the boundary conditions on the circular faces. But moreover. this is a consequence of the relations 6. we have Z h=2 0 cos( 0mz) cos( 0nz) dz = nmgm ?h=2 . m 2 Zg of solutions. the equation v cos v + 0 sin v = 0 (6. The rst equation has the form u sin u ? 0 cos u = 0 (6. and for the same theoretical reasons.5.6. HEATING PROCESSES 00 nm 305 C = K ( 00 2 + k2 ) m n Now. The orthogonality of the sine family with respect to cosine is obvious over a symmetrical interval.26) admits an in nite discrete family fvm . then we have 0 = 2u =h m m The same way. a]. 0m and 00 must respectively verify m 0 3 0 h sin( 0 h=2) ? 4 T0 h cos( 0 h=2) = 0 m2 m m 2K 0 3 00 h cos( 00 h=2) + 4 T0 h sin( 00 h=2) = 0 m2 m m 2K Let us introduce the new radiation constant 0 h=2a. m 2 Zg of solutions. m 2 Ng and fsin(2vmz=h). For the two last cases. n 2 Ng form an orthogonal complete basis for functions of r de ned on r 2 0. as already remarked.25 and 6. the boundary conditions impose kn = n =a where the n have the same de nition (6.25) It admits an in nite discrete family fum .26.5) as in the whole present chapter. m 2 Ng on z 2 ?h=2. h=2]. and 00 = 2vm =h m It is essential to note that the functions fJ0( nr=a). the functions fcos(2umz=h).

owing to the orthogonality of the sin( 00 z) and the cos( 0mz). This is done depending on the initial condition on the temperature. z) = tn(z) J0(kn r) 0 gm n where requiring T (0. z) = ? nm 1 ? e nm cos( m z )+ 00 nm n. g00 = h 1 ? sin(2vm ) = 2 m 2um 2 2vm 0 00 At this point. r.306 Z h=2 ?h=2 CHAPTER 6. HEATING ISSUES sin( 00 z) sin( 00 z) dz = m n 00 nm gm " # " # h 1 + sin(2um ) . all constants are determined except mn and mn . The temperature eld is then Xh 0 ?t= 0 0 T (t.m m ?h=2 00 nm i 00 1 ? e?t= nm sin( 00 z) J0(kn r) m (6. this m gives Z 0 = ? 1 h=2 t (z ) cos( 0 z ) dz n nm m g0 1 Z h=2 t (z) sin( 00 z) dz = ? g00 n m m ?h=2 which completes the determination.27) . Assume for instance the excess temperature is zero at t = 0. r. The steady state temperature (see preceding sections) is generally known under the form X T1(r. z) = 0 yields X X 0 0 tn(z) J0(kn r) + nm cos( m z ) + n n.m 00 00 nm sin( m z )] J0 (kn r) = 0 Owing to the orthogonality of the J0(kn r) this is equivalent to X 0 00 00 0 tn (z) + nm cos( m z ) + nm sin( m z )] = 0 m and now.

6. we nd 1 Z h=2 sin( 00 z) exp( m 00 gm ?h=2 n z=a) dz 2 cos( ) 0 Cnm = 1 + sin(2uum=2u e n m) m and so that 2 sin( 00 Cnm = 1 ? sin(2vvm))=2v e n m m 0 nm n + 0 ? ( n ? 0) e?2 n 2 2 n + um n + 0 + ( n ? 0) e?2 n 2 2 n + um cos(u ) 1 = hpn 1 + sin(2u m)=2u 2 + u2 K m m n m sin(vm) 1 hpn 00 nm = ? K 1 ? sin(2v )=2v 2 + v2 m m n m and the temperature eld is (by substituting the explicit expression of pn ) : 2 2 4 P X n exp(? nw2=8a2) T (t. HEATING PROCESSES 307 case of coating absorption In the case of heating by dissipation in the coating. We clearly need the following parameters : 1 Z h=2 cos( 0 z) exp( z=a) dz 0 Cnm g0 n m m ?h=2 00 Cnm where n and after some algebra. we have seen that ?2 n n z=a + ? nz=a tn (z) = pn a e? n ( n ? )e )2 e ( + ( )n2e?4 )e n z=a K ( n+ ? n? n h=2a.5. r.m ( n + )J0 ( n ) " cos(um) 0 ?t= 0 0 nm 1 ? e nm cos( m z )? 1 + sin(2um )=2um # sin(vm) 00 00 1 ? e?t= nm sin( 00 z ) J0 ( n r=a) (6.28) m 1 ? sin(2v )=2v nm m m . z) = MC 2 2 2 n.

16. where M = a2h is the mass of the mirror.00 0.6. Short dashed line : Stationary case of g6. one can see the time scale of the evolution of the lens pro le.04 -0. The steady state is reached after hours. The thermal lens. We get the time variable focal length f (t) de ned by 4 2 1 = ? dn Ph X n exp(? n w2=4a2) 2 f (t) dT MCa2 n.e-06 -0. r. z) dz obviously. Long dashed line : beam pro le. de ned by dn Z h=2 Z (t.29) 0 n nm 1 + sin(2um )=2um On Fig. CHAPTER 6.16: Transient thermal lensing in a standard Virgo mirror.06 -0. case of coating absorption. r) = dT ?h=2 T (t. r) = dT MC 2 2 2 n.308 0.06 Radial coordinate [m] Figure 6. only the even part contributes.m ( n + )J0 ( n ) sin(2um )=2um 0 1 ? e?t= nm J ( r=a) 0 (6.02 0.m ( n + 2)J02( n ) .e-06 t = 500 s t = 5000 s -2.e-06 t = 20000 s Stationary limit -3.04 0.6. giving 2 2 dn 2 Ph X n exp(? nw2=8a2) Z (t.02 0. HEATING ISSUES t = 100 s Excess optical thickness [m/W] t = 200 s -1.

5. HEATING PROCESSES 10 6 309 Thermal focal length [W. and spends further time growing uniformly without noticeably changing the gradients.17: Evolution of the thermal focal length (heat source on the coating) sin(2um)=2um 0 1 ? e?t= nm 0 (6.m] 10 5 1s 10 4 1 mn 10 3 1h Steady state focal length = 426. as seen above. m. the focal length takes only a few minutes (see Fig. but is automatically corrected by the servo loops. 6. " # a2 pn 1 ? 2 e? n cosh( n z=a) tn(z) = K 2 + n + ( ? n)e?2 n n . The situation is di erent for the piston. This is due to the fact that the temperature eld reaches soon its nal pro le.17)to reach its stationary value.30) nm 1 + sin(2um )=2um It is interesting to realize that though the heating process takes hours to reach the steady state. case of bulk absorption In this case.W 10 2 10-1 10 0 10 1 Time [s] 10 2 10 3 10 4 Figure 6.6. which follows the evolution of the temperature on a long time scale.

m ( 2 + n )J02( n ) (sin um=um )2 0 1 ? e?t= nm 0 (6.33) nm 1 + sin(2um)=2um Fig.310 CHAPTER 6.32) 0 n nm 1 + sin(2um)=2um Fig. HEATING ISSUES so that. r.6. " # Z h=2 2 0 z ) dz = a h pn sin um ? h cos um tn(z) cos( m 2 2 K n um 2a u2 + n ?h=2 m a2 m 2 = K h pn n sin2u+=u2m um n If we de ne Z h=2 0 = g1 nm 0 ?h=2 tn (z ) cos( m z ) dz m we have.6. The focal length is de ned by 4 2 1 = ? dn h2P X n exp(? n w2=4a2) 2 f (t) dT MCa2 n. show almost exactly the same behavior as in the case of coating absorption. using previous results.m ( + n )J0 ( n ) (6. after susbtitution of the expression for pn : 2 2 1 sin um=um Ph2 n exp(? nw2=8a2) nm = 2 2 ( 2 + 2 )J0 ( n ) 1 + sin(2um )=2um u2 + 2 2 Ka n m n 0 and nally.6.18. using the de nition of nm : 2 2 hP X n exp(? n w2=8a2) T (t. z) = 2MC 2 2 2 n.m ( + n )J0 ( n ) sin um=um 0 ?t= 0 0 nm 1 ? e nm cos( m z ) J0( n r=a) 1 + sin(2um)=2um where M = a2h is the mass of the mirror. r) = dT 2MCP 2 2 2 n. . The thermal lens is 2 2 dn h2 X n exp(? nw2=8a2) Z (t.17 and the same comments apply.19 is almost identical to Fig.31) (sin um=um )2 0 1 ? e?t= nm J ( r=a) 0 (6.

5.02 0.2 Transient thermoelastic deformations . case of bulk absorption. Short dashed line : Stationary case of g6.04 -0. This is the quasi-static regime. The basic elastodynamics equations are the equilibrium equations. that we can neglect the inertial forces.5. HEATING PROCESSES 0.06 -0.e-06 t = 500 s -2.04 0. t = 100 s 311 Excess optical thickness [m/W] t = 200 s -1.02 0. the relevant theory is elastodynamics. We shall start from the expression of the time dependent temperature eld which is never harmonic (even in the case of 6. modi ed in order to take into account inertial forces and generalizing Newton's second law : div = @t2 u The boundary conditions remain the same as in elasticity. It will be assumed for nding the slow evolution of the shape of the mirrors faces.e-06 -0.6. about one m in tens of minutes.00 0.06 Radial coordinate [m] Figure 6.9.e-06 t = 5000 s t = 20000 s Stationary limit -3. Considering motions of matter caused by a constant low rate heating. When time enter elasticity problems. except that the time enters as an evolution parameter through temperature. Long dashed line : beam pro le. the velocities of matter are so small. The equations return to the form of static elasticity.18: Transient thermal lensing in a standard Virgo mirror.

z) J1(knr) uz (t. so that we must take the general solution depending on two arbitrary functions of t. r. r. r. Cn and Dn : A0n + kn Bn = kn Cn cosh(kn z) + kn Dn sinh(kn z) (6.m] 10 5 1s 10 4 1 mn 10 3 1h Steady state focal length = 412. HEATING ISSUES Thermal focal length [W. z) = Pn Bn (t. z) J0(kn r) We know from the study of distortions caused by bulk absorption that the unknown functions An and Bn obey 2 @z2 ? kn (A0n + kn Bn ) = 0 In general. z) = tn(t. we search the displacement vector under the form ( ur (t. and assume that the temperature eld is given under the form X T (t. z) J0(kn r) As in the static study. z) = Pn An(t.34) n . m. there is no symmetry with respect to the meridian plane.19: Transient thermal focal length (heat source in the bulk) coating dissipation) so that we must tackle the thermoelastic equations.W 10 2 10-1 10 0 10 1 Time [s] 10 2 10 3 10 4 Figure 6.312 10 6 CHAPTER 6.

6. involving two more arbitrary functions of t is An(t. h=2) + 2 cosh n Mn + 2 sinh n Pn = +2 " # " # + + + 2 n sinh n ? + 2 cosh n Cn? + 2 n cosh n ? + 2 sinh n Dn 0 t. depending on 4 families of constants to be determined from the boundary conditions.5. namely " # " # + + + 2 n sinh n ? + 2 cosh n Cn+ + 2 n cosh n ? + 2 sinh n Dn 0 2 n (t. HEATING PROCESSES 313 then we are led to solve the di erential equation 2 2 ? k 2 A = kn ( + ) C sinh(k z ) + D cosh(k z )] ? kn tn @z n n n n n n +2 +2 the general solution of which. z) = Mn sinh(kn z) + Pn cosh(kn z)+ + 2( + 2 ) knz Cn cosh(kn z) + Dn sinh(kn z)] ? kn 2n + + where n (t.34 : + Bn (t. These boundary conditions on the circular faces (recall that we neglect conditions on the edge) give a rank 4 linear system. z) = (Cn ? Mn ) cosh(kn z) + (Dn ? Pn ) sinh(knz) + 2( + 2 ) 0 n Cn (cosh(knz) + kn z sinh(kn z)) + Dn (sinh(kn z) + kn z cosh(knz))] + + 2 The stress tensor is now explicitly de ned. z) represents a special solution of 2 @z2 ? kn n = tn then it is possible to deduce Bn (t. z) from 6. + 2 cosh n Mn ? 2 sinh n Pn = 2 n (+ ?h=2) 2 " # " # + + sinh n ? + 2 n cosh n Cn + cosh n ? + 2 n sinh n Dn .

h=2) ? n(t. r. ?h=2)] 2 0 0 e0n (t) = 1 n(t. In the case of absorption in the coating. r) of the mirror's surface (at z = ?h=2) is X Z (t. h=2) + n (t. r) = Bn(t.314 CHAPTER 6. ?h=2) Jn ( n r=a) n case of coating absorption The time dependent temperature eld has been derived in a preceding section.m ( n + )J0 ( n ) . n n h=2a. we found 2 2 4 P X n exp(? nw2=8a2) T (t. ?h=2)] 2 Then the shape Z (t. one can express the displacement amplitude : Bn (t. ?h=2)] 2 0 0 o0n (t) = 1 n(t. ?h=2)] 2 on (t) = 1 n(t. z) = MC 2 2 2 n. HEATING ISSUES ( ? 2 sinh n Mn ? 2 cosh n Pn = ? 2 kn +t. After solving the system and some tedious but elementary algebra. ?h=2) = + sinh( )cosh n ) ? sinh n e0n(t) ? cosh n kn o(t)] ? n cosh( n n sinh n cosh n o0n(t) ? sinh n kn e(t)] + sinh( n ) cosh( n ) + n where we have introduced the even and odd parts of the temperature eld and its gradients : en (t) = 1 n(t. h=2) + n (t. ?h=2) n n n n +2 where as usual. h=2) ? n(t. h=2) 2 " # " # + + ? sinh n ? + 2 n cosh n Cn + cosh n ? + 2 n sinh n Dn 2 sinh M ? 2 cosh P = ? 2 kn n(t.

6 Thermoelastic coupling : Coating absorption Deformation of the wavefront after re ection on a mirror due either to thermal lensing or to distortion of the re ecting coating obviously a ect the tuning of cavities involving such temperature sensitive mirrors. The tuning . THERMOELASTIC COUPLING : COATING ABSORPTION 315 " cos(um) 0 ?t= 0 0 nm 1 ? e nm cos( m z )? 1 + sin(2um )=2um # sin(vm) 00 00 1 ? e?t= nm sin( 00 z ) J ( r=a) (6. the temperature eld is. r) = ? P (1 + 2 )h Ka ( " 0 # X sinh n ( 0 cosh n + n sinh n ) cos2 um 1 ? exp(?t= nm) + pn 2 sinh n cosh n + n 1 + sin(2um )=2um (u2 + n )2 n. z) = 2MC 2 2 n.m ( 2 + n )J0 ( n ) case of bulk absorption sin um=um 0 ?t= 0 0 nm 1 ? e nm cos( m z ) J0 ( n r=a) 1 + sin(2um )=2um and the corresponding apex equation for the time-evoluting surface is hP (1 + )h2 X p sinh n ( 0 cosh n + n sinh n) Z (t.6. 2 2 hP X n exp(? nw2=8a2) T (t. according to the previous section.35) 0 n m 1 ? sin(2vm)=2vm nm using the preceding principles gives the apex equation 2 Z (t.m m " 00 #) sin2 vm 1 ? exp(?t= nm ) J ( r=a) cosh n ( 0 sinh n + n cosh n ) 0 n 2 2 sinh n cosh n ? n 1 ? sin(2vm)=2vm (vm + n)2 In this case.6.m 0 sin(2um)=2um 1 ? exp(?t= nm ) J ( r=a) 2 1 + sin(2um)=2um (u2 + n )2 0 n m 6. r) = ? 2 Ka2 n sinh n cosh n + n n. r.

z) = 0 We separate the space variable by taking T (!. With no internal source of heat.316 CHAPTER 6.36 becomes h 2 2i @z ? t(!. as a feedback loop. r. and the resulting dynamical thermal lens.36) + i !KC T (!.6.6. the second step is to derive the dynamical distortion of the re ecting face. r. and the situation is equivalent to a varying power ux. We are thus faced with the question of time varying power uxes on mirrors. The causes may be technological (an unperfect power stabilization) or fundamental : The absorbed power uctuates due to shot noise. the stored power and thus the heating rate of the mirrors. z)J0(kr) where J0 is the Bessel function. r. The rst step is to study the time dependent temperature eld. Let us rstly evaluate the result from a coating absorption. and its density. z) = 0 Let us recall that C is the speci c heat of the material. and k an arbitrary constant.1 Dynamical temperature Temperature eld We assume here an incoming optical beam having for any reason a time varying integrated power. Eq. z) = 0(!) exp(? z) + 00(!) exp( z) . z) = 0 where The general solution is: s = k2 ? i !KC t(!. 6. z) = t(!. The Fourier-heat equation reads: C@t ? K ] T (t. HEATING ISSUES a ects. We take the time-Fourier transform: (6. K its thermal conductivity.

485. a. z) = 0 (! ) exp(? z) + 00(!) exp( z)] J0(kr) The various arbitrary constants and functions can now be determined by the boundary conditions as usual. n = 1. We shall therefore consider the solution of Eq. We rstly address the condition of an outgoing radiating heat ux o the circular edge (r = a). z) = 4 0T03T (!. a. A consequence of this quantization is that now. 2.6.36 as a sum over all indices n: X 0 00 T (!. z) = (6. We have namely: Za 2 2 2 2 J0( n r=a) J0( m r=a) r dr = a ( +2 n)J0 ( n) nm 2 0 n (see for instance 20] p. r. r.6.36 under the form: T (!. the axial coordinate is such that 0 z h (h is the thickness of the mirror).38) n (! ) exp(? n z ) + n (! ) exp( n z )] J0( n r=a) n where n = s 2 n a2 ? i !KC . z) @r The z dependent factor cancels out from the equation. THERMOELASTIC COUPLING : COATING ABSORPTION 317 where 0 and 00 are two arbitrary functions.4.37 has an in nite family of discrete solutions we note f n. We take the coordinates as follows: The radial coordinate is such that 0 r a (a is the radius of the cylindrical mirror). The absorbing coating is assumed at z = 0. the family of functions J0( nr=a) form an othonormal and closed family of functions on which any reasonably behaviored function (for instance an optical intensity) can be expanded.37) An equation like 6. This yields: ?K @T (!. by setting ka = and = 4 0T03a=K : J1 ( ) ? J 0 ( ) = 0 (6. and we are left with KkJ1(ka) = 4 0T03J0(ka) or.5).6. formula 11.6. :::g. We have now a solution of Eq.

318 CHAPTER 6. p(!. r. The boundary condition is now: 0 00 2 0 3 0 00 n K ( n ? n ) = P (! )pn =2 a ? 4 T0 ( n + n ) The boundary condition on the face z = h is simply: 00 0 ? 00 0 3 0 ? n K ( n e n ? n e n ) = 4 T0 ( n e n + n e n ) where n n h. r) is the Fourier transform of the incoming absorbed intensity ow on the absorbing face. r) = 2Pw!) exp(?2r2=w2) 2 We get: 2 2 pn. We have: 0 ? nz 0 n (z?2h) P !) X T (!. we have: ( I (!. r.40) In the case of a gaussian beam of half width w. except that now some quantities are complex. 0) + I (!. 0) = ? 4 0T03T (!.gauss = ( 2 + 22nJ 2( ) exp(?w2 n =8a2) ) n 0 n . HEATING ISSUES Now we can address the boundary condition on the face z = 0: ?K @T (!. r) @z Where I (!. By introducing the constant 0 4 0T03h=K we get the system: ( 0 ( n + 0) n ? ( n ? 0)00 = P (!)pn h=2 Ka2 n (6. and quite analogous to the one obtained in the static domain.39) 0 00 ( n ? 0 ) e? n n ? ( n + 0 ) e n n = 0 so that the solution is now fully determined. r) = P (!2) pn J0( n r=a) 2 a n where P (!) refers to the integrated absorbed power ow. r) can be expanded in a Dini series: X I (!. r. z) = 2 (Kah pn ( n(+ +)e 0)2 ?+( ( n?? 0)2)e ?2 n J0( n r=a) 2 e n n n (6.

we have: ( 2 I (r) = P (!)= b (r b) 0 otherwise so that 4 pn.20 the distribution of the transfer function jT j=P (!) for f = 0:1 Hz.41) 2 a2 n ( 2 + n )J02( n ) ( n + 0)e? n z + ( n ? 0)e n(z?2h) J ( r=a) 0 n ( n + 0)2 ? ( n ? 0)2 e?2 n In the case of an ideally at beam of half width b. z) = 2PabKh ( 2n+1(2)J 2( ) ) n n 0 n (6. z) dz 0 . is signi cant only in the close neighborhood of the hot spot. We already see that the dynamic temperature eld has a much lower amplitude in the case of a at beam. and on Fig. 6.6. We can see the results for an ideally at mode on Figs.23.21 the same for 1 Hz. The apex equation Z (!. r. at = b( a2 n+J1( )nb=a) ) 2 2 n J0 ( n and the solution is explicitly: (!) X J nb=a TFlat(!. r.6. as the frequency increases.6. Moreover. z) = P (!K (6. r. THERMOELASTIC COUPLING : COATING ABSORPTION 319 so that the solution corresponding to a gaussian beam is explicitly: 2 2 )h X n exp(?w2 n =8a2) TGauss(!.42) ( n + 0)e? n z + ( n ? 0)e n(z?2h) J ( r=a) 0 n ( n + 0)2 ? ( n ? 0)2 e?2 n The temperature eld. as can be seen. Thermal lens The rst e ect of the varying temperature eld is to create a variable thermal lensing. r) = dT T (!. See on Fig. This is: dn Z h Z (!.6. the temperature eld tends to a pure skin e ect. r) giving the lens pro le is obtained by integrating the temperature along the optical path. at frequencies higher than 1 Hz.22.6.

