Preface This volume presents five review articles of areas of current research interest in several branches of modern o...
85 downloads
1195 Views
5MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
Preface This volume presents five review articles of areas of current research interest in several branches of modern optics. The first article, by U. Keller, discusses the progress which has been made in recent years in ultrafast pulse generation, using solid-state lasers. In the last decade, the performance of compact ultrafast solid-state lasers has been improved by several orders of magnitude in pulse duration, in average power, in pulse energies and in pulse repetition rates. In this article these and other breakthroughs are discussed. The article provides both the expert and the non-expert with an overview of this rapidly advancing field. The second article, by A.V. Shchegrov, A.A. Maradudin and E.R. Méndez, reviews research on the scattering of electromagnetic waves from randomly rough surfaces. Unlike earlier investigations in this field, the article covers the more difficult subject of multiple scattering from such surfaces and discusses more recently discovered effects such as enhanced backscattering and enhanced transmission. In the next article, by Y. Ishii, an interesting advance in interferometry is described, namely a phase-measuring technique which uses direct frequency modulation of a laser diode source by changing the current. This technique has been implemented in holographic interferometry and in phase conjugate interferometry, for example. The fourth article, by J. Gea-Banacloche, reviews the theory of quantum teleportation and the ways in which this intriguing quantum phenomenon has been experimentally demonstrated in optical systems. Among the topics covered are the relationship between teleportation of discrete and continuous variables, entanglement and teleportation, the difficulty of Bell measurements, and schemes for near-deterministic teleportation with linear optics. The limitations of current experiments and some questions of interpretation are also discussed. The concluding article, by H.J. Carmichael, G.T. Foster, L.A. Orozco, J.E. Reiner and P.R. Rice, uses intensity–field correlation functions of the electromagnetic field as a tool for studying quantum fluctuations of light. The relationship between the correlation functions and quadrature squeezing is noted and conditions are developed to distinguish between classical and non-classical field fluctuations. The v
vi
Preface
theoretical analysis is illustrated by examples such as the optical parametric oscillator, a cavity QED system and the composite system of a single atom coupled to an optical parametric oscillator. The results of experimental measurements on a cavity QED system are also reviewed.
Emil Wolf Department of Physics and Astronomy and the Institute of Optics University of Rochester Rochester, NY 14627, USA February 2004
Contents Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
v
Chapter 1. Ultrafast solid-state lasers, Ursula Keller (Zürich, Switzerland) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
List of symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . § 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . § 2. Definition of Q-switching and mode locking . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Q-switching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2. Mode locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . § 3. Overview of ultrafast solid-state lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Overview of solid-state laser materials . . . . . . . . . . . . . . . . . . . . . . . . . 3.2. Design guidelines of diode-pumped solid-state lasers . . . . . . . . . . . . . . . . . § 4. Loss modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1. Optical modulators: acousto-optic and electrooptic modulators . . . . . . . . . . . 4.2. Saturable absorber: self-amplitude modulation (SAM) . . . . . . . . . . . . . . . . 4.3. Semiconductor saturable absorbers . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4. Effective saturable absorbers using the Kerr effect . . . . . . . . . . . . . . . . . . § 5. Pulse propagation in dispersive media . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1. Dispersive pulse broadening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2. Dispersion compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . § 6. Mode-locking techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2. Haus’s master equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3. Active mode locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4. Passive mode locking with a slow saturable absorber and dynamic gain saturation . 6.5. Passive mode locking with a fast saturable absorber . . . . . . . . . . . . . . . . . 6.6. Passive mode locking with a slow saturable absorber without gain saturation and soliton formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.7. Soliton mode locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.8. Design guidelines to prevent Q-switching instability . . . . . . . . . . . . . . . . . 6.9. External pulse compression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . § 7. Pulse characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1. Electronic techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2. Optical autocorrelation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3. FROG and SPIDER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . § 8. Carrier envelope offset (CEO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . § 9. Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
3 7 10 10 11 15 15 39 43 43 44 49 52 54 54 56 65 65 66 70 74 77 80 81 87 89 90 90 91 92 93 95 98
viii
Contents
Chapter 2. Multiple scattering of light from randomly rough surfaces, Andrei V. Shchegrov (Rochester, NY, USA), Alexei A. Maradudin (Irvine, CA, USA) and Eugenio R. Méndez (Ensenada, Mexico) . . . . . . . . . . .
117
§ 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . § 2. Characterization of randomly rough surfaces . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Two-dimensional random surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2. One-dimensional random surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . § 3. Equations for electromagnetic fields and observable quantities . . . . . . . . . . . . . . 3.1. Physical quantities studied in rough surface scattering problems . . . . . . . . . . . 3.2. Rayleigh method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3. Surface integral equations for electromagnetic fields . . . . . . . . . . . . . . . . . § 4. Weak localization effects in the multiple scattering of light from randomly rough surfaces. Enhanced backscattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1. The nature of enhanced backscattering effect . . . . . . . . . . . . . . . . . . . . . 4.2. Theoretical methods employed in the study of multiple-scattering phenomena . . . 4.3. Experimental techniques used in the study of multiple scattering effects, including the enhanced backscattering effect . . . . . . . . . . . . . . . . . . . . . . . . . . . § 5. Angular intensity correlation functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1. Memory and reciprocal memory effects: theory and experiment . . . . . . . . . . . 5.2. The correlation function C (10) (q, k|q , k ): theory and experiment . . . . . . . . . 5.3. Correlations in film systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . § 6. Multiple-scattering effects in the scattering of light from complex media bounded by a rough surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1. Coherent effects associated with the interference of nonreciprocal optical paths . . 6.2. Scattering from the random surface of an amplifying medium . . . . . . . . . . . . 6.3. Scattering from a nonlinear medium bounded by a rough surface . . . . . . . . . . § 7. Near-field effects: localization phenomena for surface waves . . . . . . . . . . . . . . . § 8. Spectral changes induced by multiple scattering . . . . . . . . . . . . . . . . . . . . . . § 9. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 3. Laser-diode interferometry, Yukihiro Ishii (Sagamihara, Japan) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . § 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . § 2. Laser-diode operation . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Single-mode laser diodes . . . . . . . . . . . . . . . . . . . . . 2.2. Wavelength tunability in laser diodes . . . . . . . . . . . . . . . § 3. Modulation methods in laser-diode interferometers . . . . . . . . . . § 4. Laser-diode phase-shifting interferometers . . . . . . . . . . . . . . 4.1. Single-wavelength phase-shifting interferometry . . . . . . . . 4.2. Phase-shift calibration . . . . . . . . . . . . . . . . . . . . . . . 4.3. Laser-diode Fizeau interferometry . . . . . . . . . . . . . . . . 4.4. Phase-extraction algorithm insensitive to changes in LD power 4.5. Two-wavelength phase-shifting interferometry . . . . . . . . . § 5. Sinusoidal phase-modulating interferometry . . . . . . . . . . . . . § 6. Feedback interferometry . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
119 121 121 125 126 126 129 131 133 133 137 169 180 184 186 188 189 189 200 203 217 221 229 233 233
243 . . . . . . . . . . . . .
245 246 247 248 250 251 253 260 264 273 277 283 285
Contents
ix
§ 7. Heterodyne interferometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1. Single-wavelength heterodyne interferometry . . . . . . . . . . . . . . . . 7.2. Two-wavelength heterodyne interferometry . . . . . . . . . . . . . . . . . § 8. Optical coherence function synthesized by tunable LD . . . . . . . . . . . . . . § 9. Holographic interferometer and phase-conjugate interferometer by tunable LD § 10. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
Chapter 4. Optical realizations of quantum teleportation, Julio GeaBanacloche (Fayetteville, AR, USA) . . . . . . . . . . . . . . . . . . . . . .
311
§ 1. A brief primer on quantum teleportation . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2. Teleportation of a two-state system (qubit) . . . . . . . . . . . . . . . . . . . . . . 1.3. Generalization to an N -state system . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4. Continuous-variable teleportation . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5. Imperfect teleportation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.6. Possible applications and extensions . . . . . . . . . . . . . . . . . . . . . . . . . . § 2. Optical teleportation of discrete variables . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. The difficulties of Bell-state measurements . . . . . . . . . . . . . . . . . . . . . . 2.2. Teleportation using entanglement between different degrees of freedom of the same photon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3. Teleportation with less than a full Bell-state measurement (conditional teleportation) 2.4. Low-efficiency teleportation with complete Bell-state measurement . . . . . . . . . 2.5. Near-deterministic teleportation of photonic qubits using only linear optics . . . . § 3. Optical teleportation of continuous variables . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Squeezed states as EPR states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2. Teleportation fidelity for coherent states . . . . . . . . . . . . . . . . . . . . . . . . 3.3. The role of “classical” fields, and the nature of laser fields . . . . . . . . . . . . . . § 4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Chapter 5. Intensity-field correlations of non-classical light, H.J. Carmichael (Auckland, New Zealand), G.T. Foster (New York, NY, USA), L.A. Orozco (Stony Brook, NY, USA), J.E. Reiner (Stony Brook, NY, USA) and P.R. Rice (Oxford, OH, USA) . . . . . . . . . . . . . . . . . . . § 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . § 2. Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. The intensity-field correlation function hθ (τ ) . . . 2.2. Classical bounds for hθ (τ ) . . . . . . . . . . . . . 2.3. Time reversal properties of hθ (τ ) . . . . . . . . . 2.4. Intensity-field correlations in classical optics . . . § 3. Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1. Optical parametric oscillator . . . . . . . . . . . . 3.2. Cavity QED . . . . . . . . . . . . . . . . . . . . . 3.3. Two-level atom in an optical parametric oscillator
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
286 287 294 299 301 302 303 303
313 313 313 317 319 321 324 326 326 329 332 336 337 342 342 345 347 350 350
355 . . . . . . . . . .
357 359 360 363 364 365 366 369 372 381
x
Contents
§ 4. Experiment in cavity QED . . . . . . . . . . . . . . . . . . . 4.1. Cavity QED apparatus . . . . . . . . . . . . . . . . . . . 4.2. Conditional homodyne detector . . . . . . . . . . . . . . 4.3. Measurements . . . . . . . . . . . . . . . . . . . . . . . § 5. Equal-time cross- and auto-correlations . . . . . . . . . . . . 5.1. Cross-correlations . . . . . . . . . . . . . . . . . . . . . § 6. Quantum measurements and quantum feedback . . . . . . . . 6.1. Weak measurements . . . . . . . . . . . . . . . . . . . . 6.2. Vacuum state squeezing versus squeezed classical noise 6.3. Application of hθ (τ ) to quantum feedback . . . . . . . § 7. Conclusion and outlook . . . . . . . . . . . . . . . . . . . . . Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
Author index for Volume 46 . . . . Subject index for Volume 46 . . . . Contents of previous volumes . . . Cumulative index – Volumes 1–46
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . .
. . . .
. . . .
. . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . .
. . . .
. . . .
. . . .
. . . . . . . . . . . . .
385 385 386 388 393 394 396 397 398 400 402 403 403
405 423 427 437
E. Wolf, Progress in Optics 46 © 2004 Elsevier B.V. All rights reserved
Chapter 1
Ultrafast solid-state lasers by
Ursula Keller Swiss Federal Institute of Technology (ETH), Physics Department, Zürich, Switzerland
ISSN: 0079-6638
DOI: 10.1016/S0079-6638(03)46001-0 1
Contents
Page List of symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
§ 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7
§ 2. Definition of Q-switching and mode locking . . . . . . . . . . . . . .
10
§ 3. Overview of ultrafast solid-state lasers . . . . . . . . . . . . . . . . .
15
§ 4. Loss modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
§ 5. Pulse propagation in dispersive media . . . . . . . . . . . . . . . . . .
54
§ 6. Mode-locking techniques . . . . . . . . . . . . . . . . . . . . . . . .
65
§ 7. Pulse characterization . . . . . . . . . . . . . . . . . . . . . . . . . .
90
§ 8. Carrier envelope offset (CEO) . . . . . . . . . . . . . . . . . . . . . .
93
§ 9. Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
98
2
List of symbols
pulse envelope normalized such that |A(z, t)|2= P (z, t) (eq. (6.2)) A˜ is the Fourier transformation of A, i.e. A˜ = A(t)e−iωt dt (eqs. (6.8) and (6.9)) laser mode area on saturable absorber AA AL laser mode area in laser gain media Ap pump mode area b depth of focus or confocal parameter of a Gaussian beam D dispersion parameter (eq. (6.18) and Table 7), i.e. half of the total group delay dispersion per cavity roundtrip gain dispersion (eq. (6.10)) and Table 7) Dg Dp width of the pump source (i.e. approximately the stripe width of a diode array or bar) DR(t) impulse response of a saturable absorber mirror measured with standard pump probe (fig. 4) E electric field of the electromagnetic wave is the Fourier transformation of E, i.e. E = E(t)e−iωt dt (eqs. (6.6) E E and (6.7)) intracavity pulse energy Ep Ep,out output pulse energy Esat,A absorber saturation energy (Table 3) Esat,L laser saturation energy Esat,L = Fsat,L · AL P (t ) f (t) f (t) = Ep , with f (t) dt = 1 (eq. (4.7)) frep pulse repetition frequency F fluence, F = I (t) dt, in units of cmJ 2 Fsat,A absorber saturation fluence (Table 3) Fsat,L laser saturation fluence (eqs. (3.1) and (3.2)) Fp,A incident pulse fluence on saturable absorber (Table 3) g saturated amplitude laser gain coefficient small signal amplitude laser gain g0
A A˜
3
4
hν I IA Isat,A k kn l
lout ls La Lg M Ms M2 2 Mslow 2 Mfast n n2 P P0 Pabs Pav,out q q0 qp qs R S t tD T
Ultrafast solid-state lasers
[1
photon energy intensity incident intensity on saturable absorber absorber saturation intensity (Table 3) vacuum wave number, i.e. k = 2π/λ wave number in a dispersive media, i.e. kn = nk total saturated amplitude loss coefficient. l includes the output coupler, all the residual cavity losses and the unsaturated loss of the saturable absorber amplitude loss coefficient of output coupler amplitude loss coefficient of soliton due to gain filtering and absorber saturation (eq. (6.53)) absorption length length of laser gain material modulation depth of loss modulator (eq. (6.11)) curvature of loss modulation (eq. (6.11) and Table 7) M 2 factor defining the laser beam quality (eq. (3.3)) M 2 factor in the “slow” axis, parallel to the pn junction of the diode laser M 2 factor in the “fast” axis, perpendicular to the pn junction of the diode laser refractive index of a dispersive media nonlinear refractive index (eq. (6.20)) power peak power of pulse absorbed pump power average output power saturable amplitude loss coefficient (i.e. nonsaturable losses not included) (eq. (4.2)) unsaturated amplitude loss coefficient or maximal saturable amplitude loss coefficient (eq. (4.1)) total absorber loss coefficient which results from the fact that part of the excitation pulse needs to be absorbed to saturate the absorber residual saturable absorber amplitude loss coefficient for a fully saturated ideal fast absorber with soliton pulses (eq. (6.46)) reflectivity for intensity Fp,A (Section 4.2.1) saturation parameter S = Fsat,A time time shift (eq. (6.37)) time that develops on a time scale of TR (eq. (6.1))
1]
List of symbols
5
Tout TR Vp W0,G W0,opt x z
intensity transmission of the laser output coupler cavity roundtrip time pump volume beam waist of a Gaussian beam (eq. (3.3)) optimized beam waist for efficient diode pumping (eq. (3.5)) chirp parameter (eqs. (6.30) and (6.31)) pulse propagation distance
z0 A R Rns T Tns λg νg δL φ φnl φs (z) φs φnl (z)
Rayleigh range of a Gaussian beam, i.e. z0 = λ 0 change in the pulse envelope modulation depth of a saturable absorber mirror (fig. 5) nonsaturable reflection loss of saturable absorber mirror (fig. 5) modulation depth of saturable absorber in transmission nonsaturable transmission loss of saturable absorber FWHM gain bandwidth ν λ FWHM gain bandwidth, i.e. ν0g = λ0g SPM coefficient (eq. (6.22) and Table 7) phase shift, φ = kn · z, z: propagation distance nonlinear phase shift per cavity roundtrip (eq. (6.33)) phase shift of the soliton during propagation along the z-axis (eq. (6.51)) phase shift of the soliton per cavity round trip (eq. (6.54)) nonlinear phase shift of a pulse with peak intensity I0 during propagation through a Kerr media along the z-axis, i.e. φnl (z) = kn2 I0 z phase shift (eq. (6.28)) absorber coefficient (eq. (4.13) and Table 7) vacuum wavelength of light center vacuum wavelength wavelength in a dispersive media with refractive index n, i.e. λn = λ/n effective wavelength (eq. (3.4)) frequency pump photon frequency radian frequency center radian frequency modulation frequency in radians/second half-width-half-maximum (HWHM) gain bandwidth of laser in radians/seconds, i.e. Ωg = πνg (eq. (6.3)) divergence angle of a pump source (i.e. the beam radius increases approximately linearly with propagation distance, defining a cone with half-angle θ ) divergence angle of a Gaussian beam, i.e. θG = πWλ0,G (eq. (3.3))
ψ γA λ λ0 λn λeff ν νpump ω ω0 ωm Ωg θ
θG
πW 2
6
σA σL σLabs τA τAu τc τL τp τp,min
Ultrafast solid-state lasers
absorber cross section gain cross section absorption cross section recovery time of saturable absorber FWHM of intensity autocorrelation pulse photon cavity lifetime upper state lifetime of laser gain material FWHM intensity pulse duration minimal τp
[1
§ 1. Introduction Since 1990 we have observed tremendous progress in ultrashort pulse generation using solid-state lasers (fig. 1). Until the end of the 1980s, ultrashort pulse generation was dominated by dye lasers which produced pulses as short as 27 fs with a typical average output power of about 20 mW (Valdmanis, Fork and Gordon [1985]). Shorter pulse durations, down to 6 fs, were only achieved through additional amplification and fiber-grating pulse compression at much lower repetition rates (Fork, Cruz, Becker and Shank [1987]). The tremendous success of ultrashort dye lasers in the 1970s and 1980s diverted research interest away from solid-state lasers. In 1974 the first sub-picosecond passively mode-locked dye lasers (Shank and Ippen [1974], Ruddock and Bradley [1976], Diels, Stryland and Benedict [1978]) and in 1981 the first sub-100-fs colliding pulse mode-locked (CPM) dye lasers (Fork, Greene and Shank [1981]) have been demonstrated. The CPM dye laser was the “work horse” all through the 1980s for ultrafast laser spectroscopy in physics and chemistry.
Fig. 1. Development of shortest reported pulse duration over the last three decades. Circles, dye-laser technology; triangles, Ti:sapphire laser systems; solid symbols, results directly obtained from an oscillator; open symbols, results achieved with additional external pulse compression. 7
8
Ultrafast solid-state lasers
[1, § 1
Fig. 2. Different modes of operation of a laser, cw: continuous-wave. Single and multi mode refers to longitudinal modes.
The development of higher average-power diode lasers in the 1980s stimulated a strong interest in solid-state lasers again. Diode laser pumping provides dramatic improvements in efficiency, lifetime, size and other important laser characteristics. For example, actively mode-locked diode-pumped Nd:YAG and Nd:YLF lasers generated 7–12 ps pulse durations for the first time (Maker and Ferguson [1989a], Maker and Ferguson [1989b], Keller, Li, Khuri-Yakub, Bloom, Weingarten and Gerstenberger [1990], Weingarten, Shannon, Wallace and Keller [1990], Juhasz, Lai and Pessot [1990]). In comparison, flashlamp-pumped Nd:YAG and Nd:YLF lasers typically produced pulse durations of ≈ 100 ps and ≈ 30 ps, respectively. Before 1990, all attempts to passively mode-lock solidstate lasers with long upper state lifetimes (i.e. > 100 µs) resulted however in Q-switching instabilities which at best produced stable mode-locked pulses within longer Q-switched macropulses (i.e. Q-switched mode locking) (fig. 2). In Q-switched mode locking, the mode-locked pico- or femtosecond pulses are inside much longer Q-switched pulse envelopes (typically in the µs-regime) at much lower repetition rates (typically in the kHz-regime). The strong interest in an all-solid-state ultrafast laser technology was the driving force and formed the basis for many new inventions and discoveries. Today, reliable self-starting passive mode locking for all types of solid-state lasers is obtained with semiconductor saturable absorbers, first demonstrated in 1992 (Keller, Miller, Boyd, Chiu, Ferguson and Asom [1992]). Since then, more than a decade later, the performance of compact ultrafast solid-state lasers has been improved by several orders of magnitude in pulse duration, average power, pulse energy and pulse repetition rate, based on semiconductor saturable ab-
1, § 1]
Introduction
9
sorbers that are integrated into a cavity laser mirror (i.e. generally referred to as semiconductor saturable absorber mirrors – SESAMs) (Keller, Weingarten, Kärtner, Kopf, Braun, Jung, Fluck, Hönninger, Matuschek and Aus der Au [1996], Keller [1999]). Diode-pumped SESAM-mode-locked solid-state lasers offer very reliable and compact ultrafast lasers with unsurpassed performance (Paschotta and Keller [2003], Keller [2003]). Currently, pulse duration can range from picoseconds to a few femtoseconds depending on the different laser materials and saturable absorber parameters. The average power has been increased to 60 W directly from a mode-locked diode-pumped laser with pulse energies larger than 1 µJ (Innerhofer, Südmeyer, Brunner, Häring, Aschwanden, Paschotta, Hönninger, Kumkar and Keller [2003]). The pulse repetition rate has been increased to more than 100 GHz (Krainer, Paschotta, Lecomte, Moser, Weingarten and Keller [2002]). The breakthrough of femtosecond solid-state lasers came about with the discovery of the Ti:sapphire laser (Moulton [1986]), the first laser able to support sub-10-femtosecond pulses. The existing passive mode-locking techniques, primarily developed for dye lasers, were inadequate due to the much longer upper state lifetime (i.e. in the µs regime) and smaller gain cross-section (i.e. in the 10−19 cm2 regime) of Ti:sapphire compared to dyes (i.e. in the ns and 10−16 cm2 regime). Therefore, passive pulse generation techniques had to be re-evaluated with new laser material properties in mind. Kerr lens mode locking (KLM) (Spence, Kean and Sibbett [1991]) of Ti:sapphire lasers was discovered in 1991 and is still the only successful technique to push the frontier in ultrashort pulse duration into the few femtosecond regime. Today, the shortest laser pulses with less than 6 fs duration (Sutter, Steinmeyer, Gallmann, Matuschek, Morier-Genoud, Keller, Scheuer, Angelow and Tschudi [1999], Morgner, Kärtner, Cho, Chen, Haus, Fujimoto, Ippen, Scheuer, Angelow and Tschudi [1999]) are generated directly from the laser cavity without any additional external cavity pulse compression. Slightly shorter pulses have been demonstrated with external pulse compression (Baltuska, Wei, Pshenichnikov, Wiersma and Szipöcs [1997], Nisoli, Stagira, Silvestri, Svelto, Sartania, Cheng, Lenzner, Spielmann and Krausz [1997]) and continuum generation together with parametric optical amplification (Shirakawa, Sakane, Takasaka and Kobayashi [1999]). Adaptive pulse compression finally resulted in a new world record of only 3.8 fs pulse duration (Schenkel, Biegert, Keller, Vozzi, Nisoli, Sansone, Stagira, De Silvestri and Svelto [2003]). Although very different in technical detail, all of these sub-6-fs pulse generation techniques rely on the same three main components: nonlinear Kerr effect, higher-order dispersion control and ultrabroadband amplification (Steinmeyer, Sutter, Gallmann, Matuschek and Keller [1999]).
10
Ultrafast solid-state lasers
[1, § 2
KLM however has serious limitations because the mode-locking process is generally not self-starting and critical cavity alignment is required to obtain stable pulse generation. Thus, the laser cavity has to be optimized for best KLM and not necessarily for best efficiency and output power – this sets serious constraints on the cavity design, which becomes even more severe at higher average output powers and more compact monolithic cavities. Thus, passively mode-locked solid-state lasers using intracavity SESAMs have become a very attractive alternative to KLM and are more widely used today. The purpose of this chapter is to give an updated review of the progress in ultrafast solid-state lasers over the past ten years. The goal is also to give the non-expert an efficient starting position to enter into this field without providing all the detailed derivations. Relevant and useful references for further information are provided and a brief historic perspective is given throughout this paper. A basic knowledge in lasers is required. We put the emphasis on solid-state lasers because they will dominate the field in the future. More extended reviews and books summarize the dye laser era (Shank [1988], Diels [1990]). Fiber and semiconductor lasers will not be treated, but some useful references to recent review articles and book chapters will be provided.