The beam that crosses the mirror substrate undergoes a global length change. one can see the frequency dependence of the modulus transfer function. r) = dT h P (!2) 2 Ka n n n + 0 ? ( n ? 0)e? n ] J0( nr=a) (6. r) I (r) r dr where I (r) is the normalized intensity pro le of the beam. or a gaussian beam of radius w = 2 cm.24. and Z (!) as the spectral density of path length uctuations. This results here in: 2 X p2 (1 ? e? n )( 2 + n )J0( n)2 dn 2 n (6.43) We can nally address the question of e ective length variations. the spectral density of absorbed power uctuation is given by: q P (!) = 2 P0hP 0 . This dependence is clearly in 1=f for frequencies larger than a fraction of a Hz. HEATING ISSUES Where dn=dT is the temperature refractive index coe cient.320 CHAPTER 6. the TF is much lower for a at beam. This gives: X dn 2 pn (1 ? e? n ) Z (!. One sees moreover that the knee frequency is di erent for the two types of mode.44) Z (!) = dT h P (!2) 2 4 Ka n n n + 0 ? ( n ? 0)e? n ] n If we interpret P (!) as the spectral density of absorbed power uctuations.6. as seen in apreceding chapter. we see that the two SD are related by the transfer function: 2 dn 2 X p2 (1 ? e? n )( 2 + n )J0( n)2 n F (!) = dT 4 h 2 2 Ka n n n + 0 ? ( n ? 0)e? n ] n On Fig. given by: Z1 Z (!) = 2 Z (!. it is: ?10 F (f ) 6:7 10 m=absorbedW: f while for a at beam of radius b = 10 cm: ?11 F (f ) 2:7 10 m=absorbedW: f If we assume the power uctuations caused by the shot noise (the power is absorbed by quanta in the coating).

r. instead of 6. we have ! Ca2 2:3 105 K 2 which is to be compared with n n2 2. z) = 2 (Kah pn exp(? z) J0( n r=a) (6.6. and we can write: s = ?i !KC = k(1 ? i) n = and the same way = kh(1 ? i) n = q ! C=2K . If we consider the paramters s 2 n ?i ! C n = a2 K we see that in general. is never reached. i. and we have simwhere k ply.e. Namely. Expressions like exp(? h) vanish. coating thermal losses of about 1 ppm: 4 10?20 m:Hz?1=2 Z (f ) f This represents the optical path uctuations by passing for instance through the Fabry-Perot input mirrors.40: P !) X T (!. We assume the Silica parameters already given. but should be reexamined in an advanced detector with high recycled power.6. 20 kW. even for a frequency of 10 Hz. the second imaginary contribution will be much larger than the real one. THERMOELASTIC COUPLING : COATING ABSORPTION 321 where P0 is the nominal power of the incoming beam (even highly stable!). The exact model presented above can be hugely simpli ed in some realistic cases. and a power of the order of magnitude of that stored in the long cavities. If the pn are rapidly decreasing (as in the case of a gaussian beam). the index at which the real contribution becomes non negligible compared to the imaginary. This allows to give an order of magnitude for the optical path uctuations caused by the shot noise. and the losses in the mirror due to thermal dissipation.45) 2 n Asymptotic solution . This is negligible in the present con guration of Virgo.

47) 2 and we see explicitly the dependence in 1=f . but the accuracy is su cient for further purposes. we have Z IFlat(r)2r dr d = 1 2 b so that the formula is the same with w replaced by b. but normalized to 1 W. r) = dT K1 2 I (r) For the e ective displacement: Z dn Z (!) = dT K1 2 I (r)I0(r)2r dr d where I0(r) is the same intensity pro le. .40 and 6. in the case of a gaussian beam of half-width w. there are some di erences due to the weak decreasing rate of the pn in this case.46) Recall that I (r) is the absorbed intensity pro le. A simpli ed version of the thermal lens immediately follows: dn Z (!.Gauss(r)2r dr d = w2 so that dn ( Z (!) = i dT wPK!)C! (6.322 and due to the fact that CHAPTER 6. z) = K I (r)e? z (6. we have: Z 1 I0. Numerical tests show that there is no di erence between 6. whereas in the case of a at beam of radius b. r. HEATING ISSUES X I (r) = P (!2) pn J0( n r=a) 2 a n we have nally 1 Tasymp(!.46 in the case of a gaussian beam. In the case of an ideally at beam. and we see that the optical path uctuations are reduced by a factor of (w=b)2.

r. resulting in surface cutuations. z) = Pn Bn (!. z) = Pn An (!. z) J0(kn r) The elastodynamics equations reduce to i h 2 2 @z ? kn + !2 An ? kn ( + )(@z Bn + kn An) ? tn ] = 0 (6.6. as an in nite medium. We shall therefore consider the mirror in this regime.6. We have thus: Bn = Mn e? T. the temperature eld is negligible outside a thin neighborhood of the beam spot. THERMOELASTIC COUPLING : COATING ABSORPTION 323 The obvious other e ect of a uctuating temperature eld in the substrate is to induce uctuating distortions in the bulk.48) i h 2 2 @z ? kn + !2 Bn + @z ( + )(@z Bn + kn An) ? tn] = 0 (6. and consequently to a uctuation of the e ective position of the mirror. z) J1(kn r) uz (!.n z ? @zkAn n n 6.49) from what we get 2 @z2 ? kn + !2= (@z An + kn Bn ) = 0 We have seen in the preceding section that even for frequencies as low as a few Hz. We assume the displacement vector to decrease exponentially.n z where s !2 2 T. r. z) J0(kn r) The temperature eld being given under the form X T (!. r. and we take the solution of the preceding equation as: @z An + kn Bn = k Mn e? T. We consider again the elastodynamics equation: div = @t2u We take as usual the displacement vector under the form (after a Fourier transform) : ( ur(!.n = kn ? is the transverse elastic wave vector.6.2 Dynamical thermal surface distortions . z) = tn(!.

50) k T.n z + 2k T. the only boundary conditions are the vanishing of the axial pressure on the heated surface.e.n ( + )Mn e? T.n z ? ( + n2 )( (2 + )2 ) Mn e? T.52) Xn Having An.324 and by substituting in 6. i.51) + T.n ( +2 ) so that: An = Qn e? L.n Bn = k Qn e? L.n s !2 2 = kn ? + 2 L.n ? L. r.53) + n n The medium being assumes in nite.n z + 2X Mn e? T. z = 0) = zz (!.n ? kn 2n X + with the notation n n where (6.zz = ? tn + kn An + ( + 2 )@z Bn .48 we obtain: h i 2 ( + 2 ) @z2 ? kn + !2 An = ?kn The solution of which is: CHAPTER 6. But + 2 ? 2 = ? !2 T. We have n. HEATING ISSUES T.n An = Qn e? L.n z + @z 2n (6. z = 0) = 0 (6.n z ? k tn (6. r.n L. we can calculate Bn : !2 2 2kn 1 L.n z ? kn 2n (6.: rz (!.n is the longitudinal elastic wave vector.54) this leads to two equations allowing to determine the Qn and the Mn .rz = (@z An ? kn Bn ) and n.

57) L.n Q = ? n 2Xn kn n +2 kn @z n (0) 325 (6.56) 2 + =kn (1 Xn L.58) L.n T.n T. z = 0)J0(kn r) n We have: 1 L. Recall that 2 2 Xn = 2! a2 n The largest Xn is obviously X1.n 2kn Xn Mn + (1 ? Xn ) Qn = +2 (1 ? Xn )kn n (0) 1?Xn M + L.n @ (0) + (1 ? X )2 k (0) n Qn = ? + 2 kn z n =k2 ? (1 ? X n)2n (6.n =kn ? (1 ? Xn ) = Xn n +2 .n and T. THERMOELASTIC COUPLING : COATING ABSORPTION We get the system: ( T.n Bn(!.n T. in which 1 1. this is. at a frequency of 10 Hz. We have: + + O(X 2) 2 2 L. For the silica parameters.n n n Our target is the displacement of the surface. 0) = 2X Mn + k Qn + + 2 @z n(0) n n by substituting the values found for the Qn and the Mn . 0) = + 2 Xn (1 ? Xn ) =kz2 n(0) + X )2n(0)] (6.n Bn(!.n ?(0) )2 (6.55) The solution of which is: ? n Mn = 2 Xn (1 2 Xn ) @z n (0) +?L. after some algebra: @ L. It is easily seen that the parameters Xn are very small in realistic cases. we have X1 4 10?6 it is therefore quite allowed to compute Bn at the lowest order in Xn .6.n n ? (1 ? n Let us now consider some gures. or in other words the function X Bn (!.n T.6.

0) = (6.n n(0)] Now. we have: )P B (!. we obtain: ( ) Bn(!.n = 2 . in the asymptotic regime: P !) ? nz tn(!.n and nally.59) + @z n(0) + L. We have namely: @z n (0) + L. For the e ective displacement.326 so that: CHAPTER 6. we have: Z Z (!) = Z (!. z) = 2 (Kah e + 0 2 ? 2 pn n 2 P !) 1 = ? 2 (Kah + 0 +1 pn 2 n n L. 0) = ?i + 2 P 2!C! pn a If we express this in terms of the linear dilatation coe cient and the Poisson ratio . z) = 2 (Kah e + 0 pn 2 n we have P !) ? n z 1 (!. The surface distortion is thus proportional to the temperature eld. r) I (r) r dr d 4 10?6 fn n By keeping only the leading terms. for the surface apex equation: Z (!.60) a L. r) = ?i 2 (1 +C!)P (!) I (r) where I (r) is the absorbed intensity. 0) = ?i (1 +2 C!(!) pn (6.n Recall now that the heat wave vector n is very large compared to the elastical ones.n n (0) so that n n L. if we introduce the temperature eld found in the preceding section. HEATING ISSUES Bn(!.

The intensity being constant along z.1 Dynamical temperature We again assume an incoming light beam of power P (!. As usual. z) = P (!).61) C! For a gaussian beam of half-width w.7 Thermoelastic coupling : Bulk absorption The same work can be carried out in the case where the incident power is dissipated in the bulk material. and the axial coordinate is ?h=2 z h=2. the mirror is assumed to have a radius a and a thickness h.62) C! to be speci c. in order to bene t from the symmetry of the problem. and normalized intensity pro le I (r).6.7.7. 6. The radial coordinate is r a. the formula is the same. We have thus a new de nition of the axial coordinate. 6. with w replaced by b. the transfer function from the power variations to the mirror displacement is: (1 Z (!)=P (!) = ? 2i w2 + ) (6. As usual. in the case of silica. this is ?10 Z (f )=P (f ) 2 10 m=W f In the case of a at beam of radius b. We follow the same scheme as in the preceding section. we get a temperature eld which generates a thermal lens and a distortion of the solid. so that its intensity is assumed constant with respect to z : P (!. As usual in this chapter. The beam is weakly absorbed during its crossing the mirror substrate. for w 2 cm. z) (either a Fourier component or a spectral density). the normalized intensity pro le can be expanded on the basis Temperature eld . the result is that the temperature eld will be symmetrical with respect to z. THERMOELASTIC COUPLING : BULK ABSORPTION or 327 2i (1 + )P (!) Z I (!. r)2r dr d Z (!) = ? (6.

we can write: X T (!. z) = ? 2 PKa) pn (6. These conditions are the vanishing of heat ows on the faces and on the edge.64) n 2 where q 2 = kn ? i C!=K The z-symmetrical solution is obviously: P( tn(z) = An cosh( nz) + 2 Ka!) 2 pn (6. it obeys the inhomogeneous Fourierheat equation: i! C ? K ] T (!. In order to separate the variables. We have: X I (r) = 2 1a2 pn J0(knr) n If we note T (!. We have exchanged the partial di erential equation for a set of di erential equations: (! (@z2 ? 2 )tn(!. z) J0(kn r) n where the functions tn(!. where the family of constants kn are to be determined. we get the condition ?K @T = 4 0T03 T @r (the notations are the same as throughout all this chapter). On the edge.63) where is as usual the linear absorption coe cient (m?1). r.65) 2 n and the arbitrary constants An are to be determined by the boundary conditions. z) the temperature eld. z) = tn(!. HEATING ISSUES of the Bessel functions J0(kn z). r. z) = P (!)I (r) (6. namely: kn = n =a where the n are the zeroes of the equation: J1 ( ) ? J 0 ( ) = 0 ( 4 0T03a=K ) . r. z) remain to be determined. This gives the same equation as in the preceding section and determines the kn.328 CHAPTER 6.

h=2a (6. owing to the fact that 1: (6. THERMOELASTIC COUPLING : BULK ABSORPTION 329 so that the pn are the same as in all preceding sections. thanks to symmetry to one condition on the face z = h=2: ?K @T = 4 0T03 T @z and the result determines the An: 00 P An(!) = ? 2 K (!)pn 2 a2 n where the notation being: n n h=2 n sinh 00 n 1 + 0 cosh n (6. the n are almost all equal: q ?i!C =K = ) n h=2 n so that the temperature eld reproduces the same pro le as the beam intensity: " # 00 cosh( z ) P (!) 1 ? T (!. r) = dT PC!)h I (r) .67) Again.69) (6.68) dn (! Z (!.7. z) = 2 Ka2 n 2 n # 00 cosh( n z ) J0( n r=a) n sinh n + 00 cosh n Thermal lens By integrating along z. we note that for frequencies larger than a few Hz.66) .6. r. the temperature eld is: " P (!) X pn 1 ? T (!. we nd the thermal lens: " # 00 sinh dn P (!)h 1 ? Z (!. r) = dT K 2 ( sinh + 00 cosh ) I (r) or simply. z) K 2 sinh + 00 cosh I (r) And nally. Now the conditions on each circular faces reduce. r.

with w replaced by b. z) = X n Bn (!.6.47. the two contributions to thermal lensings are equal. z)J1(kn r) (6. In the coating absorption case.72) n and uz (!. z) an odd function of z.71) C! This is identical to eq. Now. the formula is the same.330 CHAPTER 6. r. In the case of the Silica parameters. z) = tn(!. r) r dr d (6. and Bn(!. the transfer function is: " # Z (!)= hP (!) 6:7 10?10 1Hz m=absorbedW f which gives: 6. in the case of a gaussian beam of half-width w. r. the temperature eld was localized in the neighborhood of the hot spot.7. the temperature eld extends throughout the mirror. z) J0(kn z) n . z)J0(kn r) where An(!. which shows that for equal absorbed power in the coating and in the bulk.70) dn P Z (!) = i dT w2(!)h (6. z) = An(!. The temperature eld is assumed expanded as: X T (!. z) is assumed an even function of z. and write the displacement vector as: X ur (!.2 Dynamical thermal distortions General solution We have to carry out the same calculations as in the coating absorption case. HEATING ISSUES The e ective length of the path in the substrate is as usual the average of the thermal lens weighted by the intensity pro le: Z Z (!) = I (r)Z (!. In the case of a at beam of radius b. except that the de nition of pm is changed. and that the symmetry is di erent. r. We take the same coordinate system as above.

We therefore consider only boundary conditions on the two circular faces of the mirror (in nite slab of nite width).n( 2 +? ) 2 ) Qn cosh( L.6. THERMOELASTIC COUPLING : BULK ABSORPTION 331 In all the following calculations.6.n z ) ? kn n +2 @z n +2 (6.n and we choose the odd solution: @z An + kn Bn = kQn sinh( T.74) 2 !2=2kn .6. owing to eq. The condition rz = 0 (z = h=2) n h= = @z +(2 2) yields the equation: L.n (6.n 2kn Xn Qn cosh( T. allowing to determine the Qn and the Mn . there are only two equations.n z ) ? T.73.7. z) is assumed a particular solution of (@z2 ? 2 ) L.n z ) (6. The elastodynamic equations are identical to eq. and due to symmetry.n )( T.n n T.49. An = Mn cosh( L.48. the kn are the same as in the preceding section.nz ) + (6.48 and 6. this gives in turn: 1 L.77) .n By using again the notation Xn we get: T. for frequencies higher than a few Hz. in practice.73) and by substituting in eq.76) = t We have seen that the temperature eld. and taking the even solution. has the same pro le as the beam's intensity.75) and. We nd again that: (@z2 ? 2 )(@z An + kn Bn) = 0 T. we get: (@z2 ? 2 )An = ( +k2 T.6.n L.n z ) ? kn tn +2 (6.n ? ? 1 2XXn Qn sinh n T.n kn Mn sinh L.nz) + 2X Qn sinh( n n where (!.n Bn = ? k Mn sinh( L.

n (6. we get: 0 Qn = Xn (1 ? Xn ) ( + 2 )D cosh L.n cosh T.n n ? cosh L.n 2kn Xn CHAPTER 6.332 and the condition yields the equation (1?Xn ) Mn cosh with the notation is: L. After some straightforward algebra.n n n L.n Mn ? 2X sinh T.78) Qn cosh T.81) 0 sinh L.77.n n] 0 The constants n and n are derived from the preceding section: " # 00 cosh n P (!) pn 1 + 1 n = ? 2? 2 00 2 K 2 a2 2 n L.n Bm(!.n kn n (6.n sinh L.L. The solution of the system (6.n T.n n] (6.n ? (1 ? Xn ) 2 cosh L. owing to the symmetry). (or h=2 as well. ?h=2) = k sinh L.n n sinh n + cosh n and 00 sinh n P (!) pn n 0 = ? n 2 K a2 ( 2 ? 2 ) sinh + 00 cosh L.80) n But we are interested in the displacement of the surface z = ?h=2.n . ?h=2) = ( + 2 )D Xn (1 ? Xn ) sinh T.n ? L.n 0 Mn = ( + 2 )D k cosh T.78) T.79) n n 0 (we have written n for @z n(h=2). and n for n(h=2) ) we have also set: Dn = The same way.T. it comes: Bn (!.n n L.6.n sinh T.n n ? (1 ? Xn )2 sinh T.n zz T.n n ? L. We need: 0 1 L.n h=2.n L.n 2 kn sinh L.n Qn ? + n 2 n n 0 (this because @z n(?h=2) = ? n). HEATING ISSUES = 0 (z = h=2) n h= = (1?Xn ) kn +(2 2) (6.n n n n n T.

leads to: P ) ? 2 K (!apn n 2 2 L and P (!) pn 00 00 1 0 ? 2 K 2a2h n 0 so that in the combination involved in Bn(!. At the lowest order we nd: + Dn = + 2 Xn (xn + sinh xn cosh xn) where xn kn h=2.83) xn n n n) For an input Virgo mirror.n. the surface pro le does not reproduces the intensity pro le.6. n can be neglected in regard of n. the same considerations on the order of magnitude of n that becomes practically independent of its index at frequencies larger than a few Hz. An example of the surface pro le is shown on g.26. By taking the average of the surface weighted by the intensity pro le. and much larger than T.L. we get the e ective displacement. we nd the asymptotic apex equation: (1 + ) P (!)h X sinh2 xn pn J0( n r=a) Z (!. THERMOELASTIC COUPLING : BULK ABSORPTION 333 Asymptotic regime Now.6. where it can be seen that contrarily to the previous asymptotic cases. ?h=2). r) = ? 2 K 2a2 n xn (xn + sinh xn cosh xn ) (6. The transfer function from the absorbed power P (!)h to the e ective displacement Z (!) is: 2 2 J 2( n sinh2 2 + X n 0 Z (!)= hP (!) = ?i 4 (1C!a)2 (1 + (x= +)sinh x ) cosh xxn pn (6.84) f .82) The di erence between the exact calculation and the preceding asymptotic formula is negligible for frequencies lager than 1 Hz. We also remark that the parameters Xn are very small. The asymptotic expression for Bn is now: P (!)pn sinh2 xn Bn ? + 2 K 2a2k x + sinh x cosh x n n n n Finally. this is: ?11 Z (f )= hP (f ) = 1:3 10 m=absorbedW (6. so that an expansion of the various expressions is needed.7.

we get: ?13 Z (f )= hP (f ) = 8:8 10 m=absorbedW f . HEATING ISSUES (6.85) and for a at beam of radius 10 cm.334 CHAPTER 6.