§ 2. Definition of Q-switching and mode locking 2.1. Q-switching The history of Q-switching goes back to 1961, when Hellwarth [1961] predicted that a laser could emit short pulses if the loss of an optical resonator was rapidly switched from a high to a low value. Experimental proof was produced a year later (McClung and Hellwarth [1962], Collins and Kisliuk [1962]). The technique of Q-switching allows the generation of laser pulses of short duration (from the nanosecond to the picosecond range) and high peak power. The principle of the technique is as follows. Suppose a shutter is introduced into the laser cavity. If the shutter is closed, laser action cannot occur and the population inversion can reach a value far in excess of the threshold population that would have occurred if the shutter were not present. If the shutter is now opened suddenly, the laser will have a gain that greatly exceeds the losses, and the stored energy will be released in the form of a short and intense light pulse. Since this technique involves switching the cavity Q-factor from a low to a high value, it is known as Q-switching. Ideally, Q-switched lasers operate with only one axial mode because strong intensity noise is observed in a multi-mode Q-switched laser.
1, § 2]
Definition of Q-switching and mode locking
11
In passive Q-switching the shutter is replaced by an intracavity saturable absorber. The saturable absorber starts to bleach as the intensity inside the laser continues to grow from noise due to spontaneous emission. Thus, the laser intensity continues to increase which in turn results in stronger bleaching of the absorber, and so on. If the saturation intensity of the absorber is comparatively small compaired to the gain, the inversion still left in the laser medium after the absorber is bleached is essentially the same as the initial inversion. Therefore, after bleaching of the saturable absorber the laser will have a gain well in excess of the losses and if the gain cannot saturate fast enough, the intensity will continue to grow and stable Q-switching can occur. A large modulation depth of the saturable absorber then results in a high Q-switched pulse energy. Typically, the pulse repetition rate in Q-switched solid-state lasers is in the hertz to few megahertz regime, always much lower than the cavity round-trip frequency. Today, picosecond pulse durations can be obtained with Q-switched diodepumped microchip lasers (Zayhowski and Mooradian [1989], Zayhowski and Harrison [1997]) with pulses as short as 115 ps for active Q-switching using electrooptic light modulators (Zayhowski and Dill [1995]) and 37 ps for passive Q-switching using SESAMs (Spühler, Paschotta, Fluck, Braun, Moser, Zhang, Gini and Keller [1999]). For LIDAR applications passively Q-switched Er:Yb:glass microchip lasers around 1.5 µm are particularly interesting (Fluck, Häring, Paschotta, Gini, Melchior and Keller [1998], Häring, Paschotta, Fluck, Gini, Melchior and Keller [2001]). The performance of Q-switched microchip lasers bridges the gap between Q-switching and mode locking both in terms of pulse duration (nanoseconds to a few tens of picoseconds) and pulse repetition rates (kilohertz to a few tens of megahertz).
2.2. Mode locking Mode locking is a technique to generate ultrashort pulses from lasers. In cw mode locking the pulses are typically much shorter than the cavity round trip and the pulse repetition rate (from few tens of megahertz to a few hundreds of gigahertz) is determined by the cavity round trip time. Typically, an intracavity loss modulator (i.e. a loss modulator inside a laser cavity) is used to collect the laser light in short pulses around the minimum of the loss modulation with a period given by the cavity round trip time TR = 2L/νg , where L is the laser cavity length and νg the group velocity (i.e. the propagation velocity of the peak of the pulse intensity). Under certain conditions, the pulse repetition rate can be some integer multiple of the fundamental repetition rate (i.e. harmonic mode locking)
12
Ultrafast solid-state lasers
[1, § 2
(Becker, Kuizenga and Siegman [1972]). We distinguish between active and passive mode locking. For active mode locking, an external signal is applied to an optical loss modulator typically using the acousto-optic or electro-optic effect. Such an electronically driven loss modulation produces a sinusoidal loss modulation with a period given by the cavity round trip time TR . For passive mode locking, a saturable absorber is used to obtain a selfamplitude modulation (SAM) of the light inside the laser cavity. Such an absorber introduces some loss to the intracavity laser radiation, which is relatively large for low intensities but significantly smaller for a short pulse with high intensity. Thus, a short pulse then produces a loss modulation because the high intensity at the peak of the pulse saturates the absorber more strongly than its low intensity wings. This results in a loss modulation with a fast initial loss saturation (i.e. reduction of the loss) determined by the pulse duration and typically a somewhat slower recovery which depends on the detailed mechanism of the absorption process in the saturable absorber. A homogeneously broadened laser normally lases at one axial mode at the peak of the gain. However, the periodic loss modulation transfers additional energy phase-locked to adjacent modes separated by the modulation frequency. This modulation frequency is normally adapted to the cavity round trip frequency. The resulting frequency comb with equidistant axial modes locked together in phase forms a short pulse in the time domain. Mode locking was first demonstrated in the mid-1960s using a HeNe-laser (Hargrove, Fork and Pollack [1964]), ruby laser (Mocker and Collins [1965]) and Nd:glass laser (De Maria, Stetser and Heynau [1966]). The passively mode-locked lasers were also Q-switched, which means that the mode-locked pulse train was strongly modulated (fig. 2). This continued to be a problem for passively modelocked solid-state lasers until the first intracavity saturable absorber was designed correctly to prevent self-Q-switching instabilities in solid-state lasers with microsecond or even millisecond upper state lifetimes (Keller, Miller, Boyd, Chiu, Ferguson and Asom [1992]). Passive mode locking starts from normal noise fluctuations in a laser. One noise spike is strong enough to sufficiently start to saturate the absorber, which results in lower loss and therefore more gain. Thus, this noise spike begins to grow and becomes shorter until a stable pulse duration is obtained. For reliable passive cw mode locking, the pulse generation starts from normal laser noise within less than about 1 ms. Q-switching instabilities are a serious issue with passively mode-locked solidstate lasers. The parameters of the saturable absorber have to be chosen such that
1, § 2]
Definition of Q-switching and mode locking
13
the mode locking is self-starting (i.e. starting from the normal intensity noise of the laser) and stable, i.e. without any Q-switching instabilities. For example, if the loss modulation becomes too large it can drive the laser unstable: The loss saturation increases the intensity inside the laser cavity. The gain then needs to saturate more strongly to compensate for the reduced loss and to keep the intensity inside the laser cavity constant. If the gain cannot respond fast enough, the intensity continues to increase as the absorber is bleached which leads to selfQ-switching instabilities or in the best case to stable Q-switched mode locking. In the latter case, the mode-locked pulse train is strongly modulated at close to the relaxation oscillation frequency of the laser (typically in the kHz rate) (fig. 2). A large modulation depth of the saturable absorber results in shorter pulses but an upper limit is set by the onset of Q-switching instabilities which is a serious challenge in solid-state lasers (Hönninger, Paschotta, Morier-Genoud, Moser and Keller [1999]). Passive mode-locking mechanisms are well-explained by three fundamental models: slow saturable absorber mode locking with dynamic gain saturation (New [1972, 1974]) (fig. 3a), fast saturable absorber mode locking (Haus [1975a], Haus, Fujimoto and Ippen [1992]) (fig. 3b), and slow saturable absorber mode locking without dynamic gain saturation stable both in the picosecond domain (Paschotta and Keller [2001]) and in the femtosecond domain described by soliton mode locking (Kärtner and Keller [1995], Kärtner, Jung and Keller [1996])
Fig. 3. Passive modelocking mechanisms explained by three fundamental models: (a) slow saturable absorber modelocking with dynamic gain saturation; (b) fast saturable absorber modelocking; (c) slow saturable absorber modelocking without dynamic gain saturation, which in the femtosecond regime is described by soliton modelocking.
14
Ultrafast solid-state lasers
[1, § 2
(fig. 3c). In the first two cases, a short net-gain window forms and stabilizes an ultrashort pulse. In fig. 3a, an ultrashort net-gain window can be formed by the combined saturation of absorber and gain for which the absorber has to saturate and recover faster than the gain, while the recovery time of the saturable absorber can be much longer than the pulse duration. Dynamic gain saturation means that the gain experiences a fast, pulse-induced saturation that then recovers again between consecutive pulses (fig. 3a). For solid-state lasers we cannot apply slow saturable absorber mode locking as shown in fig. 3a, because no significant dynamic gain saturation is taking place due to the small gain cross-section and the long upper state lifetime of the laser. The upper state lifetime of solid-state lasers is typically in the µs to ms regime, much longer than the pulse repetition period, which is typically in the nanosecond regime. In addition, the gain cross-section is 1000 or even more times smaller than for dye lasers. We therefore do not observe any significant dynamic gain saturation and the gain is only saturated to a constant level by the average intracavity intensity. This is not the case for dye, semiconductor and color center lasers for which fig. 3a describes most mode-locking processes. Therefore it was assumed that without other pulse forming mechanisms (such as soliton pulse shaping) a fast saturable absorber is required for solid-state lasers. Kerr lens mode locking is an almost ideal example of fast saturable absorber mode locking. However, during the last 10 years we have seen that a slow saturable absorber can support significantly shorter pulses even though a net gain window remains open after the short pulse (fig. 3c). At first this seems surprising, because on the trailing edge of the pulse there is no shaping action of the absorber and, even worse, one would expect that the net gain after the pulse would destabilize the pulse. However, we have shown that in the femtosecond regime soliton formation is actually the dominant pulse forming mechanism and the slow saturable absorber needs only to be fast enough to stabilize this soliton – which is referred to as soliton mode locking (Kärtner and Keller [1995], Jung, Kärtner, Brovelli, Kamp and Keller [1995], Kärtner, Jung and Keller [1996]). In the picosecond regime without soliton formation, a more strongly saturated slow saturable absorber can stabilize a shorter pulse because the pulse is constantly delayed by the absorber and therefore swallows any noise growing behind it (Paschotta and Keller [2001]). This means that even with solid-state lasers we can work with relatively slow saturable absorbers that have approximately a recovery time in the range of 10 to 30 times the absorber recovery time.
1, § 3]
Overview of ultrafast solid-state lasers
15
§ 3. Overview of ultrafast solid-state lasers 3.1. Overview of solid-state laser materials Solid-state lasers can be grouped in two types: transition-metal doped Cr3+ , Cr4+ , Ti3+ , Ni2+ , Co2+ and rare-earth doped (Nd3+ , Tm3+ , Ho3+ , Er3+ , Yb3+ ) solidstate lasers (Table 1). Color center lasers have also supported ultrashort pulse durations, but they require cryogenic cooling (Mitschke and Mollenauer [1987], Islam, Sunderman, Soccolich, Bar-Joseph, Sauer, Chang and Miller [1989]). A similar wavelength range can be covered with Cr:YAG lasers, for example. Tables 1 and 2 summarize the laser parameters of these solid-state lasers and the performances that have been demonstrated with these lasers up to date. Because of their complexity we do not include results that are based on coupled-cavity modelocking techniques. For further information see the detailed summary in Table 2. The rare-earth-doped solid-state lasers have 4f-electrons responsible for the laser transition, which are shielded from the crystal host. Thus, the gain bandwidth is normally not very large and pulse durations are limited to a few 100 fs (Tables 1 and 2). However, inhomogeneously broadened rare-earth-doped lasers in glass hosts, for example, can produce significantly shorter pulses in the sub100-fs regime (Tables 1 and 2). Transition-metal-doped solid-state lasers have 3d-electrons responsible for the laser transition, which are not very well shielded from the host material. Thus, these lasers are strongly phonon broadened and can support much shorter pulses than the rare-earth-doped crystal. Presently, the shortest pulses generated from a laser are based on Ti:sapphire, a transition-metal-doped solid-state laser (Table 1). Mode-locked and Q-switched ion-doped fiber lasers also showed a lot of progress during the last ten years. More recent reviews on mode-locked fiber lasers are given in book chapters and review articles by Duling et al. (Duling [1995]), by Fermann (Fermann [1994], Fermann [1995], Fermann, Galvanauskas, Sucha and Harter [1997], Fermann [2003]), and by Haus (Haus, Tamura, Nelson and Ippen [1995], Haus [1995]). Generally, mode-locked fiber lasers generate significantly lower pulse energies and longer pulse durations than bulk lasers. However, recent progress in mode-locked fiber lasers resulted in Er/Yb-doped fiber lasers that generate 2.7 nJ pulses at 32 MHz with 100-fs pulse duration (Nelson, Fleischer, Lenz and Ippen [1996]). Much shorter pulses but also at much lower pulse energies have been obtained in Nd-doped fiber lasers with pulse durations as short as 38 fs (Hofer, Ober, Haberl and Fermann [1992]) and in erbium fiber lasers with pulses as short as 84 fs (Fermann, Andrejco, Stock, Silberberg and Weiner [1993]) and 63 fs (Tamura, Ippen and Haus [1995]). Fiber lasers require
Laser material
Ref.
λ0 [nm]
σL [10−20 cm2 ]
τL [µs]
λg [nm]
τp,min
16
Table 1 Relevant laser materials for short and ultrashort pulse generation. λ0 : center lasing wavelength. σL : gain cross-section (σem –σabs for 3 level lasers). τL : upper state lifetime. λg : FWHM gain bandwidth. τp,min : minimal pulse duration that has been demonstrated so far (coupled cavity mode-locking schemes are not considered because of their increased complexity, for further information with regards to coupled cavity mode-locking results see Table 2) Ref.
Transition metal doped solid-state lasers (Cr3+ , Cr4+ , Ti3+ , Ni2+ , Co2+ ) [86Mou], [96Koe] [89Pay], [91Sch], [96Koe] [88Pay] [92Smi] [92Cha] [79Wal], [89Sam], [96Koe] [88Pet], [93Car], [96Koe] [88Ang], [95Küc]
790 850 758 835 ≈ 800 750 1240 1378
41 5 1.23 3.3 0.76 14.4 33
3.2 67 175 88 80 260 2.7 4.1
≈ 230 180 115 190 ≈ 155 ≈ 100 170 ≈ 250
5–6 fs 9.9 fs 9 fs 14 fs 90 fs 3 ps 14 fs 20 fs
[99Sut], [01Ell] [03Uem] [02Wag] [97Sor] [94Wan] [84Pes] [01Chu] [02Rip]
6.7 ps/6.8 ps 1.5 ps 5.7 ps 2.7 ps/2.8 ps 4.6 ps 1.9 ps 1.6 ps 150 fs
[97Hen]/[94Kel] [90Mal] [96Flu2] [00Kra]/[99Cou] [96Flu] [98Kel] [96Bra] [95Kop]
Rare earth doped solid-state lasers (Nd3+ , Tm3+ , Ho3+ , Er3+ ) Nd:YAG Nd:YLF
[64Geu], [97Zay] [69Har], [97Zay], [96Koe]
Nd:YVO4
[97Zay]
Nd:YAlO3 Nd:LSB (10 at.%) Nd:phosphate glass (LG-760) Nd:silicate glass (LG-680)
[91Kut], [94Mey] [99Sch]
1064 1047 1314 1064 1342 930 1064 1054
[99Sch]
1059.7
33 18 300 60 4.1 13 4.5 2.54
230 480 480 100 100 160 118 323
0.6 1.2
2.5 4 24.3
361
35.9
0.8
68 fs
Ultrafast solid-state lasers
Ti3+ :sapphire Cr3+ :LiSAF Cr3+ :LiCAF Cr3+ :LiSGAF Cr3+ :LiSCAF Cr3+ :Alexandrite Cr4+ :Forsterite Cr4+ :YAG
[97Aus] [1, § 3
(continued on next page)
λ0 [nm]
σL [10−20 cm2 ]
τL [µs]
[78Sch]
1054
2.6
495
26.1
60 fs
[93Fan] [97Kul1], [97Kul2], [03Kul] [97Kul1], [97Kul2], [02Puj], [00Dem] [99Mou] [02Dru]
1030 1026 1025
2.0 2.8 3.0
950 ≈ 250 ≈ 250
6.3 ≈ 25 ≈ 25
340 fs 112 fs 71 fs
[96Kel], [99Hön] [00Bru] [01Liu]
1045 1062 1025−1060 1025−1060 1025−1060 1535
0.36 0.2 0.049 0.093 0.158 0.8
2600 1100 1300 1100 1300 7900
44 60 62 77 81 55
90 fs 69 fs 58 fs 61 fs 60 fs 380 fs
[00Dru] [02Dru] [98Hön], [99Hön] [98Hön], [99Hön] [98Hön], [99Hön] [01Was]
Laser material
Ref.
Nd:fluorophosphate glass (LG-812) Yb:YAG Yb:KGW Yb:KYW
[96Kig]
λg [nm]
τp,min
Ref. [97Aus]
[64Geu] Geusic, Marcos and Uitert [1964], [69Har] Harmer, Linz and Gabbe [1969], [78Sch] Schott Glass Technologies [1978], [79Wal] Walling, Jenssen, Morris, O’Dell and Peterson [1979], [84Pes] Pestryakov, Trunov, Matrosov and Razvalyaev [1984], [86Mou] Moulton [1986], [88Ang] Angert, Borodin, Garmash, Zhitnyuk, Okhrimchuk, Siyuchenko and Shestakov [1988], [88Pay] Payne, Chase, Newkirk, Smith and Krupke [1988], [88Pet] Petricevic, Gayen and Alfano [1988], [89Pay] Payne, Chase, Smith, Kway and Newkirk [1989], [89Sam] Sam [1989], [90Mal] Malcolm, Curley and Ferguson [1990], [91Kut] Kutovoi, Laptev and Matsnev [1991], [91Sch] Scheps, Myers, Serreze, Rosenberg, Morris and Long [1991],
17
(continued on next page)
Overview of ultrafast solid-state lasers
Yb:GdCOB Yb:BOYS Yb:phosphate glass Yb:silicate glass Yb:fluoride phosphate Er,Yb:glass
1, § 3]
Table 1 (Continued)
18
Table 1 (Continued)
[1, § 3
(continued on next page)
Ultrafast solid-state lasers
[92Cha] Chai, Lefaucheur, Stalder and Bass [1992], [92Smi] Smith, Payne, Kway, Chase and Chai [1992], [93Car] Carrig and Pollock [1993], [93Fan] Fan [1993], [94Kel] Keller [1994], [94Mey] Meyn, Jensen and Huber [1994], [94Wan] Wang, Kam Wa, Lefaucheur, Chai and Miller [1994], [95Kop] Kopf, Kärtner, Keller and Weingarten [1995], [95Küc] Kück, Petermann, Pohlmann and Huber [1995], [96Bra] Braun, Hönninger, Zhang, Keller, Heine, Kellner and Huber [1996], [96Flu] Fluck, Zhang, Keller, Weingarten and Moser [1996], [96Kel] Keller, Weingarten, Kärtner, Kopf, Braun, Jung, Fluck, Hönninger, Matuschek and Aus der Au [1996], [96Kig] Kigre Inc. [1996], [96Koe] Koechner [1996], [97Aus] Aus der Au, Kopf, Morier-Genoud, Moser and Keller [1997], [97Hen] Henrich and Beigang [1997], [97Kul1] Kuleshov, Lagatsky, Podlipensky, Mikhailov and Huber [1997], [97Kul2] Kuleshov, Lagatsky, Shcherbitsky, Mikhailov, Heumann, Jensen, Diening and Huber [1997], [97Sor] Sorokina, Sorokin, Wintner, Cassanho, Jenssen and Szipöcs [1997], [97Zay] Zayhowski and Harrison [1997], [98Hön] Hönninger, Morier-Genoud, Moser, Keller, Brovelli and Harder [1998], [98Kel] Kellner, Heine, Huber, Hönninger, Braun, Morier-Genoud and Keller [1998], [99Cou] Couderc, Louradour and Barthelemy [1999], [99Hön] Hönninger, Paschotta, Graf, Morier-Genoud, Zhang, Moser, Biswal, Nees, Braun, Mourou, Johannsen, Giesen, Seeber and Keller [1999], [99Mou] Mougel, Dardenne, Aka, Kahn-Harari and Vivien [1999], [99Sch] Schott Glass Technologies [1999], [99Sut] Sutter, Steinmeyer, Gallmann, Matuschek, Morier-Genoud, Keller, Scheuer, Angelow and Tschudi [1999],
1, § 3]
Table 1 (Continued)
Overview of ultrafast solid-state lasers
[00Bru] Brunner, Spühler, Aus der Au, Krainer, Morier-Genoud, Paschotta, Lichtenstein, Weiss, Harder, Lagatsky, Abdolvand, Kuleshov and Keller [2000], [00Dem] Demidovich, Kuzmin, Ryabtsev, Danailov, Strek and Titov [2000], [00Dru] Druon, Balembois, Georges, Brun, Courjaud, Hönninger, Salin, Aron, Mougel, Aka and Vivien [2000], [00Kra] Krainer, Paschotta, Moser and Keller [2000b], [01Chu] Chudoba, Fujimoto, Ippen, Haus, Morgner, Kärtner, Scheuer, Angelow and Tschudi [2001], [01Ell] Ell, Morgner, Kärtner, Fujimoto, Ippen, Scheuer, Angelow, Tschudi, Lederer, Boiko and Luther-Davis [2001], [01Liu] Liu, Nees and Mourou [2001], [01Was] Wasik, Helbing, König, Sizmann and Leuchs [2001], [02Dru] Druon, Chenais, Raybaut, Balembois, Georges, Gaume, Aka, Viana, Mohr and Kopf [2002], [02Puj] Pujol, Bursukova, Güell, Mateos, Sole, Gavalda, Aguilo, Massons, Diaz, Klopp, Griebner and Petrov [2002], [02Rip] Ripin, Chudoba, Gopinath, Fujimoto, Ippen, Morgner, Kärtner, Scheuer, Angelow and Tschudi [2002], [02Wag] Wagenblast, Morgner, Grawert, Schibli, Kärtner, Scheuer, Angelow and Lederer [2002], [03Kul] Kuleshov, Private communication [2003], [03Uem] Uemura and Torizuka [2003]
19
Laser material
ML technique
λ0
τp
Pav,out
frep
Active-AOM CCM-APM
814 nm
150 fs 1.4 ps
600 mW 300 mW
80.5 MHz
CCM-RPM Dye sat. absorber
860 nm 750 nm
2 ps 140 fs
90 mW
250 MHz
KLM
880 nm
60 fs
300 mW
852 nm
73 fs
240 mW
Remarks
20
Table 2 CW mode-locked solid-state lasers using different mode-locking techniques. “Best” means in terms of pulse duration, highest average output power, highest pulse repetition rate, etc. – the result for which “best” applies is in bold letters. The lasers are assumed to be diode-pumped, if not stated otherwise (except Ti:sapphire laser). ML: mode locking. CCM: coupled cavity mode locking. APM: additive pulse mode locking. RPM: resonant passive mode locking. KLM: Kerr lens mode locking. SESAM: passive mode locking using semiconductor saturable absorber mirrors (SESAMs). Soliton-SESAM: Soliton mode locking with a SESAM. AOM: acousto-optic modulator. EOM: electro-optic phase modulator. λ0 : center lasing wavelength. τp : measured pulse duration. Pav,out : average output power. frep : pulse repetition rate Ref.
Ti:sapphire
47 fs 32 fs
110 mW 320 mW
134 MHz
≈ 100 MHz
KLM started with dye saturable absorber (not understood – assumed to have a CPM Ti:sapphire laser) First demonstration of KLM (but KLM not understood) First experimental evidence for KLM, self-starting due to RPM KLM, a Gaussian approximation LaFN28-glass prisms, 2 cm Ti:sapphire crystal thickness
[91Cur] [89Goo]
[90Kel2] [91Sar]
Ultrafast solid-state lasers
813 nm
Highly chirped 1.4 ps output pulses externally compressed down to 200 fs
[91Spe] [91Kel]
[91Sal] [92Riz1] [92Hua1] [1, § 3
(continued on next page)
Laser material
ML technique
λ0
τp
Pav,out
1, § 3]
Table 2 (Continued) frep
33 fs 22 fs
950 mW
100 MHz
817 nm
17 fs
500 mW
80 MHz
775 nm
12.3 fs
780 nm
11 fs
500 mW
11 fs 8.5 fs
300 mW ≈ 1 mW
100 MHz
850 nm 800 nm 800 nm 800 nm
8.2 fs 7.5 fs 6.5 fs
100 mW 150 mW 200 mW
80 MHz 86 MHz
≈ 800 nm
5.8 fs
300 mW
85 MHz
≈ 800 nm
< 6 fs
200 mW
90 MHz
Ref.
Schott F2 prisms, 8 mm Ti:Sapphire crystal thickness Schott LaK31 prisms, 2 cm Ti:Sapphire crystal thickness Schott LaKL21 prisms, 9 mm Ti:Sapphire crystal thickness Fused silica prisms, 4 mm Ti:Sapphire crystal thickness Fused silica prisms, 4.5 mm Ti:Sapphire crystal thickness Chirped mirrors, no prisms Metal mirrors and fused silica prisms Chirped mirrors only Chirped mirrors, ring cavity Fused silica prisms and double-chirped mirrors, KLM is self-starting with SESAM Fused silica prisms and double-chirped mirrors, KLM is self-starting with SESAM, pulse duration measured with SPIDER CaF2 prisms, double chirped mirrors, pulse duration estimated with fit to IAC (not very accurate)
[92Kra] [92Lem] [92Hua2] [93Cur] [93Asa] [94Sti] [94Zho] [95Sti] [96Xu] [97Jun]
[99Sut], [00Mat]
Overview of ultrafast solid-state lasers
804 nm
Remarks
[99Mor]
21
(continued on next page)
Laser material
λ0
ML technique
τp
Pav,out
frep
≈ 800 nm
≈ 5 fs
120 mW
65 MHz
780 nm
8.5 fs
1W
75 MHz
850 nm 800 nm ≈ 780 nm 782 nm
13 fs 16.5 fs 80 fs 23 fs
1.5 W 170 mW 360 mW 300 mW
110 MHz 15 MHz 4 MHz 2 GHz
≈ 800 nm
9.5 fs
180 mW
85 MHz
840 nm
34 fs
140 mW
98.9 MHz
90 fs 810 nm
13 fs
80 mW
85 MHz
800–880 nm
150 fs
50 mW
82 MHz
≈ 840 nm
50 fs
135 mW
Remarks
Ref.
CaF2 prisms, double chirped mirrors, pulse duration estimated with fit to IAC 1.5 MW peak, focused intensity 5 × 1013 W/cm2 1 MW peak, 13 nJ out 0.7 MW peak 90 nJ pulse energy Ring laser, ML is self-starting due to feedback from external mirror KLM self-starting with novel broadband fluoride SESAM First SESAM design with a single quantum well absorber in a Bragg reflector New acronym for special SESAM design: SBR Shortest pulse with soliton mode locking and no KLM
[01Ell]
[98Xu] [99Bed] [99Cho] [01Cho] [99Bar]
[02Sch] [95Bro]
Ultrafast solid-state lasers
Soliton-SESAM
22
Table 2 (Continued)
[95Tsu] [96Kär]
Cr:LiSAF KLM
[92Mil]
[92Riz2]
(continued on next page)
[1, § 3
First mode-locked femtosecond Cr:LiSAF laser, Kr-pumped Ar-ion pumped, intracavity dye absorber for starting KLM
Laser material
ML technique
λ0 880–920 nm
860 nm 880 nm 850 nm 875 nm 880 nm 850 nm ≈ 850 nm Soliton-SESAM 840 nm 850 nm 865 nm
Pav,out
50 fs
150 mW
33 fs
25 mW
93 fs
5 mW
300 ps
≈ 1 mW
220 fs 90 fs
≈ 10 mW < 20 mW
frep 85 MHz
90 MHz
97 fs 34 fs 55 fs 40 fs (24 fs) 18 fs 14.8 fs 12 fs 9.9 fs 98 fs
2.7 mW 80 MHz 42 mW 80 MHz 10 mW 70 mW (≈ 1 mW) ≈ 1 mW 70 mW 70 MHz 23 mW 200 MHz 50 mW
120 MHz
45 fs 100 fs
105 mW 11 mW
176 MHz 178 MHz
Remarks
Ref.
Ar-ion pumped, regenerative AOM for starting KLM Ar-ion pumped, intracavity dye absorber for starting KLM Ar-ion pumped, KLM started with intracavity SESAM First diode-pumped mode-locked Cr:LiSAF laser, AOM or RPM for starting KLM KLM started by SESAM Pumped by a SHG diode-pumped actively mode-locked Nd:YLF laser, prism mode locker to start KLM KLM started by regen. AOM KLM started by regen. AOM KLM self-starting
[92Eva]
Kr-ion-laser pumped ML not self-starting Strong wings in IAC First soliton mode locking, no KLM required
[92Riz3] [93Riz] [93Fre]
[94Mel] [94Lin1]
[94Dym] [95Dym] [95Fal] [95Mel] [97Dym] [97Sor] [99Uem] [03Uem] [94Kop1]
Overview of ultrafast solid-state lasers
870 nm
τp
1, § 3]
Table 2 (Continued)
[95Kop2], [97Kop1] [95Tsu] 23
(continued on next page)
Laser material
ML technique
Pav,out
24
Table 2 (Continued) τp
frep
842 nm
160 fs
25 mW
868 mW
70 fs
100 mW
860 nm 875 nm
60 fs 50 fs
125 mW 340 mW
176 MHz 150 MHz
875 nm
110 fs
500 mW
150 MHz
857 nm
146 fs
3 mW
1 GHz
800 nm 793 nm 820 nm ≈ 850 nm
170 fs 52 fs 20 fs 9 fs
100 mW 75 mW 13 mW 220 mW
90 MHz 95 MHz 95 MHz 97 MHz
35 mW
71 MHz
Remarks
Ref.
Compact cavity design with no prisms for dispersion compensation (dispersive SESAM) 0.5 W diffraction limited MOPA pump
[96Kop1]
Low-brightness 0.9-cm wide, 15 W diode laser array Low-brightness 0.9-cm wide, 15 W diode laser array
[96Tsu] [97Kop1] [97Kop2] [97Kop2] [01Kem]
Cr:LiCAF KLM
Kr-pumped
Ti:sapphire laser pumped, only fit to IAC measurement
[92LiK] [98Gab] [98Gab] [02Wag]
Ultrafast solid-state lasers
λ0
Cr:LiSGaF KLM
830 nm 835 nm 842 nm 895 nm
100 fs (50 fs) 64 fs 44 fs 14 fs
200 mW 100 mW
80 MHz 70 MHz
[95Yan] [96Sor1] [96Sor2] [97Sor]
(continued on next page)
[1, § 3
Kr-ion-laser pumped Kr-ion-laser pumped, GTI Kr-ion-laser pumped, chirped mirror
Laser material
ML technique
λ0
τp
Pav,out
1, § 3]
Table 2 (Continued) frep
Soliton-SESAM
839 nm
61 fs
78 mW
119 MHz
KLM
860 nm
90 fs
100 mW
140 MHz
Remarks
Ref. [97Loe]
Cr:LiSCAF Kr-ion-laser pumped
[94Wan]
Nd:YAG laser pumped, KLM self-starting with AOM Nd:YAG laser pumped Nd:YAG laser pumped Nd:YVO4 laser pumped, not self-starting Double chirped mirrors KLM selfstarting due to Semiconductor-doped silica films Nd:YVO4 laser pumped Nd:YVO4 laser pumped Double-clad fiber pumped Nd:YAG laser pumped
[93Sen]
Cr:Forsterite
Soliton-SESAM
1.23 µm (1.21–1.27 µm) 1.24–1.27 µm 1.27 µm 1.28 µm
48 fs
380 mW
81 MHz
50 fs 25 fs 20 fs
45 mW 300 mW
82 MHz 80 MHz
1.3 µm 1.3 µm
14 fs 25 fs
80 mW
100 MHz
1.29 µm 1.29 µm 1.27 µm 1.26 mW
40 fs 36 fs 80 fs 78 fs
60 mW 60 mW 68 mW 800 mW
90 MHz 83 MHz
1.52 µm
120 fs
360 mW
81 MHz
1.51 µm 1.55 µm
70 fs 60 fs
50 mW 50 mW
235 MHz
[93Sea] [93Yan] [97Zha2] [01Chu] [02Pra]
[97Zha1] [97Zha2] [98Liu] [98Pet]
Overview of ultrafast solid-state lasers
KLM
Cr:YAG KLM
Nd:YAG laser pumped, regen. AOM for starting KLM Nd:YAG laser pumped Nd:YAG laser pumped
[94Sen] [94Con] [94Ish] 25
(continued on next page)
Laser material
ML technique
λ0
τp
Pav,out
26
Table 2 (Continued) frep
53 fs 43 fs 75 fs
250 mW 200 mW 280 mW
1.52 µm 1.54 µm 1.45 µm Soliton-SESAM 1.541 µm 1.5 µm 1.52 µm
55 fs 115 fs 20 fs 110 fs 114 fs 200 fs
150 mW 400 mW 70 mW 94 mW 82 mW
1.46 µm
400 fs
230 mW
1.52 µm
44 fs
65 mW
Active AOM
1.064 µm
25 ps
Active FM Active EOM Active AOM
1.064 µm 1.32 µm 1.32 µm
12 ps 8 ps 53 ps
65 mW 240 mW 1.5 W
350 MHz 1 GHz 200 MHz
CCM-APM
1.064 µm 1.064 µm
1.7 ps 2 ps
25 mW 110 mW
136 MHz 125 MHz
70 MHz 1 GHz 1.2 GHz 2.64 GHz 110 MHz 185 MHz 0.9, 1.8, 2.7 GHz
152 MHz
Ref.
Nd:YAG laser pumped Nd:YVO4 laser pumped Nd:YVO4 laser or Yb:fiber laser pumped Nd:YVO4 laser pumped Nd:YVO4 laser pumped Nd:YVO4 laser pumped Nd:YVO4 laser pumped Nd:YVO4 laser pumped Nd:YVO4 laser pumped, 0.9 GHz fundamental, then double and triple harmonics Nd:YAG laser pumped, single prism for dispersion comp. Nd:YVO4 laser pumped
[96Ton] [97Ton] [98Mel]
Lamp-pumped: Pulse shortening due to intracavity etalon
[86Ros]
[00Tom] [01Tom] [02Rip] [96Col] [97Spä] [97Col]
[98Cha] [99Zha]
Ultrafast solid-state lasers
1.54 µm 1.54 µm 1.52 µm
Remarks
Nd:YAG
Lamp-pumped and harmonic mode-locked
[89Mak2] [91Zho] [88Kel]
(continued on next page)
[1, § 3
[90Goo] [91McC]
Laser material
ML technique
KLM
Polarization switching in nonlinear crystal
τp
Pav,out
frep
1.064 µm 1.32 µm 1.064 µm 1.064 µm
6 ps 10 ps 8.5 ps
2.4 W 700 mW 1W
1.064 µm 1.064 µm 1.064 µm 1.064 µm 1.064 µm 1.064 µm
6.7 ps 8.7 ps 6.8 ps 16 ps 19 ps 23 ps
675 mW 100 mW 400 mW 10.7 W 27 W 4W
106 MHz 100 MHz 217 MHz 88 MHz 55 MHz 150 MHz
1.053 µm 1.053 µm 1.047 µm 1.047 µm 1.047 µm 1.053 µm 1.053 µm 1.3 µm 1.047 µm
37 ps 18 ps 9 ps 7 ps 6.2 ps 4 ps 4.5 ps 8 ps 200 ps
6.5 W 12 mW 150 mW 135 mW 20 mW 400 mW 400 mW 240 mW 27 mW
100 MHz 230 MHz 500 MHz 2 GHz 1 GHz 237.5 MHz 2.85 GHz 1 GHz 100 MHz
100 MHz 100 MHz 100 MHz
Remarks
Ref.
Lamp-pumped Lamp-pumped
[90Liu1] [90Liu1] [92Liu] [95Chu]
Lamp-pumped, severe instabilities Ti:sapphire laser pumped
Multiple laser heads Lamp-pumped, KTP crystal
[97Hen] [93Wei] [94Kel] [99Spü] [00Spü] [99Kub]
Nd:YLF Active AOM
Active EOM
Active MQW
Lamp-pumped
Ti:sapphire laser pumped
Semiconductor multiple quantum well (MQW) modulator
[87Bad] [89Mak1] [90Kel1] [90Wei] [90Wal] [92Wei] [92Wei] [91Zho] [95Bro2]
27
(continued on next page)
Overview of ultrafast solid-state lasers
SESAM
λ0
1, § 3]
Table 2 (Continued)
Laser material
ML technique KLM
SESAM
CCM-RPM
τp
Pav,out
1.047 µm 1.047 µm
3 ps 2.3 ps
250 mW 800 mW
1.047 µm 1.047 µm 1.047 µm 1.3 µm 1.053 µm 1.053 µm 1.047 µm 1.047 µm 1.047
3.3 ps 5.1 ps 2.8 ps 5.7 ps 3.7 ps 1.7 ps 1.5 ps 4 ps 3.7 ps
700 mW 225 mW 460 mW 130 mW 7W
1.064 µm 1.064 µm 1.064 µm 1.064 µm 1.064 µm 1.064 µm 1.064 µm 1.064 µm
7 ps 8.3 ps 6.8 ps 33 ps 21 ps 4.8 ps 2.7 ps 2.7 ps 13.7 ps 4.6 ps 2.7 ps 7.9 ps
4.5 W 198 mW 81 mW 4.4 W 20 W 80 mW 65 mW 45 mW 2.1 W 50 mW 1.1 W 1.35 W
20 mW 1W 550 mW
frep 82 MHz 220 MHz 100 MHz 220 MHz 98 MHz 76 MHz 103 MHz 123 MHz 120 MHz 250 MHz
Remarks
Ref.
Lamp-pumped, microdot mirror Ti:sapphire laser pumped
Lamp-pumped
Pump laser for PPLN OPO Ti:sapphire laser pumped
[94Lin2] [93Ram] [92Kel2] [93Wei] [94Kel] [96Flu] [90Liu2] [99Jeo] [90Mal] [99Lef] [92Kel1]
Nd:YVO4 SESAM
Ti:sapphire laser pumped
Ti:sapphire laser pumped Ti:sapphire laser pumped Ti:sapphire laser pumped
[97Ruf] [99Kra1] [99Kra2] [99Gra] [00Bur] [00Kra1] [00Kra2] [02Kra2] [02Kra2] [96Flu] [02McC] [97Agn]
(continued on next page)
[1, § 3
1.3 µm CCM-APM 1.064 µm Nonlinear mirror 1.064 µm ML
84 MHz 13 GHz 29 GHz 235 and 440 MHz 90 MHz 39, 49, 59 GHz 77 GHz 157 GHz 10 GHz 93 MHz 90.7 MHz 150 MHz
Ultrafast solid-state lasers
CCM-APM
λ0
28
Table 2 (Continued)
Laser material
τp
Pav,out
Intensity-dependent 1.064 µm polarization rotation
2.8 ps
670 mW
130 MHz
SESAM
1.062 µm 1.062 µm
1.6 ps 2.8 ps
210 mW 400 mW
240 MHz 177 MHz
Active-AOM Active-FM
1.070 µm 1.070 µm 1.070 µm
7.5 ps 2.9 ps 3.9 ps
230 mW 30 mW 30 mW
250 MHz 238 MHz 20 GHz
Active-FM CCM-APM SESAM
1.067 µm 1.067 µm 1.067 µm
12 ps 2.3 ps 6.3 ps
850 mW 1W
200 MHz 76.5 MHz 64.5 MHz
Active-AOM
1.054 µm 1.054 µm 1.054 µm 1.063 µm
7 ps ≈ 10 ps 9 ps 310 fs
20 mW 30 mW 30 mW 70 mW
240 MHz 240 MHz
1.054 µm
9 ps
14 mW
235 MHz
ML technique
λ0
1, § 3]
Table 2 (Continued) frep
Remarks
Ref. [99Cou]
Nd:LSB Ti:sapphire laser pumped
Nd:BEL Harmonic mode locking Harmonic mode locking
[91Li] [91God] [91God]
Nd:KGW [95Flo] [02Maj1] [02Maj2]
Nd:glass Nd:phosphate
Nd:phosphate
Active-FM
Ar-ion laser pumped
Ti:sapphire laser pumped, regeneratively actively mode-locked
Overview of ultrafast solid-state lasers
[96Bra] [96Bra]
[86Yan] [88Bas] [92Hug] [94Kop2]
[91Hug] 29
(continued on next page)
30
Table 2 (Continued) Pav,out
CCM-APM 1.054 µm Soliton-SESAM 1.054 µm 1.054 µm 1.054 µm
122 fs 130 fs 150 fs 120 fs
200 mW 160 mW 110 mW 30 mW
180 MHz 150 MHz
1.057 µm 1.054 µm 1.065 µm 1.064 µm
175 fs 275 fs 60 fs 130 fs
1W 1.4 W 84 mW 80 mW
117 MHz 74 MHz 114 MHz 180 MHz
1.03 µm
540 fs
100 mW
81 MHz
1.03 µm 1.03 µm 1.03 µm
340 fs 2.2 ps 730 fs
170 mW 8.1 W 16.2 W
63 MHz 35 MHz
1.03 µm
3.3–89 ps 0.83–1.57 ps 810 fs
12 W
28 MHz
60 W
34.3 MHz
ML technique
Nd:phosphate Nd:phosphate
Nd:fluorophosphate Nd:silicate
λ0
frep
Remarks
Ref.
Kr-ion laser pumped Ti: sapphire laser pumped
[91Spi] [93Kel] [95Kop1] [96Kop2]
Single-prism for dispersion compensation
[98Aus] [00Pas] [97Aus] [95Kop1]
Yb:YAG Soliton-SESAM
1.03 µm
First passively mode-locked Yb:YAG laser
First passively mode-locked thin disk laser Tunable pulse duration Thin disk laser, 1.75 µJ pulse energy, 1.9 MW peak power
[95Hön] [99Hön] [99Aus] [00Aus]
Ultrafast solid-state lasers
τp
Laser material
[01Bru] [03Inn]
Yb:KGW Soliton-SESAM 1.037 µm 1.046 µm
176 fs 112 fs
1.1 W 200 mW
86 MHz 86 MHz
[00Bru] [00Bru] [1, § 3
(continued on next page)
1, § 3]
Table 2 (Continued) Laser material
ML technique
λ0
τp
Pav,out
frep
Remarks
Ref.
Thin disk laser, 0.9 µJ pulse energy, 3.3 MW peak power
[02Bru]
Yb:KYW Soliton-SESAM
1.028 µm
240 fs
22 W
25 MHz
KLM
1.057 µm
71 fs
120 mW
110 MHz
[01Liu]
Soliton-SESAM
1.045 µm
90 fs
40 mW
100 MHz
[00Dru]
Soliton-SESAM
1.062 µm
69 fs
80 mW
113 MHz
[02Dru]
58 fs 61 fs
65 mW 53 mW
112 MHz 112 MHz
[98Hön] [98Hön]
Yb:GdCOB
Yb:glass Yb:phosphate glass Yb:silicate glass
Soliton-SESAM
1.025–1.065 µm 1.03–1.082 µm
Active AOM Active FM
≈ 1.53 µm 1.53 µm ≈ 1.5 µm 1.533 µm
90 ps 9.6 ps 9.6–30 ps 48 ps
7 mW 3 mW 3 mW 1 mW
1.535 µm 1.534 µm
2.5 ps 380 fs 3.8 ps
4 mW 12 mW
Er:Yb:glass
SESAM
100 MHz 2.5 GHz 2.5 and 5 GHz 5 GHz
100 MHz 10 GHz
Dual wavelength with 165 GHz separation
[94Cer] [94Lon] [95Lap] [98Lon]
Overview of ultrafast solid-state lasers
Yb:BOYS
[99Spü2] [01Was] [02Kra1] 31
(continued on next page)
Laser material
ML technique
λ0 1.535 µm
1.528–1.561 µm
τp
Pav,out
frep
2.5 ps 2.5 ps
50 mW > 5 dBm
1.9 ps
25 mW
25 GHz
4.3 ps
18 mW
40 GHz
10 GHz 40 GHz MUX
Remarks
Ref.