175 335 0.1 Hz.10 Figure 6.035 0.20: Opto-thermal transfer function: f = 0.05 0.035 -4.070 0.7.000 0.140 0.91 -0.070 -9.00 -20. case of gaussian beam w = 2 cm .140 -14.97 -0.12 -0.105 0. THERMOELASTIC COUPLING : BULK ABSORPTION 0.6.00 0.175 0.105 -0.94 -0.

336 0.175 CHAPTER 6.09 -0. HEATING ISSUES 0.05 0.00 -20.000 -0.10 Figure 6.19 -0.28 -0.140 -15.105 0.140 0.00 0.21: Opto-thermal transfer function: f = 1 Hz.070 -10.175 0.070 0.035 0. case of a gaussian beam w = 2 cm .105 -0.035 -5.38 -0.

070 -10.05 0.39 -0.1 Hz.6. case of a at beam.7. THERMOELASTIC COUPLING : BULK ABSORPTION 0.175 0.140 -15. b = 10 cm .175 337 0.17 -0.070 0.78 -0.00 -20.10 Figure 6.22: Opto-thermal transfer function: f = 0.000 -1.00 0.035 0.035 -6.105 -0.105 0.57 -0.140 0.

175 CHAPTER 6.00 -20..55 -0. HEATING ISSUES 0. b = 10 cm .52 -0. case of a at beam.140 -15.035 -6.00 0.105 0.03 -0.10 Figure 6.05 0.070 -11.105 -0.140 0.23: Opto-thermal transfer function: f = 1 Hz.338 0.000 -2.07 -0.175 0.035 0.070 0.

7.(absorbedW)-1] 10-7 10-8 10-9 10-10 10-11 10-12 10-13 10-6 10-5 10-4 10-3 10-2 10-1 10 0 10 1 10 2 frequency [Hz] Figure 6. b = 10 cm . THERMOELASTIC COUPLING : BULK ABSORPTION 339 10-5 10-6 transfer function [m.6.24: Opto-thermal transfer function. w = 2 cm. Solid line: gaussian beam. Dashed line: at beam.

Solid line: asymptotic solution. Dashed line: intensity pro le for comparison . -0.10 -0.05 0.05 -0. HEATING ISSUES 2.340 CHAPTER 6.10 radial coordinate [m] Figure 6.08 -0.08 0.25: Bulk absorption.00 0. Surface distortion at 1 Hz for a gaussian beam of width w = 2 cm.02 0.e-11 m/absorbedW 1.02 0.e-11 0.

10 0.00 0.15 -0.10 -0. -0.15 0.26: Bulk absorption. THERMOELASTIC COUPLING : BULK ABSORPTION 341 1. Solid line: asymptotic solution.7.05 0.e-13 0. Surface distortion at 1 Hz for a at beam of radius b = 10 cm.6. Dashed line: intensity pro le for comparison .20 radial coordinate [m] Figure 6.e-12 m/absorbedW 5.20 -0.05 0.

342 CHAPTER 6. HEATING ISSUES .

the state of the body can be represented by a linear superposition of all the modes. A re ecting face can move either because it is displaced by its suspension system. and. At nite temperature. of a noise. At thermal equilibrium. there is a discrete in nity of such stationary waves. and if we consider the re ecting face of a mirror. the two e ects are possible. re ecting on the faces and the onset of stationary waves. in other words. We address here the internal stresses. Consider a massive body at temperature T . 343 . the same energy kB T (kB is the Boltzmann constant). Estimation of the resulting spectral density of phase noise is the internal thermal noise problem in massive mirrors.Chapter 7 Mirrors standard thermal noise standard thermal noise is the phase noise caused by random motions of the re ecting faces of mirrors in a GW interferometer. If T > 0. The motion of atoms near a limiting surface of the body will slightly modify its shape. due to the energy equipartition theorem. with random relative phases. or because it undergoes internal stresses. for a nite body (like for instance a cylinder of silica). each corresponding to a particular elastic normal mode. The fact that they are strongly coupled to neighboring atoms makes possible propagation of elastic waves of various types. the atoms constituting the body are excited and have random motions around their equibrium position. a surface distortion is a possible cause of phase change in the re ected beam. One can show that.

the result is still much too high. we couple the oscillator to the heat bath : the driving force expresses action of the external world on the oscillator. whereas the damping factor releases the received energy. Assume for instance a frequency of 2 1000 Hz and an equivalent mass of 10 kg (in fact the masses equivalent to modes are even smaller). the displacementlike variable x is. each being equal to kB T=2 (kB 1:38 10?23 J:K?1 is the Boltzmann constant). Even by cooling at very low temperature (say 0. If we introduce simultaneously a random driving force (Langevin force) F (t) and a damping factor accounting for dissipation. at room temperature much larger than GW induced (xgw 10?18 m displacements. this seems to de nitely forbid any GW detection. but the amplitude is a scalar x obeying a dynamical equation analogous to the harmonic oscillator's. this is : d2x + !2x = 0 dt2 0 where !0=2 is the eigenfrequency of the mode. For the potential energy. we have 1 2 EP = 2 m!0 x2 by taking the expectation value. At thermal equilibrium with the environment (the heat bath). its geometry and its amplitude.01 K).344 CHAPTER 7. Determination of the eigenfrequencies and of the eigenmodes of an arbitrary body is in general di cult. and assuming a zero mean of x.1 Damped harmonic oscillator Each internal mode is characterized by its eigenfrequency. MIRRORS STANDARD THERMAL NOISE 7. so that the energy of the oscillator is statistically stationary. the amplitude follows a random walk so that the potential and kinetic energies have equal means. If we consider the decoupled and undamped oscillator. we get a standard deviation (x) = 3 10?15 m At rst sight. In fact this is not true if we take into account the frequency distribution of the noise. this gives kB T V (x) = m!2 0 It is important to understand that though very small. The motion equation is (case of viscous damping) : d2x + dx + !2 x = F (t)=m dt2 dt 0 .

1. its value can be determined by requiring that Z1 kB T Sx(f ) df = V (x) = m!2 0 0 We have obviously Z1 1 Z 1 S (!) d! Sx(f ) df = 2 ?1 x 2 0 so that Z1 d! 4 kB Tm SF 2 ? ! 2 )2 + 2! 2 = (! !2 For carrying out the integration.7. DAMPED HARMONIC OSCILLATOR By taking the Fourier transform. it is convenient to set ?1 0 0 2 (!2 ? !0 )2 + 2!2 = (!2 ? 2)(!2 ? 2) q 2 2 = !0 ? 2=2 + i !0 ? 2=4 so that q 2 = !0 ? 2=4 + i =2 then the integral can be splitted into two terms. giving Z1 1 1 d! 4 kB Tm SF 2 2 ? 2 ? !2 ? 2 2? 2 = !0 ?1 ! The Cauchy theorem gives (provided that > 0) : Z 1 d! i 2? 2 = ?1 ! 2 where . this is 345 ~ ) 1 x(!) = m !2 ?F (2!+ i ! ~ 0 ! The relation between the spectral densities of x and F must therefore be : 1 Sx(f ) = SF (f ) m2 (!2 ? !212 + 2!2 0) SF is a constant (white noise) .

7. which becomes more and more narrow.1. saying that for the elementary dynamical system described by a degree of freedom x and a driving thermal force F . one sees the general philosophy of thermal noise. This is why high-Q material and xations are searched for. On Fig. in GW experiments. we can de ne a velocity v = i!x. The integral of the spectral density Sx is independent on Q. but by increasing Q. and the root spectral density is 1=Q the value at resonance. Then Sx(f ) = 4kB2T <e Z ] (7. and reduce the thermal noise outside the resonance.2 The FD theorem There is a more general derivation of the spectral density.7. due to Callen and Welton 26]. MIRRORS STANDARD THERMAL NOISE 2 so that the integral reduces to = !0 .346 CHAPTER 7.3) ! . Heavy test masses and low temperatures have been also obviously proposed a number of times.2) The main features are: 4kB T ! ! 0 ) Sx(f ) ! mQ!3 0 kB ! ! !0 ) Sx(f ) ! 4m!Q 3 0 kB T! ! ! 1 ) Sx(f ) ! 4mQ!40 so that the spectral density is a constant for low frequencies.1) 0 + The mechanical quality factor Q is de ned as Q = !0 = (7. and a mechanical impedance Z = v=F . based on the Fluctuation-Dissipation Theorem. we can concentrate the SD in the neighborhood of the resonance. and the nal result is SF = 4kB Tm The spectral density of displacement is nally: 4 BT Sx(f ) = (!2 ?k!2)2 =m 2!2 (7.

1 follows directly. But this approach allows to obtain results more di cult to derive by other means. A very simple model of thermoelastic dissipation is given by a complex elastic sti ness.1: sqrt of spectral density of thermal displacement : viscous damping In the preceding case. For instance. dissipation of the elastic energy is not caused by viscosity. for instance. the motion equation being in some frequency domain: h 2 2 i ~ ~ ?! + !0 (1 + i ) x(!) = F=m . if we consider a solid resonator. as for instance a mirror substrate. but rather by thermoelastic processes: stressed regions are heated.2.7. we had !=m Z (!) = !2 ?i!2 + i ! 0 from where 7. THE FD THEOREM 10-12 347 10-13 SD of displacement [m. and there is a heat ow from hot to cold regions due to nite thermal consuctivity leading to irreversibility.Hz-1/2] 10-14 Q = 10 10-15 10-16 Q = 1000 10-17 10-18 1 10 frequency [Hz] 100 Figure 7.

7. by using the FD theorem (eq. because we know that the eq. causing a phase noise. associate such a model to each corresponding mode (the question of the e ective mass of the mode is raised).3 The Levin generalized coordinate method We can now address the problem of internal degrees of freedom in the mirrors. 4kB T ! ! 0 ) Sx(f ) ! mQ!!2 0 kB ! ! !0 ) Sx(f ) ! 4m!Q 3 0 kB T!2 ! ! 1 ) Sx(f ) ! 4mQ!50 (see Fig. Anyway. MIRRORS STANDARD THERMAL NOISE where is the so-called loss angle. 7. often considered as independent on the frequency. It is essential to note the very di erent behavior of this thermoelastic spectral density with respect to the viscoelastic. each being viewed as a thermoelastically damped harmonic oscillator. Internal elastic waves eventually distort the re ecting surface. we get 1 T!2 Sx(f ) = 4kBm! 0 (!2 ? !2)2 + !4 2 0 0 This formula clearly holds above some cut-o frequency.348 CHAPTER 7.2) This is a common behavior for all internal modes of solid resonators. However. We have already discussed the way of obtaining the information on . we can no more use the direct approach of integrating over frequencies to recover the variance.3). It is possible to numerically compute resonance frequencies of a cylindrical solid (as the mirror substrates). It may be seen as the inverse of the quality factor. We have thus: ~ x(!) = !2 ? F=m i !2 ~ (7. the increase of the thermal noise at low frequency is presently the main limitation to GW detectors.7.4 is only valid in some frequency domain.7. and sum up to nd the global noise.4) 2 0 ! + 0 In order to determine the function F .

x. y) is the normalized light intensity distribution in the TEM00 mode assumed to be the readout beam. THE LEVIN GENERALIZED COORDINATE METHOD 10-12 349 10-13 SD of displacement [m.Hz-1/2] 10-14 Q = 10 10 -15 Q = 1000 10 -16 10-17 10-18 1 10 frequency [Hz] 100 Figure 7. This case is . x. The interaction energy is E = ?F (t) x(t) or ZZ E = uz (t. y) be the z component of the displacement vector of matter at the surface of the mirror. y) I (x.3. We address now the case of low frequencies. y) F (t) I (x. y) dx dy where I (x. x.7.y) dx dy where the displacement u may be thought of as beeing caused by the pressure distribution F I . The equivalent displacement (generalized coordinate x) is ZZ x(t) = uz (t.2: sqrt of spectral density of thermal displacement : thermoelastic damping the surface relevant for the beam. Let F (t) be the corresponding driving force. Let uz (t. We now follow the method proposed by Levin ( 24]).

5) f . it is nevertheless extremely interesting to have the low frequency tail. W is proportional to F 2. The strain energy is de ned in classical elasticity theory by 1ZZ W = 2 uz (x. y) = (1 ? i )uz (x. y) dx dy Z (f ) = i! F so that RR <e Z ] = ! uz (x. so that U W=F 2 is the strain energy for a static pressure normalized to 1 N. the frequency will be lower than the cut-o for any standing waves. In the Fourier domain. x. y) is the pressure distribution causing the displacement uz (x. in the low frequency regime. of the form uz (t. Thus. although a general knowledge on internal thermal noise is useful. This can be obtained as follows. y) the impedance is (1 ? i ) R R uz (x. y) dx dy where p(x. x. the pressure F I will produce an oscillating stationary displacement u. The phase represents a retardation e ect that dissipation may cause. If we consider a force F (t) = F ei!t oscillating at very low frequency. y) F:I (x. MIRRORS STANDARD THERMAL NOISE very relevant. y) this is equivalent to neglecting inertial forces in the motion of matter. The SD of displacement takes the general (low frequency) form : Sx(f ) = 4kB T U (7. because resonances of mirrors are at relatively high frequencies (several kHz) and the region where internal thermal noise is disturbing lies long before the rst resonance. We can thus write for the spectral density of displacement : W Sx(f ) = 4kB T F 2 f in fact. y) I (x. this is uz (!. y) dx dy F2 where the numerator of the fraction appears as the elastic energy stored in the solid stressed by the pressure distribution F:I . y) at the surface where it is applied. y)p(x.350 CHAPTER 7. y) = ei(!t? )u(x.

7. BASIC LINEAR ELASTICITY 351 The problem is reduced to the computation of U .4.4. 7. They are related to the Young modulus Y and the Poisson ratio by = (1 + Y ? 2 ) )(1 = 2(1Y ) + i=1 Eii . The strain tensor Eij is de ned as Eij = 1 (@iuj + @j ui) 2 Its trace is E= 3 X The stress tensor ij is linearly related to the strain tensor in a way generalizing Hooke's law.4 Basic linear elasticity We recall here the principles and master formulas of the linear elasticity theory. the relation is very simple : ij = ij E + 2 Eij the two parameters ( . 7.1 displacement. strain. It is however possible to obtain analytic solutions in the case of axial symmetry. This can be di cult in the general case of an arbitrary solid.z) coordinate system by its reference state. For isotropic solids (like silica). ) are called Lame coe cients. and its deformed state xi ! xi + ui(xk ) The vector u is called displacement vector.y. stress Let a solid be decribed in the (x. but numerical nite element codes are able to give more or less accurate estimates.

352

CHAPTER 7. MIRRORS STANDARD THERMAL NOISE

7.4.2 Elastodynamics equation
The elastodynamics equation is :

@j

ij

= @t2ui
ij

which in the static case reduces to the equilibrium equation

@j

= 0

In polar coordinates (r; ; z), the strain tensor has coordinates

Err ; Er ; Erz ; E ; Ez ; Ezz Err = @r ur 1 Er = 2 @ru ? ur + 1 @ ur r Erz = 1 (@r uz + @z ur ) 2 E = 1 @ u + ur r r Ez = 1 1 @ uz + @z u 2 r Ezz = @z uz The elastodynamics equation reads in detail 8 1 1 > @r rr + r ( rr ? ) + r @ r + @z rz = @t2ur < 2) 1@ +r + @z z = @t2u > (@r + r ) r + 1 @ : (@r + 1 rz z + @z zz = @t2 uz r r
de ned by :

(7.6)

In the special case of static axial symmetry, the system reduces to the equilibrium equation :

(

@r rr + 1 ( rr ? 1 r (@r + r ) rz + @z

zz

) + @z = 0

rz

= 0

(7.7)

7.5. MIRROR AS A HALF-SPACE

353

7.4.3 Boundary conditions
j

The boundary conditions express the balance between internal stresses and external pressures at the limiting surfaces : X ij nj ] = pi

where ni is the normal to surface

7.5 Mirror as a half-space
If the spot of the readout beam on a mirror is centered and small compared to the mirror's dimensions (radius, thickness), we can consider the substrate as an in nite half-space limited by a plane (the optical curvature is negligible here). The problem obeys the axial symmetry and it is easy to verify that there is a solution of 7.7 of the form : ! + 2 + kz e?kz J (kr) ur (r; z) = ? + 1 ! uz (r; z) = + + + kz e?kz J0(kr) where ( ; ; k) are arbitrary constants. The Jn are the Bessel functions. The region occupied by the substrate is supposed to extend from z = 0 till in nity. The boundary conditions are
rz ]z=0

= 0

= p(r) where p(r) is the gaussian pressure having the beam's pro le and normalized to 1 N (the integral over the whole plane of a pressure is a force): 2 p(r) = w2 e?2r =w It is easy to compute the stresses :
zz ]z=0
2 2

and

rz

= 2 k ( ? ? kz)J1(kr)

354

CHAPTER 7. MIRRORS STANDARD THERMAL NOISE
zz

= ?2 k ( + kz)J0(kr) The rst boundary condition gives = . The solution depends now on two arbitrary constants ( ; k). In fact the most general solution will be an integral over k : ! Z1 ur (r; z) = (k) ? + kz e?kz J1(kr)k dk + 0 ! Z1 + 2 + kz e?kz J (kr)k dk uz (r; z) = (k) 0 + 0 and now, (k) refers to an arbitrary function of k. The zz stress component becomes : Z1 (k)J0(kr) k2 dk zz (r; z = 0) = ?2 0 so that the last boundary condition becomes Z1 (k)J0(kr) k2 dk = ? 21 p(r) (7.8) 0 This expresses a Bessel transform. Recall that for functions admitting a Fourier transform, the two reciprocal Fourier transforms become, for axially symmetrical functions, reciprocal Bessel transforms : Z1 f~( ) = J0( r)f (r) r dr

f (r) = 0 J0( r)f~( ) d we have thus, inverting the Bessel transform in 7.8 : 1 Z 1 p(r) J (kr) r dr k (k) = ? 2 0 0 It is possible to carry out the integration (see 20], Eq. 11.4.29), obtaining (k) = ? 4 1 k e?k w =8 and consequently a displacement + 2 1 Z 1 e?k w =8 J (kr) dk uz (r; z = 0) = ? ( + ) 4 (7.9) 0 0
2 2 2 2

and

Z1

0

7.5. MIRROR AS A HALF-SPACE
0.0 -0.1 -0.2 -0.3

355

uz(r,z=0)/uz max

-0.4 -0.5 -0.6 -0.7 -0.8 -0.9 -1.0 -3 -2 -1 0 r/w 1 2 3

Figure 7.3: Displacement of the surface of an in nite substrate under gaussian pressure. The dashed line recalls the beam pro le. The surface is assumed in nite (radius much larger than the beam width) the integral can be found in tables of Bessel transforms 27], then converting ( , ) into (Y , ) leads to : s 1 ? 2 2 I (r2=w2) e?r =w uz (r; z = 0) = ? Y w2 0
2 2

where I0 refers to the modi ed Bessel function. The pro le of the displacement is shown in Fig.7.3. But we are interested in the strain energy, which can be calculated using 1 Z Z u (r; z = 0) p(r) r dr d U = ?2 z that is Z1 2 2 Z1 U = 1? r dr e?2r =w dk e?k w =8J0(kr) Y w2 0 0
2 2 2 2

356

2 p = 21 ? Y w And nally, the spectral density of internal thermal noise takes the very simple expression 2 p Sz (f ) = 4kB T 21 ? Y w (7.10) f

CHAPTER 7. MIRRORS STANDARD THERMAL NOISE 1 ? 2 2 Z 1 dk e?k w =8 Z 1 r dr e?2r =w J (kr) = Y 0 w2 0 0 2 2 Z1 2 = 1? dk e?k w =8 w e?k w =8 Y w2 0 4
2 2 2 2 2 2 2 2

with values such that Y 7:3 0:17, w = 0:02 m, and a loss angle of 10?6 , we get a root spectral density " #1=2 1=2(f ) 10?18 1Hz m Hz?1=2 Sz f

; 1010Nm?2,

7.6 Finite mirrors
The preceding calculation does not allow to study the e ect of the aspect ratio of the actual mirror on the spectral density of thermal noise. We propose here an approximate model for a cylindrical mirror having a radius a and a thickness h. This model has been published in the BHV paper 34] with one wrong boundary condition. Then Yuk Tung Liu et al. (YT) have derived a correction to the BHV result.