Timing jitter 80 fs rms over 10 Hz and 1 MHz, 40 GHz multiplexed (MUX) Gain equalized frequency comb. with 36 discrete channels, tunable over C-band
[02Spü]
[03Spü]
[03Zel]
(continued on next page)
[1, § 3
[86Ros] Roskos, Robl and Seilmeier [1986], [86Yan] Yan, Ling, Ho and Lee [1986], [87Bad] Bado, Bouvier and Coe [1987], [88Bas] Basu and Byer [1988], [88Kel] Keller, Valdmanis, Nuss and Johnson [1988], [89Goo] Goodberlet, Wang, Fujimoto and Schulz [1989], [89Mak1] Maker and Ferguson [1989a], [89Mak2] Maker and Ferguson [1989b], [90Goo] Goodberlet, Jacobson, Fujimoto, Schulz and Fan [1990], [90Kel1] Keller, Li, Khuri-Yakub, Bloom, Weingarten and Gerstenberger [1990], [90Kel2] Keller, Knox and Roskos [1990], [90Liu1] Liu, Huxley, Ippen and Haus [1990], [90Liu2] Liu and Chee [1990], [90Mal] Malcolm, Curley and Ferguson [1990], [90Wal] Walker, Avramopoulos and Sizer II [1990], [90Wei] Weingarten, Shannon, Wallace and Keller [1990], [91Cur] Curley and Ferguson [1991], [91God] Godil, Hou, Auld and Bloom [1991], [91Hug] Hughes, Barr and Hanna [1991],
Ultrafast solid-state lasers
1.534 µm
32
Table 2 (Continued)
1, § 3]
Table 2 (Continued)
Overview of ultrafast solid-state lasers
[91Kel1] Keller, ’t Hooft, Knox and Cunningham [1991], [91Li] Li, Sheridan and Bloom [1991], [91McC] McCarthy, Maker and Hanna [1991], [91Sal] Salin, Squier and Piché [1991], [91Sar] Sarukura, Ishida and Nakano [1991], [91Spe] Spence, Kean and Sibbett [1991], [91Spi] Spielmann, Krausz, Brabec, Wintner and Schmidt [1991], [91Zho] Zhou, Malcolm and Ferguson [1991], [92Eva] Evans, Spence, Sibbett, Chai and Miller [1992], [92Hua1] Huang, Kapteyn, McIntosh and Murnane [1992], [92Hua2] Huang, Asaki, Backus, Murnane, Kapteyn and Nathel [1992], [92Hug] Hughes, Phillips, Barr and Hanna [1992], [92Kel1] Keller and Chiu [1992], [92Kel2] Keller, Miller, Boyd, Chiu, Ferguson and Asom [1992], [92Kra] Krausz, Spielmann, Brabec, Wintner and Schmidt [1992], [92Lem] Lemoff and Barty [1992], [92LiK] LiKamWa, Chai and Miller [1992], [92Liu] Liu, Flood, Walker and van Driel [1992], [92Mil] Miller, LiKamWa, Chai and Van Stryland [1992], [92Riz1] Rizvi, French and Taylor [1992a], [92Riz2] Rizvi, French and Taylor [1992b], [92Riz3] Rizvi, French and Taylor [1992c], [92Wei] Weingarten, Godil and Gifford [1992], [93Asa] Asaki, Huang, Garvey, Zhou, Kapteyn and Murnane [1993], [93Cur] Curley, Spielmann, Brabec, Krausz, Wintner and Schmidt [1993], [93Fre] French, Mellish, Taylor, Delfyett and Florez [1993], [93Kel] Keller, Chiu and Ferguson [1993], [93Ram] Ramaswamy, Gouveia-Neto, Negus, Izatt and Fujimoto [1993],
33
(continued on next page)
34
Table 2 (Continued)
Ultrafast solid-state lasers
[93Riz] Rizvi, French, Taylor, Delfyett and Florez [1993], [93Sea] Seas, Petricevic and Alfano [1993], [93Sen] Sennaroglu, Pollock and Nathel [1993], [93Wei] Weingarten, Keller, Chiu and Ferguson [1993], [93Yan] Yanovsky, Pang, Wise and Minkov [1993], [94Cer] Cerullo, De Silvestri, Laporta, Longhi, Magni, Taccheo and Svelto [1994], [94Con] Conlon, Tong, French, Taylor and Shestakov [1994], [94Dym] Dymott and Ferguson [1994], [94Ish] Ishida and Naganuma [1994], [94Kel] Keller [1994], [94Kop1] Kopf, Weingarten, Brovelli, Kamp and Keller [1994], [94Kop2] Kopf, Kärtner, Weingarten and Keller [1994], [94Lin1] Lincoln, Dymott and Ferguson [1994], [94Lin2] Lincoln and Ferguson [1994], [94Lon] Longhi, Laporta, Taccheo and Svelto [1994], [94Mel] Mellish, French, Taylor, Delfyett and Florez [1994], [94Sen] Sennaroglu, Pollock and Nathel [1994], [94Sti] Stingl, Spielmann, Krausz and Szipöcs [1994], [94Wan] Wang, Kam Wa, Lefaucheur, Chai and Miller [1994], [94Zho] Zhou, Taft, Huang, Murnane, Kapteyn and Christov [1994], [95Bro1] Brovelli, Jung, Kopf, Kamp, Moser, Kärtner and Keller [1995], [95Bro2] Brovelli, Lanker, Keller, Goossen, Walker and Cunningham [1995], [95Chu] Chung and Siegman [1995], [95Dym] Dymott and Ferguson [1995], [95Fal] Falcoz, Balembois, Georges and Brun [1995], [95Flo] Flood, Walker and van Driel [1995], [95Hön] Hönninger, Zhang, Keller and Giesen [1995], [95Kop1] Kopf, Kärtner, Keller and Weingarten [1995],
[1, § 3
(continued on next page)
1, § 3]
Table 2 (Continued)
Overview of ultrafast solid-state lasers
[95Kop2] Kopf, Weingarten, Brovelli, Kamp and Keller [1995], [95Lap] Laporta, Longhi, Marchesi, Taccheo and Svelto [1995], [95Mel] Mellish, Barry, Hyde, Jones, French, Taylor, van der Poel and Valster [1995], [95Sti] Stingl, Lenzner, Spielmann, Krausz and Szipöcs [1995], [95Tsu] Tsuda, Knox, de Souza, Jan and Cunningham [1995], [95Yan] Yanovsky, Wise, Cassanho and Jenssen [1995], [96Bra] Braun, Hönninger, Zhang, Keller, Heine, Kellner and Huber [1996], [96Col] Collings, Stark, Tsuda, Knox, Cunningham, Jan, Pathak and Bergman [1996], [96Flu] Fluck, Zhang, Keller, Weingarten and Moser [1996], [96Kär] Kärtner, Jung and Keller [1996], [96Kop1] Kopf, Zhang, Fluck, Moser and Keller [1996], [96Kop2] Kopf, Spühler, Weingarten and Keller [1996], [96Sor1] Sorokina, Sorokin, Wintner, Cassanho, Jenssen and Noginov [1996], [96Sor2] Sorokina, Sorokin, Wintner, Cassanho, Jenssen and Szipöcs [1996], [96Ton] Tong, Sutherland, French, Taylor, Shestakov and Chai [1996], [96Tsu] Tsuda, Knox and Cundiff [1996], [96Xu] Xu, Spielmann, Krausz and Szipöcs [1996], [97Agn] Agnesi, Pennacchio, Reali and Kubecek [1997], [97Aus] Aus der Au, Kopf, Morier-Genoud, Moser and Keller [1997], [97Col] Collings, Bergman and Knox [1997], [97Dym] Dymott and Ferguson [1997], [97Hen] Henrich and Beigang [1997], [97Jun] Jung, Kärtner, Matuschek, Sutter, Morier-Genoud, Zhang, Keller, Scheuer, Tilsch and Tschudi [1997], [97Kop1] Kopf, Prasad, Zhang, Moser and Keller [1997] [97Kop2] Kopf, Weingarten, Zhang, Moser, Emanuel, Beach, Skidmore and Keller [1997], [97Loe] Loesel, Horvath, Grasbon, Jost and Niemz [1997], [97Ruf] Ruffing, Nebel and Wallenstein [1997], [97Sor] Sorokina, Sorokin, Wintner, Cassanho, Jenssen and Szipöcs [1997],
35
(continued on next page)
36
Table 2 (Continued)
[1, § 3
(continued on next page)
Ultrafast solid-state lasers
[97Spä] Spälter, Böhm, Burk, Mikulla, Fluck, Jung, Zhang, Keller, Sizmann and Leuchs [1997], [97Ton] Tong, French, Taylor and Fujimoto [1997], [97Zha1] Zhang, Torizuka, Itatani, Kobayashi, Sugaya and Nakagawa [1997a], [97Zha2] Zhang, Torizuka, Itatani, Kobayashi, Sugaya and Nakagawa [1997b], [98Aus] Aus der Au, Loesel, Morier-Genoud, Moser and Keller [1998], [98Cha] Chang, Maciejko, Leonelli and Thorpe [1998], [98Gab] Gäbel, Rußbüldt, Lebert and Valster [1998], [98Hön] Hönninger, Morier-Genoud, Moser, Keller, Brovelli and Harder [1998], [98Liu] Liu, Qian, Wise, Zhang, Itatani, Sugaya, Nakagawa and Torizuka [1998], [98Lon] Longhi, Sorbello, Taccheo and Laporta [1998], [98Mel] Mellish, Chernikov, French and Taylor [1998], [98Pet] Petrov, Shcheslavskiy, Mirtchev, Noack, Itatani, Sugaya and Nakagawa [1998], [98Xu] Xu, Tempea, Spielmann, Krausz, Stingl, Ferencz and Takano [1998], [99Aus] Aus der Au, Schaer, Paschotta, Hönninger, Keller and Moser [1999], [99Bar] Bartels, Dekorsky and Kurz [1999], [99Bed] Beddard, Sibbett, Reid, Garduno-Mejia, Jamasbi and Mohebi [1999], [99Cho] Cho, Bouma, Ippen and Fujimoto [1999], [99Cou] Couderc, Louradour and Barthelemy [1999], [99Gra] Graf, Ferguson, Bente, Burns and Dawson [1999], [99Hön] Hönninger, Paschotta, Graf, Morier-Genoud, Zhang, Moser, Biswal, Nees, Braun, Mourou, Johannsen, Giesen, Seeber and Keller [1999], [99Jeo] Jeong, Kang and Nam [1999] [99Kra1] Krainer, Paschotta, Aus der Au, Hönninger, Keller, Moser, Kopf and Weingarten [1999], [99Kra2] Krainer, Paschotta, Spühler, Moser and Keller [1999], [99Kub] Kubecek, Couderc, Bourliaguet, Louradour and Barthelemy [1999], [99Lef] Lefort, Puech, Butterworth, Svirko and Hanna [1999], [99Mor] Morgner, Kärtner, Cho, Chen, Haus, Fujimoto, Ippen, Scheuer, Angelow and Tschudi [1999], [99Spü1] Spühler, Paschotta, Keller, Moser, Dymott, Kopf, Meyer, Weingarten, Kmetec, Alexander and Truong [1999], [99Spü2] Spühler, Gallmann, Fluck, Zhang, Brovelli, Harder, Laporta and Keller [1999],
1, § 3]
Table 2 (Continued)
37
(continued on next page)
Overview of ultrafast solid-state lasers
[99Sut] Sutter, Steinmeyer, Gallmann, Matuschek, Morier-Genoud, Keller, Scheuer, Angelow and Tschudi [1999], [99Uem] Uemura and Torizuka [1999], [99Zha] Zhang, Nakagawa, Torizuka, Sugaya and Kobayashi [1999], [00Aus] Aus der Au, Spühler, Südmeyer, Paschotta, Hövel, Moser, Erhard, Karszewski, Giesen and Keller [2000], [00Bru] Brunner, Spühler, Aus der Au, Krainer, Morier-Genoud, Paschotta, Lichtenstein, Weiss, Harder, Lagatsky, Abdolvand, Kuleshov and Keller [2000], [00Bur] Burns, Hetterich, Ferguson, Bente, Dawson, Davies and Bland [2000], [00Dru] Druon, Balembois, Georges, Brun, Courjaud, Hönninger, Salin, Aron, Mougel, Aka and Vivien [2000], [00Kra1] Krainer, Paschotta, Moser and Keller [2000a], [00Kra2] Krainer, Paschotta, Moser and Keller [2000b], [00Mat] Matuschek, Gallmann, Sutter, Steinmeyer and Keller [2000], [00Pas] Paschotta, Aus der Au, Spühler, Morier-Genoud, Hövel, Moser, Erhard, Karszewski, Giesen and Keller [2000], [00Spü] Spühler, Südmeyer, Paschotta, Moser, Weingarten and Keller [2000], [00Tom] Tomaru and Petek [2000], [01Bru] Brunner, Paschotta, Aus der Au, Spühler, Morier-Genoud, Hövel, Moser, Erhard, Karszewski, Giesen and Keller [2001], [01Cho] Cho, Kärter, Morgner, Ippen, Fujimoto, Cunningham and Knox [2001], [01Chu] Chudoba, Fujimoto, Ippen, Haus, Morgner, Kärtner, Scheuer, Angelow and Tschudi [2001], [01Ell] Ell, Morgner, Kärtner, Fujimoto, Ippen, Scheuer, Angelow, Tschudi, Lederer, Boiko and Luther-Davis [2001], [01Kem] Kemp, Stormont, Agate, Brown, Keller and Sibbett [2001], [01Liu] Liu, Nees and Mourou [2001], [01Tom] Tomaru [2001], [01Was] Wasik, Helbing, König, Sizmann and Leuchs [2001], [02Bru] Brunner, Südmeyer, Innerhofer, Morier-Genoud, Paschotta, Kisel, Shcherbitsky, Kuleshov, Gao, Contag, Giesen and Keller [2002], [02Dru] Druon, Chenais, Raybaut, Balembois, Georges, Gaume, Aka, Viana, Mohr and Kopf [2002], [02Kra1] Krainer, Paschotta, Spühler, Klimov, Teisset, Weingarten and Keller [2002], [02Kra2] Krainer, Paschotta, Lecomte, Moser, Weingarten and Keller [2002], [02Maj1] Major, Langford, Graf and Ferguson [2002], [02Maj2] Major, Langford, Graf, Burns and Ferguson [2002], [02McC] McConnell, Ferguson and Langfrod [2002],
38
[02Pra] [02Rip] [02Sch] [02Spü]
Prasankumar, Chudoba, Fujimoto, Mak and Ruane [2002], Ripin, Chudoba, Gopinath, Fujimoto, Ippen, Morgner, Kärtner, Scheuer, Angelow and Tschudi [2002], Schön, Haiml, Gallmann and Keller [2002], Spühler, Dymott, Klimov, Luntz, Baraldi, Kilburn, Crosby, Thomas, Zehnder, Teisset, Brownell, Weingarten, Dangel, Offrein, Bona, Buccafusca, Kaneko, Krainer, Paschotta and Keller [2002], [02Wag] Wagenblast, Morgner, Grawert, Schibli, Kärtner, Scheuer, Angelow and Lederer [2002], [03Inn] Innerhofer, Südmeyer, Brunner, Häring, Aschwanden, Paschotta, Hönninger, Kumkar and Keller [2003], [03Spü] Spühler, Golding, Krainer, Kilburn, Crosby, Brownell, Weingarten, Paschotta, Haiml, Grange and Keller [2003], [03Uem] Uemura and Torizuka [2003], [03Zel] Zeller, Krainer, Spühler, Weingarten, Paschotta and Keller [2003]
Ultrafast solid-state lasers
Table 2 (Continued)
[1, § 3
1, § 3]
Overview of ultrafast solid-state lasers
39
somewhat different saturable absorber parameters than bulk lasers. However, as has been demonstrated early on, the SESAM parameters can be adjusted for stable cw mode locking of fiber lasers (Zirngibl, Stulz, Stone, Hugi, DiGiovanni and Hansen [1991], Ober, Hofer, Keller and Chiu [1993]). Interested readers are referred to the review articles given above. Mode-locked and gain-switched semiconductor lasers compared to solid-state lasers are still low in output power (milli-watt regime) and also have significantly longer pulse durations (a few hundreds of femtosecond at best). They have, however, the advantage of being electrically pumped. Mode-locked semiconductor lasers are compact, easy to use and cost effective. They can be used in applications that do not require high powers. Higher output power can be achieved from semiconductor optical amplifiers (Delfyett, Florez, Stoffel, Gmitter, Andreadakis, Silberberg, Heritage and Alphonse [1992], Delfyett [1995]). A more extensive recent review of ultrashort pulse generation with semiconductor lasers is given in (Jiang and Bowers [1995], Vail’ev [1995], Delfyett [2003]). More recently, diode-pumped vertical external cavity semiconductor lasers have been passively mode-locked (Hoogland, Dhanjal, Roberts, Tropper, Häring, Paschotta, MorierGenoud and Keller [2000]) with an intracavity saturable absorber for the first time. Average power scaling is not limited and 950 mW of average power in 15ps pulses with a 6-GHz repetition rate (Häring, Paschotta, Aschwanden, Gini, Morier-Genoud and Keller [2002]) has been generated, with the promise of even higher powers in the multi-GHz regime.
3.2. Design guidelines of diode-pumped solid-state lasers An all-solid-state ultrafast laser technology is based on diode-pumped solid-state lasers. These lasers have to be optimized to support stable pulse generation. The discussion in the following sections will show that a small saturation energy of the laser medium results in a lower tendency of self-Q-switching instabilities. The saturation fluence of a four-level laser system is hν mσL and for a three-level system Fsat,L =
Fsat,L =
hν m(σL + σLabs )
(3.1)
,
(3.2)
where hν is the lasing photon energy, σL is the gain cross section, σLabs is the absorption cross section of the three-level gain medium and m is the number
40
Ultrafast solid-state lasers
[1, § 3
of passes through the gain medium per cavity round-trip. In case of a standing wave laser cavity this factor is m = 2, in a unidirectional ring laser cavity it is m = 1. A small saturation energy, low pump threshold and good beam quality is obtained with a small pump and cavity mode volume while maintaining good spatial overlap of the pump laser and laser mode. This can be easily obtained when a diffraction-limited pump laser is used, as, for example, in a Ti:sapphire laser. The lower limit of the pump volume is then set by diffraction, and ultimately pump-induced damage to the crystal. However, diode laser arrays or bars do not generate diffraction limited pump beams, which makes the situation a bit more complicated and is therefore explained next. The propagation of diffraction-limited Gaussian laser beams is extensively described in many text books, see for example Saleh and Teich [1991] and Siegman [1986]. A beam quality factor M 2 was introduced to describe the propagation of non-diffraction limited beams (Siegman [1986], Sasnett [1989]). The objective was to provide propagation equations for non-diffraction limited beams that retain the simplicity of the fundamental Gaussian mode beam equations. The M 2 -factor is given by M2 ≡
θ , θG
with θG ≡
λ , πW0,G
(3.3)
where θ is the actual far-field divergence angle of any beam with any mixtures of modes, θG the far-field Gaussian beam divergence angle, W0,G the beam waist of a Gaussian beam which is set equal to the beam waist of the non-diffraction limited beam. The quantity M 2 is then a numerical expression of (inverse) beam quality with 1 being a perfect Gaussian beam and higher values indicating “poorer” quality. This is entirely equivalent to the “number of times diffraction limit” quantity introduced by Siegman [1986]. The beam quality does not give any information about the details of higher-order mode content in the beam. The propagation of laser beams with M 2 larger than 1, can be reduced to standard Gaussian beam propagation after substituting the wavelength λn with a new “effective wavelength” λeff given by λeff = M 2 · λn ,
(3.4)
where λn is the wavelength in the dispersive medium (i.e. λn = λ/n) in which the beam is propagating. Physically, this means that non-diffraction-limited beams propagate like an ideal diffraction-limited Gaussian beam but with the new, longer “effective wavelength”. Beams with larger M 2 have larger “effective wavelengths”, and therefore a smaller depth of focus for a given beam waist. The output beam of a laser diode array or broad-stripe diode suffers from poor beam quality. In the so-called “fast” axis, perpendicular to the pn-junction of the
1, § 3]
Overview of ultrafast solid-state lasers
41
diode laser, the light diverges with a large angle (25 to 40◦ , typically) from a narrow aperture of ≈ 1 µm. However, in this direction the light is nearly diffraction2 ≈ 1. Thus, even though the light in the fast axis is highly dilimited with Mfast vergent, it can be efficiently collected with a “fast” high-numerical aperture lens and tightly focussed due to its diffraction-limited nature. In the “slow” axis, parallel to the pn-junction of the diode laser, the beam typically has a divergence of ≈ 10◦ . For single-stripe diodes, the emitting aperture is ≈ 3 µm, resulting in a beam close to diffraction-limited. For higher-power “arrays” of such apertures, the divergence is also ≈ 10◦ , but the total aperture has increased from typically 50 µm to more than 200 µm, or in case of “arrays of arrays” (i.e. bars) to a width of approximately 1 cm. The diode laser light in the slow axis is therefore many times worse than diffraction limited. High brightness diode arrays with 5 W out2 of ≈ 20. More recently put power and a 100 µm stripe width can have an Mslow diode arrays of 1.5 W and a 50 µm strip width can have an M 2 ≈ 12. Low bright2 ness bars with ≈ 20 W and ≈ 1 cm stripe width have an Mslow > 1000. The slow axis ultimately limits the spot size of focussed pump due to the requirements of mode matching to the laser mode. The good news is that the brightness of diode pump lasers continues to improve. With such pump lasers, the lowest pump threshold can be achieved with the following optimized mode matching (OMM) design guidelines applied to both the fast and slow axis of the diode pump laser (Fan and Sanchez [1990], Kopf, Weingarten, Zhang, Moser, Emanuel, Beach, Skidmore and Keller [1997]): (i) Determine M 2 for the pump beam (eq. (3.3)) where 2W0,G is set equal to the width of the pump source Dp . The width of the pump source is approximately given by the stripe width of a diode array or bar or more accurately by the second-order intensity moment. (ii) Determine the “effective wavelength” λeff (eq. (3.4)). (iii) Set the depth of focus or confocal parameter b of the pump beam approximately equal to the absorption length La of the pump beam in the laser medium, i.e. b ≈ La . (iv) Determine the smallest pump beam waist W0,opt for which a good mode overlap over the absorption length of the pump and the cavity mode can be obtained. This is the minimum pump spot size in the gain medium that still guarantees good laser beam quality and therefore determines the lowest pump threshold: Calculate optimum beam waist radius W0,opt using Rayleigh range formula for an ideal Gaussian beam (i.e. b = 2z0 , where z0 is the Rayleigh range of a Gaussian beam) with the “effective wavelength” given in eq. (3.3) and a confocal parameter b given in (iii):
42
Ultrafast solid-state lasers
W0,opt =
[1, § 3
λeff b = 2π
M 2 λn La . 2π
(3.5)
From eq. (3.5) it becomes clear that for a small spot size the absorption length La in the gain medium should be as short as possible. The absorption length, however, limits the maximum pump power at which some thermal effects will start to degrade the laser’s performance. This will be more severe for “thermally challenged” lasers which exhibit a low thermal heat conductivity and/or upper state lifetime quenching. Low thermal conductivity results in large thermal lenses and distortions, which limit the maximum pump power. Such a thermally challenged laser material is Cr:LiSAF which is interesting for an all-solid-state femtosecond laser. Upper state lifetime quenching as observed in Cr:LiSAF results in the following: As the temperature in the laser medium increases, the upper state lifetime of the laser drops, and the pump threshold increases. Beyond a critical temperature, the laser actually switches off. If the absorption length is too short for these materials, this critical temperature occurs at relatively low pump powers. There is an optimum doping level for best mode matching to the available pump diodes and for minimizing pump-induced upper-state lifetime quenching. In standard diode pumping, we use high-brightness diode arrays (i.e. brightness as high as possible) and apply OMM (eqs. (3.1)–(3.5)) only in the slow axis of the diodes and weaker focussing in the fast axis. This results in an approximately circular pump beam that becomes slightly elliptical when the laser crystal is pumped at a Brewster angle. Standard diode pumping is explained in more detail by the example of a diode-pumped Cr:LiSAF laser (Kopf, Weingarten, Brovelli, Kamp and Keller [1994]). With this standard pumping approach, the average output power was limited by the above-mentioned thermal problems to 230 mW cw and 125 mW mode-locked with 60 fs pulses (Kopf, Prasad, Zhang, Moser and Keller [1997]). Standard diode pumping has also been successfully used with most other solid-state lasers such as Nd:YAG. Such lasers are not “thermally challenged”, and much higher average output power has been achieved with this approach. Significantly more output power can be obtained with a diode-pumped Cr:LiSAF laser for which OMM (eqs. (3.1)–(3.5)) is applied to both axes in combination with a long absorption length and efficient cooling (Kopf, Keller, Emanuel, Beach and Skidmore [1997], Kopf, Aus der Au, Keller, Bona and Roentgen [1995]). Optimized mode matching in both axes results in a highly elliptical laser mode in the crystal, because the pump beam can be focused to a much smaller beam radius in the diffraction limited fast axis compared to the slow axis. Additionally, we can extract the heat very efficiently with a thin crystal of ≈ 1 mm height and obtain approximately a one-dimensional heat flow. With a cylindrical
1, § 4]
Loss modulation
43
cavity mirror we still obtained nearly ideal TEM00 output beams. Using a 15 W, 2 2 = 1 (Skidmore, = 1200 and Mfast 0.9 cm wide diode pump array with an Mslow Emanuel, Beach, Benett, Freitas, Carlson and Solarz [1995]), the average output power of such a diode-pumped Cr:LiSAF laser was 1.4 W cw, 500 mW modelocked with 110 fs pulses, and 340 mW mode-locked with 50 fs pulses (Kopf, Prasad, Zhang, Moser and Keller [1997]). Combined with a relatively long absorption length, we pumped a thin sheet volume with a width of approximately 1 mm, a length of La ≈ 4 mm and a thickness of ≈ 80 µm in the laser crystal. This approach was also applied to a diode-pumped Nd:glass laser, resulting in an average output power of about 2 W cw and 1 W mode-locked with pulses as short as 175 fs (Aus der Au, Loesel, Morier-Genoud, Moser and Keller [1998]) and more recently 1.4 W with pulses as short as 275 fs (Paschotta, Aus der Au, Spühler, Morier-Genoud, Hövel, Moser, Erhard, Karszewski, Giesen and Keller [2000]). In addition, a diode-pumped Yb:YAG laser with the same approach produced 3.5 W average power with 1-ps pulses and 8.1 W with 2.2-ps pulses (Aus der Au, Schaer, Paschotta, Hönninger, Keller and Moser [1999]).
§ 4. Loss modulation 4.1. Optical modulators: acousto-optic and electrooptic modulators Many textbooks, for example, Siegman [1986], Yariv and Yeh [1984], Svelto [1998], have reviewed optical modulators for pulse generation. Today the most important optical modulators for short pulse generation are the acousto-optic and electrooptic modulators. The acousto-optic modulators have the advantage of low optical insertion loss and can readily be driven at high repetition rates. They are typically used for cw mode locking. However, for Q-switching their loss modulation is limited and the switching time is rather slow. Therefore, acousto-optic modulators are primarily used for repetitive Q-switching of cw-pumped lasers (e.g., Nd:YAG) and electrooptic modulators are used for Q-switching in general. For mode locking, the acousto-optic modulator typically consists of an acoustooptic substrate (typically fused quartz) and a transducer that launches an acoustic wave into the substrate. An acoustic resonator is formed when opposite to the transducer the crystal substrate is air backed. Then the acoustic wave is reflected and an acoustic standing wave is formed which produces a light modulator at twice the microwave drive frequency. At higher frequencies (a few hundreds of megahertz to a few gigahertz) the loss modulation is strongly reduced by the acoustic attenuation in the substrate. Thus, at higher modulation frequencies a
44
Ultrafast solid-state lasers
[1, § 4
sapphire (Keller, Li, Khuri-Yakub, Bloom, Weingarten and Gerstenberger [1990], Weingarten, Shannon, Wallace and Keller [1990]), or a GaP (Walker, Avramopoulos and Sizer II [1990]) substrate has been used successfully.
4.2. Saturable absorber: self-amplitude modulation (SAM) Saturable absorbers have been used to passively Q-switch and mode-lock many lasers. Different saturable absorbers, such as organic dyes, color filter glasses, dye-doped solids, ion-doped crystals and semiconductors have been used. Independent of the specific saturable absorber material, we can define a few macroscopic absorber parameters that will determine the pulse generation process. The relevant macroscopic properties of a saturable absorber are the modulation depth, the nonsaturable loss, the saturation fluence, the saturation intensity and the impulse response or recovery times (Table 3). These parameters determine the operation of a passively mode-locked or Q-switched laser. In our notation we assume that the saturable absorber is integrated within a mirror structure, thus, we are interested in the nonlinear reflectivity change R(t) as a function of time and R(Fp,A ) as a function of the incident pulse energy fluence on the saturable absorber. If the saturable absorber is used in transmission, we simply characterize the absorber by nonlinear transmission measurements, i.e. T (t) and T (Fp,A ). Both the saturation fluence Fsat,A and the absorber recovery time τA are determined experimentally without the need to determine the microscopic properties of the nonlinearities. The saturation fluence of the absorber does not only depend on material properties but also on the specific device structure in which the absorber is integrated, which gives significantly more design freedom. Table 3 Saturable absorber quantities, their defining equations and units Quantity
Symbol
Defining equation or measurement
Unit
Saturation fluence Recovery time Incident beam area Saturation energy Saturation intensity Modulation depth Nonsaturable loss Incident pulse energy Incident pulse fluence Incident intensity
Fsat,A τA AA Esat,A Isat,A R or T Rns or Tns Ep Fp,A IA (t)
Measurement R(Fp,A) or T (Fp,A ) (fig. 5) Measurement R(t) or T (t) (fig. 4) Measurement Esat,A = AA Fsat,A Isat,A = Fsat,A /τA Measurement R(Fp,A) or T (Fp,A ) (fig. 5) Measurement R(Fp,A) or T (Fp,A ) (fig. 5) Measurement Fp,A = Ep /AA Fp,A = IA (t) dt
J/cm2 s cm2 J W/cm2
J J/cm2 W/cm2
1, § 4]
Loss modulation
45
Fig. 4. Typical measured impulse response of a SESAM measured with standard degenerate pump-probe measurements using two different excitation pulse durations. The saturable absorber was grown at low temperature, which reduced the recovery time to about 20 ps. The short intraband thermalization recovery time results in negligible modulation depth with a 4-ps excitation pulse. Thus, only the slower recovery time due to carrier trapping is important in the picosecond regime.
Standard pump-probe techniques determine the impulse response R(t) and therefore τA . In the picosecond regime we normally only have to consider one picosecond recovery time, because much faster femtosecond nonlinearities in the saturable absorber give negligible modulation depth. This is shown with a semiconductor saturable absorber in fig. 4, where the differential impulse response DR(t) was measured for two different excitation pulse durations. For excitation with a picosecond pulse the pump-probe trace clearly shows no significant modulation depth with a fast time constant. In the femtosecond pulse regime we normally have to consider more than one absorber recovery time. In this case the slow component normally helps to start the initial pulse formation process. The modulation depth of the fast component then determines the pulse duration at steady state. Further improvements of the saturable absorber require a better understanding of the underlying physics of the nonlinearities which is very interesting and rather complex. A more detailed review about the microscopic properties of ultrafast semiconductor nonlinearities for saturable absorber applications is provided by Keller [1999] and Siegner and Keller [2000]. Ultrafast semiconductor dynamics in general are discussed in more detail by Shah [1996]. However, a basic knowledge of the macroscopic properties of the absorber and how set them to a certain value is normally sufficient for stable pulse generation. The saturation fluence Fsat,A is determined and defined by measuring the nonlinear change in reflectivity R(Fp,A ) as a function of increased incident pulse fluence (fig. 5). The common travelling wave rate equations (Agrawal and Olsson [1989]) in the slow absorber approximation usually give a very good fit and deter-
46
Ultrafast solid-state lasers
[1, § 4
Fig. 5. Measured nonlinear reflectivity as a function of incident pulse fluence on a SESAM. Theoretical fit determines the macroscopic saturable absorber parameters: saturation fluence Fsat,A , modulation depth R and nonsaturable loss Rns .
mine the saturation fluence Fsat,A , the modulation depth R and the nonsaturable losses Rns of the absorber (Brovelli, Keller and Chiu [1995], Keller [1999]). The modulation depth is typically small to prevent Q-switching instabilities in passively mode-locked solid-state lasers (Hönninger, Paschotta, Morier-Genoud, Moser and Keller [1999]). Thus it is reasonable to make the following approximation: R = 1 − e−2q0 ≈ 2q0 ,
q0 1,
(4.1)
where q0 is the unsaturated amplitude loss coefficient. The saturation of an absorber can be described with the following differential equation (Agrawal and Olsson [1989]): q(t) − q0 q(t)P (t) dq(t) =− − , dt τA Esat,A
(4.2)
where q(t) is the saturable amplitude loss coefficient that does not include any nonsaturable losses. Note that eq. (4.2) may not be precise for strongly saturated real absorbers. For example, strongly saturated semiconductor absorbers can exhibit additional effects such as two-photon absorptions, free carrier absorption, thermal and even various damage effects. Two-photon-absorption only starts to become significant in the femtosecond pulse-width regime and results in an earlier roll-off of the nonlinear reflectivity at high incident pulse fluences (Thoen, Koontz. M. Joschko, Langlois, Schibli, Kärtner, Ippen and Kolodziejski [1999]). This well-known effect has been used in power-limiting devices before (Walker, Kar, Ji, Keller and Smith [1986]). In this regime, however, the absorber is operated more closely to the damage threshold which needs to be evaluated separately. Our
1, § 4]
Loss modulation
47
experience is that damage in semiconductor saturable absorbers typically occurs at around 100 times the saturation fluence of the absorber with long-term degradation observed close below this damage threshold. Therefore, we normally operate the device at least an order of magnitude below this damage threshold, ideally at an incident pulse fluence of 3 to 5 times the saturation fluence of the absorber. We therefore neglect these very high-fluence effects in the following discussion. At any time t the reflected (or transmitted) intensity Iout (t) from the saturable absorber is given by Iout (t) = R(t)Iin (t) = e−2q(t )Iin (t). Then the total net reflectivity is given by Iout (t) dt 2 Fout =1− = q(t)Iin (t) dt. Rtot = Fin Fin Iin (t) dt
(4.3)
(4.4)
This determines the total absorber loss coefficient qp , which results from the fact that part of the excitation pulse needs to be absorbed to saturate the absorber: Rtot = e−2qp ≈ 1 − 2qp . From eqs. (4.4) and (4.5) it then follows 1 qp = q(t)Iin (t) dt = q(t)f (t) dt, Fin where 1 Iin (t) Pin (t) = , with f (t) dt = f (t) ≡ Iin (t) dt = 1. Fin Ep,in Fin
(4.5)
(4.6)
(4.7)
We then distinguish between two typical cases: a slow and a fast saturable absorber. 4.2.1. Slow saturable absorber In the case of a slow saturable absorber, we assume that the excitation pulse duration is much shorter than the recovery time of the absorber (i.e. τp τA ). Thus, we can neglect the recovery of the absorber during pulse excitation and eq. (4.2) reduces to: q(t)P (t) dq(t) ≈− . (4.8) dt Esat,A This differential equation can be solved and we obtain for the self-amplitude modulation (SAM): t Ep f (t ) dt . q(t) = q0 exp − (4.9) Esat,A 0
48
Ultrafast solid-state lasers
[1, § 4
Equation (4.6) then determines the total absorber loss coefficient for a given incident pulse fluence Fp,A : Fsat,A 1 − e−Fp,A /Fsat,A . qp (Fp,A ) = q(t)f (t) dt = q0 (4.10) Fp,A It is not surprising that qp does not depend on any specific pulse form because τp τA . It is useful to introduce a saturation parameter S ≡ Fp,A /Fsat,A . For strong saturation (S > 3), we have qp (Fp,A ) ≈ q0 /S (eq. (4.10)) and the absorbed pulse fluence is about ≈ Fsat,A R. 4.2.2. Fast saturable absorber In the case of a fast saturable absorber, the absorber recovery time is much faster than the pulse duration (i.e. τp τA ). Thus, we can assume that the absorption instantaneously follows the absorption of a certain power P (t) and eq. (4.2) reduces to 0=−
q(t) − q0 q(t)P (t) − . τA Esat,A
(4.11)
The saturation of the fast absorber then follows directly from eq. (4.11): q(t) =
q0 1+
IA (t ) Isat,A
,
(4.12)
where we used the fact that Psat,A = Esat,A /τA and P (t)/Psat,A = IA (t)/Isat,A . In the linear regime we can make the following approximation in eq. (4.12): q(t) ≈ q0 − γA P (t),
with γA ≡
q0 . Isat,A AA
(4.13)
The total absorber loss coefficient qp (eqs. (4.5)–(4.7)) now depends on the pulse form and for a sech2 -pulse shape we obtain for an incident pulse fluence Fp,A and the linear approximation of q(t) for weak absorber saturation (eq. (4.13)):
1 1 Fp,A q(t)IA (t) dt = q0 1 − . qp (Fp,A ) = (4.14) Fp,A 3 τ Isat,A We will later see that we only obtain an analytic solution for fast saturable absorber mode locking, if we assume an ideal fast absorber that saturates linearly with pulse intensity (eq. (4.13)) – which in principle only applies for weak absorber saturation in a real absorber. For a maximum modulation depth, we then can assume that q0 = γA P0 (assuming an ideal fast absorber over the full modulation depth). We then obtain with eq. (4.14) a residual saturable absorber loss of q0 /3 suffered by the pulse to fully saturate the ideal fast saturable absorber.
1, § 4]
Loss modulation
49
4.3. Semiconductor saturable absorbers Semiconductor saturable absorbers have been used as early as 1974 in CO2 lasers (Gibson, Kimmitt and Norris [1974]) and 1980 for semiconductor diode lasers (Ippen, Eichenberger and Dixon [1980]). A color center laser was the first solidstate laser that was cw mode-locked with an intracavity semiconductor saturable absorber (Islam, Sunderman, Soccolich, Bar-Joseph, Sauer, Chang and Miller [1989]). However, for both the diode and color center laser, dynamic gain saturation supported pulse formation and the recovery time of the slow saturable absorber was not relevant for pulse generation (fig. 3a). In addition, because of the large gain cross section (i.e. approximately 10−14 cm2 for diode lasers and 10−16 cm2 for color center lasers) Q-switching instabilities were not a problem. This is not the case for most other solid-state lasers, such as ion-doped solid-state lasers, which have typically 1000 or even more times smaller gain cross sections. Thus, the semiconductor saturable absorber parameters (figs. 4 and 5) have to be chosen much more carefully for stable cw mode locking. We typically integrate the semiconductor saturable absorber into a mirror structure, which results in a device whose reflectivity increases as the incident optical intensity increases. This general class of devices is called semiconductor saturable absorber mirrors (SESAMs) (Keller, Miller, Boyd, Chiu, Ferguson and Asom [1992], Keller, Weingarten, Kärtner, Kopf, Braun, Jung, Fluck, Hönninger, Matuschek and Aus der Au [1996]). Detailed descriptions and guidelines for designing a SESAM for either passive mode locking or Q-switching for different laser parameters have been given by Keller [1999] and Paschotta and Keller [2003]. SESAMs are well-established as useful devices for passive mode locking and Q-switching in many kinds of solid-state lasers. The main reason for its utility is that both the linear and nonlinear optical properties can be engineered over a wide range, allowing for more freedom in the specific laser cavity design. In addition, semiconductor saturable absorbers are ideally suited for passive modelocking solid-state lasers because the large absorber cross section (in the range of 10−14 cm2 ) and therefore small saturation fluence is ideally suited for suppressing Q-switching instabilities (Hönninger, Paschotta, Morier-Genoud, Moser and Keller [1999]). Initially, semiconductor saturable absorber mirrors for solid-state lasers were used in coupled cavities (Keller, Knox and Roskos [1990], Keller and Chiu [1992]), because these early SESAM designs introduced too much loss for the laser cavity (fig. 6a). In 1992, this work resulted in a new type of intracavity saturable absorber mirror, the antiresonant Fabry–Perot saturable absorber (A-FPSA) (Keller, Miller, Boyd, Chiu, Ferguson and Asom [1992], Keller
50
Ultrafast solid-state lasers
[1, § 4
Fig. 6. Historical evolution of different SESAM designs: (a) Initially the semiconductor saturable absorber was used inside a nonlinear coupled cavity, termed resonant passive modelocking (RPM) (Keller, Knox and Roskos [1990]). (b) First intracavity saturable absorber to passively modelock diode-pumped solid-state lasers without Q-switching instabilities: antiresonant Fabry–Perot saturable absorber (A-FPSA) (Keller, Miller, Boyd, Chiu, Ferguson and Asom [1992]). (c) Scaling of the A-FPSA resulted in a single quantum well saturable absorber integrated into a Bragg mirror (Brovelli, Jung, Kopf, Kamp, Moser, Kärtner and Keller [1995]) – later also referred to as saturable Bragg reflector (SBR) (Tsuda, Knox, de Souza, Jan and Cunningham [1995]). (d) General concept of semiconductor saturable absorber mirror (SESAM) without any restrictions on the mirror design (Keller and Kopf [1995], Keller, Weingarten, Kärtner, Kopf, Braun, Jung, Fluck, Hönninger, Matuschek and Aus der Au [1996]).
[1994]), where the absorber was integrated inside a Fabry–Perot structure of which the bottom reflector was a high reflector (i.e. approximately 100%) (fig. 6b). This was the first intracavity saturable absorber design that allowed for passive mode-locking of diode-pumped solid-state lasers without Q-switching instabilities. The Fabry–Perot was operated at antiresonance to obtain broad bandwidth and low loss. The A-FPSA mirror was mainly based on absorber layers sandwiched between the lower semiconductor and the higher SiO2 /TiO2 dielectric Bragg mirrors. The top reflector of the A-FPSA provides an adjustable parameter that determines the intensity entering the semiconductor saturable absorber and therefore the saturation fluence of the saturable absorber device. Therefore, this design allowed for a large variation of absorber parameters by simply changing absorber thickness and top reflectors (Brovelli, Keller and Chiu [1995], Jung, Brovelli, Kamp, Keller and Moser [1995]). This resulted in an even simpler SESAM design with a single quantum well absorber layer integrated into a Bragg mirror (Brovelli, Jung, Kopf, Kamp, Moser, Kärtner and Keller
1, § 4]
Loss modulation
51
[1995], Tsuda, Knox, de Souza, Jan and Cunningham [1995]) (fig. 6c) – this was also referred to as saturable Bragg reflectors (SBRs) (Tsuda, Knox, de Souza, Jan and Cunningham [1995]). In the 10-femtosecond regime with Ti:sapphire lasers we have typically replaced the lower semiconductor Bragg mirror with a metal mirror to support the required large reflection bandwidth (Fluck, Jung, Zhang, Kärtner and Keller [1996], Jung, Kärtner, Matuschek, Sutter, MorierGenoud, Shi, Scheuer, Tilsch, Tschudi and Keller [1997]). However, more recently an ultrabroadband monolithically-grown fluoride semiconductor saturable absorber mirror was demonstrated that covers nearly the entire gain spectrum of the Ti:sapphire laser. Using this SESAM inside a Ti:sapphire laser resulted in 9.5fs pulses (Schön, Haiml, Gallmann and Keller [2002]). The reflection bandwidth was achieved with a AlGaAs/CaF2 semiconductor Bragg mirror (Schön, Haiml and Keller [2000]). Keller and Kopf [1995] further realized that the intracavity saturable absorber can be integrated in a more general mirror structure that allows for both saturable absorption and negative dispersion control, which is now generally referred to as a semiconductor saturable absorber mirror (SESAM) (fig. 6d). In a general sense we can then reduce the design problem of a SESAM to the analysis of multilayered interference filters for a given desired nonlinear reflectivity response for both the amplitude and phase. The A-FPSA (Keller, Miller, Boyd, Chiu, Ferguson and Asom [1992]), the saturable Bragg reflector (SBR) (Brovelli, Jung, Kopf, Kamp, Moser, Kärtner and Keller [1995], Tsuda, Knox, de Souza, Jan and Cunningham [1995], Knox [1995]), and the dispersive saturable absorber mirror (DSAM) (Kopf, Zhang, Fluck, Moser and Keller [1996]) are examples of SESAM designs. In this more general class of design we do not restrict ourselves to Bragg mirror structures, which are defined by a stack of quarter-wave layers with alternating high and low refractive indices (e.g., Knox [1995], Kim, Garmire, Hummel and Dapkus [1989]). For example, we have demonstrated with many examples that non-quarter-wave layers in mirrors give more design freedom for integrating the absorber layers into the mirror structure. Furthermore, double-chirped semiconductor mirror structures can provide very broadband negative dispersion (Paschotta, Spühler, Sutter, Matuschek, Keller, Moser, Hövel, Scheuer, Angelow and Tschudi [1999]). In addition, low-field enhancement (LFR) SESAM designs (Weingarten, Spühler, Keller and Krainer [2001]) can be used to further reduce saturation fluence without the detrimental effects of strongly resonant structures such as bistability and narrow bandwidth. Such a LFR-SESAM design has a lowfinesse resonant structure such that the field strength is substantially higher in the spacer layer containing the absorber and therefore reducing the saturation fluence further. Thus, different absorber materials and parameters have been used
52
Ultrafast solid-state lasers
[1, § 4
in many passively mode-locked solid-state lasers which resulted in a much better understanding of the absorber and laser design necessary to obtain stable passive mode locking (Hönninger, Paschotta, Morier-Genoud, Moser and Keller [1999]) or Q-switching (Spühler, Paschotta, Fluck, Braun, Moser, Zhang, Gini and Keller [1999]) of many different solid-state lasers. With a few exceptions (Braun and Keller [1995]), the SESAM is mostly used as an end mirror of a standing-wave cavity. Very compact cavity designs have been achieved, for example, in passively Q-switched microchip lasers (Braun, Kärtner, Keller, Meyn and Huber [1996], Spühler, Paschotta, Fluck, Braun, Moser, Zhang, Gini and Keller [1999], Häring, Paschotta, Fluck, Gini, Melchior and Keller [2001], Häring, Paschotta, Gini, Morier-Genoud, Melchior, Martin and Keller [2001]) and passively mode-locked miniature lasers (Krainer, Paschotta, Spühler, Moser and Keller [1999], Krainer, Paschotta, Moser and Keller [2000b]), where a short laser crystal defines a simple monolithic cavity. The SESAM attached directly to the laser crystal then formed one end-mirror of this laser cavity. As the laser can not be pumped through the SESAM, the laser output needs to be separated from the collinear pump by a dichroic mirror. These examples suggest that there is need for a device which combines the nonlinear properties of the SESAM with an output coupler. This has been demonstrated before for a passively mode-locked fiber laser (Sharp, Spock, Pan and Elliot [1996]) and more recently for solid-state lasers (Spühler, Reffert, Haiml, Moser and Keller [2001]). Semiconductor-doped dielectric films have been demonstrated for saturable absorber applications (Bilinsky, Fujimoto, Walpole and Missaggia [1999]). In this case, InAs-doped thin-film rf-sputtering technology was used which offers similar advantages as SESAMs, i.e. the integration of the absorber into a mirror structure. However, the saturation fluence of ≈ 10 mJ/cm2 is rather high for stable solidstate laser mode locking. In comparison, epitaxially grown SESAMs typically have a saturation fluence in the range of 10 µJ/cm2 depending on the specific device structure.
4.4. Effective saturable absorbers using the Kerr effect The extremely rapid response and the broad bandwidth of the Kerr nonlinearity are very attractive for a mode-locking process. For high intensities, the polarization inside a dielectric medium does not proportionally follow the electric field anymore. This gives rise to an index change proportional to intensity. Offresonance, this nonlinear optical effect is extremely fast, with estimated response times in the few-femtosecond range. The transverse and longitudinal effects resulting from the intensity dependence are shown schematically in fig. 7. The
1, § 4]
Loss modulation
53
Fig. 7. The Kerr effect gives rise to an increase of the refractive index with intensity, causing a retardation of the most intense parts of a pulse (i.e. for n2 > 0). In its longitudinal form, the Kerr effect causes self-phase modulation and in its transverse form, a nonlinear lens is formed in the central part of the beam profiles (i.e. Kerr lens).
transverse Kerr effect retards the central and most intense part of a plane wavefront and thus acts as a focusing lens, referred to as the Kerr lens. Along the axis of propagation, the longitudinal Kerr effect retards the center of an optical pulse, producing a red shift of the leading part of the pulse, and a blue shift in the trailing part. Consequently, the longitudinal Kerr effect has been named self-phase modulation (SPM). The longitudinal Kerr effect can also be used to produce the same effect as a fast saturable absorber. To do this, the phase nonlinearity provided by the longitudinal Kerr effect has to be converted into an effective amplitude nonlinearity. The earliest mode-locking schemes based only on SPM used a coupled cavity to convert SPM into SAM. In the soliton laser (Mollenauer and Stolen [1984]), pulses compressed by SPM and anomalous dispersion in the coupled cavity are directly coupled back into the main laser cavity. This provides more gain for the center of the pulse. Pulses as short as 19 fs have been demonstrated with color center lasers (Mitschke and Mollenauer [1987]). Later, the SPM-to-SAM conversion with a coupled cavity was demonstrated even when the pulses inside the coupled cavity were broadened due to positive group velocity dispersion (Kean, Zhu, Crust, Grant, Landford and Sibbett [1989]). In this case, no compressed pulse was fed back into the main cavity. An effective SAM was obtained because SPM inside the coupled cavity generates a phase modulation on the pulse that adds constructively at the peak of the pulse in the main cavity and destructively in the wings, thus shortening the pulse duration inside the main cavity. This was also later referred to as additive pulse mode locking (APM) (Ippen, Haus and Liu [1989], Haus, Fujimoto and Ippen [1991]). Although very powerful in principle, these coupled-cavity schemes have the severe disadvantage that the auxiliary cavity has
54
Ultrafast solid-state lasers
[1, § 5
to be stabilized interferometrically. An alternative method for converting the reactive Kerr nonlinearity into an effective saturable absorber was discovered in 1991: Kerr-lens mode locking (KLM) (Spence, Kean and Sibbett [1991]). The discovery of Kerr lens mode locking has been a breakthrough in ultrashort pulse generation (Spence, Kean and Sibbett [1991]). Initially, the mode-locking mechanism was not understood and was something of a mystery. But a short time after the initial discovery it became clear that the transverse Kerr effect provides a fast saturable absorber. In KLM, the transverse Kerr effect produces a Kerr lens (fig. 7) that focuses the high intensity part of the beam more strongly than the low intensity part. Thus, combined with an intracavity aperture, the Kerr lens produces less loss for high intensity and forms an effective fast saturable absorber (Keller, ’t Hooft, Knox and Cunningham [1991], Salin, Squier and Piché [1991], Negus, Spinelli, Goldblatt and Feugnet [1991]). A similar mode-locking effect can be obtained without a hard aperture when the Kerr lens produces an increased overlap of the laser mode with the pump profile in the gain medium (Piché and Salin [1993]). The Kerr lens provides the strongest advantage for the pulsed operation when the cavity is operated close to the stability limit. Optimization guidelines for SAM produced by the Kerr lens in different cavities can be found in (Magni, Cerullo, de Silvestri and Monguzzi [1995]). Unfortunately, the transverse Kerr effect couples the mode-locking process with the laser cavity mode. In contrast, the use of only the longitudinal Kerr effect in mode locking totally decouples the mode-locking process from the laser mode. This allows optimum cavity design for scaling the laser to higher powers and to higher pulse repetition rates without being constrained by the Kerr lens. In fiber lasers a different Kerr-effect-based effective saturable absorber has been used to generate pulses as short as 38 fs (Hofer, Ober, Haberl and Fermann [1992]) – the shortest pulses generated directly from a fiber laser so far. An effective fast saturable absorber is obtained with a Kerr-induced nonlinear polarization rotation in a weakly birefringent fiber combined with a polarization dependent loss. Previously, a similar idea has been used to “clean up” high intensity pulses by reducing the low-intensity pulse pedestals (Tapié and Mourou [1992], Beaudoin, Chien, Coe, Tapié and Mourou [1992]).
§ 5. Pulse propagation in dispersive media 5.1. Dispersive pulse broadening Dispersion compensation is important in ultrashort pulse generation. The laser material and other elements inside the laser cavity normally produce positive dis-
1, § 5]
Pulse propagation in dispersive media
55
persion which broadens a transform-limited pulse. Dispersion is a linear effect and therefore is additive and can be treated in the spectral domain. A very good summary of linear pulse propagation in dispersive media is given in Chapter 9 of (Siegman [1986]). Assuming a Gaussian pulse shape, an analytical solution can be determined for the pulse broadening due to second-order dispersion (Siegman [1986]):
τp (z) 4 ln 2d2 φ/dω2 2 = 1 + , (5.1) τp (0) τp2 (0) where it is assumed that the incident pulse is transform limited, i.e. the timebandwidth product of the Gaussian pulse is τp (0)νp = 0.4413, where νp is the FWHM (full width half maximum) spectral width of the pulse intensity. Only second-order dispersion (i.e. d2 φ/dω2 , Table 5) and higher orders are broadening the pulse. The first-order dispersion gives the group delay, i.e. the delay of the peak of the pulse envelope (Table 5). It is important to note that in the linear pulse propagation regime the spectrum of the pulse remains unchanged, only the spectral content of the pulse is redistributed in time. With positive dispersion the long wavelength part of the spectrum is in the leading edge of the pulse and the short wavelength part in the trailing edge of the pulse, i.e. “red is faster than blue” d2 φ 2 (fig. 8). In the regime of strong pulse broadening, i.e. dω 2 τp (0), we can reduce eq. (5.1) to d2 φ (5.2) ωp , dω2 where ωp = 2πνp is the FWHM spectral width (in radians/second) of the pulse intensity. τp (z) ≈
Fig. 8. Dispersive pulse broadening through a material with positive dispersion for the pulse envelope A(t) and/or intensity I (t) and the electric field E(t).