7.6.1 A solution to the equilibrium equations
We consider a cylindrical mirror limited by: 0 r a; 0 z h The re ecting face is assumed at z = 0 In the case of a nite solid, we expect the displacement vector to be a discrete sum of Bessel modes, of the form: 8 > ur(r; z) = Pm Am(z)J1(kmr) < (r; z) = 0 (7.11) > u (r; z) = P B (z)J (k r) : uz 0 m m m

7.6. FINITE MIRRORS

357

Where Am; Bm are arbitrary functions of z, and km arbitrary constants. The equilibrium equations however imply for each order: ( 00 2 0 (Am ? kmAm) ? ( + )km (Bm + km Am) = 0 (7.12) 00 ? k 2 Bm ) + ( + )(B 00 + km A0 ) = 0 (Bm m m m so that by combining the two, we get:
2 @z2 ? km](A0m + kmBm ) = 0

the solution of which is:

A0m + kmBm = km

?k z k z m e m + me m

where m; m are arbitrary constants. This allows to substitute Bm in the rst of eq.7.12, and yields: + 2 2 A00 ? km Am = ? + 2 km me?km z ? mekmz m the solution of which is: + Am(z) = me?km z + mekmz + 2( + 2 ) km z me?km z + mekmz (7.13) introducing two new series ( m ; m) of arbitrary constants. Now Bm is determined: ! ! +3 +3 ?km z + km z Bm(z) = 2( + 2 ) m + m e 2( + 2 ) m ? m e + + + 2( + 2 ) kmz m e?kmz ? mekmz (7.14) The stress tensor has the following non zero components of order m: 8 0 > m;rr = (Bm + kmAm)J0(kmr) + 2 km AmJ10 (kmr) > < m; = (Bm + kmAm)J0(kmr) + +2 Am J1(km r) 0 r (7.15) 0 > m;zz = ( + 2 )Bm + km Am] J0(km r) > : m;rz = (A0 ? kmBm )J1(km r) m

358

CHAPTER 7. MIRRORS STANDARD THERMAL NOISE

7.6.2 Boundary conditions

The boundary conditions we assume are: No shear on the cylindrical edge, i.e. rz (r = a; z ) = 0 this can be satis ed by requiring that km a is a zero of J1 km m =a where the m are the strictly positive zeros of J1. No shear on the two circular faces, i.e. rz (r; z = 0) = 0; rz (r; z = h) = 0 Pressure of the beam on the rst face: zz (r; z = 0) = ?p(r) No pressure on the second face: zz (r; z = h) = 0 No radial stress on the cylindrical edge: rr (r = a; z ) = 0

(7.16) (7.17) (7.18) (7.19)

To the preceding constraints, Yuk Tung Liu et al. have pointed out that the pressure acting on the face z = h results in a global force accelerating the solid, so that an acceleration eld must be added to the equilibrium equations. This will be treated later. Now the pressure distribution can be expanded on the orthogonal family of functions J0( mr=a): X p(r) = p0 pm J0( m r=a)
m

where p0 = 1= a2 is a normalisation constant such that the pm are dimensionless. The orthogonality relations are: Za 1 J0( m r=a)J0( nr=a) r dr = 2 a2J02( m ) 0

7.6. FINITE MIRRORS

359

so that the pm are obtained as: Za pm = J 22 ) p(r)J0( m r=a) r dr (7.20) 0( m 0 The rz and zz components of the stress tensor are easily found from Am and Bm : ! ! km z ? 2 + ?k z m;rz =km = 2 m ? m e m m e m + +2 +2 and + ? + 2 kmz
m;zz =km

?k z k z me m ? me m

(7.21)

= ( m ? 2 m )ekmz ? ( m + 2 m )e?km z ? + ? + 2 km z me?km z + mekmz The boundary conditions provide 4 equations. The two rst are:
m?2 m? m?2 m

(7.22) (7.23) (7.24)

= ?pm=km

and

" # 1 +3 + pm m = 4 + 2 m ? + 2 m + km The next two boundary conditions imply: " # " # km h ? 2 + ?km h 2 m? +2 m e m +2 m e + i + h +km h + 2 mekmh ? m e?kmh = 0

2 m? +2 m?2 m? +2 m = 0 They allow to compute m and m in terms of m; m: " # 1 + +3 pm m = 4 +2 m? +2 m+ k
m

29) > u (r. YT pointed out that the component of spatial frequency zero of the pressure has not been taken into account. At this point. z) = 0 p r ) : uz (r. z) = 2 (3p +2 ) (1 ? z=h) < (7.25) (7. MIRRORS STANDARD THERMAL NOISE ?k h me m + k h h ekm h + +2 m m by substituting the values found for m .360 and ( CHAPTER 7. m.27) m = ?p0 2km (1 ? qm)2 ? 4qm x2 m pm qm 2x2 ? + (1 ? qm + 2xm) m (7. the preceding displacement has a zero average on the strained face. Because the series involves only strictly positive zeros of J1. so that an acceleration eld should be added in the equilibrium equations (recall that our mirrors are practically free falling in the z direction).28) m = ?p0 2km (1 ? qm)2 ? 4qmx2 m with the notation xm kmh and qm exp(?2xm). This can be done by adding to the preceding displacement an extra displacement of the form: 8 r > ur (r. z) = 4 h(3 +2 ) ? ((3++2p ) (z ? z2=2h) 0 0 2 0 This extra displacement contributes only the axial stress: zz = ?p0(1 ? z=h) . we nd pm( + 2 ) 1 ? qm + 2qm xm m = p0 km ( + ) (1 ? qm)2 ? 4qmx2 m k h ?k h m ? 2 m )e m ? ( m + 2 m )e m ? m i = 0 (7. But this force produces an acceleration. One must consider the resulting force acting on the body under the uniform pressure p0 = 1= a2 producing a force of 1 N after integration on the disk.26) then p 2 q (1 2 = p0 km ( ( ++ ) (1m? q?)q2m?+ q xm2) ) 4 mxm m m pm 2qmx2 + + (1 ? qm + 2qmxm) m (7.

11 and the extra displacement 7. z ) 0 = (km Am(z) + Bm(z))J0( m ) + 2 km Am(z)J10 ( m) but due to the fact that J10 ( m ) = J0( m ). z) = 2 (3+2 ) (c0r + c1rz) +2 < u (r.6. except for a radial contribution: rr (z ) ?qm (1 ? qm + 2xm (1 ? xm)) ekmz ? i ?kmzqm(1 ? qm + 2xm)ekm z ? km z(1 ? qm + 2qm xm)e?kmz = c 0 + c1 z .7.30) It is numerically easy to check that this function of z is not very di erent from linear.rr (r = a. FINITE MIRRORS 361 all other stress components are identically zero. and after substituting the explicit values of Am and Bm . z) = ? (3 +2 ) (c0z + c1z2=2) ? 4 (3+2 ) c1r2 +2 This displacement induces zero stresses. We have indeed m. where the strain energy is likely weak. Now the sum of the displacement 7. except the vanishing of the radial stress on the cylindrical edge. we compensate for the mean stress and torque on the edge. The second extra displacement is of the form: 8 > ur (r. and thus leaves unchanged the boundary conditions. It has even a vanishing average. z) = p0 (1 ? J0()2m?p4q x2 (1 ? qm + 2qmxm(1 + xm)) e?km z ? m. and the resulting solution is very accurate everywhere in the body.31) > : uz (r.rr qm m m (7. z) = 0 (7. we get h )m (r = a.29 satis es all boundary conditions. It is therefore possible to nd an approximate solution of the problem by using the De Saint-Venant principle: If we add to our displacement vector one more extra displacement giving a linear stress with suitable parameters. except maybe in the neighborhood of the edge. The equilibrium equations remain satis ed: 1N = @z zz = p0 =h = (1 N )=M = z a2 h where M is the mirror mass and the density.

3 Strain Energy The global displacement vector has the form 8 > ur (r. z)dz I0 = h 0 rr 1 Z h (r = a.30) allows to compute I0.7. Q. The strain components are: X Err (r. one nds I0 = 0. z) = P B (z)J ( r=a) + Wr2 + Tz + Sz2 :u z m m 0 m where P. T. z)z dz I1 = h2 0 rr we have the values of c0. S are known coe cients related to the two extra displacement terms de ned above.m = p0 J0( m )pm=km h2 so that I1 = p0 s where a2 X s = h2 pm J02( m) m m c0 = 6sp0 . z) (eq. z) = kmAm(z)J10 ( mr=a) + P + Qz m . c1 = 6(I0 ? 2I1)=h The explicit expression of rr(r = a. W. z) = 0 > u (r. z ) + rr (z )] dz 0 If we de ne 1 Z h (r = a. z). c1 = ?12sp0 =h 7. z) = Pm Am(z)J1( mr=a) + Pr + Qrz < (r. I1. and secondly 2 I1.6. We require for instance a minimum value for the integral Zh 2 rr (r = a. Firstly. MIRRORS STANDARD THERMAL NOISE then This linear stress can be adjusted to compensate for the rst moments of the residual stress rr (r = a.362 CHAPTER 7. c1: c0 = 6I1 ? 4I0.

FINITE MIRRORS E (r. z) = (Am(z) ? kmBm (z)) J1( mr=a) Am(z) J1( mr=a) + P + Qz r the trace of the strain tensor is thus: X 0 E (r.32) The squares of the stress components involve the squares of the main stresses. There is thus a perfect decoupling.and the extra terms in the displacement vector result in corrections to the global energy.6. z) = 363 X X m 0 Ezz (r. Now we can compute the main contribution. z) = (Bm(z) + km Am(z)) J0( mr=a) + 2P + T + 2(Q + S )z m m The strain energy per N2(our target) is given by Z2 Zh Za U = d dz r dr w(r. the squares of the extra stresses.7. We recall the following orthogonality relations: Za 2 J0( mr=a)J0( m r=a)r dr = a J02( m ) 2 0 Za 2 J1( mr=a)J1( m r=a)r dr = a J02( m ) 2 0 For the Bessel modes contribution we have thus: a2 X J ( )2 Z h U (z) dz U = 2 0 m m 0 m Main contribution to the strain energy . plus crossed terms. z) = Bm(z)J0( mr=a) + T + 2Sz m X 0 Erz (r. It is possible to show that crossed terms vanish in the r integration. z) 0 0 0 where the energy density w is de ned as: i 1h 2 2 2 w = 2 E 2 + 2 Err + E 2 + Ezz + 2Erz (7.

364 where CHAPTER 7. the integration is straightforward.N?2. using the Young modulus Y and the Poisson ratio instead of the Lame coe cients: 2 X J 2 ( m )p2 1 ? q 2 + 4qm xm 0 m m U = 1? (7. and the result is: 2 a3 2 X 2 ? 2 +4 U = 4 ( ++ ) J0 ( m )pm (11? qqm)2 ? qm xm2 4q x or as well.33) aY m (1 ? qm)2 ? 4qmx2 m m The dimension of U is J. MIRRORS STANDARD THERMAL NOISE 02 1 0 2 Um = (Bm + km Am)2 + 2 km A2 + Bm + 2 (A0m ? km Bm)2 m All the terms being known. m m m m m Correction to strain energy 0 The contribution of the extra stresses to the strain energy is: Z2 Za Zh U = d r dr w(z)dz 0 0 where w(z) is the extra density: h i w(z) = 1 (((2P + T + 2(Q + S )z)2 + 2 2(P + Qz)2 + (T + 2Sz)2 2 The coe cients are: P = 2 (3 p0+ 2 ) ( + 6s( + 2 )) 0 Q = ? 2 (3 p+ 2 )h ( + 12s( + 2 )) 0 T = ? (3 p+ 2 ) ( + + 6s ) 0 S = 2 (3 p+ 2 )h ( + + 12s ) The result is: 2 2 h i U = 6 (3a hp0 ) 6 s + + + 36( + 2 )s2 +2 .

7. See the displacement pro le on gure 7. we have 2 2 Z a exp(?2r2=w2)J ( r=a) r dr pm = J ( )2 w2 0 m 0 0 m . FINITE MIRRORS After replacing the Lame coe 2 a2 4 U = 6 h3 Y with cients by Y. ?qm (1 ? qm + 2xm (1 ? xm)) e mz=a? i zh ? m a qm(1 ? qm + 2xm )e mz=a + (1 ? qm + 2qmxm)e? mz=a Case of gaussian beams if the beam intensity comes from a TEM00 wave of width w. we have as seen above (7.30): h 1 X J0( m)pm ? m z=a ? rr (r = a.6. z ) = 2 2 ? 4qm x2 (1 ? qm + 2qm xm (1 + xm )) e a m>0 (1 ? qm) m A plot of rr(r = a.5) shows the its quasi-linear behavior. this is: 3 ! h 2 + 72(1 ? ) 25 a 365 (7.4 For the stress on the cylindrical edge before correction. justifying a posteriori the De Saint-Venant approximation.34) X m>0 2 pmJ0( m )= m It is interesting to have the explicit expression for the re ecting surface displacement: 2 X ? 2 +4 uz (r. ! h 4 + 12 a . z) ( g. z = 0) = 2(1 ? ) (11? qqm)2 ? qm xm2 pm J0( m r=a) + aY m>0 4qmxm m m " # r2=a2 + 12 a2 (1 ? ) + 2 hY h2 Explicit coating displacement and edge stress units are m/N.7.

4. h=0.35) aY m>0 mJ0( m )2 (1 ? qm)2 ? 4qm x2 m the parameter involved in expression 7. -1.33 for U takes the special form 2 X exp(? 2 w2=4a2 ) 1 ? q 2 + 4qm xm m m UGauss = 1 ? (7.00 -0.e-10 0. 11.29).60 -0.6.20 0.4: Displacement of the surface of a nite substrate under a gaussian pressure The upper integration bound can be replaced by +1 if. MIRRORS STANDARD THERMAL NOISE 4.e-10 -1. the cylinder has negligible di raction losses.175 m. " 2 # 1 exp ? m w2 pm = J ( )2 8a2 0 m The expansion of p(r) on the orthogonal family J0( m r=a) is rapidly convergent. A plot of p(r) reconstructed from only 12 terms is shown on g. as a mirror.366 5.34 for U takes the special form: 2 w2 2 X = exp(? Jm( =)8a ) Gauss 2 m 0 m m>0 .60 1.1 m 2.20 0.e-10 displacement [m/N] 3.00 radial parameter r/a Figure 7.e-10 CHAPTER 7.7.e-10 a=0. The expression 7. Then the result can be found in 20] (eq.e-10 1. A good accuracy is obtained for all the numerical calculations with only 50 terms.

Solid line : rr (r = a. w =0.0 axial parameter z/h Figure 7.5 0. 0. 20.0 0. a=0.7 0. -10. 7.7. h=0. -30.1 0.175 m. 0.8 0.5: Radial stress along the edge of the cylindrical solid. of the half-space (in nite mirror) appproximation.1 m 367 Θrr(r=a.175m.6.2 0. h =0. we can plot the ratio for varying aspect ratios (see. values of a as small as possible are desirable. we get 10?10 J:N?2 U 2:08 10?11 J:N?2 = U + U 2:02 10?10 J:N?2 . z ). If we note UHS the corresponding strain energy and UFM that of the nite mirror.6 0.02m.9 1.4 0. a =0. Gong-like mirrors are worse than bar-like ones. FINITE MIRRORS 30.??). -20. Dashed line: linear t c0 + c1 z It is interesting to compare the results with the case discussed in the preceding section.4 Some numerical results U Utot 1:81 For a Virgo input mirror.z) [m-2N-1] 10.6. and it is clear that for a given thickness h.1m.3 0.

02 0. The corresponding root spectral density of thermal noise is given by s 4kB T U 1=2 = Sx(f ) f tot so that we nd (the loss angle being 10?6 : Sx(f )1=2 U 1:03 10?19 m:Hz?1=2 at 100 Hz 5:55 1:75 10?11 J:N?2 10?11 J:N?2 For a Virgo end mirror (a =0.01 0. MIRRORS STANDARD THERMAL NOISE 1600 Pressure [P/N] 1200 reconstructed with 12 orders 800 400 0 0.04 0.0554m) we nd: U .00 0.6: Solid line : gaussian pressure.175m. Dashed line : reconstruction The in nite mirror approximation was: U1 1:88 10?10 J:N?2 so that U=U1 1:07.05 radial coordinate [m] Figure 7. w =0. h =0.368 2000 CHAPTER 7.1m.03 0.

2 1.8. and 7:30 10?11 J:N?2 10?11 J:N?2 The beam width has changed (w=5. long dashed line: a=0. so that 6:77 Sx(f )1=2 6:21 10?20 m:Hz?1=2 at 100 Hz Finally.15 h [m] 0.25 Figure 7.2m.10 0.05 Half-space approx.20 0. short dashed line: a=0.7.7: relative spectral density of Thermal noise for various aspect ratios.3 1.7. Solid line: a=0.6.9 0. FINITE MIRRORS 1.4 1. This is shown on Fig. it is interesting to check the convergence of Utot to the in nite mirror approximation when the size of the mirror increases.8 1.175m.54cm).7 1.6 1.0 0.1 1.5 369 UFM/UHS 1.15m Utot U1 so that U=U1 1:08. 0. .

8e-10 Utot 1.7.8 0.8: Convergence of the nite mirror model to the in nite.9 1. when the size of the mirror increases.7e-10 1. 7. the general expression of (k) is.370 2.5 0.5e-10 0. in order to reduce the thermoelastic noise. but a quantitative model is obviously needed.7 0.0 0.3 0.6 0.1 Half-space approximation . 1.6e-10 1. z = 0) = ? 2 ( + ) 0 0 0 0 7.1 0.7 Non gaussian beams It has been suggested ( 38]) to use light beams with a at pro le in the long cavities instead of gaussian modes. For any pressure pro le p(r).4 0. The idea is convincing. MIRRORS STANDARD THERMAL NOISE 1.0e-10 CHAPTER 7.9e-10 Half-space approx. It is expected that widen the beam will average the surface uctuations and. as already seen: 1 Z 1 p(r) J (kr) r dr k (k) = ? 2 0 0 so that the displacement of the surface of the half-space is: + 2 Z 1 dk J (kr) Z 1 r0dr0 J (kr0) p(r0 ) uz (r.2 0. We have addressed the question of how to generate these modes in a preceding section.0 Mirror size: a=h [m] Figure 7.