56
Ultrafast solid-state lasers
[1, § 5
Table 4a Sellmeier equations for different materials. The wavelength λ is given in units of µm Material
Defining Sellmeier equation Aλ2 λ2 −λ21
Fused quartz
n2 = 1 +
+
SF10 glass
n2 = a0 + a1 λ2 +
Sapphire
n2 = 1 +
Bλ2 λ2 −λ22
+
Constants Cλ2 λ2 −λ23
a2 a a a + 34 + 46 + 58 λ2 λ λ λ
a λ2 a1 λ2 a λ2 + 22 + 23 λ2 −b1 λ −b2 λ −b3
A = 0.6961663 λ1 = 0.0684043 B = 0.4079426 λ2 = 0.1162414 C = 0.8974794 λ3 = 9.896161 a0 = 2.8784725 a1 = −0.010565453 a2 = 3.327942 × 10−2 a3 = 2.0551378 × 10−3 a4 = −1.1396226 × 10−4 a5 = 1.6340021 × 10−5 a1 = 1.023798 a2 = 1.058264 a3 = 5.280792 b1 = 0.00377588 b2 = 0.0122544 b3 = 321.3616
Material dispersion is normally described with Sellmeier equations for the refractive index as a function of the wavelength, i.e. n(λ). With the Sellmeier equations (Table 4a) all the necessary dispersive quantities can be calculated (Table 5).
5.2. Dispersion compensation The challenge in ultrashort pulse generation is dispersion compensation over a large bandwidth, that is, to compensate for the dispersive pulse broadening that occurs in the gain material and other elements inside the laser cavity. Dispersion compensation is important because, for example, a 10 fs (1 fs) Gaussian pulse at the center wavelength of 800 nm is broadened to 100 fs (1 ps) after only 1 cm of fused quartz due to second-order dispersion. This follows from eq. (5.2) for the regime of strong pulse broadening and Table 4. In addition, in femtosecond lasers the pulses are ideally soliton pulses for which a constant negative dispersion over the full spectral width of the pulse balances the chirp of the self-phase modulation. The negative dispersion required for a certain pulse duration follows from eq. (6.52). It is necessary that all higher-order dispersion terms are negligibly small.
1, § 5]
Pulse propagation in dispersive media
57
Table 4b Examples of material dispersions calculated from the Sellmeier equations given in Table 4a and the equations given in Table 5 Material
Refractive index n at a center wavelength of 800 nm
Fused quartz
n(0.8 µm) = 1.45332 ∂n 1 ∂λ 800 nm = −0.017 µm ∂2n = 0.04 1 2 ∂λ2 800 nm µm ∂3n 1 3 800 nm = −0.24 3 µm
∂λ
SF10 glass
n (0.8 µm) = 1.71125 ∂n 1 ∂λ 800 nm = −0.0496 µm 2 ∂ n = 0.176 1 2 ∂λ2 800 nm µm 1 ∂3n = −0.997 3 800 nm 3 µm
∂λ
Sapphire
n(0.8 µm) = 1.76019 ∂n 1 ∂λ 800 nm = −0.0268 µm ∂2n = 0.064 1 2 ∂λ2 800 nm µm ∂3n 1 = −0.377 3 800 nm 3 µm
∂λ
Propagation constant kn at a center wavelength of 800 nm ∂kn −9 s = 4.84 ns m m ∂ω 800 nm = 4.84 × 10 2 ∂ kn −26 s2 = 36.1 fs2 = 3.61 × 10 m mm ∂ω2 800 nm 3 ∂ 3 kn fs3 = 2.74 × 10−41 sm = 27.4 mm ∂ω3 800 nm
∂kn −9 s = 5.70 ns m m ∂ω 800 nm = 5.70 × 10 2 2 ∂ kn fs2 s −25 = 1.59 × 10 m = 159 mm ∂ω2 800 nm 3 ∂ 3 kn fs3 = 1.04 × 10−40 sm = 104 mm ∂ω3 800 nm ∂kn −9 s = 5.87 ns m m ∂ω 800 nm = 5.87 × 10 ∂ 2 kn −26 s2 = 58 fs2 = 5.80 × 10 m mm ∂ω2 800 nm
3 ∂ 3 kn fs3 = 4.21 × 10−41 sm = 42.1 mm ∂ω3 800 nm
Table 5 Dispersion quantities, their defining equations and units. kn : wavevector in the dispersive media, i.e. kn = kn = n2π /λ, where λ is the vacuum wavelength. z: a certain propagation distance. c: vacuum light velocity. ω: frequency in radians/second Quantity
Symbol
Defining equation
Defining equation using n(λ)
Phase velocity
υp
Group velocity
υg
ω kn dω dkn
Group delay
Tg
dφ Tg = υzg = dω , φ ≡ kn z
c n 1 c n 1− dn λ dλ n nz 1 − dn c dλ nz 1 − dn c dλ λ3 z d2 n 2π c2 dλ2
Dispersion: 1st order Dispersion: 2nd order Dispersion: 3rd order
dφ dω d2 φ dω2 d3 φ dω3
λ n λ n
−λ4 z 3 d2 n + λ d3 n 4π 2 c3 dλ2 dλ3
For optimum soliton formation of a sub-10 fs pulse inside the laser, only a very small amount of a constant negative total intracavity dispersion is necessary to form a stable soliton pulse (eq. (6.52)). For example, we estimated the neces-
58
Ultrafast solid-state lasers
[1, § 5
sary dispersion to be only −10 fs2 for a Ti:sapphire laser producing 6.5 fs pulses (Jung, Kärtner, Matuschek, Sutter, Morier-Genoud, Zhang, Keller, Scheuer, Tilsch and Tschudi [1997], Sutter, Jung, Kärtner, Matuschek, Morier-Genoud, Scheuer, Tilsch, Tschudi and Keller [1998]). Here we assumed an estimated self-phase modulation coefficient of about 0.07/mW, an average output power of 200 mW using a 3% output coupler and a pulse repetition rate of 86 MHz. This results in an intracavity pulse energy of 77.5 nJ. With eq. (6.52) then follows an estimated negative group delay dispersion of −10 fs2 in a cavity round trip. There are different methods for dispersion compensation, such as Gires– Tournois Interferometer (GTI) grating pairs, prism pairs and chirped mirrors. The dispersion compensation is summarized in Table 6 with the symbols defined in fig. 9.
(a)
(b)
(c) (continued on next page) Fig. 9. Dispersion compensation techniques: (a) Gires–Tournois Interferometer (GTI); (b) grating pair; (c) prism pair; (d) chirped mirror (i.e. shown here a double chirped mirror).
1, § 5]
Pulse propagation in dispersive media
59
(d) Fig. 9. (Continued)
5.2.1. Gires–Tournois Interferometer (GTI) A GTI (Gires and Tournois [1964], Kafka, Watts and Pieterse [1992]) is a very compact dispersion compensation element, which basically replaces one flat laser cavity mirror. The negative dispersion is obtained due to the Fabry–Perot structure (fig. 9a). Normally, in a GTI the bottom mirror is a 100% reflector and the top mirror is a relatively low reflector, typically with a reflectivity of a few percent. The spacer layer in the Fabry–Perot should contain a non-absorbing material and is very often air, such that the thickness can be easily changed. In Table 7 the material dispersion in the GTI spacer layer was neglected. The bandwidth compared to the other techniques is limited, thus a GTI is typically used for pulse durations above 100 fs. There is a trade-off: a broader bandwidth is obtained with a smaller Fabry–Perot thickness but then the amount of negative dispersion is strongly reduced: d2 φ ∝ d 2, dω2
bandwidth of
d2 φ 1 ∝ . dω2 d
(5.3)
60
Ultrafast solid-state lasers
[1, § 5
Table 6 Dispersion compensation, their defining equations and figures. c: light velocity in vacuum, λ: wavelength in vacuum, λ0 : center wavelength of pulse spectrum, ω: frequency in radians/second Quantity
Defining equation
Gires–Tournois Interferometer (GTI) (fig. 9a)
d: thickness of Fabry–Perot n: refractive index of material inside Fabry–Perot (airspaced n = 1). Note: material dispersion is neglected. t0 = 2nd c : roundtrip time of the Fabry–Perot Rt : intensity reflectivity of top reflector of Fabry–Perot (bottom reflector is assumed to have a 100%-reflectivity)
Dispersion: 2nd order Four-grating compressor (fig. 9b)
Dispersion: 2nd order Dispersion: 3rd order Four-prism compressor (fig. 9c)
Dispersion: 2nd order
Dispersion: 3rd order
√ −2t02 (1−Rt ) Rt sin ωt0 d2 φ √ = dω2 (1+Rt −2 Rt cos ωt0 )2
Lg : grating pair spacing Λ: grating period θi : angle of incidence at grating λ 2 −3/2 λ 3 Lg d2 φ 2 =− 2 2 1 − Λ − sin θi dω
πc Λ
λ sin θ −sin2 θ 2 1+ Λ d3 φ i i = − d φ2 6πc λ 2 λ dω3 dω 1− Λ −sin θi
n: refractive index of prisms θB : angle of incident of prism is at Brewster angle θB = arctan[n(λ0 )] α: apex angle of prism, α = π − 2θB sin θ θ2 (λ) = arcsin n(λ) sin π − 2θB − arcsin n(λ)B L: apex-to-apex prism distance h: beam insertion into second prism h cos θ2 sin β = L cos(α/2) 3 2 d2 φ = λ 2 d P2 dω2 2π c dλ d2 P = 2 ∂ 2 n ∂θ2 + ∂ 2 θ2 ∂n 2 L sin β − 2 ∂θ2 ∂n 2 L cos β ∂λ ∂n ∂λ dλ2 ∂λ2 ∂n ∂n2 ∂2n ∂n 2 ∂n 2 1 ≈ 4 2 + 2n − 3 ∂λ L sin β − 8 ∂λ L cos β ∂λ n 4 2 3 d3 φ = −λ2 3 3 d P2 + λ d P3 dω3 4π c dλ dλ
d3 P ≈ 4 d3 n L sin β − 24 dn d2 n L cos β dλ dλ2 dλ3 dλ3
For example, an air-spaced GTI with 80 µm thickness and a top reflectivity of 4% produces a negative dispersion of about −0.13 ps2 at a wavelength of 799 nm. In comparison a 2.25 µm thick air space results in about −100 fs2 at a wavelength of 870 nm.
Laser cavity element
Linearized operator
New constants
Constants
Gain
(6.10)
2 A ≈ g + Dg ∂ 2 A
Dg ≡ g2
Dg : gain dispersion (eq. (6.10)) g: saturated amplitude gain coefficient Ωg : HWHM of gain bandwidth in radians/second νg : FWHM of gain bandwidth, i.e. νg = Ωg /π
Loss modulator
(6.12)
A ≈ −Ms t 2 A
Ms ≡
∂t
Ωg
2 Mωm 2
A ≈ −lA
Constant loss
l: amplitude loss coefficient
A ≈ iψA
Constant phase shift
Ms : curvature of loss modulation (eq. (6.11)) ωm : loss modulation frequency in radians/second 2M: peak-to-peak modulation depth for amplitude loss coefficient
ψ: phase shift q
Fast saturable absorber
(6.14)
A ≈ γA |A|2 A
γA ≡ I 0 A sat,A A
γA : absorber coefficient (eq. (4.13)) q0 : maximum saturable amplitude loss coefficient Isat,A : saturation intensity AA : laser mode area in saturable absorber
Dispersion: 2nd order
(6.19)
2 A ≈ iD ∂ 2 A
D ≡ 12 kn z
D: dispersion parameter (half of the total group delay dispersion per cavity roundtrip (eq. (6.18)))
∂t
Pulse propagation in dispersive media
Eq.
1, § 5]
Table 7 Linearized operators that model the change in the pulse envelope A(t) for each element in the laser cavity and their defining equations. The pulse envelope is normalized such that |A(z, t)|2 is the pulse power P (z, t) (eq. (6.2)). kn : wavevector in the dispersive media, i.e. kn = kn = n2π/λ, where λ is the vacuum wavelength. z: the relevant propagation distance for negative dispersion or SPM, respectively. c: vacuum light velocity. ω: frequency in radians/second
2 kn = d k2n
dω
SPM
(6.25)
A ≈ −iδL |A|2 A
kn z
δL ≡ A 2 L
δL : SPM coefficient (eq. (6.22)) n2 : nonlinear refractive index AL : laser mode area inside laser material (Note: here we assume that the dominant SPM occurs in the laser material. Then z is equal to 2 times the length of the laser crystal in a standing wave cavity) 61
62
Ultrafast solid-state lasers
[1, § 5
5.2.2. Grating pairs Grating pairs (Treacy [1969]) produce negative dispersion due to the wavelengthdependent diffraction (fig. 9b). To obtain a spatially coherent beam 2 paths through the grating pair are required. Normally gratings introduce too much loss inside a solid-state laser resonator. Thus, they have only been used for external pulse compression schemes (Tomlinson, Stolen and Shank [1984]). The grating pairs alone can only be used to compensate for second-order dispersion. Higherorder dispersion limits pulse compression in the ultrashort pulse-width regime. Therefore, a combination of a grating and prism compressor was used to generate the long-standing world record of 6-fs pulses with dye lasers (Fork, Cruz, Becker and Shank [1987]). There are mainly two types of gratings: ruled or holographic. Removing material from a master substrate with a precise instrument called a ruling engine produces ruled gratings. Replicas of the ruled grating are then pressed and the pressings are coated. Holographic gratings are produced by interfering two laser beams on a substrate coated with a photoresist, which is subsequently processed to reproduce the sinusoidal interference pattern. Generally, replicas cannot be produced from holographic masters, i.e. they are more expensive. Higher damage threshold can be obtained with gold coated ruled gratings (i.e. > 500 mJ/cm2 at 1 ps). The diffraction efficiency in this case is from around 88% to 92% depending on the grating. Because 4 paths through the gratings are required for dispersion compensation and a spatially coherent beam, this amounts to at least about 30% loss. 5.2.3. Prism pairs Prism pairs (Fork, Martinez and Gordon [1984]) are well established for intracavity dispersion compensation. Negative dispersion is obtained with the wavelength dependent refraction (fig. 9c). To obtain a spatially coherent beam 2 paths through the prism pairs are required. The insertion loss is very small because the angle of incidence is the Brewster angle. The prism apex angle is chosen such that the incident beam at Brewster angle is also at minimum deviation. Prism pairs offer two advantages. First, the pulse width can be varied by simply moving one of the prisms (see fig. 9c) and second, the laser can be tuned in wavelength by simply moving a knife edge at a position where the beam is spectrally broadened. Both properties are often desired for spectroscopic applications, for example. However, the prism pair suffers from higher-order dispersion, which is the main limitation in ultrashort pulse generation in the sub-10-fs regime. Different prism materials introduce different amounts of higher-order dispersion. For compact lasers with pulse durations of few tens to hundreds of femtoseconds the more dispersive
1, § 5]
Pulse propagation in dispersive media
63
SF10-prisms are better because they require a smaller prism separation than fused quartz prism, for example. But a smaller prism separation comes at the expense of a larger higher-order dispersion. Fused quartz is one of the best materials for ultrashort pulse generation with minimal higher-order dispersion. The higher-order dispersion of the prism pairs is dominated by the prism spacing, which is not changed significantly when we adjust the dispersion by inserting the prisms into the laser beam. For example, at a fused quartz prism spacing of 40 cm, the total dispersion produced by a double pass through the prism pairs amounts to −862 fs2 and a thirdorder dispersion of −970 fs3 at a center wavelength of 800 nm, assuming zero prism insertion into the beam (Table 6). We can reduce the negative dispersion by moving the prisms into the laser beam: Each additional millimeter of prism insertion produces a positive dispersion of 101 fs2 but only a third-order dispersion of 78 fs3 per cavity round trip. Thus, the prism pairs can generally only be used to compensate for either second or third-order dispersion. With fused quartz prisms alone pulses as short as 10 fs can be produced with a Ti:sapphire laser (Asaki, Huang, Garvey, Zhou, Kapteyn and Murnane [1993], Fluck, Jung, Zhang, Kärtner and Keller [1996]). Slightly shorter pulses can be achieved if the Ti:sapphire laser is operated at a center wavelength of approximately 850 nm, where nearzero second and third-order dispersion can be obtained for a Ti:sapphire crystal of about 2 mm thickness. In this case 8.5 fs have been generated (Zhou, Taft, Huang, Murnane, Kapteyn and Christov [1994]). Another dispersion compensation is then required for shorter pulse generation. Therefore, the chirped mirrors were designed to show the inverse higher-order group delay dispersion of the dispersion of the prism pair plus laser crystal to eliminate the higher-order dispersion and obtain a slightly negative but constant group delay dispersion required for an ideal soliton pulse. It is interesting to note that under special cavity design a single prism can also be used for dispersion compensation (Kopf, Spühler, Weingarten and Keller [1996]). 5.2.4. Chirped mirror Chirped mirrors (Szipöcs, Ferencz, Spielmann and Krausz [1994]) have been established in different set-ups for the generation of ultrashort laser pulses. Figure 9d shows a typical chirped mirror structure which schematically shows the path of a long-wavelength (i.e. 1000 nm) and short-wavelength (i.e. 650 nm) beam. This results in negative dispersion, because the long wavelengths are made to penetrate deeper into the mirror structure than short wavelengths. Figure 9d also shows the standing wave electric field patterns in a chirped mirror structure
64
Ultrafast solid-state lasers
[1, § 5
versus wavelength. The negative dispersion of the mirrors is clearly illustrated by the dependence of penetration depth on wavelength for the range of 650 to 1050 nm. The highly transmissive region around 500 nm is used for pumping the Ti:sapphire laser through these chirped mirrors. According to (Szipöcs, Ferencz, Spielmann and Krausz [1994]), chirping means that the Bragg wavelength λB is gradually increased along the mirror, producing a negative group delay dispersion (GDD). However, no analytical explanation of the unwanted oscillations typically observed in the group delay and GDD of such a simple-chirped mirror was given. These oscillations were mostly minimized purely by computer optimization. An exact coupled-mode analysis (Matuschek, Kärtner and Keller [1997]) was used to develop a double-chirp technique. The strong periodic variation in the group delay of the original chirped mirrors occurs due to impedance mismatch between the incident medium (i.e. typically air) and the mirror stack and also within the mirror stack (Kärtner, Matuschek, Schibli, Keller, Haus, Heine, Morf, Scheuer, Tilsch and Tschudi [1997], Matuschek, Kärtner and Keller [1999]). Using the accurate analytical expressions for phase, group delay, and GDD (Matuschek, Kärtner and Keller [1999]), double-chirped mirrors (DCM) could be designed and fabricated with a smooth and custom tailored GDD suitable for generating pulses in the two-cycle regime directly from a Ti:sapphire laser (Sutter, Steinmeyer, Gallmann, Matuschek, Morier-Genoud, Keller, Scheuer, Angelow and Tschudi [1999], Morgner, Kärtner, Cho, Chen, Haus, Fujimoto, Ippen, Scheuer, Angelow and Tschudi [1999]). A double chirped mirror (DCM) (Kärtner, Matuschek, Schibli, Keller, Haus, Heine, Morf, Scheuer, Tilsch and Tschudi [1997], Matuschek, Kärtner and Keller [1999]) is a multilayer interference coating that can be considered as a composition of at least two sections, each with a different task. The layer materials are typically SiO2 and TiO2 . The first section is the AR coating, typically composed of 10 to 14 layers. It is necessary because the theory is derived assuming an ideal matching to air. The other section represents the actual DCM structure, as derived from theory. The double-chirp section is responsible for the elimination of the oscillations in the GDD. Double-chirping means that in addition to the local Bragg wavelength λB the local coupling of the incident wave to the reflected wave is independently chirped as well. The local coupling is adjusted by slowly increasing the high-index layer thickness in every pair so that the total optical thickness remains λB /2. This corresponds to adiabatic matching of the impedance. The AR-coating, together with the rest of the mirror, is used as a starting design for a numerical optimization program. Since theoretical design is close to the desired design goal, a local optimization using a standard gradient algorithm is sufficient. At this point, only the broadband AR-coating sets a limitation on the reduction of
1, § 6]
Mode-locking techniques
65
the GDD oscillations. An AR coating with a residual reflectivity of less than 10−4 is required for a DCM at a center wavelength of around 800 nm, which results in a bandwidth of only 250 nm. This bandwidth limitation cannot be removed with more layers in the mirror structure. To overcome this AR-coating limitation, we demonstrated a new technique for the design of chirped mirrors with extremely smooth dispersion characteristics over an extended ultrabroadband wavelength range: back-side coated (BASIC) chirped mirrors (Matuschek, Gallmann, Sutter, Steinmeyer and Keller [2000]). Dispersion oscillations are significantly reduced by coating the chirped mirror structure on the back side of a substrate, providing ideal impedance matching between coating and ambient medium. The chirped mirror structure and the AR coating are non-interfering due to slightly wedged substrates or substrates with differently curved surfaces (e.g., focusing mirror) for example. Thus, chirped mirror and AR-coating multilayer structures and can be independently designed and optimized. The separation of both coating sections provides a much better solution for the impedance-matching problems than previous approaches to chirped mirror design. The AR-coating may be added on the front side of the substrate, geometrically separated from the chirped mirror, to simply reduce the intracavity losses. Thus, the AR coating does not need to be as good and we can easily obtain a much broader bandwidth. We showed by a theoretical analysis and numerical simulations that minimum dispersion oscillations are achieved if the index of the substrate is identical to the index of one of the coating materials and if double chirping is used for the chirped mirror structure. Based on this analysis, we designed a back-side coated double-chirped mirror (BASIC DCM) that supports a bandwidth of 220 THz with group delay dispersion oscillations of about 2 fs2 (rms), an order-of magnitude improvement compared to previous designs of similar bandwidth.
§ 6. Mode-locking techniques 6.1. Overview Passive mode-locking mechanisms are well-explained by three fundamental models: slow saturable absorber mode locking with dynamic gain saturation (New [1972, 1974]) (fig. 3a), fast saturable absorber mode locking (Haus [1975a], Haus, Fujimoto and Ippen [1992]) (fig. 3b), and slow saturable absorber mode locking without dynamic gain saturation which in the femtosecond regime is described by soliton mode locking (Kärtner and Keller [1995], Jung, Kärtner, Brovelli, Kamp
66
Ultrafast solid-state lasers
[1, § 6
and Keller [1995], Kärtner, Jung and Keller [1996]) and in the picosecond regime by Paschotta and Keller [2001] (fig. 3c). The physics of most of these techniques can be well explained with Haus’s master equation formalism as long as at steady state the changes in the pulse envelope during the propagation inside the cavity are small. At steady state the pulse envelope has to be unchanged after one roundtrip through the cavity. Passive mode locking, however, can only be analytically modelled in the weak saturation regime, which is typically not the case in SESAM mode-locked solidstate lasers. However, this formalism still provides a useful approach to describe mode-locking techniques in a unified fashion. Recent numerical simulations show that analytical results with fast saturable absorbers only slightly underestimate numerical solutions and correctly describe the dependence on saturated gain, gain bandwidth and absorber modulation, taking into account more strongly saturated absorbers and somewhat longer saturation recovery times in SESAM modelocked solid-state lasers (Paschotta and Keller [2001]). A short introduction to Haus’s formalism is given in Section 6.2 and Table 7. Afterwards we will describe all mode-locking techniques using this formalism and summarize the theoretical prediction for pulse shape and pulse duration (Table 9). For solid-state lasers self-Q-switching instabilities in passive mode locking is a serious challenge. Simple guidelines to prevent those instabilities and obtain stable cw mode locking are presented in Section 6.8.
6.2. Haus’s master equations Haus’s master equation formalism (Haus [1995], Haus [2000]) is based on linearized differential operators that describe the temporal evolution of a pulse envelope inside the laser cavity. At steady-state we then obtain the differential equation: ∂A(T , t) TR (6.1) = Ai = 0, ∂T i
where A is the pulse envelope, TR is the cavity round-trip time, T is the time that develops on a time scale of the order of TR , t is the fast time of the order of the pulse duration, and Ai are the changes of the pulse envelope due to different elements in the cavity, such as gain, loss modulator or saturable absorber, dispersion, etc. (Table 7). Equation (6.1) basically means that at steady state after one laser roundtrip the pulse envelope cannot change and all the small changes due to the different elements in the cavity have to sum up to zero. Each element is modeled as a linearized operator, which will be discussed in more detail below.