1 = 38 2 2 ! Z1 J1(x) 2 = 4 dx x 3 0 . 2 . ? 1 . thus expressible in terms of a hypergeometric series: ! Z1 J1(x) 2 = 1 F 1 . representing a simpli ed version of a realistic mode (which would be only almost at).7. ? 1 . 2. NON GAUSSIAN BEAMS 371 or as well.36) uz (r. 1 dx x 2 2 2 0 now (see 20] p.487).556).7. In the special case of a distribution uniform on the disk r < b. we have ( 2 1 b r p(r) = 0= (r (b)< b) so that kb p(k) = J1(kb ) ~ and the energy integral reduces to ! 2(1 ? 2) Z 1 dx J1(x) 2 U = Yb 0 x the integral is of the Weber-Schafheitlin type (see 20] p. using the Poisson ratio and the Young modulus: 2(1 ? 2) Z 1 dk J (kr) Z 1 r0dr0 J (kr0) p(r0 ) (7. z = 0) = ? Y 0 0 0 0 It is easy to see that the strain energy per N2 is then given by 2 Z1 dk p(k)2 ~ U = 2 (1Y? ) 0 where Z1 p(k) = ~ r dr p(r) J0 (kr) 0 is nothing but the Fourier transform of the pressure distribution. so that we have F 1 .

and a theory with a nite mirror radius is needed. because the reduction of thermal noise operates only if a is much larger than w. b. and give the apex equation of the surface as: 81 (r = 0) > > 1 1 2 =b2 2 <F . 1 . denoted by U at with the gaussian value. 2 . the pressure distribution takes however signi cant values probably near the edge of the mirror. z) denotes the Gauss hypergeometric series.7. MIRRORS STANDARD THERMAL NOISE 2 U = 8(1 ?Y b ) 3 2 which yields the nal result: It is worth to compare this value.9. ? . Approximate representation of the mirror as an in nite half-space is thus questionable in this case. and though it is of no practical interest for our present purpose (but any result may always be re-used one day in a di erent context). the gain in thermal noise could be s U at 0:44 U Gauss which means a factor better than 2 in sensitivity. . b2=r2 =2r (r > b) 2 2 where F (a. we show the (virtual) distorted surface on Fig.372 CHAPTER 7. For curious readers. and outside the mirror for end mirrors of radius 35 cm. c. therefore 1 order of magnitude in the analyzed volume of space in the frequency band around 100 Hz. In this case. . denoted by UGauss: U at = 16 w :96 w UGauss 3 3=2 b b If it is possible to establish a at mode of radius 10 cm where a gaussian mode of half-width 2 cm was used. and secondly because the size of actual mirrors has been de ned as the minimum consistent with small di raction losses. so that say 5 times w is near the physical edge for input mirrors. z = 0) = ? 2(1 ? b ) > 2= 2 2 (r = b) Y > 1 1 : bF . r (0 < r < b) uz (r.

7. But The series giving U and are still convergent.2 Finite test mass approximation The model developped for a nite mirror of radius a and thickness h can be extended to the case of a at pressure ( p(r) = 1= b2 (r b) p(r) = 0 (r > b) representing approximately a at mode. these new pm are decreasing like 1= m.8 -1.0 -3 -2 -1 0 r/b 1 2 3 Figure 7. so that reconstruction of p(r) is numerically di cult. despite the new values for the p pm. units) -0. NON GAUSSIAN BEAMS 0.9: Displacement of the surface of an in nite substrate under a pressure uniform on the disk r < b. so that the formal series giving p(r) is valid in the sense of the distribution theory.2 -0. The surface is assumed in nite 7. But the terms in . In fact.0 373 Surface distortion (arb.4 -0. The pressure coe cients are: aJ ( pm = 2b 1J 2mb=a) m 0 ( m) The pm decrease much less rapidly for increasing m than in the case of a gaussian pro le.6 -0.7.7.

The displacement of the re ecting surface is much less than in the gaussian case ( g.10: Radial stress and corresponding linear t.7.z) [m-2N-1] 10.175 m.0 0. and in the series de ning are nevertheless decreasing like 1=m3.0 axial parameter z/h Figure 7.374 30.12). showing that the De Saint-Venant correction is still realistic. 0.10). -30. The following plot (7. z) with the new coe cients. 0. MIRRORS STANDARD THERMAL NOISE a=0. 20. so that ordinary convergence is secured.13) shows again the large gain that could be achieved by increasing the spot radius. apart from the new values for pm . we get the following plot ( g.1m the series 7. It is interesting to remark that the case b = a (the at mode has the same radius as the mirror) leads to U = 0 . Comparison is made with a gaussian beam of width 2cm. -20.5 0.8 0.2 0.7 0.4 0.1 0. the linear correction seems to make sense (see g. Even with a spot radius of 0.3 0.1m. It is especially intersting to compare the spectral densities of thermal noise in the gaussian mode regime to the at mode regime. h=0. b=0. If we compute the stress rr(r = a.11).7. and distortion is very similar to the in nite case. It is nevertheless necessary to check that the correction for the radial stress on the edge is still reasonable. CHAPTER 7.15m.33 fo r U .7.1 m Θrr(r=a. -10. All the formulas derived in the preceding section are unchanged.175m and a pressure at in a disk of radius 0. in the case of a mirror of radius 0.9 1.6 0.

we nd: U 1:60 10?11 J:N?2 U 1:06 10?11 J:N?2 Utot 2:65 10?11 J:N?2 The in nite mirror approximation was: U1 3:59 10?11 J:N?2 .15m and = 0.34 to U = h=6 a2Y 7. -10.z) [m-2N-1] 10. 20. For the input mirrors.7.11: Radial stress and corresponding linear t. h=0.4 0.0 axial parameter z/h Figure 7. a=0.3 0. and the strain energy reduces. -20. -30.7.1 0.0 0.175 m.7 0.3 Numerical results Let us assume such a at mode in the Virgo cavities whose mirrors are assumed identical in size to the current situation.1 m 375 Θrr(r=a.7.6 0.2 0.8 0. In this case the solution is exact.9 1. 0.b=0. according to 7. 0.5 0. NON GAUSSIAN BEAMS 30.

(a=0.e-11 1. we would have an amplitude .e-11 -1.376 5. h=0. The more realistic model proposed by D'ambrosio et al. h=0.60 1.175m.e-11 -4. knowing that the mode has almost exactly the same spot size after propagation. MIRRORS STANDARD THERMAL NOISE a=0.175m.1m. for the sake of de niteness. It is thus necessary to check that taking a more realistic at mode does not destroy the preceding conclusions.e-11 CHAPTER 7. so that the numerical results are identical.e-11 4.1m) so that U=U1 0:74. 38] consists in a superposition of elementary gaussian modes of waist w0 on a disk of radius b.00 radial parameter r/a Figure 7. as represented by an ideal at top function is unrealistic from an optical point of view. If we adapt the model to the Virgo parameters. We nd the spectral density (the loss angle being still 10?6 : Sx(f )1=2 3:74 10?20 m:Hz?1=2 at 100 Hz For the end mirrors.20 b/a 0.12: Displacement of the re ecting surface under a pressure uniform in a disk of radius 0.e-11 -3.4 Realistic modes The preceding approach is still questionable because the pressure distribution.e-11 0.e-11 -5.e-11 3.00 -0. b=0.1m.60 -0.1m displacement [m/N] 2. 7. -1.20 0.e-11 -2.7.

40 0. 0) dx0 dy0 where is the disk of radius b.1m) on the at input mirror of the form: Z A(x.90 377 relative SD of TN : (Uflat/UGauss)1/2 0.00 0.00 0.175m. with a gaussian-like edge Parameter w0 determines the sharpness of this edge. w is the beam half-width after zR = w 0 propagation on the distance L: q 2 w = w0 1 + L2=zR .30 0. y. z) a gaussian TEM00 wave: ! x2 + y 2 (x.10 radius b of the flat mode [m] 0.80 0.15 Figure 7. L) / exp ?Z ( ? 0)2 exp(?2Z 0) I0(2Z 0 ) 0 d 0 0 2 = being the Rayleigh parameter. h=0. y.13: gain in SD of thermal noise vs spot radius. 0) = exp ? w2 0 The resulting amplitude has a quite at maximum.20 0. It is easy to show that after propagation at a distance L.10 0. 0) = (x ? x0. y ? y0. NON GAUSSIAN BEAMS 1. y.7.05 0.7. the amplitude is (up to a normalization factor): Z b=w h i A(x.00 0.60 0. y.50 0.(a=0. and (x.70 0.

we have plotted the mode intensity pro le for the following parameters: w0 3:2 cm. Dashed line : Pro le on the far mirror (3 km away). we can compute numerically the pm from jAj2 after normalization. Knowing A.06 radial distance [m] 0. On Fig. at the two ends of a cavity of length 3 km. MIRRORS STANDARD THERMAL NOISE Intensity [arb. because the intensity distribution is weakly modi ed by di raction for not too small w0.15. The corresponding strain energies are almost the same.14. L = 0) and for the end mirrors ('mexican hat' wavefront. Even for smaller w0 resulting in more distorted intensity pro les on the end mirror.378 CHAPTER 7. because the function exp(?z)I0(z) has an easy behavior. There is no better analytical expression for the amplitude. I (z) is the modi ed Bessel function of R 0 . but a numerical integration is straightforward. This can be done for the input mirrors ( at wavefront. units] 0. L =3 km).12 0. In Fig.00 0. giving a too px2 + y2=w. the rst kind. It is clear that by decreasing the parameter w0 (sharping the edge). Solid line : Pro le on the input at cavity mirror. we plot the values found for several particular radii and several values of the parameter w0.18 Figure 7.1 m.7. However a too sharp edge is not desirable from an optical point of view. and Z 1 ? iL=z .7. b = 0.14: Intensity pro le in the at mode. we get more and more close to the ideally at model. the strain energies are nearly identical at the two ends.

z) = umain. 0.175 m. z) uz (r.16 0. thickness 0.18 radius of the flat mode [m] Figure 7. in the limit of reasonable parameters.7.e-11 0. z) = umain.1 m and ideally at mode.z (r. 7. z) . Circles: same nite mirror with realistic mode w0 3:2 cm.12 0. z) + uz (r. MIRROR DISTORTIONS AND ENERGY MAPS 5. Dashed line: In nite mirror and ideally at mode.14 0.e-11 379 4. Triangles: same nite mirror with realistic mode w0 2 cm.8. The expressions of the displacement vector components in the case of nite cylindrical mirrors are: ur (r.e-11 1. Diamonds: same nite mirror with realistic mode w0 1 cm distorted wavefront (and consequently unfeasible mirrors).15: Strain energy U vs radius of the at mode.e-11 Infinit e med ium a pprox . z) + ur(r. Solid line : Mirror of radius 0.08 0. we remark a good agreement between the ideal and realistic models.r (r. 2. However.8 Mirror distortions and energy maps It is interesting to write explicitly the solution of the elastical problem.e-11 Strain energy [J/N2] 3.10 0.

z) = 1 + aY m>0 m Dm Pm (z) with Dm (1 ? qm)2 ? 4qm x2 .16) the distorted shape of the solid.380 CHAPTER 7.j (r. and to check that the distortion is minimized by the at mode. z) = Emain. z) = 2 a1 hY + 12s(1 ? )] r2 + (1 + 24s )z2 ? 2(1 + 12s )hz 2 (with s a2=h2).j (r. ur(r. this is extremely fast to compute (see the "Heating issues" chapter for algorithmic details). we get: where m>0 m m Ei. By derivating the preceding expressions.7.z (r. It is also possible to give the strain components. These formulas allow to draw (see Fig. z) = ar2Y f + 6s(1 ? ) ? z + 12s(1 ? )] =hg X pmJ1( m r=a) umain. z) . MIRRORS STANDARD THERMAL NOISE with the following expressions: n o uz (r.i. z) + Ei.j (r. z) = (1 + ) pm J0(D r=a) Qm(z) aY 1 Q (z) = h(1 ? )(1 ? q + 2q x ) ? q x2 )i exp(? z=a)+ m m m m m m 2 m h i +qm (1 ? )(1 ? qm + 2xm) + x2 exp( m z=a)+ m + m 2za (1 ? qm + 2qmxm) exp(? mz=a) ? qm(1 ? qm + 2xm ) exp( m z=a)] Despite the apparent complexity.r(r. and m h i ?Pm(z) = 2qmx2 + (1 ? 2 )(1 ? qm + 2qm xm) exp(? mz=a)+ m h i +qm 2x2 ? (1 ? 2 )(1 ? qm + 2xm) exp( mz=a)? m ? m z (1 ? qm + 2qmxm) exp(? mz=a) + qm(1 ? qm + 2xm) exp( m z=a)] a and X m umain.

16: Distorted mirror for 1 N normalized pressure. (exaggerated by a factor of 6 107 ) .54cm. at mode of radius b =10cm. From left to right: Gaussian mode. w=2cm. MIRROR DISTORTIONS AND ENERGY MAPS 381 Figure 7.7.8. Gaussian mode w=5.

z) = 1 a2Y m 1 m m>0 Dm + X pm J1( mr=a) P (z) Emain.rr (r. z) = 0 E (r. z) = a12Y + 6s(1 ? ) ? z ( + 12s(1 ? )) =h] E (r. (r. z) = 1 a2Y m m r=a m>0 Dm + X pm J ( r=a) a Q0 (z) Emain. z) = a2Y 0 m m m m m>0 Dm The functions Pm (z). Moreover we have: a Q0 (z)+ P (z) = ?2(1?2 ) h(1 ? q + 2q x )e? mz=a ? q (1 ? q + 2x )e mz=ai m m m m m m m m m For the extra contributions. z) = a2Y 1 m m 2 m m m>0 Dm " # 1 + X pm J ( r=a) P (z) + a Q0 (z) Emain(r. z) = a12Y (1 + 24s )z=h ? 1 ? 12s )] Erz (r. in detail: CHAPTER 7. z) = 1 a2Y 0 m m m m>0 Dm " # 1 + X pm J ( r=a) 1 a P 0 (z) ? Q (z) Emain.382 with. z) Ezz (r.zz (r. we have Err (r. z) = Err (r. MIRRORS STANDARD THERMAL NOISE + X pm J 0 ( r=a) P (z) Emain. z) = ? 1 ?22 1 ? 12s ? (1 ? 24s)z=h] aY and " # ! 1 a P 0 (z) ? Q (z) = 2q x2 ? (1 ? q + 2q x ) m z e? mz=a ? m m m m m m 2 m m a ! 2 ? q (1 ? q + 2x ) m z e mz=a ? 2qm xm m m m a .rz (r. Qm(z) have been de ned above.

5 cm.00 0.000 -0.140 -0.175 0. Logarithmic scale .070 0.105 -0. of thickness h=10 cm.035 -0.7.10 Figure 7.105 0.8.035 0.140 0.070 -0.17: Distribution of strain energy in a cylindical mirror of radius a=17.05 0. MIRROR DISTORTIONS AND ENERGY MAPS 0.175 383 0. under a gaussian pressure w=2 cm.

54 cm.175 0.05 0.00 0.070 0.000 -0.5 cm.384 0. under a gaussian pressure w=5.105 -0.070 -0.140 0.035 0. of thickness h=10 cm.175 CHAPTER 7.140 -0. MIRRORS STANDARD THERMAL NOISE 0.Logarithmic scale .10 Figure 7.18: Distribution of strain energy in a cylindical mirror of radius a=17.035 -0.105 0.

035 -0.070 -0.070 0.5 cm.105 -0.105 0.05 0.00 0.140 0.035 0.10 Figure 7.175 385 0. MIRROR DISTORTIONS AND ENERGY MAPS 0.175 0.8.19: Distribution of strain energy in a cylindical mirror of radius a=17.140 -0. under a at top pressure b=15 cm.7. of thickness h=10 cm.Logarithmic scale .000 -0.

105 0.070 -0. under a realistically at pressure b=15 cm. .175 CHAPTER 7.175 0. of thickness h=10 cm.00 0.05 0.386 0.070 0.5 cm.140 0.140 -0.035 -0.105 -0. MIRRORS STANDARD THERMAL NOISE 0.20: Distribution of strain energy in a cylindical mirror of radius a=17.10 Figure 7.035 0.000 -0.

and of a realistic at beam (b=0.15 m) (solid lines) .m-3.06 0. MIRROR DISTORTIONS AND ENERGY MAPS 387 10-4 10-5 Strain energy density [J.07 0.09 0. h=10 cm).08 0.8. Case of a gaussian (w=2 cm) beam (dashed lines).01 0.N-1] 10-6 10-7 10-8 10-9 10-10 axis edge axis edge 10-11 10-12 0.00 0.21: Distribution of strain energy on the axis and on the edge of a cylindrical mirror (a=17.10 axial coordinate z [m] Figure 7.02 0.7.04 0.03 0.05 0.5 cm.

locally lower than the at beam average. The energy density is weak on the edge. It can be seen that the energy density is much lower on the edge than on the hot point on the axis.7.7. MIRRORS STANDARD THERMAL NOISE so that it is possible to compute explicitly the strain energy density: " # Y E (r. and this is even more clear for wider w and a fortiori.7. namely a gaussian beam of width 2 cm. even if there is a sharp minimum for the gaussian beam. z)2 + 2E (r. z)2 w(r.21 the energy density on the axis and on the edge in the two extreme cases. in average for at beams and realistic at beams.20) show the distribution of w in some cases examined above. in both cases.7.7.17. and a realistic at mode of radius 15 cm.18. We show in Fig.z The following pictures (Fig.19. z)2 + E (r. . z)2 + E (r.388 CHAPTER 7. z) = 2(1 + ) 1 ? 2 rr zz r. z)2 + E (r.

For 389 .Chapter 8 Thermoelastic noise 8. and use it to compute the dissipated energy. As in the preceding chapter. Obviously. we had W = 2U! as average dissipated energy. being a global loss angle and U the static strain energy. the temperature eld itself depends on the strain eld. and use the Levin approach already presented. For the standard thermal noise. producing eventually random motions of the surface. These uctuations are called thermodynamical and can couple with strain via the thermal dilatation constant . the readout noise) as depending on the energy dissipated when the body is under a virtual pressure having the same pro le as the optical beam and excited at low frequency.1) ! where W is the average dissipated energy.1 Introduction The brownian motion of matter inside the substrates is not the only cause of noise in the optical readout.e. In this case. then we compute the resulting temperature eld. we rst solve the static linear elastical problem (it is done in the preceding chapter). we shall consider the low frequency tail of the spectral density of the e ective motion of the surface (i. the spectral density is still of the form (Levin's formula): Sx(f ) = 4kB2T W (8. A good way for modeling this kind of noise is to start from the general thermodynamical formulas as detailed by Landau and Lifshitz 37]. Using the same approach as used in ??. But now W must be interpreted as the energy dissipated via coupling of the strain with the temperature eld in the bulk. There is another cause due to temperature uctuations in a nite volume of material.

as well: @S = ? 1 div(~) @t T q The total entropy variation in the body is therefore: dStot = ? Z 1 div(~) dV dt T q where the integral is extended to the whole body. The variations of the entropy density are related to the heat ux ~ by q requiring conservation of the energy in the body: T @S = ? div(~) q (8.2) @t where ~ = ?K gradT . and we have: dStot = ? Z 1 ~:grad T dV dt T2 q but using the de nition of ~. this is: q dStot = Z K (grad T )2 dV dt T2 so that the energy variation is : dStot = Z K (grad T )2 dV W = T dt (8.4) T . This becomes: dStot = K Z (grad T )2 dV W = T dt (8.3) T We shall say now that the temperature gradient eld is caused by the small deformations of the body that we have computed precedingly. the rst integral vanishes.Landau and Lifshitz ??). while T is the mean temperature. THERMOELASTIC NOISE computing the dissipated energy. this is as well: dStot = ? Z div ~ dV + Z ~:grad 1 dV q q dt T T Owing to the fact that the heat ux is zero on the surface of the body. we use the time dependence of the entropy S . K being the thermal conductivity of the material q (cf. Or.390 CHAPTER 8.

The resulting temperature eld obeys the Heat (Fourier) equation: (8. INTRODUCTION 391 Where we have replaced T by a T in the gradient for more clarity. it is well known (cf. . We have after the preceding chapter on standard thermal noise all the material for computing W .7) (see ??).In fact. one being the entropy in the reference state. Now we reach the relevant equation for the dissipated energy: K 2T Z (grad E )2 dV W = 2C 2 (8. Landau-Lifshitz) that the total entropy is the sum of two terms. Finally: " #2 Z Y W = KT (1 ? 2 ) C (grad E )2 dV (8. so that there is a trivial solution: T T = CE The boundary conditions (null heat ux on the surfaces) are considered satis ed in time average ( T is assumed oscillating at a few tens of Hz). and a second one proportional to the trace E of the strain tensor: S = S0 + E being the thermoelastic coe cient. On the other hand. so that there is in the bulk material a power source given by P = T: dS = T dE dt dt where E is the trace of Eik .5) ( C@t ? K ) T = T dE dt The trace of the strain tensor Eik found in the preceding chapter is in any case a harmonic function. and the Poisson ratio.1. they are exactly satis ed on the circular edge of the mirror.8.6) is related to the linear dilatation coe cient by = 1 ?Y2 where Y is the Young modulus.

z) = u(k) kz ? 1 + 2 ] exp(?kz) J1(kz) k dk (8.2 Case of in nite mirrors Let us recall the results obtained in the preceding chapter on standard thermal noise.8) R Z uz (r. the displacement vector is: Z ur (r. that E (r. Let us compute the gradient of E : @E = 2(1 ? 2 )(1 + ) Z p(k) exp(?kz) J (kz) k2 dk ~ 1 @r Y R @E = 2(1 ? 2 )(1 + ) Z p(k) exp(?kz) J (kz) k2 dk ~ 0 @z Y R Now. ~ 2(1 ? 2 )(1 + ) Z p(k) exp(?kz) J (kr) k dk ~ E (r. in passing.9) R so that: Z E (r. using the closure relation Z 0 J (kr) J (k0r) r dr = (k ? k ) (8. z) = u(k) kz + 2 ? 2 ] exp(?kz) J0(kz) k dk (8.10) u R The function u(k) is determined by the virtual pressure distribution p(r).11) u(k) = ? 1 + p(kk) Y where p(k) is the Fourier-Bessel transform of p(r). z) = ? 0 Y R Which shows.392 CHAPTER 8. and therefore the volume integration of its square will be problematic. the gradient will involve Dirac distributions.12) k R . THERMOELASTIC NOISE 8. Namely: ~ (8. z) = div ~ (r. Under beam pressure. As a result. z) = ?2(1 ? 2 ) u(k) exp(?kz) J0(kz) k2 dk (8. 0) = ? 2(1 ? 2 Y)(1 + ) p(r) We can thus already foresee that in the case of an ideally at top beam.

using (8. For the end mirrors (w = 5. 36].7): 2 2 2 p Sx(f ) = 4kB KT2 C 2(1 2+ 3 ) (8.2.15) fw This result has been found rstly by Braginsky et al. 8. but almost signi cant. For silica parameters: K 1:4 W:m?1:K?1 5:4 10?7 K?1 2. This is a strong requirement ior better than k on the Fourier transform of the pressure distribution. 1. this is: " # 1=2 = 5:81 10?21 1 Hz m:Hz?1=2 Sx(f ) f so that the spectral density of thermoelastic noise is. ?].8.14) .2. it is possible to carry out the volume integration: Z1 Z1 2 Z )2(1 ~ 2 r dr dz (gradE )2 = 8 (1 ? 2 Y 2 + ) p(k)2 k2 dk (8.1) and (8. then by Liu et al. using the preceding approach. we have seen that: h i p(k) = 21 exp ?k2w2=8 ~ giving Z 2 2 2 ~ E )2 dV = 4(1 ?p ) 2(1 3+ ) (grad Y w (8.54 cm) . 500 J:kg?1:K?1 on nds: " # 1=2 = 2:68 10?20 1 Hz m:Hz?1=2 Sx(f ) f For a gaussian pro le of half width w.13) ~ 0 0 R This expression shows that the function p(k) must have an asymptotic behav~ ?3=2 for the integral to converge. 202kg:m?3 C 7. CASE OF INFINITE MIRRORS 393 for = 0.1 Gaussian beams which is lower than the standard thermal noise.

the model developped for standard thermal noise provides the explicit expressions for the trace E of the strain tensor: E (r. z) with h i X pm J0( mr=a) um e? mz=a ? vm e mz=a E0(r. 8. THERMOELASTIC NOISE If we now consider a at beam modeled by its ideal representation: ( 2 1 b r p(r) = 0= (r (b)< b) we have the Fourier-Bessel transform: kb p(k) = J1(kb ) ~ which shows that the requirement on the decreasing rate for large k is not ful lled.2 Flat beams 8.394 CHAPTER 8. z) + E (r. z) = ? 2(1 ? 2a2)(1 + ) Y m>0 Dm . We meet two conclusions: the rst is that we must carry out a numerical integration with the "realistic" at modes detailed in the preceding chapter.2. we note that "Mathematica" nevertheless gives a nite (and rather strange) result: Z1 J1(x)2 dx = Ln(64)2? 4 + 2 0 ( = Euler's constant). the second is that we must be cautious with results of symbolic computation softwares. J (k) having an asymptotic behavior in k?1=2.3 Case of nite mirrors In the case of nite mirrors. For fun. z) = E0(r.This is the consequence of our preceding remark on the discontinuity of the pressure. If we try to compute the integral. we get: Z 2 2Z ~ E )2 dV = 8(1 ? 2 ) 3(1 + ) 1 J1(x)2 dx (grad b 0 wich is a divergent integral.

vm = qm(1 ? qm + 2xm ) qm and xm have also the same de nitions. The um. E (r. Moreover. Z ?2 2 ~ (grad E )2 dV = (1 a2hY ) (1 ? 24s)2 2 ~ ~ (NB: grad E and gradE0 are orthogonal in the r integration). We have successively: " # 4KT 2 (1 + )2 X w + (1 ? 24s)2 a W = a3 2 C 2 m 4h And for the spectral density: " # 4kB KT 2 2 (1 + )2 X w + (1 ? 24s)2 a Sx(f ) = a3 2C 2f 2 m 4h m>0 m>0 (8. z) = ? 1 ?22 1 ? 12s ? (1 ? 24s)z=h] aY so that the gradient of E is: @E0 = 2(1 ? 2 )(1 + ) X pm m J ( r=a) hu e? mz=a ? v e mz=ai 1 m m m @r a3 Y m>0 Dm (8.16) @E0 = 2(1 ? 2 )(1 + ) X pm m J ( r=a) hu e? mz=a + v e mz=ai 0 m m m @z a3Y m>0 Dm (8.18) @z a2hY Owing to the orthogonality relations for the J ( mr=a).17) @ E = 1 ? 2 (1 ? 24s) (8. vm are: um = 1 ? qm + 2qmxm. CASE OF FINITE MIRRORS 395 where the pm are the Fourier-Bessel coe cients of the pressure distribution.20) . we get Z 2 2 X ~ (gradE0)2 dV = 4(1 ? 2 ) (1 + ) wm (8.8.3. and where the Dm have been de ned in the preceding chapter.19) a3 Y 2 m>0 where 2 m wm = pD2m J0( m)2 (1 ? qm) m h i 2 ) + 8q (1 ? q )x + 4q (1 + q )x2 (1 ? qm)(1 ? qm m m m m m m and obviously.

THERMOELASTIC NOISE For gaussian beams. Note the two "hot" points corresponding to the regions where the gradient is the largest. we have represented the distribution of (gradE )2 for a realistic at mode (b = 10 cm. For w = 5. The sharp edge generates high spatial frequencies that forbid the Fourier-Bessel coe cients pm to have a decreasing rate able to secure the convergence of the series.8.1 Gaussian beams 8.5 cm. this is " #2 1=2(f ) = 4:92 10?21 m:Hz?1=2 1 Hz Sx f On Fig.3.3. 8. When sharping the edge. we nd: " # 1=2(f ) = 2:76 10?20 1 Hz Sx f slightly worse than the in nite case.2 Flat modes . for w0 = 1 cm. The result for a Virgo-like mirror (a = 17. h = 10 cm.2. a = 17. For the parameters corresponding to Virgo input mirrors (w =2 cm. which is worse than the in nite case. To be speci c.21) Sx f It is weakly dependent on the parameter w0 (sharpness of the beam's edge). is: # " 1=2(f ) = 4:89 10?21 m:Hz?1=2 1 Hz (8. w0 = 1 cm).54 cm (end mirrors): " # 1 Hz 1 Sx=2(f ) = 8:20 10?21 f in Fig. these points become hoter and hoter.396 CHAPTER 8. h = 10 cm) and for a realistic mode (b = 10 cm. w0 = 3.1. yielding a singularity in the limit of an ideal at top beam. One more time we have to numerically compute the pm for realistic at modes.5 cm.2 cm). The same drawback happens in the case of ideally at modes. one sees the distribution of (gradE )2 in the case of an input Virgo mirror. we substitute the pm's in the preceding formulae.8.

070 4.05 0.8.000 7.75 -0.43 -0.00 1.3.035 0.10 Figure 8.140 2.80 -0.11 -0.1: Distribution of the square gradient of the temperature in the case of a gaussian beam.140 0.06 0. (Logarithmic scale. CASE OF FINITE MIRRORS 0.175 397 0. arbitrary units) .070 0.105 -0.105 0.035 6.175 0.

14 -0.070 0.175 0.140 0.000 4.140 1.19 -0.2: Distribution of the square gradient of the temperature in the case of a realistic at beam.00 0.23 0.035 3.10 Figure 8.05 0.070 2.175 CHAPTER 8.105 -0.10 -0.105 0.035 0. THERMOELASTIC NOISE 0. arbitrary units) .05 -0. (Logarithmic scale.398 0.

for the modulator 399 . In the present study. During the rst test runs of the new instrument. in a way easily done numerically. we show how to systematically construct these transfer functions. then the proper algebra being de ned. it will be necessary to permanently control the stationarity of statistical parameters. starting from elementary objects like mirrors and space between them. These elementary objects are so simple that one can rely on them. A correct association between noise characteristics and parts of the instrument requires information about the transfer functions relating elementary perturbations of these elements and output at the di erent ports of the interferometer. transfer functions for the motions of mirrors. many ltering techniques requiring a good knowledge of the spectral density of the instrument noise will be running on line. Moreover. increasing the reliability of the results with respect to long special analytical formulas. for laser frequency uctuations. and the correct result is automatically obtained.Chapter 9 Modulation and Transfer functions 9. one has simply to code the simple general analytical formulas. During the regular exploitation period. We shall present as rst results and examples. that are the commissionning of the instrument and the signal processing. identi cation of special types of noise will be a valuable aid to diagnostics and correction for bias.1 Introduction Analysis of the statistical structure of noise at the output of gravitational wave interferometers like LIGO or Virgo is essential regarding two di erent tasks.

For instance this has been shown in detail 30]. and for a gravitational wave. we consider only the projection of the amplitudes on the TEM00 mode. This is not a loss of generality because other perturbations can be shown to eventually reduce to an equivalent longitudinal displacement. In the case of scattered light. The modulation/demodulation system plays an essential role in the derivation of transfer functions. 9. a displacement of the re ecting surface along its normal. or equivalent motions. We restrict our attention to motions.g. for distortions globally equivalent to a displacement of the hot spot on the coating (e. substrate thermal noise) or for any phase perturbation mathematically equivalent to a displacement (e.400 CHAPTER 9. so that a rst order expansion is 9. MODULATION AND TRANSFER FUNCTIONS frequency uctuations. and we give the expressions for a lock in detection with a given phase lag. 31] for excited internal degrees of freedom of a real mirror if coupled to the beam phase (internal thermal noise). Small displacements of mirrors must be considered either for actual displacements (e. and changes in the vacuum properties due to a passing GW. This comprises a large class of phenomena. A special attention is devoted to the question of modulated quantum noise. The last one is necessary for evaluating the spectral densities of gravitational amplitudes equivalent to the various sources of noise and add them in a consistent way for obtaining the spectral sensivity of the instrument. x(t) of amplitude very small compared to a wavelength.2 Elementary perturbations and audio sidebands The basic elements constituting an interferometer are the mirrors and the vacuum space in between them.g. This approach is especially well adapted to object-oriented coding. We shall assume in nite at mirrors.2.g. scattered light recombination). The possible perturbations reduce consequently to two kinds : changes in position of the mirrors.1 Perturbation of mirrors by small displacements . and as only allowed perturbation. In the present approach. pendulum thermal noise). In other words. the noisy recombination e ect takes place on the mirror's surface and generate a phase that cannot be distinguished from a displacement phase. we shall describe the light beams circulating in the instrument as plane waves and mirrors as at surfaces.

and 2 f . Thanks to the linarity of the calculations. Obviously. if the incident wave now comes from the right. Assume a wave of the preceding form is re ected by our moving mirror. and all further computations are linear with respect to the displacement x.2. because the re ected wave experiences then an extra delay.9. it is allowed and convenient to consider x(f ) as the amplitude of a Fourier component of the motion at frequency f . whatever the cause of the phase uctuation is. (f ) = 4 x(f )= is the RSD of phase equivalent to the displacement. except on the splitter. the re ected amplitude B(t) is given by B (t) = i r A t ? 2x(t) cos =c].1) A(t) = A0 + 1 (f ) A1e?i t + 2 (f ) A2ei t e?i!Lt 2 where !L=2 is the laser frequency. ELEMENTARY PERTURBATIONS AND AUDIO SIDEBANDS 401 allowed. which leads us to consider the general case).3) . of the form 1 (9. and study the situation created by this elementary harmonic perturbation : The result is to add two sidebands to the main wave. Because we intend to study the noise in the detection band (a few Hz to a few kHz). the incidence angle being (almost all incidence angles in the interferometer are zero. we have to replace x by ?x. we call these "audio" sidebands. We have B (t) = i r A0 e?i!L(t?2xcos =c) + 1 (f ) A1ei (!L+ )(t?2x cos =c) 2 + 1 (f ) A2 ei (!L? )(t?2x cos =c) (9. so that the amplitude of light anywhere in the interferometer is modulated. We shall moreover consider x(t) as a zero mean random process of spectral density x2(f ) and will refer to x(f ) as its "root spectral density" (RSD). Substituting x(t) = x(f ) cos( t) and expanding this expression at rst order leads to ! ?i!L t 1 + 2 i x(f ) cos e?i t + 2 i x(f ) cos ei t B (t) = A0 e 1 + 2 (f ) A1 e?i (!L+ )t + 1 (f ) A2e?i (!L? 2 )t (9.2) 2 The factor of i is inserted for taking into account the necessary relative phase of =2 between the re ected and the transmitted wave at each partial re ection (all transmission coe cients will be thus taken real). provided that the incident wave is propagating to the right. i.e. This form will hold quite generally.

of the form ! h A e?i t + h A ei t e?i!t A(t) = A0 + 2 1 2 2 .e. Consider now a wave already modulated at the gravitational frequency f . having two sidebands proportional to h. and a right side operator R = R?1 . and we can write. With this operator notation. having the + polarization.2 Perturbation of a vacuum by a gravitational wave A passing gravitational wave will perturb light-distance measurements due to small changes in the space-time metrics.4) @ A i cos 0 1 If the wave is coming from the right. (common modes. A photon travelling along the x or y direction and detected at time t after a round trip of length 2L. But if we intend to study coherent motions of pairs of mirrors. was emitted at the retarded time L tr = t ? 2c ? hL sinc( L=c) cos (t ? L=c)] c where = 1 depending on the direction x or y ( 28]). A1. we have to change the sign of the non diagonal elements. MODULATION AND TRANSFER FUNCTIONS This veri es that the structure (carrier+2 sidebands) is stable and consequently it is allowed representing the modulated amplitudes by 3-vectors as : A = (A0. R10 for instance. This means that a mirror has two operators : a left side re ection operator R.2. or di erential modes. and is of no consequence if we consider only one perturbed mirror in a given con guration. 9. Re ection is then a linear operator.402 CHAPTER 9. for instance). is a transfer function relating the upper sideband amplitude to the RSD of phase (f ) = 2kx(f ). for a mirror of photometric re ectivity r. i. A2). Assume a GW of amplitude h and frequency f = =2 propagating along the z direction. This sign convention is quite arbitrary. and for a wave coming from the left : B = irRA the operator R having the form 0 1 1 00 R = B i cos 1 0 C (9. we have to be careful with the signs.

According to the above outlined philosophy.e. L ? f respectively. B (t) = A(tr) so that we obtain ! h B e?i t + h B ei t e?i!t B (t) = B0 + 2 1 2 2 with (we set K L=c): B0 = e2ikL A0 B1 = e2i(k+K)L A1 ? i kL sinc(KL)ei(2k+K)L A0 B2 = e2i(k?K)L A2 ? i kL sinc(KL)ei(2k?K)L A0 This can be represented as the action of the operator (see 1]) : 1 0 2ikL e 0 0 P (2L) = B i ei(2k+K)Lsinc(KL) e2i(k+K)L 0C (9. the o -diagonal terms will be considered only in the case of propagation in the km long Fabry-Perot cavities. however they may be signi cantly di erent from zero only on very long distances. the diagonal terms express the phase factor corresponding to ordinary propagation in a vacuum of waves of frequency L. a matter of algebra.3 Algebra of rst order perturbations Any history of modulated light through the interferometer is thus represented by a product of propagation and re ection rank 3 operators.2. ELEMENTARY PERTURBATIONS AND AUDIO SIDEBANDS 403 The propagated amplitude B (t) is nothing but the incoming wave taken at the retarded time. and consequently. The general form of any operator is : 0 1 O00 0 0 C O = B O10 O11 0 A (9.5) A @ i(2k?K )Lsinc(KL) 2i(k?K )L i e 0 e on vector amplitudes. Obviously. 9. L + f.2.6) @ O20 0 O22 . whereas P (2L)10 must be understood as the transfer function relating the upper sideband amplitude to the RSD of phase (f ) = kh(f )L. and are the seed of the whole detection process in an interferometer. i.9. Though the preceding expression is valid in general. the o -diagonal terms evaluate the creation of sidebands by the GW.

A Fabry-Perot cavity consists of two mirrors. 9. The equation relating the incoming eld Ain and the intracaviy eld B is B = t1Ain ? r1r2R1P (L)R2P (L) B (9. it is very simple. (22)) components can be deduced from (01) (resp (11)) components by simply changing into ? . In a numerical scheme. very fast (there is no need of a general matrix inversion) . separated by a vacuum gap of length L. owing to the fact that a motion of the transparent sustrate of the 9. and easy to implement in a numerical code. and consequently using rank 3 operators (in practice 5 components objects) is not a waste of time nor memory. isomorphous to the algebra of rst order expansions. We have presented for the sake of clarity the full three dimensional version of this algebra.404 CHAPTER 9. p2. t2. t1. MODULATION AND TRANSFER FUNCTIONS The product of two operators A. it is possible using A to compute global transmission and re ection operators between an input point and any output point. it is anyway necessary to evaluate all the components. the transmission coe cient t and the loss rate p. The power balance reads r2 + t2 = 1 ? p.3. however.1 Cavity . and one could argue a redundancy. B is 0 1 (AB )00 = A00B00 0 0 0C (AB ) = B (AB )10 = A10B00 + A11B10 (AB )11 = A11B11 @ A (AB )20 = A20B00 + A22B20 0 (AB )22 = A22B22 (9.7) ?1 of any operator A.8) This is a non commutative algebra we call A for brevity. due to the fact that (02) (resp. p1 and a maximum re ectance mirror M2 of parameters r2.9) in this expression we note that the transmission is represented by a pure scalar.3 Interferometer operators The basic parameters of a mirror are the re ection coe cient r. a coupling mirror M1 of parameters r1. we have : For the inverse A 0 ?1 1 (A )00 = 1=A00 0 0 0C A?1 = B (A?1)10 = ?A10=A00A11 (A?1)11 = 1=A11 @ A (A?1)20 = ?A20=A00A22 0 (A?1)22 = 1=A22 (9. We show hereafter how for any complex optical scheme.

we have here. We shall consider separately the e ects of perturbations. We have the following general structure : 0 1 F0 0 0 F = B G1 F1 0 C (9. and R1 at the right side of M1). we note that the R2 operator has been conjugated (or inverted) because the normal to M2 is in the opposite direction with respect to that of M1. We have B = 1 + r1r2R1P (L)R2P (L) t1Ain (9.14) M1 and M2 are pure scalars and P (2L) is the perturbed propagator.12) We can say that F is the Fabry-Perot re ectance operator.15) 1 + r1 2 ( (ordinary re ectance for the carrier and the two sidebands). 1) (9. or M1 or M2 is moving and then P (L) is diagonal. 1) (9. 2i(k+ K )L F = r1 + (1 ?rp1e)2ri2ke+ K)L ( = ?1.16) r1r2 1 r2 ?1 . We nd Aref = i F Ain with the general formula : h ih i?1 F = R1 r1 + (1 ? p1)r2R1P (L)R2P (L) 1 + r1r2R1P (L)R2P (L) (9.9. so that either P (L) is gravitationnally perturbed and then R1. instead of long special analytical formulas.3.10) Concerning the eld Aref re ected o the cavity. Though it is quite useless to know the details of the operators in a numerical scheme.13) @ A G?1 0 F?1 GW event : F = r1 + (1 ? p1)r2P (2L)] 1 + r1r2P (2L)]?1 (9. INTERFEROMETER OPERATORS 405 mirror does not a ect the phase of a transmitted wave. It contains three possible pertubations we enumerate below. where we stick to synthetic algebraic expressions as the preceding one. 2 i(2k+ K )L KL) G = i (1 + rr2t1ee2ikL) (1 +sinc(e2i(k+ K)L) ( = ?1. R2 reduce to identity. 0.11) 1 (remember that R1 corresponds to a re ection at the left side. we have Aref = i r1R1Ain + i t2r2P (L)R2 P (L) B (9. it is nevertheless interesting to see the e ect of these perturbations on the cavity A operator.