1, § 6]
Mode-locking techniques
67
The pulse envelope is normalized such that |A(z, t)|2 is the pulse power P (z, t): 2 E(z, t) ∝ A(z, t)ei[ω0 t −kn (ω0 )z] , with A(z, t) ≡ P (z, t), (6.2) where E(z, t) is the electric field, ω0 the center frequency in radians/second of the pulse spectrum and kn = nk with k = 2π/λ the wave number with λ the vacuum wavelength and n the refractive index. Before we discuss the different mode-locking models we briefly discuss the linearized operators for the differential equations. Gain. A homogeneously broadened gain medium is described by a Lorentzian lineshape for which the frequency dependent gain coefficient g(ω) is given by
g ω2 g(ω) = (6.3) , for (ω − ω0 /Ωg )2 1, ≈g 1− 0 2 Ωg2 1 + ω−ω Ωg where ω = ω − ω0 , g is the saturated gain coefficient for a cavity round trip and Ωg is the HWHM (half width half maximum) of the gain bandwidth in radians/seconds. In the frequency domain the pulse envelope after the gain medium is given by A˜ out (ω) = eg(ω) A˜ in (ω) ≈ 1 + g(ω) A˜ in(ω), for g 1, (6.4) ˜ where A(ω) is the Fourier transformation of A(t). Equations (6.3) and (6.4) then give g A˜ out (ω) = 1 + g − 2 ω2 A˜ in (ω) ⇒ Ωg (6.5) g ∂2 Aout (t) = 1 + g + 2 2 Ain (t), Ωg ∂t where we used the fact that a factor of ω in the frequency domain produces a time derivative in the time domain. For example, for the electric field we obtain: ∂ ∂ 1 1 iωt iωt (6.6) E(t) = dω = dω E(ω)e E(ω)iωe ∂t ∂t 2π 2π and
∂2 1 1 ∂2 iωt 2 iωt E(t) = 2 ]e dω E(ω)e dω = E(ω)[−ω 2π ∂t 2 ∂t 2π
(6.7)
˜ ˜ 0 + ω) and similarly for the pulse envelope. Note that A(ω) ≡ A(ω) = E(ω ∂ ∂ 1 1 iωt iωt ˜ ˜ (6.8) A(t) = dω = dω A(ω)e A(ω)iωe ∂t ∂t 2π 2π
68
Ultrafast solid-state lasers
[1, § 6
and
∂2 ∂2 1 1 iωt 2 iωt ˜ ˜ A(t) = dω = ]e dω. A(ω)e A(ω)[−ω 2π ∂t 2 ∂t 2 2π (6.9) For the change in the pulse envelope A = Aout − Ain after the gain medium we then obtain: ∂2 g A ≈ g + Dg 2 A, Dg ≡ 2 , (6.10) ∂t Ωg
where Dg is the gain dispersion. Loss modulator. A loss modulator inside a laser cavity is typically an acoustooptic modulator and produces a sinusoidal loss modulation given by a time dependent loss coefficient: 2 Mωm (6.11) , 2 where Ms is the curvature of the loss modulation, 2M is the peak-to-peak modulation depth and ωm the modulation frequency which corresponds to the axial mode spacing in fundamental mode locking. In fundamental mode locking we only have one pulse per cavity round trip. The change in the pulse envelope is then given by Aout (t) = e−l(t ) Ain (t) ≈ 1 − l(t) Ain (t) ⇒ A ≈ −Ms t 2 A. (6.12)
l(t) = M(1 − cos ωm t) ≈ Ms t 2 ,
Ms ≡
Fast saturable absorber. In case of an ideal fast saturable absorber we assume that the loss recovers instantaneously and therefore shows the same time dependence as the pulse envelope (eqs. (4.12) and (4.13)): q0 q0 q(t) = (6.13) ≈ q0 − γA P (t), γA ≡ . 1 + I (t)/Isat,A Isat,A AA The change in the pulse envelope is then given by Aout (t) = e−q(t ) Ain (t) ≈ 1 − q(t) Ain (t) ⇒
A ≈ γA |A|2A.
(6.14)
Group velocity dispersion (GVD). The wave number kn (ω) in a dispersive material depends on the frequency and can be approximately written as: 1 kn (ω) ≈ kn (ω0 ) + kn ω + kn ω2 + · · · , 2 2
(6.15)
∂ kn n where ω = ω −ω0 , kn = ∂k ∂ω |ω=ω0 and kn = ∂ω2 |ω=ω0 . In the frequency domain the pulse envelope in a dispersive medium after a propagation distance of z is
1, § 6]
Mode-locking techniques
69
given by ˜ ω) = e−i[kn (ω)−kn (ω0 )]z A(0, ˜ ω) ≈ 1 − i kn (ω) − kn (ω0 ) z A(0, ˜ ω), A(z, (6.16) where we used the slowly varying envelope approximation (which is applicable for pulse durations of more than 10 fs in the near infrared wavelength regime). Taking into account only the first- and second-order dispersion terms we then obtain: 1 2 ˜ ω). ˜ A(z, ω) ≈ 1 − ikn ωz − i kn ω z A(0, (6.17) 2 The linear term in ω determines the propagation velocity of the pulse envelope (i.e. the group velocity νg ) and the quadratic term in ω determines how the pulse envelope gets deformed due to second-order dispersion. The influence of higher-order dispersion can be considered with more terms in the expansion of kn (ω) (eq. (6.15)). However, higher-order dispersion only becomes important for ultrashort pulse generation with pulse durations below approximately 30 fs depending how much material is inside the cavity. Normally we are only interested in changes of the pulse envelope and therefore it is useful to restrict our observation to a reference system that is moving with the pulse envelope. In this reference system we only need to consider second- and higher-order dispersion. In the time domain we then obtain for second-order dispersion: 1 1 d2 φ ∂2 , A(z, t) ≈ 1 + iD 2 A(0, t), D ≡ kn z = (6.18) 2 2 dω2 ∂t where D is the dispersion parameter which is half of the total group delay dispersion per cavity roundtrip. Therefore we obtain for the change in the pulse envelope: A ≈ iD
∂2 A. ∂t 2
(6.19)
Self-phase modulation. The Kerr effect introduces a space and time dependent refractive index: n(r, t) = n + n2 I (r, t),
(6.20)
where n is the linear refractive index, n2 the nonlinear refractive index and I (r, t) the intensity of the laser beam, typically a Gaussian beam profile. For laser host materials, n2 is typically of the order of 10−16 cm2 /W and does not change very much for different materials. For example, for sapphire n2 =
70
Ultrafast solid-state lasers
[1, § 6
3 × 10−16 cm2 /W, fused quartz n2 = 2.46 × 10−16 cm2 /W, Schott glass LG760 n2 = 2.9 × 10−16 cm2 /W, YAG n2 = 6.2 × 10−16 cm2 /W, and YLF n2 = 1.72 × 10−16 cm2 /W. The nonlinear refractive index produces a nonlinear phase shift during pulse propagation: 2 φ(z, r, t) = −kn(r, t)z = −k n + n2 I (r, t) z = −knz − δL A(r, t) , (6.21) where δL is the self-phase modulation coefficient (SPM coefficient): δL ≡ kn2 z/AL ,
(6.22)
where AL is the laser mode area inside the laser medium. Here we assume that the dominant SPM inside the laser occurs in the gain medium. In this case, z is equal to twice the laser material length. Of course, the mode area can also be very small in other materials. In this case, we will have to add up all the SPM contributions inside the laser resonator. The laser mode area AL is an “averaged value” in case the mode is changing within the gain medium. The electric field during propagation is changing due to SPM: E(z, t) = eiφ E(0, t) ∝ e−iδL |A(t )| A(0, t)eiω0 t −ikn (ω0 )z . 2
(6.23)
For δL |A|2 1 we obtain:
2 2 A(z, t) = e−iδL |A(t )| A(0, t)e−ikn (ω0 )z ≈ 1 − iδL A(t) A(0, t)e−ikn (ω0 )z . (6.24)
After one cavity roundtrip we then obtain A ≈ −iδL |A|2 A.
(6.25)
6.3. Active mode locking Short pulses from a laser can be generated with a loss or phase modulator inside the resonator. For example, the laser beam is amplitude-modulated when it passes through an acousto-optic modulator. Such a modulator can modulate the loss of the resonator at a period equal to the round-trip time TR of the resonator (i.e. fundamental mode locking). The pulse evolution in an actively mode-locked laser without self-phase modulation (SPM) and group-velocity dispersion (GVD) can be described by the master equation of Haus (Haus [1975b]). Taking into account gain dispersion and loss modulation we obtain with eqs. (6.10) and (6.12) (Table 7) the following differential equation:
1 ∂2 ωm t 2 (6.26) Ai = g 1 + 2 2 − l − M A(T , t) = 0. Ωg ∂t 2 i
1, § 6]
Mode-locking techniques
71
Typically, we obtain pulses which are much shorter than the roundtrip time in the cavity and which are placed in time at the position where the modulator introduces the least amount of loss. Therefore, we were able to approximate the cosine modulation by a parabola (eq. (6.11)). The only stable solution to this differential equation is a Gaussian pulse shape with a pulse duration: τp = 1.66 × 4 Dg /Ms , (6.27) where Dg is the gain dispersion (eq. (6.10) and Table 7) and Ms is the curvature of the loss modulation (eq. (6.11) and Table 7). Therefore, in active mode locking the pulse duration becomes shorter until the pulse shortening of the loss modulator balances the pulse broadening of the gain filter. Basically, the curvature of the gain is given by the gain dispersion Dg and the curvature of the loss modulation is given by Ms . The pulse duration is only scaling with the fourth root of √ √ the saturated gain (i.e. τp ∝ 4 g ) and the modulation depth (i.e. τp ∝ 4 1/M ) √ and with the square root of the modulation frequency (i.e. τp ∝ 1/ωm ) and the gain bandwidth (i.e. τp ∝ 1/ωg ). A higher modulation frequency or a higher modulation depth increases the curvature of the loss modulation and a larger gain bandwidth decreases gain dispersion. Therefore, we obtain shorter pulse durations in all cases. At steady-state the saturated gain is equal to the total cavity losses, therefore a larger output coupler will result in longer pulses. Thus, higher average output power is generally obtained at the expense of longer pulses (Table 2). The results obtained in actively mode-locked flashlamp pumped Nd:YAG and Nd:YLF lasers can be very well explained by this result. For example, with Nd:YAG at a lasing wavelength of 1.064 µm we have a gain bandwidth of λg = 0.45 nm. With a modulation frequency of 100 MHz (i.e. ωm = 2π · 100 MHz), a 10% output coupler (i.e. 2g ≈ Tout = 0.1) and a modulation depth M = 0.2, we obtain a FWHM pulse duration of 93 ps (eq. (6.27)). For example, with Nd:YLF at a lasing wavelength of 1.047 µm and a gain bandwidth of λg = 1.3 nm we obtain with the same mode-locking parameters a pulse duration of 53 ps (eq. (6.27)). The same result for the pulse duration (eq. (6.27)) has been previously derived by Kuizenga and Siegman [1970a, 1970b] where they assumed a Gaussian pulse shape and then followed the pulse once around the laser cavity, through the gain and the modulator. They then obtain a self-consistent solution when no net change occurs in the complete round trip. The advantage of Haus’s theory is that no prior assumption has to be made for the pulse shape. His theory then predicts a Gaussian pulse shape for actively mode-locked lasers, which then in principle justifies Kuizenga and Siegman’s assumption. Equation (6.27) shows that increasing the modulation frequency is an effective method to shorten the pulses. Harmonic mode locking is a technique in which the
72
Ultrafast solid-state lasers
[1, § 6
cavity is modulated at a frequency that is some integer multiple of the fundamental pulse repetition rate with a period given by the cavity round trip time. This technique was first introduced and analyzed by Becker, Kuizenga and Siegman [1972] in a phase mode-locked Nd:YAG laser. Second harmonic mode locking of a flashlamp pumped Nd:YAG laser at 1.06 µm resulted in less than 50 ps (Johnson and Simpson [1983], Johnson and Simpson [1985]) and at 1.32 µm in 53 ps (Keller, Valdmanis, Nuss and Johnson [1988]). It has been well known for quite some time that the addition of a nonlinear index medium to a passively (Martinez, Fork and Gordon [1984], Martinez, Fork and Gordon [1985]) or actively (Haus and Silberberg [1986]) mode-locked laser system can lead to shorter pulses. The bandwidth limitation that results from gain dispersion can be partially overcome by the spectral broadening caused by the nonlinearity. We can extend the differential equation (eq. (6.26)) with the additional terms for SPM (eq. (6.25) and Table 7):
1 ∂2 ωm t 2 − iδL |A|2 + iψ A(T , t) = 0. Ai = g 1 + 2 2 − l − M Ωg ∂t 2 i (6.28) In this case, however, we have to include an additional phase shift ψ to obtain a self-consistent solution. So far, we always assumed ψ = 0. This phase shift is an additional degree of freedom because the boundary condition for intracavity pulses only requires that the pulse envelope is unchanged after one cavity round trip. The electric field, however, can have an arbitrary phase shift ψ after one round trip. This is normally the case because the phase and the group velocity are not the same. It is important to note, that for ultrashort pulses in the one to two cycle regime this becomes important and stabilization of the electric field with respect to the peak of the pulse envelope is required (Telle, Steinmeyer, Dunlop, Stenger, Sutter and Keller [1999]). We can obtain an analytic solution of eq. (6.28) if we assume a parabolic approximation for |A|2 :
t2 2 2 |A|2 = |A0 |2 e−t /τ ≈ P0 1 − 2 . (6.29) τ The solution of eq. (6.28) is then a chirped Gaussian pulse 1 t2 A(t) = A0 exp − 2 (1 − ix) 2τ
(6.30)
with the chirp parameter x=
τ 2 φnl , 2Dg
(6.31)
1, § 6]
Mode-locking techniques
and the FWHM pulse duration τp = 1.66 · τ = 1.66 ·
4
Dg
2 /4D Ms + φnl g
,
73
(6.32)
where φnl is the nonlinear phase shift per cavity roundtrip at peak intensity. Typically, the beam diameter is very small in the laser gain material, thus the dominant SPM contribution comes from propagation through the gain material: φnl = 2kLg n2 I0,L
(6.33)
and Lg is the length of the laser gain material (assuming standing wave cavity) and I0,L the peak intensity inside the SPM medium, i.e. the laser gain medium. This analytical result can explain the much shorter 7–12 ps pulses in actively mode-locked diode-pumped Nd:YAG (Maker and Ferguson [1989a]) and Nd:YLF (Maker and Ferguson [1989b], Keller, Li, Khuri-Yakub, Bloom, Weingarten and Gerstenberger [1990], Weingarten, Shannon, Wallace and Keller [1990], Juhasz, Lai and Pessot [1990]) lasers because the laser mode area in those diodepumped lasers is very small which results in significant SPM pulse shortening (eq. (6.32)). For example, our experiments with an actively mode-locked diodepumped Nd:YLF laser (Braun, Weingarten, Kärtner and Keller [1995]) are very well explained with eq. (6.32). In this case the lasing wavelength is 1.047 µm, the gain bandwidth is λg = 1.3 nm, the pulse repetition rate is 250 MHz, the output coupler is 2.5%, the average output power is 620 mW, the mode radius inside the 5 mm long Nd:YLF crystal is 127 µm × 87 µm and the loss modulation of the acousto-optic mode locker is about 20%. We then obtain a FWHM pulse duration of 17.8 ps (eq. (6.32)) which agrees well with the experimentally observed pulse duration of 17 ps. Without SPM we would predict a pulse duration of 33 ps (eq. (6.27)). Equation (6.32) would predict that more SPM continues to reduce the pulse duration. However, too much SPM will ultimately drive the laser unstable. This has been shown by the numerical simulations of Haus and Silberberg [1986] which predict that pulse shortening in an actively mode-locked system is limited by roughly a factor of 2 in the case of SPM only. They also showed that the addition of negative GVD can undo the chirp introduced by SPM, and therefore both effects together may lead to stable pulse shortening by a factor of 2.5. However, experimental results with fiber lasers and solid-state lasers indicate that soliton shaping in the negative GVD regime can lead to pulse stabilization and considerable more pulse shortening. We have extended the analysis of Haus and Silberberg by investigating the possible reduction in pulse width of an actively mode-locked laser as a result of soliton-like pulse formation, i.e. the pres-
74
Ultrafast solid-state lasers
[1, § 6
ence of SPM and an excessive amount of negative GVD (Kärtner, Kopf and Keller [1995]). We show, by means of soliton perturbation theory, that beyond a critical amount of negative GVD a soliton-like pulse is formed and kept stable by an active mode locker. If the bandwidth of the gain is large enough, the width of this solitary pulse can be much less than the width of a Gaussian pulse generated by the active mode locker and gain dispersion alone. We established analytically that the pulse shortening possible by addition of SPM and GVD does not have a firm limit of 2.5. Numerical simulations and experiments with a regeneratively actively mode-locked Nd:glass laser (Kopf, Kärtner, Weingarten and Keller [1994]) confirm these analytical results. The pulse-width reduction achievable depends on the amount of negative GVD available. For an actively mode-locked Nd:glass laser a pulse shortening up to a factor of 6 may result, until instabilities arise. 6.4. Passive mode locking with a slow saturable absorber and dynamic gain saturation Dynamic gain saturation can strongly support pulse formation in passive mode locking and has allowed pulses with a duration much shorter than the absorber recovery time. Dynamic gain saturation means that the gain undergoes a fast pulseinduced saturation that recovers between consecutive pulses. This technique has been used to produce sub-100-fs pulses with dye lasers and dye saturable absorbers even though the absorber recovery time was in the nanosecond regime. Dynamic gain saturation can only help if the following conditions are fulfilled (fig. 3a): (i) The loss needs to be larger than the gain before the pulse: q0 > g0 , (6.34) where q0 is the unsaturated loss coefficient (eq. (4.1)) and g0 is the small signal gain coefficient in the laser (Weingarten, Braun and Keller [1994]). (ii) The absorber needs to saturate faster than the gain. From eq. (4.9) (i.e. slow saturable absorber and gain) then follows that AA AL Esat,A < Esat,L ⇔ (6.35) < , σA σL where Esat,A and Esat,L are the saturation energy of the absorber and the gain; σA and σL the absorber and gain cross section; and AA and AL the laser mode area in the absorber and gain. (iii) The absorber has to recover faster than the gain: τA < τL ,
(6.36)
where τA is the absorber recovery time and τL the upper state lifetime of the gain medium (Table 1).
1, § 6]
Mode-locking techniques
75
Table 8 Femtosecond Rhodamine 6G dye laser with its saturable absorber DODCI at a wavelength of 620 nm Rhodamine 6G DODCI Photoisomer
σL = 1.36 × 10−16 cm2 σA = 0.52 × 10−16 cm2 σ˜ A = 1.08 × 10−16 cm2
τL = 4 ns τA = 2.2 ns τ˜L = 1 ns
Passively mode-locked dye lasers are based on this mode-locking technique and a more extensive review of ultrashort pulse generation with dye lasers is given in Shank [1988] and Diels [1990]. The most important dye laser for sub-100-fs pulse generation is the colliding pulse mode-locked (CPM) laser (Fork, Greene and Shank [1981]). The design considerations of such a laser are very well described in Valdmanis and Fork [1986]. This laser is based on Rhodamine 6G as the gain medium and on DODCI as the absorber (Ippen, Shank and Dienes [1972]). From Table 8 follows that the conditions (eqs. (6.34)–(6.36)) are only fulfilled if the mode area in the absorber jet is smaller than in the gain jet. In addition, a sixmirror ring cavity design where the absorber and the gain jet are separated by 1/4 of the resonator round trip gives the two counter propagating pulses the same gain and the absorber is more strongly saturated because of the two pulses colliding inside the saturable absorber jet (i.e. colliding pulse mode locking – CPM) (Fork, Greene and Shank [1981]). This effectively shortens the pulses and increases the stability. The best performance of this laser was only obtained after both the gain and the absorber dyes have been freshly prepared, because both photodegrade within a relatively short time (i.e. 1–3 weeks). The best result was 27 fs pulses at a center wavelength of 620 nm with an average output power of 20 mW each in two output beams at a pulse repetition rate of 100 MHz (Valdmanis, Fork and Gordon [1985], Valdmanis and Fork [1986]). More typically, this laser produced slightly below 100 fs pulses. This laser was the “work horse” for all the pioneering work on sub-100-fs spectroscopy for nearly 10 years in the 1980s. However, the rather large maintenance of this laser explains the success of the Ti:sapphire laser at the beginning of the 1990s. Today, solid-state lasers with much shorter pulses, higher output powers and better stability have replaced practically all ultrafast dye laser systems. Another important application of this mode-locking technique is the semiconductor laser, which also has an upper state lifetime in the nanosecond regime and a large gain cross section in the range of 10−14 cm2 . More extensive reviews of ultrashort pulse generation with semiconductor lasers have been provided by (Jiang and Bowers [1995]) and (Vail’ev [1995]). The CPM technique, as developed for dye lasers, has also been used for passively mode-locked semiconductor
76
Ultrafast solid-state lasers
[1, § 6
lasers (van der Ziel, Logan and Mikulyak [1981], Vasil’ev, Morzov, Popov and Sergeev [1986]). Wu, Chen, Tanbun-Ek, Logan, Chin and Raybon [1990] and Chen and Wu [1992] monolithically incorporated the CPM technique in quantum well lasers. Pulses of 0.64 ps at a repetition rate of 710 MHz were generated (Chen, Wu, Tanbun-Ek, Logan and Chin [1991]). Compared with other modelocked solid-state lasers and mode-locked dye lasers, semiconductor lasers still have low output power (milli-watt regime) and long pulse width (a few hundreds of femtoseconds). The master equation for this mode-locking mechanism (fig. 3a) is given by (Haus [1975c], Ippen [1994]): d g0 d2 A(T , t) = 0. Ai = g(t) − q(t) + 2 2 + tD (6.37) Ωg dt dt i
For a self-consistent solution we will have to include a time shift tD of the pulse envelope due to the saturation of the absorber. g(t) and q(t) are given by the slow saturation approximation (eq. (4.9)). For an analytical solution we have to expand the exponential function up to the second order, i.e. ex ≈ 1 + x + x 2 /2: t 2 σA A(t ) dt q(t) = q0 exp − AA hν −∞ ≈ q0
σA 1− AA hν
t −∞
2 A(t ) dt +
σA2 2(AA hν)2
t −∞
2 A(t ) dt
2 ,
(6.38) and analogous for g(t). In this case we can obtain an analytical solution with a sech2 -pulse shape:
t A(t) = A0 sech (6.39) , τ and a FWHM pulse duration τp ≈ 1.76 ×
4 1 , π νg
(6.40)
where νg is the FWHM gain bandwidth of the laser. In eq. (6.40) the conditions (eqs. (6.34)–(6.36)) are assumed and in addition Esat,L Esat,A and Ep Esat,A (i.e. a fully saturated absorber). For the example of Rhodamine 6G and DODCI (Table 8) we then obtain for a gain bandwidth of νg ≈ 4 × 1013 Hz a pulse duration of about 56 fs (eq. (6.40)). In contrast, pulses as short as 27 fs have been demonstrated (Valdmanis, Fork
1, § 6]
Mode-locking techniques
77
and Gordon [1985]). However, it was recognized early on that SPM together with negative dispersion results in soliton formation and further reduces pulse duration by about a factor of 2 in dye lasers (Martinez, Fork and Gordon [1984, 1985]). This would explain the difference in the theoretical prediction of eq. (6.40) from the experimentally demonstrated 27 fs. However, at that time an analytic solution for the pulse shortening effect was not presented. For semiconductor lasers we typically observe strongly chirped pulses (Jiang and Bowers [1995]). Therefore, we would have to include dispersion and selfphase modulation in the rate equation. This is not so easy because we would also have to include the refractive index change that occurs during gain saturation.
6.5. Passive mode locking with a fast saturable absorber In passive mode locking the loss modulation is obtained by self-amplitude modulation (SAM), where the pulse saturates an absorber. In the ideal case, the SAM follows the intensity profile of the pulse. This is the case of an ideal fast saturable absorber. In this case, SAM produces a much larger curvature of loss modulation than in the sinusoidal loss modulation of active mode locking, because the modelocked pulse duration is much shorter than the cavity roundtrip time. Therefore, we would expect from the previous discussion of active mode locking, that we obtain much shorter pulses with passive mode locking. This is indeed observed. In the fast saturable absorber model no dynamic gain saturation is required and the short net-gain window is formed by a fast recovering saturable absorber alone (fig. 3b). This was initially believed to be the only stable approach to passively mode-locked solid-state lasers with long upper state lifetimes. Additive pulse mode locking (APM) was the first fast saturable absorber for such solidstate lasers (Section 4.4). However, APM required interferometric cavity length stabilization. Kerr lens mode locking (KLM) (Spence, Kean and Sibbett [1991]) (Section 4.4) was the first useful demonstration of an intracavity fast saturable absorber for a solid-state laser and because of its simplicity replaced coupled cavity mode-locking techniques. KLM is very close to an ideal fast saturable absorber, where the modulation depth is produced either by the decreased losses because of self-focusing through a hard aperture (Keller, ’t Hooft, Knox and Cunningham [1991], Salin, Squier and Piché [1991], Negus, Spinelli, Goldblatt and Feugnet [1991]) or by increased gain in the laser as a result of an increased overlap of the laser mode with the pump mode in the laser crystal (Piché and Salin [1993]). Only in the ultrashort pulse regime of a few optical cycles more complicated space– time coupling occurs and wavelength dependent effects start to limit further pulse reduction.