2 Michelson . Note the close similarity between the GW case and an M2 far mirror motion of amplitude 1 x(f ) = 2 h(f )L. ps . but now : ! r2t2e2ikL 1 G = ? i (1 + r r e2ikL) (1 + r r e2i(k+ K)L) ? F0 (9. with respect to the incoming carrier's.18) 1 2 1 2 Motion of the far mirror M2 h ih i?1 F = r1 + (1 ? p1)r2P (L)R2 P (L) 1 + r1r2P (L)R2P (L) (9.17) P (2L) is diagonal and M2 scalar. Motion of the coupling mirror M1 F = R1 r1 + (1 ? p1)r2R1P (2L)] 1 + r1r2R1P (2L)]?1 (9.406 CHAPTER 9. The elements F are still unchanged. Fwest of both cavities. MODULATION AND TRANSFER FUNCTIONS which represents the sidebands amplitudes created by the cavity while the GW event.1) then the transmittance is 9. L due to unavoidable asymmetries. it is easy to compute the transmittance Tmic and the re ectance Rmic of the Michelson. maybe having di erent parameters M1. ts. At low frequencies the sinc function can be replaced by 1 and the two formulas become identical. by a and b the short distances (see Fig. For more clarity we can denote by "North" and "West" the directions of the arms without loss of generality. M2. The components F are the same as in the precedent item.3. A GW Michelson interferometer like LIGO or Virgo involves two arms containing each a Fabry-Perot cavity. and the G are now : 2 t2 i(2k+ K )L (9. We denote by Ms the splitter of parameters rs. Knowing the operators Fnorth.19) P (L) is diagonal and M1 scalar.20) G = i (1 + r r e2rikL1)e(1 + r r e2i(k+ K)L) 1 2 1 2 (The change of sign with respect to the coupler formula comes from the opposite orientation of the normal).9.

22) In these expressions. we get 2 Rmic = t2P (a)FnorthP (a) ? rs Rs P (b)FwestP (b)Rs s (9. P (a) and P (b) can be understood as diagonal.21) Let us note that the motion of any mirror can be considered as the sum of a motion along its normal.3. In the operator Rs. Only the normal part gives rise to a phase lag.9. the incidence angle is taken as =4.1: General setup for a Power recycled interferometer Tmic = ?rsts R?1P (a)FnorthP (a) + P (b)FwestP (b)Rs s (9. neglecting a possible GW perturbation on so short distances. INTERFEROMETER OPERATORS West F2 407 light in l reflected light b a F1 North transmitted light Figure 9. There is therefore no ambiguity in the de nition of the left and right sides of the splitter Ms : these are determined with respect to the right oriented normal. and a motion orthogonal to it. For the re ectance. .

408 CHAPTER 9. It is straightforward. a code can build the cavity operators. We have simply for the global transmission of the system : Tglob = Titf TMC because mode-cleaners are designed (ring cavities) for suppressing re ected waves.24) At this point.23) Titf = tr TmicP (l) 1 + rr R?1 P (l)RmicP (l) r ih i?1 h Ritf = Rr rr + (1 ? pr )R?1 P (l)RmicP (l) 1 + rr R?1 P (l)RmicP (l) r r (9. there is no cavity between the MC and the interferometer. Moreover. and the detector receiving the light coming from the unavoidable spurious re ection o the back face (coated for anti-re ection) of the splitter (port #5).3 Recycled interferometer transmittance and reectance A recycling interferometer is a cavity involving a recycling mirror Mr of parameters rr . MODULATION AND TRANSFER FUNCTIONS 9. The operator associated with port #2 is clearly T2.3.1). without handling intricate analytical formulas : Once given the elementary operators (mirrors. and moreover the amplitude of the sidebands created inside by the motion of any mirror or a passing GW. pr at a distance l of a Michelson (see Fig. For instance. propagator). it is possible to associate a global operator describing the transfer from input light to any point of the interferometer. TMC being the mode-cleaner A operator. tr . For instance. namely the detector receiving a part of the light re ected by the interferometer (port #2). apart from the main port on which we have focused unitl now.9.24). we can do it by a constructive approach.2).glob = Ritf TMC . Elementary calculations lead to the transmittance Titf and the re ectance Ritf of the whole system. we are able to compute the transmission of a wave of arbitrary frequency through the interferometer.23. at least two other ports are of interest (see Fig. Any other port inside the interferometer can be treated the same way. The same way.9. the Michelson operators and eventually the interferometer's by using A and synthetic expressions like (9. using the same principles. h i?1 (9.9. to insert an input or output mode-cleaner.

to describe how the various cavities and the interferometer itself are tuned.itf TMC 9.mic = iPa Fnorth Pa This allows to compute the corresponding operator when recycling is applied (formula formally identical to 9.micP (l) 1 + rr R?1P (l)Rmic P (l) r then the operator associated with port 5 is simply T5. a new operator : T5.23) : h i?1 T5. In .9.itf = tr T5.glob = T5.4 Tuning the interferometer It is essential for understanding the results that we report below. baside the operators Rmic and Tmic related to the Michelson part.2: Location of the interferometer ports (this is why it is necesary to compute Ritf ). For port #5. All information on the tuning is contained in the various propagators encountered. TUNING THE INTERFEROMETER W 409 N 2 reflection off interferometer 1 main port 5 spurious reflection on back face of splitter Figure 9. it is convenient to de ne.4.

9. 046:558. we loss any precision on the ne tuning. by the modulation. The parameter allows to examine special modes of operation of the interferometer assuming detuned cavities. To obtain this result. The propagator's phase kL appearing in P (L) (see Eq. Resonance obviously corresponds to = 0. MODULATION AND TRANSFER FUNCTIONS these propagators. 748. We try therefore to extract from all phases a large static value that proves irrelevant in the calculations. The result is " ! # kL = 2 + + 1+ (9. value in which a variation of the last digit is enough to change the phase re ectance of a high nesse cavity by =2. In other words.4.064 m. We may assume L0 = (4n + 1) =4 where n is the largest integer less than 4L= . If we keep the full value of phases. or to represent DC servo errors. plus a small detuning L that we can express as a fraction of the linewidth of the cavity (of nesse F ) : 2F with ?1 1. especially in long (km) distance operators. 983 Rd. we get kL 17. and consider only o sets with respect to that frequency for describing for instance the transmission of sidebands.5) is thus : ! 2 ( + ) L + kL = c L 0 2F where represents an o set with respect to the laser frequency caused for instance.1 Tuning long cavities The optical length L of a long cavity can be considered as the sum of a resonant part L0. as 2 LL=c appear. huge phases involving the laser frequency.410 CHAPTER 9. and 3 km propagation. 715. and are di cult to numerically take into account without a special care. This way it is possible to get rid of the exact carrier frequency. Possible departures from resonance will be expressed with respect to the resonant length.25) F (mod 2 ) FSR L L = . For a light of wavelength 1. as will be seen in the next section.9. it is di cult to handle the same way kilometers and picometers. we shall consider that the lengths of the various cavities present in the interferometer are kept nearly resonant for the laser light (carrier).

b are integer multiples of the laser wavelength plus a small o set : a = a0 + =4 . This is obtained if 1 = 0 + 2 + Arg Fwest]00( L) ? Arg Fnorth]00( L) (9.26) FSR At the main output of the interferometer. kb of the propagators P (a).4. namely Tmic]00 = rsts e2ika Fnorth]00 + e2ikb Fwest]00 Assume that the short arms lengths a.4. The relevant information on this is contained in the (00) component of the Tmic operator.28) b kb = 2 c ? 2 1 + L (mod 2 ) 0 0 9. TUNING THE INTERFEROMETER FSR 411 = c=2L is the free spectral range of the cavity. the phase factors ka. The argument KL appearing in some components of P (L) is simply f KL = (mod 2 ) (9.9. This being calculated. P (b) are given by ka = 2 ac + 2 1 + L (mod 2 ) (9. we have j Tmic]00( L)j = rsts j Fnorth]00( L)j + ei' j Fwest]00( L)j where ' = 4 c L (b ? a) ? Arg Fnorth]00( L ) + Arg Fwest]00( L) The dark fringe at the laser frequency corresponds to ' = (mod 2 ). b = b0 ? =4 with a0 = na .2 Tuning at a dark fringe . At the laser frequency ( = 0).27) 2 where 0 represents a possible o set with respect to the dark fringe caused for instance by a DC servo error. It might be thought that we failed to get rid of the laser frequency L. and it is well known that the optimum signal to noise ratio is obtained when the the optical path di erence between the two arms is such that the extinction is a maximum. but the way it enters the last formula is now much less dangerous. because is appears in a small correction factor (even negligible in certain cases). b0 = nb . and a high precision value is no more needed for it. two partial waves returning from the two arms interfere.

MODULATION AND TRANSFER FUNCTIONS 9. of length l. We get = = + Arg Rmic ]00( L ) (mod 2 ).5. We have thus for the argument kl entering the propagator P (l) : ! ! 2 l0 + Arg Rmic ]00( L) 1 + kl = c (mod 2 ) 2 0+1? L (9.3 Tuning the recycling cavity The recycling resonance allows to increase the power reaching the splitter. is at resonance when D = j1 + rrec Rmic ]00e2iklj is a minimum. Detection.412 CHAPTER 9.1 General case . It is passed through a phase modulator in order to translate the detection band in a high frequency region where the laser frequency noise is 9. At the laser frequency. The recycling cavity (recycling mirror + Michelson) . Demodulation and Transfer functions The optical wave entering the interferometer is not a simple monochromatic wave.29) 9. which yields 0+1? The minimum of D( L ) is attained when = Arg Rmic]00( L) 0 allowing to take into account a possible o set with respect to resonance. this is D( L ) = 1 + rrecj Rmic]00( L)j ei where = 4 Ll=c + Arg Rmic ]00( L) We assume a length l = l0 + =4 where l0 is an integer multiple of and an adjustable parameter.4.5 Modulation.

f ) p We can then write the amplitude B (t) at the considered port : 0 X X B (t) = Ae?i!Lt @ tp Jp( )e?ip!t + 1 (f ) t+ Jp( )e?i(p!+ )t p 2 p2Z p2Z 1 1 (f ) X t? J ( )e?i(p!? )tA +2 (9. f ) depend also on the perturbation frequency. some spurious re ections. DEMODULATION AND TRANSFER FUNCTIONS413 lower. from where the gravitational information is expected to come. The action of an ideal phase modulator is analogous to a transmittance of the form T (t) = e?i sin !t (9. treating the tp as rst order terms : 0 X P (t) = B (t)B (t) = P0 @ tptq JpJq e?i(p?q)!t+ p. but the eld re ected by the recycling mirror. The power is. de ning transfer coe cients for each discrete Fourier component of the light amplitude : tp = S00( L + p fm ) t+ = S10( L + p fm .31) p2Z The interferometer contains a number of points where it is useful to detect the light amplitude. DETECTION. the eld weakly transmitted by the end mirrors. If the laser output amplitude is Ae?i!Lt.33) p p where we see that the e ect of the perturbed interferometer is to add two audio sidebands to every rf sideband. whereas S10( . MODULATION.5. X A0(t) = A Jp( )e?i(!L+p!)t (9. We shall use the following notation. Each of these amplitudes can be computed by the constructive way outlined above.30) where ! = 2 fm .9. f ). are also of some interest for the control of the instrument.q2Z p2Z . The parameter is the modulation depth. f ) (9. S20( . fm being the modulation frequency. the modulated amplitude A0(t) is a sum of a carrier and partial waves we call "rf sidebands" because fm is of the order of a few MHz. The main is obviously the dark fringe. The component S00( ) depends only on the frequency of the light source.32) p t? = S20( L + p fm . giving a suitable operator S 2 A.

It is therefore clear that in the preceding sum. de ned as 1 (f ) = 2 a+ ei ? a? e?i ] (9.34) (tp tq + tpt+)JpJq e?i (p?q)!? ]t A q 2 p. followed by a low pass ltering suppressing frequencies equal or higher than fm . = 0 gives the in-phase demodulation current. ltered current (DFC ) at frequency f is : i 1h DFC (t) = 4i (a+ ei ? a? e?i )e?2i ft + c:c (f ) (9. MODULATION AND TRANSFER FUNCTIONS 1 (f ) X (t+t + t t?)J J e?i (p?q)!+ ]t+ pq p q p q 2 p. only terms such that p ? q = 1 will contribute the demodulated ltered current. The function (f ).37) The function DFC (t) is given up to an arbitrary amplitude depending on the tuning of the various ampli ers of the detection chain. we are going to compare one another the DFC's due to di erent causes.39) where Jp is a shorthand notation for Jp( ).36) p p ? = ? t + t t+ ) a J J (t p2Z p p?1 p p?1 p p?1 CHAPTER 9. and = =2 the quadrature. and the undetermined common amplitude plays no role in the discussion.q2Z 1 X ? 1 (f ) (9.38) is thus the (complex) transfer function relating the RSD of DFC to the RSD of special phase noise : DFC (f ) = (f ) (f ) (9. We can thus write the contributing part Peff of the detected power as : 1 Peff (t)=P0 = a0e?i!t + 2 (f ) a+ e?i(!+ )t + 1 (f ) a? e?i(!? )t + c:c 2 (9.414 The demodulated.35) where the coe cients ak have the following de nitions : a0 = Pp2Z JpJp?1tptp?1 P a+ = Pp2Z JpJp?1(t+tp?1 + tpt??1) (9. The end of the process is a mixing with a demodulation current of the form D(t) = sin(!t + ) where denotes the demodulation phase.q2Z . Anyway.

9. the tk coe cients have been calculated from operap tors all diagonal except the propagators P (2Lnorth ) and P (2Lwest ). DETECTION.40) where in x!DFC . Our discussion of the quantum noise calculation is based on the approach by Niebauer et al.41) DFC (f ) = (f ) 2 L h(f ) h!DFC where in h!DFC . and we consider the associated centered process P (t) = hP tL (n(t) ? n0(t)) Having the variance ! hP L 2 n (t) = hP L P (t) V P (t)] = 0 t t 0 9.2 The special case of quantum noise .5. Consider a time interval t. From (f ). both useful in servo loops studies. one can extract the modulus and phase transfer functions. having an expectation value E n(t)] = n0(t). so that (9. The statistical parameter n0(t) is related to the averaged power P0(t) during t by (hP denoting the Planck constant) : 0 n0(t) = Ph(t) t P L We can reverse as well the point of view and consider the detected power as a random process. around time t. MODULATION. The number n(t) of photons reaching the photodiode during this time is a random variable obeying a Poisson statistics. DEMODULATION AND TRANSFER FUNCTIONS415 (f ) = 2k x(f ) for a moving mirror.5. the tk coe cients have been calculated from operap tors all diagonal except the special one corresponding to the perturbed mirror (f ) = k h(f )L for a GW event. 32] about non stationary shot noise. So that DFC (f ) = x!DFC (f ) 4 x(f ) (9. very short compared to the modulation period 1=fm . so that its variance is V n(t)] = n0(t).

and t.42) P L where t.416 CHAPTER 9. the detection current generated by the diode is (for its centered part) : I (t) = h e P (t) P L where e is the elementary charge. it admits an expansion in a Fourier series. MODULATION AND TRANSFER FUNCTIONS Assuming a quantum e ciency of 1. and the coe cients are Z ~ J (!) = 1 J (t) ei!t 2 where T is any multiple of 1=fm and consequently much longer than t. of variance: 2 V J (t)] = D(t)2V I (t)] = D(t)2 h e t P0 (t) P L Moreover.t0 (9. t0 are the centers of two time slices.t02T E J (t)J (t0)] ei(!t?!0t0) . The output current being periodic.t = 1. the demodulated current J (t) is given by J (t) = D(t) I (t) This de nes a new centered process. so that if t. The integral is thus fairly approximated by the discrete sum X ~ J (!) = t J (t) ei!t T T T t2T and we get ~ ~ E J (!) J (!) ] = T t 2 X t. The process I (t) has a variance V I (t)] = h e t P0(t) P L D(t) being the demodulation current.t06=t = 0. we can consider the uctuations P as uncorrelated between any two di erent time intervals. we have 2 E J (t)J (t0)] = h e t D2(t)P0(t) t.

9.5. MODULATION, DETECTION, DEMODULATION AND TRANSFER FUNCTIONS417
thanks to eq.9.42, we nd 2 2 X 2P ~ ~ E J (!) J (!) ] = h e T T t D(t)2 P0(t) ei(!?!0)t = h e T Dg 0(!?!0) P L P L t2T
2 2P ~ E jJ (!)j2] = h e T Dg 0(0) (9.43) P L This result is independent on t that we can take arbitrarily small, therefore, eq.(9.43) is exact. It follows that the spectral density of demodulated current is e2 Dg (0) 2 P0 Q(!) = h (9.44) P L The mean detected power being : X P0(t) = PL Jp Jq tp; tq e?i(p?q)!t

In particular,

p;q2Z

and the squared demodulation current 1 D(t)2 = sin(!t + )2 = 4 2 ? e2i!t+2i ? e?2i!t?2i we get 0 1 X J J t t e?i(p?q)!t ? e2i X J J t t e?i(p?q?2)!t ? D2P0(t) = 4 @2 p q pq p q pq
p;q2Z p;q2Z

e?2i

X
p;q2Z

JpJq tptq

e?i(p?q+2)!tA

1

so that the Fourier coe cient of the zero frequency is 2 3 1 42 X J 2t t ? e?2i X J J t t ? e2i X J J t t 5 2P Dg 0(0) = 4 p p?2 p p?2 p p?2 p?2 p p pp
p2Z p2Z p2Z

once substituted in (9.44), the RSD of quantum noise current is determined. Note that in the calculation of the DFC's due to classical perturbations (precedent subsection) a factor of eP0=hP L was ignored, as a part of a common arbitrary scale factor. It is necessary to remember it here : ignoring

418

CHAPTER 9. MODULATION AND TRANSFER FUNCTIONS

this factor is equivalent to take 1 as the amplitude of the demodulating current, and divide all DFC's by eP0=hP L. If we keep this convention, we must nally take for the spectral density of modulated quantum noise : s P DFCqn (f ) = 2hP L qn!DFC
0

The factor of 2 is necessary for passing to a one sided spectral density. We have otherwise : 1 h2a ? b e?2i ? b e2i i qn!DFC = 4 X a = Jp2tptp p2Z X b = JpJp?2tptp?2
p2Z

9.5.3 Transfer functions to an equivalent h(f)

An essential point is to compare the various perturbations acting on the interferometer to the expected gravitational signals. One way for doing it is to express these perturbations in terms of an equivalent spectral density h(f ) of gravitational amplitude. This is often implicitly done in papers. The method we propose is to identify the DFC produced by a GW to the DFC produced by any perturbation X of RSD X (f ) :

DFCh (f ) = DFCX (f )
or, introducing the transfer functions
h!DFC

h(f ) =

X !DFC

X (f )

This allows to express the hX (f ) equivalent to X (f ) as

hX (f ) =
X !h

X !h

X (f )

and de ne a new class of transfer functions : =
X !DFC h!DFC

9.6. INTERFEROMETER NOISES
10 8 Far mirror Coupling mirror

419

TF from δx(f) to DFC(f) [arb. units]

10 7

10 6 Splitter

10 5 Recycling mirror 10 4

10 3

10 0

10 1

10 2 frequency [Hz]

10 3

10 4

Figure 9.3: Transfer functions from x(f ) to DFC (f ) for 4 types of mirrors, 5% nesse asymmetry

9.6 Interferometer noises
9.6.1 Proof masses position noise
A rst possible use of our method could be to nd the transfer functions corresponding to motions of every mirror involved in a given instrument. This could help in the commissionning phase, when we are free to test the response of the interferometer to given, calibrated excitations. We show below (Fig.9.3) the behavior of the transfer functions relating the RSD of motion x(f ) of each mirror of a perfectly tuned interferometer to the resulting DFC root spectral density. The transfer function from the GW RSD to DFC is mostly identical to that of an end mirror, as already seen, apart from an extra cause of cuto , due to the sinc(2 fL=c) factor, representing the averaging e ect of propagation inside the cavity during a time comparable to the GW period. This is why the transfer functions

420

CHAPTER 9. MODULATION AND TRANSFER FUNCTIONS

from the displacements RSD to an equivalent h(f ) are not exactly constant (see Fig.9.4). We see that the transfer functions for the cavity mirrors are almost identical, the end mirror's one being slightly larger. The splitter's and the recycler's are much smaller (at least in the detection band). If furthermore we assume the laser locked in frequency to the recycling cavity, the laser frequency is correlated with the recycler's motions (error signal is taken a port 2). A simple approach assuming an in nite gain in the servo loop is to adopt the transfer function
xR !h;1

=

xR !DFC;1 ?

h!DFC;1

xR !DFC;2 L !DFC;2

L !DFC;1

where the numerical indices refer to the corresponding port. The transfer function L !DFC;p expresses the relation between the laser frequency noise and the DFC on port p, and will be expressed in detail in a coming section. It follows that the recycler's position noise is almost cancelled (see Fig.9.4), at least in the detection band. The irreducible part of the position noise is caused by small motions of the mirrors, essentially driven by excitation of all degrees of freedom of the various oscillators coupled to each. If we restrict ourselves to the main features, we can take into account the motion of the suspension (the mirrors are suspended like pendulums) and the motion of the re ecting face resulting from excitation of the internal modes. For the pendulum thermal noise, we adopt the following model ( 33]) assuming that the dissipation occurs due to a nite thermal conductivity in the wires : 1 T 2 x(f )2 = 2kBm !w ( 2 ? !2)2 + 2!4 p w with the following de nitions : kB is the Boltzmann constant, T the temperature, = 2 f as usual. The loss angle (f ) is of the thermoelastic form (f ) = 0 + 1 + ( )2

with the Virgo parameters (case of an end mirror) ( = 3 10?3 , = 2:34 10?4 s. The frequency corresponding to the elasticity of the steel wires (pendulum frequency in zero gravity) is !w = 2 0:017 Hz. The resultant pendulum g 2 2 frequency is !p = L + !w .

9.6. INTERFEROMETER NOISES

421

10-3 End mirror Input mirror 10-4

TF from δx(f) to h(f) [m-1]

10-5

Splitter

10-6 Recycling mirror, laser unlocked 10-7

10-8

Recycling mirror, laser locked

10-9

10 0

10 1

10 2 frequency [Hz]

10 3

10 4

Figure 9.4: Transfer functions from x(f ) to h(f ) for 4 types of mirrors,5% nesse asymmetry

422

CHAPTER 9. MODULATION AND TRANSFER FUNCTIONS

where l is the pendulum length. Obviously, the parameters must be slightly modi ed according to the type of the considered mirror. For the internal thermal noise, we take a very simple model (see 34], 35]) valid for the low frequency tail : x2(f ) = 4kB T U M f Where U is proportional to the strain energy stored in the assumed cylindrical substrate when a static pressure is applied having the same pro le as the light power ux (gaussian). It depends on the size of the blank and on the radius of the light spot. This model does not take into account resonances, that are likely at frequencies (several kHz) where only thin peaks will emerge from the shot noise. For instance, for the Virgo end or corner mirror, we nd U = 7:32 10?11J:N?2. The loss angle M can be as low as 10?6 for silica mirrors. Suspension wires have also a special thermal noise spectrum (violin modes), but essentially concentrated on thin resonance lines non essential for data analysis since a number of papers 39] have been devoted to removal from data of that kind of component. By taking for a given mirror the sum of all these contributions, applying its transfer function, we get the equivalent GW amplitude hi (f ) (i enumerates the mirrors). The shot noise RSD is a constant, and therefore, the transfer function from shot noise to an equivalent hQN (f ) is the inverted transfer function from h(f ) to DFC (see Fig.9.5). QN !DFC QN !h = then the h(f ) equivalent to shot noise is
h!DFC

9.6.2 Quantum noise

hQN (f ) =

s

QN !h

2hP L P0

9.6.3 Sensitivity curve

Then the (incoherent) sum of all hi(f ) thermal contributions gives a global hTHN (f ) equivalent to thermal noise. A new incoherent sum with the hQN gives an estimate of the sensitivity of the interferometer (see Fig.9.6), to be

9.7. UPSTREAM NOISES
10-21

423

h(f) equivalent to q. noise [Hz-1/2]

10-22

10-23 10 0

10 1

10 2 frequency [Hz]

10 3

10 4

Figure 9.5: h(f) equivalent to quantum noise taken into account, for instance when evaluating the e ciency of matched lters. This sensitivity curve is very well approximated by the t function : 0 2 ! 311=2 4:5 10?43 + 9 10?37 + 3:24 10?46 41 + f 25A h(f ) = @ f f5 500 Hz

9.7 Upstream noises
Some noises are caused by perturbations acting before entrance of light in the interferometer. We consider here the three main sources of upstream noise, the laser itself and the modulator.

9.7.1 Laser frequency noise

The laser may be noisy in phase and in amplitude (in power). Let us consider these two cases. The frequency noise will be described by a noisy optical

424

CHAPTER 9. MODULATION AND TRANSFER FUNCTIONS

10-18

-1/2

] Simplified sensitivity curve [Hz
10-19

Th er

10-20

m : al nd pe ul um

10-21

10-22

Therm
10-23 10 0

al : sub

strates
10 3

Qu

u ant

m

10 1

10 2 frequency [Hz]

10 4

Figure 9.6: h(f ) equivalent to the global noise SD

(f ) is very small compared even to small values of f ). The purpose of the tp notation is to avoid confusion with the tp of the preceding section that have a di erent meaning. UPSTREAM NOISES phase (t) : 425 1 @ = (t) L + 2 @t Where L is the averaged frequency. The tp express the .7.45) The interferometer being static. After a rst order expansion (in the frequency region of interest. f ) tp? = S22( L + p fm . f ) (9. instead of its equivalent (f )=f . the transfer coe cients are : (f ) cos(2 ft) tp = S00( L + p fm ) tp+ = S11( L + p fm . S being as above the A operator associated with the comsidered port of the interferometer.9. we get the laser output amplitude as the sum of a carrier plus two sidebands : # " ?i!L t + (f ) e?i(!L + )t ? (f ) e?i(!L ? )t A(t) = A0 e 2f 2f After passing the phase modulator. We shall as usual consider a special Fourier component of the frequency noise and write (t) = so that the phase is ( (t) = !Lt + ff ) sin( t) We could have directly introduced a RSD of "laser phase noise" (f ). the transmittance is the ordinary scalar transmittance. and (t) a centered stationnary random process of RSD (f ). the amplitude becomes 2 3 X ?ip!t (f ) X J e?i(p!+ )t ? (f ) X J e?i(p!? )t5 A0(t) = A0 e?i!Lt 4 Jp e + 2f p 2f p2Z p p2Z p2Z each of these partial waves is transmitted by the interferometer according to their frequency.

426

CHAPTER 9. MODULATION AND TRANSFER FUNCTIONS

rate of creation of sidebands inside the interferometer, whereas the tp express the transmission by the interferometer of sidebands already generated The transmitted amplitude is now : 2 X X X B (t) = A0 e?i!L t 4 Jp tp e?ip!t + 2(ff ) Jp tp+ e?i(p!+ )t ? 2(ff ) Jp tp? e?i(p!? p2Z p2Z p2Z And the power reaching the photodiode : 2 X X P (t) = P0 4 Jp Jq tp tq e?i(p?q)!t + 2(ff ) Jp Jq (tp+tq ? tptq?)e?i (p?q)!+ p;q2Z p;q2Z 3 X ? 2(ff ) Jp Jq (tp?tq ? tptq+)e?i (p?q)!? ]t5 Applying the demodulation/ ltering scheme already detailed above, we obtain the RSD of DFC as : ( DFC (f ) = !DFC (f ) ff ) where the complex transfer function is de ned as in the precedent section (Eq.9.38) by 1 ?i !DFC (f ) = a+ ei ? a? e ] 2 where a have the following de nitions : a+ = P1 ?1 Jp Jp?1(tp+tp?1 ? tptp?1 ? ) p= (9.46) a? = P1 ?1 Jp Jp?1(tptp?1+ ? tp?tp?1) p= Fig.9.7 shows the behavior of the transfer function in the case where some asymmetry in the arms (di erent nesses) makes the interferometer sensitive to frequency noise Assume now uctuations of the laser power, such that the averaged power is P and the instantaneous power P (t) the sum of P plus a centered random process of RSD P (f ) : P (t) = P0 + P (f ) sin(2 ft) (9.47)
p;q2Z
]t

)

9.7.2 Laser amplitude noise

9.7. UPSTREAM NOISES

427

10-15 ∆F/F = 0.05 10-16

Tr. F. to equivalent h [Hz-1]

∆F/F = 0.01

10-17 10-18 10-19 10
-20

∆F/F = 0

10-21 10-22 10 0

10 1

10 2 frequency [Hz]

10 3

10 4

Figure 9.7: Transfer function for laser frequency noise with and without mode-cleaner

428

CHAPTER 9. MODULATION AND TRANSFER FUNCTIONS
P (t)1=2

! P (f ) sin( t) = A0 1 + 2P (9.48) 0 So that the complex amplitude can be written as ! ?i!L t 1 + i P (f ) e?i t ? i P (f ) ei t A(t) = A0e (9.49) 4P0 4P0 The quantity P (f )=2P0 plays here the role of a phase RSD. Then, a treatment similar to the preceding leads to the transfer function ( DFC (f ) = P !DFC (f ) PPf ) (9.50) 2 0 with 1 a ei ? a e?i ] P !DFC (f ) = ? 2 + where the a have the following de nitions : a+ = P1 ?1 Jp Jp?1(tp+tp?1 + tptp?1 ?) p= (9.51) a? = P1 ?1 Jp Jp?1(tp?tp?1 + tptp?1+) p=
The coe cients tp, tp have the same de nition as above (Eq.9.45). Fig.9.8 show the transfer function in the case where some detuning of the cavities makes the interferometer is sensitive to laser power noise. It uses an experimental spectral density of laser power noise measured on the Virgo laser, that can be t by the following expression : P (f ) = 3:1 10?6 + 1:82 10?9 + 8:18 10?17 f 2 P f 1:5

The modulus of the amplitude is thus

If the oscillator driving the phase modulator presents some frequency noise, some e ects could be a priori expected on the interferometer noise. These e ects should however be small, if the demodulation current comes from the same oscillator. Roughly speaking, what matters are the di erences between

9.7.3 Modulator noise

9.7. UPSTREAM NOISES
10-18 10-19 10-20

429

h(f) [Hz-1/2]

10-21 10-22 Thermal + Quantum 10-23 10-24 10-25 10 0

10 1

10 2 frequency [Hz]

10 3

10 4

Figure 9.8: Transfer function for laser amplitude noise. 10?11m detuning of the black fringe the frequencies of the rf sidebands and that of the demodulator, and they are constants unless some extra noise is fed into the demodulator. The modulator has the transmittance T (t) = e?i sin (t) where the phase (t) obeys 1 @ = f + (f ) sin(2 ft) m 2 @t so that ( (t) = !t ? ff ) cos t and (Jp Jp( ))

T (t) = e?i

sin !t ei

f

( )

f cos !t cos t

430

where (f ) = (f )=f is the RSD of phase noise. Thanks to well known properties of the Bessel functions, we can write as well : X ( X ( X T (t) = Jpe?i p!t + i 2f ) pJp e?i (p!+ )t + i 2f ) pJp e?i (p!? )t p2Z p2Z p2Z The wave transmitted by the interferometer is thus 2 X X ( X B (t) = A0 4 tp Jp e?i (!L+p!)t + i 2f ) p tp+ Jp e?i (!L+p!+ )t + p tp? Jp e?i (!L+p!? p2Z p2Z p2Z with the same de nition as above (Eq.9.45) for the tp; tp . The demodulation current must contain the frequency noise : ! (f ) cos t + D(t) = sin !t ? f After some straightforward algebra, we nd for the RSD of DFC (f ) DFC (f ) = (f ) f with as customary (f ) = 1 a+ei ? a?e?i ] 2 and in this special case : P a+ = Pp2Z JpJp?1 (ptp+tp?1 ? (p ? 1)tptp?1? ? tptp?1) (9.52) a? = ? p2Z JpJp?1 (ptp?tp?1 ? (p ? 1)tptp?1+ ? tptp?1) The third term in each parenthesis represents the demodulator's noise. The essential feature is that independently taken, the modulator and the demodulator noises have the same 1=f behavior at low frequencies, but in the above formula, assuming a perfect coherence of phase between the modulating and the demodulating currents, they almost exactly cancel each other at low frequency. This is apparent on Fig.9.9. Anyway, even with large perturbations, the noise level remains negligible, well below the sentitivity curve (see Fig.9.10). .

CHAPTER 9. MODULATION AND TRANSFER FUNCTIONS 0 1 X ?i p!t i (f ) X ?i (p+1)!t X ?i (p?1)!tA = Jpe + 2 cos t @ Jp e + Jp e
p2Z p2Z p2Z

9.7. UPSTREAM NOISES

431

10-17

Transfer function [arb. units]

10-18

10-19

Total
10
-20

PM

onl

y

M
10-21

ix er

on ly

10-22 10 0

10 1

10 2 frequency [Hz]

10 3

10 4

Figure 9.9: Transfer function for modulation frequency noise. 10?12 m detuning of 1 cavity

432

CHAPTER 9. MODULATION AND TRANSFER FUNCTIONS

10-18

] h(f) equivqlent to modulator noise [Hz

-1/2

10-19 10-20 10-21 10-22 10
-23

Mod Dem

ulato

odul

r onl

y

Virgo se

ator

nsitivity

only

10-24 10
-25

Modulator + Demodulator

10-26 10 0

10 1

10 2 frequency [Hz]

10 3

10 4

Figure 9.10: h(f) equivalent to modulator frequency noise for /1000 detuning of one cavity

Bibliography
1] Jean-Yves Vinet, Brian Meers, Catherine Nary Man and Alain

Brillet

Optimization of long-baseline interferometers for gravitational-wave detection Phys. Rev. D , Vol 38, N.2 (1988) p. 433

2] Brian J. Meers Recycling in laser-interferometric GW detectors Phys. Rev. D , Vol 38, N.2 (1988) p. 2317 3] Ronald Drever in Gravitational Radiation edited by N. Deruelle and T. Piran (North Holland 1983) p.321 4] F. Pegoraro, L.A. RAdicati, Ph. Bernard and E. Picasso Phys. Lett. A68 (1978) p.165 5] J. Mizuno, K.A. Strain, P.G. Nelson, J.M. Chen, R. Schilling,

A. Rudiger, W. Winkler and K. Danzmann Phys. Lett. A175 (1993) p.273 6] Jun Mizuno in
7] Max BORN & Emil WOLF Principles of Optics Pergamon Press 1980 433

Comparison of optical con gurations for laser-interferometric gravitational-wave detectors PHD Thesis, Max-Planck-Institut fur Quantenoptik (MPQ 203) 1995

951 (1968) 16] Erika D'Ambrosio to be published in Phys. N6....... Universite Paris-Sud (1990) .. p... p. Tukey An algorithm for the machine calculation of complex Fourier series Math.. p... Book (Oxford University Press) 86 12] H. Opt. pp.13...434 BIBLIOGRAPHY 8] Fritz Oberhettinger Tables of Bessel Transforms Springer 1972 9] Fritz Oberhettinger Tabellen zur Fourier Transformation Springer 1957 10] J. Cooley and J.1874 (1975) 14] Patrice Hello. 1996 vol 27.. Tingye Li Laser beams and resonators Appl.... Siegman Lasers Univ..W... 5.. p.. Sc. 17] Patrice Hello Modele physique et simulation de l'antenne interferometrique gravitationnelle Virgo These de Doctorat en Sciences.. Comput. Paris. Opt.. Opt. Siegman Mode calculation in unstable resonators with owing saturable gain I : Hermite-Gaussian expansion Appl....W. E........ D .265-276 15] Michael Hercher The spherical mirror Fabry-Perot interferometer Appl.19. Kogelnik..2775 (1974) Mode calculation in unstable resonators with owing saturable gain II : FFT methods Appl. Opt. jean-Yves Vinet Simulation of beam propagation in o -axis optical sytems J... pp.1550 (1966) 13] E. Rev.14. Sziklas.. A..... Opt.. 7.297-301 (1965) 11] Anthony E.

Lett. Stegun Handbook of mathematical functions Dover 435 21] Virendra N. of Photo-optical Instrumentation Engineers April 18-21. 2243 24] Yu. Mahajan Zernike circle polynomials and optical aberrations of systems with circular pupils Engineers & Laboratory Notes (Aug. Jean-Yves Vinet Thermal noise in mirrors of interferometric gravitational wave antennas Phys. 1977. Abramowitz and I. Reston (Virginia/USA) 20] M. D 57 (1998) p.659 25] Francois Bondu. A 246 (1998) pp. France 51 (1990) p.A.K. July 1969 19] B. Phys. Rev. Bergland A guided tour of the fast Fourier Transform IEEE spectrum. 1994) pp.34 . 83 (1951) p. Soc. Patrice Hello.D. Likeness Stray light simulation with advanced Monte-Carlo techniques Proc. 21-24 22] Patrice Hello and Jean-Yves Vinet Analytical models of thermal aberrations in massive mirrors heated by high power laser beams J. Levin Internal thermal noise in the LIGO test masses : a direct approach Phys. France 51 (1990) p. Rev. Callen and T.1267 23] Patrice Hello and Jean-Yves Vinet Analytical models of transient thermoelastic deformations of mirrors heated by high power cw laser beams J. Welton Phys.BIBLIOGRAPHY 18] G. 227-236 26] H.B. Phys.

D 67 102004 39] Soma Mukherjee and Lee Samuel Finn To be published in the proceedings of the third Amaldi Meeting (AIP Press) .M. F. Thorne Phys. A 198 (1995) p. Vinet Phys. A 178 (1993) P. Bondu. Rev D (gr-qc/0002055) 36] V. Rev. Marchesoni. Kip S. Vol. Thorne in press for Phys.Bondu and J-Y. A 264 (1999) 37] Landau & Lifshitz Course of theoretical physics. Lett. L. 1959 38] Erika D'ambrosio.252 29] Timoshenko S. A 43 (1991) p. Cagnoli. Vinet Phys.L. 357 32] T. Lett. Gorodetsky and S. J-Y. Rev. Niebauer et al.P. A 246 (1998) p.436 BIBLIOGRAPHY 27] Fritz Oberhettinger Tables of Bessel Transforms Springer-Verlag (1972) 28] Jean-Yves Vinet Journal de Physique (Paris) 10 (1985) p. Hello. Punturo Virgo note PJT94/032 34] F.Gillespie and F. P. Raab Phys. Phys.B Braginsky.7 Pergamon. 74-78 31] A. Kip S. Lett. M.227 35] Yuk Tung Liu. M. Vyatchanin Phys.N. and Goodier J. Gammaitoni. 5052 33] C. Lett. Theory of elasticity Mc Graw & Hill 1951 30] F.

Master your semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master your semester with Scribd & The New York Times

Cancel anytime.