78
Ultrafast solid-state lasers
[1, § 6
Besides the tremendous success of KLM, there are some significant limitations for practical or “real-world” ultrafast lasers. First, the cavity is typically operated near one end of its stability range, where the Kerr-lens-induced change of the beam diameter is large enough to sustain mode locking. This results in a requirement for critical cavity alignment where mirrors and the laser crystal have to be positioned to an accuracy of several hundred microns. Once the cavity is correctly aligned, KLM can be very stable and under certain conditions even selfstarting. However, self-starting KLM lasers in the sub-50-fs regime have not yet been demonstrated without any additional starting mechanisms. This is not surprising, since in a 10 fs Ti:sapphire laser with a 100 MHz repetition rate, the peak power changes by 6 orders of magnitude when the laser switches from cw to pulsed operation. Therefore, nonlinear effects that are still effective in the sub10 fs regime are typically too small to initiate mode locking in the cw-operation regime. In contrast, if self-starting is optimized, KLM tends to saturate in the ultrashort pulse regime or the large self-phase modulation (SPM) will drive the laser unstable. For an ideal fast saturable absorber, Haus, Fujimoto and Ippen [1992] developed an analytic solution for the pulse width starting with the following master equation:
1 ∂2 Ai = g 1 + 2 2 − l + γA |A|2 A(T , t) = 0, (6.41) Ωg ∂t i
not taking into account GVD and SPM. The solution is an unchirped sech2 -pulse shape
t P (t) = P0 sech2 (6.42) τ with a FWHM pulse width: τp = 1.7627 · τ = 1.7627
4Dg , γ A Ep
(6.43)
where P0 is the peak power of the pulse, Ep is the intracavity pulse energy and Dg is the gain dispersion of the laser medium (eq. (6.10)). The shortest possible pulses can be obtained when we use the full modulation depth of the fast saturable absorber. We only obtain an analytic solution if we assume an ideal fast absorber that saturates linearly with the pulse intensity (eq. (4.13)) over the full modulation depth. This is clearly a strong approximation because eq. (4.13) only holds for weak absorber saturation. For a maximum modulation depth and this linear approximation we then can assume that γA P0 = q0 .
1, § 6]
Mode-locking techniques
For a sech2 -shaped pulse (eq. (6.42)), the pulse energy is given by Ep = P (t) dt = 2τ P0 .
79
(6.44)
The minimal pulse width for a fully saturated ideal fast absorber then follows from eq. (6.43): 1.7627 2g . τp,min = (6.45) Ωg q0 This occurs right at the stability limit when the filter loss due to gain dispersion is equal to the residual loss a soliton undergoes in an ideal fast saturable absorber (eq. (4.14)): Dg q0 = qs = residual saturable absorber loss. = (6.46) 3 3τ 2 The residual saturable absorber loss qs results from the fact that the soliton pulse initially experiences loss to fully saturate the absorber (see Section 4.2.2). This residual loss is exactly q0 /3 for a sech2 -pulse shape and a fully saturated ideal fast saturable absorber. This condition results in a minimal FWHM pulse duration given by eq. (6.45). Including GVD and SPM, i.e. soliton formation, in the fast saturable absorber model, an additional pulse shortening of a factor of 2 was predicted. However, unchirped soliton pulses (i.e. ideal sech2 -shaped pulses) are only obtained for a certain negative dispersion value given by Filter loss =
|D| Dg = . δL γA
(6.47)
This is also where we obtain the shortest pulses with a fast saturable absorber. Here we assume that higher-order dispersion is fully compensated or negligibly small. In addition, computer simulations show that too much self-phase modulation drives the laser unstable. KLM is well described by the fast absorber mode-locking model discussed above even though it is not so easy to determine the exact saturable absorber parameters such as the effective saturation fluence. However, the linearized model does not describe the pulse generation with Ti:sapphire lasers in the sub-10fs regime very well. Pulse-shaping processes in these lasers are more complex (Brabec, Spielmann and Krausz [1991], Krausz, Fermann, Brabec, Curley, Hofer, Ober, Spielmann, Wintner and Schmidt [1992]). Under the influence of the different linear and nonlinear pulse shaping mechanisms, the pulse is significantly broadened and recompressed, giving rise to a “breathing” of the pulse width. The
80
Ultrafast solid-state lasers
[1, § 6
order of the pulse shaping elements in the laser cavity becomes relevant and the spectrum of the mode-locked pulses becomes more complex. In this case, an analytical solution can no longer be obtained. As a rough approximation, the pulses still behave like solitons and consequently these lasers are also called solitary lasers (Brabec, Spielmann and Krausz [1991]).
6.6. Passive mode locking with a slow saturable absorber without gain saturation and soliton formation Over many years we consistently observed in experiments that even without soliton effects the pulse duration can be much shorter than the absorber recovery time in SESAM mode-locked solid-state lasers. It has always been postulated that without soliton pulse shaping, we need to have a fast saturable absorber for stable mode locking, which is in disagreement with our experimental observations. More recently, we therefore performed some more detailed numerical investigations (Paschotta and Keller [2001]). For a strongly saturated slow absorber with a saturation parameter S > 3 and an absorber recovery time smaller than 10 to 30 times the pulse duration, we found a useful guideline for the predicted pulse duration g 1.5 τp,min ≈ (6.48) . νg R We neglected, similar to the ideal fast saturable absorber mode-locking model, the effects of Kerr nonlinearity and dispersion in the cavity, phase changes on the absorber and spatial hole burning in the gain medium. Compared to the analytical solution of a fully saturated absorber (eq. (6.45)) we would predict a slightly longer pulse duration given by a factor of about 1.3. Otherwise, the dependence with regards to gain saturation, gain bandwidth and absorber modulation depth has been explained very well with the analytical solution. Numerical simulations show that the pulse duration in eq. (6.48) can be significantly shorter than the absorber recovery time and has little influence on the pulse duration as long as τA < 10τp to 30τp . Numerical simulations show that this is a reasonable estimate and that with too long recovery time, the pulse does not simply become longer but unstable. At first this long recovery time might be surprising, because on the trailing edge of the pulse there is no shaping action of the absorber. There is even net gain, because the loss caused by the absorber is very small for the trailing edge (always assuming a fully saturated absorber), while the total loss for the pulse is larger and is balanced by the saturated gain in steady-state. Thus, one might expect that this
1, § 6]
Mode-locking techniques
81
net gain after the pulse would destabilize the pulse – but this is not the case. The reason is that the pulses experience a temporal shift by the absorber, which limits the time in which noise behind the pulse can be amplified. The absorber attenuates mostly the leading wing of the pulse, thus shifting the pulse center backwards in each cavity round-trip. This means that the pulse is constantly moving backward and swallows any noise growing behind itself. An upper limit on the recovery time then follows from the condition that this noise cannot grow too much. Note that weak reflections in the laser cavity could generate weak satellite pulses behind the main pulse. These satellite pulses could be stronger than the noise level and thus significantly reduce maximum tolerable recovery time of the absorber. So far we have not included any additional pulse shaping effects such as SPM or solitons. Further simulations show that SPM alone in the positive dispersion regime should always be kept small because it always makes pulses longer and even destabilizes them, particularly for absorbers with small modulation depth (Paschotta and Keller [2001]). This result might be surprising at first – but again the temporal delay caused by the absorber gives a simple explanation. SPM (with positive n2 , as is the usual case) decreases the instantaneous frequency in the leading wing and increases the frequency in the trailing wing. The absorber always attenuates the leading wing, thus removing the lower frequency components, which results in an increased center frequency and broader pulses due to the decrease in pulse bandwidth. For strong SPM the pulses become unstable. A rule of thumb is that the nonlinear phase shift for the peak should be at most a few mrad per 1% of modulation depth. It is clear that SPM could hardly be made weak enough in the sub-picosecond regime. For this reason, soliton mode locking is usually required in the sub-picosecond domain, because there the nonlinear phase changes can be much larger.
6.7. Soliton mode locking In soliton mode locking, the pulse shaping is done solely by soliton formation, i.e. the balance of group velocity dispersion (GVD) and self-phase modulation (SPM) at steady state, with no additional requirements on the cavity stability regime as, for example, for KLM. In contrast to KLM we use only the time-dependent part of the Kerr effect at the peak intensity (i.e. n(t) = n + n2 I0 (t), eq. (6.20)) and not the radially dependent part as well (i.e. n(r, t) = n + n2 I (r, t), eq. (6.20)). The radially dependent part of the Kerr effect is responsible for KLM because it forms the nonlinear lens that reduces the beam diameter at an intracavity aperture inside the gain medium. Thus, this nonlinear lens forms an effective fast saturable
82
Ultrafast solid-state lasers
[1, § 6
absorber because the intensity dependent beam diameter reduction at an aperture introduces less loss at high intensity and more loss at low intensity. Such a radially dependent effective saturable absorber however couples the mode-locking mechanism with the cavity mode. In contrast, soliton mode locking does not depend on the transverse Kerr effect and has therefore the advantage that the mode-locking mechanism is decoupled from the cavity design and no critical cavity stability regime is required – it basically works over the full cavity stability range. In soliton mode locking, an additional loss mechanism, such as a saturable absorber (Kärtner and Keller [1995]), or an acousto-optic modulator (Kärtner, Kopf and Keller [1995]), is necessary to start the mode-locking process and to stabilize the soliton pulse forming process. In soliton mode locking, we have shown that the net-gain window (fig. 3c) can remain open for significantly more than 10 times the width of the ultrashort pulse, depending on the specific laser parameters (Jung, Kärtner, Brovelli, Kamp and Keller [1995], Kärtner, Jung and Keller [1996]). This strongly relaxed the requirements on the saturable absorber and we can obtain ultrashort pulses even in the 10 fs regime with semiconductor saturable absorbers that have much longer recovery times. With the same modulation depth, one can obtain almost the same minimal pulse duration as with a fast saturable absorber, as long as the absorber recovery time is roughly less than ten times the final pulse width. In addition, high dynamic range autocorrelation measurements showed excellent pulse pedestal suppression over more than seven orders of magnitude in 130 fs pulses of a Nd:glass laser (Kopf, Kärtner, Keller and Weingarten [1995]) and very similar to or even better than KLM pulses in the 10-fs pulse-width regime (Jung, Kärtner, Henkmann, Zhang and Keller [1997]). Even better performance can be expected if the saturable absorber also shows a negative refractive index change coupled with the absorption change as is the case for semiconductor materials (Kärtner, Jung and Keller [1996]). With a slow saturable absorber as a starting and stabilizing mechanism for soliton mode locking, there remains a time window with net round-trip gain behind the pulse, where the loss of the still saturated absorber is smaller than the total loss for the pulse that is balancing the saturated gain. At first glance it may seem that the discussion in the last section about slow saturable absorbers would apply here as well. However, there is another limiting effect which usually becomes more effective in soliton mode-locked lasers: the dispersion causes the background to temporally broaden and thus permanently loosing the energy in those parts which drift into the time regions with net loss. We can describe soliton mode locking by the Haus’s master equation formalism, where we take into account GVD, SPM and a slow saturable absorber q(T , t) that recovers slower than the pulse duration (see fig. 3c) (Kärtner and Keller
1, § 6]
Mode-locking techniques
83
[1995], Kärtner, Jung and Keller [1996]): 2 ∂2 Ai = −iD 2 + iδL A(T , t) A(T , t) ∂t i
∂2 + g − l + Dg 2 − q(T , t) A(T , t) = 0. ∂t
(6.49)
This differential equation can be solved analytically using the soliton perturbation theory. The first bracket term of this differential equation determines the nonlinear Schrödinger equation for which the soliton pulse is a stable solution for negative GVD (i.e. D < 0) and positive SPM (i.e. n2 > 0):
t −iφs (z) e A(z, t) = A0 sech (6.50) . τ This soliton pulse propagates without distortion through a medium with negative GVD and positive SPM because the effect of SPM exactly cancels that due to dispersion. The FWHM soliton pulse duration is given by τp = 1.7627 · τ and the time-bandwidth product by νp τp = 0.3148. φs (z) is the phase shift of the soliton during propagation along the z-axis 1 1 φs (z) = kn2 I0 z ≡ φnl (z), (6.51) 2 2 where I0 is the peak intensity inside the SPM medium (i.e. typically the gain medium). Thus, the soliton pulse experiences during propagation a constant phase shift over the full beam profile. For a given negative dispersion and an intracavity pulse energy Ep we obtain a pulse duration τp = 1.7627
2|D| δ L Ep
(6.52)
for which the effects of SPM and GVD are balanced with a stable soliton pulse. This soliton pulse loses energy due to gain dispersion and losses in the cavity. Gain dispersion and losses can be treated as perturbation to the nonlinear Schrödinger equation for which a soliton is a stable solution (i.e. second bracket term in eq. (6.50)). This lost energy, called continuum in soliton perturbation theory, is initially contained in a low intensity background pulse, which experiences negligible SPM, but spreads in time due to GDD (fig. 10a). In soliton mode locking a stable soliton pulse is formed for all group delay dispersion values as long as the continuum loss is larger than the soliton loss (eq. (6.53)) or the pulses break up in two or more pulses (Aus der Au, Kopf, Morier-Genoud, Moser and Keller [1997]). Thus, for smaller negative dispersion the pulse duration becomes shorter
84
Ultrafast solid-state lasers
[1, § 6
(a)
(b) Fig. 10. Soliton modelocking in (a) time and (b) frequency domain. The continuum pulse spreads in time due to group velocity dispersion and thus undergoes more loss in the relatively slow absorber, which is saturated by the shorter soliton pulse. However, the longer continuum pulse has a narrower spectrum and thus undergoes more gain than the spectrally broader soliton pulse.
(eq. (6.52)) until minimal pulse duration is reached. With no pulse-break-up, the minimal pulse duration is given when the loss for the continuum pulse becomes equal to the loss of the soliton pulse. When the soliton pulse is stable, then the saturated gain is equal to the loss:
Dg Ep Esat,A 1 − exp − , g = l + ls , with ls = 2 + q0 (6.53) 3τ Ep Esat,A where l is the total saturated amplitude loss coefficient per cavity roundtrip (i.e. output coupler, residual cavity losses and nonsaturable absorber loss of the saturable absorber) and ls the additional loss experienced by the soliton as a result of gain filtering (eq. (6.46)) and the amplitude loss coefficient for saturation of the slow absorber (eq. (4.10)). Soliton perturbation theory then determines the roundtrip loss of the continuum pulse (Kärtner, Jung and Keller [1996]). The continuum is spread in time because of dispersion and therefore experiences enhanced
1, § 6]
Mode-locking techniques
85
loss in the recovering absorber that has been saturated by the much shorter soliton pulse. Solitons alone in the cavity are not stable. The continuum pulse is much longer and therefore experiences only the gain at line center, while the soliton exhibits an effectively lower average gain due to its larger bandwidth. Thus, the continuum exhibits a higher gain than the soliton. After a sufficient build-up time, the continuum would actually grow until it reached lasing threshold, destabilizing the soliton. However, we can stabilize the soliton by introducing a relatively slow saturable absorber into the cavity. This absorber is fast enough to add sufficient additional loss for the growing continuum that spreads in time during its build-up phase so that it no longer reaches lasing threshold. We then predict a minimum pulse duration for a soliton pulse (Kärtner and Keller [1995], Kärtner, Jung and Keller [1996]): 3/4
3/2 1/4 1 −1/8 τA g φs τp,min = 1.7627 √ q0 6Ωg
1 3/4 τA 1/4 g 3/8 ≈ 0.45 , (6.54) 1/8 νg R φs where φs is the phase shift of the soliton per cavity round trip (assuming that the dominant SPM occurs in the laser gain medium), νg the FWHM gain bandwidth and R ≈ 2q0 (eq. (4.1)). With eq. (6.51) then follows φs = φs (z = 2Lg ) = kn2 Lg I0,L = 12 δL P0 , where δL is the SPM coefficient (eq. (6.25)) and P0 the peak power inside the laser cavity. In eq. (6.54) we assume a fully saturated slow saturable absorber and a linear approximation for the exponential decay of the slow saturable absorber. The analytical solution for soliton mode locking (eq. (6.54)) has been experimentally confirmed with a Ti:sapphire laser where a Fabry–Perot filter has been inserted to give a well-defined gain bandwidth (Jung, Kärtner, Brovelli, Kamp and Keller [1995]). However, this equation still does not tell us what soliton phase shift would be ideal. Equation (6.54) would suggest that very high values are preferred, which actually leads to instabilities. Also this equation is not taking into account that the soliton pulse may break up into two solitons which occurs more easily if the absorber is too strongly saturated. Numerical simulations can give better estimates for these open questions (Paschotta and Keller [2001]). In soliton mode locking, the dominant pulse formation process is assumed to be soliton formation. Therefore, the pulse has to be a soliton for which the negative GVD is balanced with the SPM inside the laser cavity. The pulse duration is then given by the simple soliton solution (eq. (6.52)). This means that the pulse duration scales linearly with the negative group delay dispersion inside the laser cavity
86
Ultrafast solid-state lasers
[1, § 6
(i.e. τp ∝ |D|). In the case of an ideal fast saturable absorber, an unchirped soliton pulse is only obtained at a very specific dispersion setting (eq. (6.47)), whereas for soliton mode locking an unchirped transform-limited soliton is obtained for all dispersion levels as long as the stability requirement against the continuum is fulfilled. This fact has been also used to experimentally confirm that soliton mode locking is the dominant pulse formation process and not a fast saturable absorber such as KLM (Aus der Au, Kopf, Morier-Genoud, Moser and Keller [1997]). Higher-order dispersion only increases the pulse duration, therefore is undesirable and is assumed to be compensated. The break-up into two or even three pulses can be explained as follows: Beyond a certain pulse energy, two soliton pulses with lower power, longer duration, and narrower spectrum will be preferred, since their loss introduced by the limited gain bandwidth decreases so much that the slightly increased residual loss of the less saturated saturable absorber cannot compensate for it. This results in a lower total roundtrip loss and thus a reduced saturated or average gain for two pulses compared to one pulse. The threshold for multiple pulsing is lower for shorter pulses, i.e. with spectra which are broad compared to the gain bandwidth of the laser. A more detailed description of multiple pulsing is given elsewhere (Kärtner, Aus der Au and Keller [1998]). Numerical simulations show, however, that the tendency for pulse break-up in cases with strong absorber saturation is found to be significantly weaker than expected from simple gain/loss arguments (Paschotta and Keller [2001]). To conclude, soliton shaping effects can allow for the generation of significantly shorter pulses, compared to cases without SPM and dispersion. The improvement is particularly large for absorbers with a relatively low modulation depth and when the absorber recovery is not too slow. In this regime, the pulse shaping is mainly done by the soliton effects, and the absorber is only needed to stabilize the solitons against growth of the continuum. The absorber parameters are generally not very critical. It is important not only to adjust the ratio of dispersion and SPM to obtain the desired soliton pulse duration (eq. (6.52)), but also to keep their absolute values in a reasonable range where the nonlinear phase change is in the order of a few hundred mrad per round-trip (i.e. significantly larger than acceptable in cases without negative dispersion). Soliton formation is generally very important in femtosecond lasers, which has been already recognized in colliding pulse-mode-locked dye lasers. However, no analytic solution was presented for soliton pulse shortening. It was always assumed that for a stable solution the mode-locking mechanism without soliton effects has to generate a net gain window as short as the pulse (fig. 3a and b). In contrast to these cases, in soliton mode locking we present an analytic solution based on soliton pertur-
1, § 6]
Mode-locking techniques
87
bation theory, where soliton pulse shaping is clearly assumed to be the dominant pulse formation process, and the saturable absorber required for a stable solution is treated as a perturbation. Then, the net gain window can be much longer than the pulse (fig. 3c). Stability of the soliton against the continuum then determines the shortest possible pulse duration. This is a fundamentally different mode-locking model than previously described. We therefore refer to it as soliton mode locking, emphasizing the fact that soliton pulse shaping is the dominant factor.
6.8. Design guidelines to prevent Q-switching instability For picosecond solid-state lasers the self-amplitude modulation of a saturable absorber with a picosecond or tens of picoseconds recovery time is sufficient for stable pulse generation. A picosecond recovery time can be achieved with low-temperature grown semiconductor saturable absorbers where mid-gap defect states form very efficient traps for the photoexcited electrons in the conduction band (Haiml, Siegner, Morier-Genoud, Keller, Luysberg, Lutz, Specht and Weber [1999]). In the picosecond regime, we developed a very simple stability criteria for stable passive mode locking without Q-switching instabilities (Hönninger, Paschotta, Morier-Genoud, Moser and Keller [1999]): 2 Ep2 > Ep,c = Esat,L Esat,A R.
(6.55)
The critical intracavity pulse energy Ep,c is the minimum intracavity pulse energy, which is required to obtain stable cw mode locking, that is, for Ep > Ep,c we obtain stable cw mode locking and for Ep < Ep,c we obtain Q-switched mode locking. For good stability of a mode-locked laser against unwanted fluctuations of pulse energy, operation close to the stability limit is not recommended. Thus, a large modulation depth supports shorter pulses (Table 9), but an upper limit is given by the onset of self-Q-switching instabilities (eq. (6.55)). In the femtosecond regime, we observe a significant reduction of the tendency of Q-switching instabilities compared to pure saturable absorber mode-locked picosecond lasers (eq. (6.55). This can be explained as follows: if the energy of an ultrashort pulse rises slightly due to relaxation oscillations, SPM and/or SAM broadens the pulse spectrum. A broader pulse spectrum, however, increases the loss due to the finite gain bandwidth (eq. (6.46)), which provides some negative feedback, thus decreasing the critical pulse energy which is necessary for stable cw mode locking. The simple stability requirement of eq. (6.55) then has to
88
Ultrafast solid-state lasers
[1, § 6
Table 9 Predicted pulse duration for the different mode-locking (ML) techniques. The parameters used here are summarized and defined in Table 7. The pulse is transform-limited for proper dispersion compensation and is either an unchirped Gaussian or soliton pulse ML technique
Eq.
Active ML: Amplitude loss modulation
(6.27) Gauss
τp = 1.66 ×
Passive ML: Slow saturable absorber and dynamic gain saturation
(6.40) Soliton
1 τp ≈ 1.76 × π4 ν g
Slow saturable absorber for solid-state lasers and strongly saturated absorbers (S > 3)
1.5 (6.48) Numerical τp,min ≈ ν g simulations
Passive ML: Fast saturable absorber (FSA)
(6.43) Soliton
Fully saturated ideal fast saturable absorber
(6.45)
only transform-limited soliton pulses for a well-defined intracavity group delay dispersion (assuming negligible higher order dispersion): |D|/δL = Dg /γA Dg 2g g |D| 1.12 τp,min = 1.76 q ≈ νg R , for δ = γ Ωg
Passive ML: Soliton mode locking
(6.52) Soliton
τp = 1.76 δ2|D| Ep
(6.54)
Pulse shape Pulse duration (FWHM) 4 Dg Ms
= 1.66 ×
4 2g
M
1 ωm Ω g
g R , for τA 30τp
4D
τp = 1.76 γ Egp A
0
L
A
L
transform-limited soliton pulses for all intracavity group delay dispersion (assuming negligible higher order dispersion) 3/4 −1/8 τA g3/2 1/4 φs τp,min = 1.7627 √ 1 q0 6Ωg 1 3/4 τA 1/4 g3/8 ≈ 0.45 ν 1/8 , R g
φs
be modified as follows (Hönninger, Paschotta, Morier-Genoud, Moser and Keller [1999]): Esat,L gK 2 Ep2 + Ep2 > Esat,L Esat,A R,
(6.56)
where K is given by K≡
0.315 2πn2 Lg . 1.76 DAL λ0 νg
(6.57)
1, § 6]
Mode-locking techniques
89
Here we assume a standing wave cavity and that the dominant SPM is produced in the laser medium. In other cases we have to add all other contributions as well.
6.9. External pulse compression SPM generates extra frequency components. The superposition with the other frequency components in the pulse can spectrally broaden the pulse (i.e. for positively chirped pulses assuming n2 > 0). SPM alone does not modify the pulse width, but with proper dispersion compensation a much shorter pulse can be generated with the extra bandwidth (Mollenauer, Stolen and Gordon [1980]). The positively chirped spectrally broadened pulse then can be compressed with appropriate negative dispersion compensation, where the “blue” spectral components have to be advanced relative to the “red” ones. A careful balance between a nonlinear spectral broadening process and negative dispersion is needed for efficient compression of a pulse. Typically, self-phase modulation in a single-mode fiber with optimized length was used to linearly chirp the pulse, which is then compressed with a grating pair compressor (Tomlinson, Stolen and Shank [1984]). Ultimately, uncompensated higher-order dispersion and higher-order nonlinearities limit compression schemes. For pulses shorter than 100 fs, compression is typically limited to factors of less than 10. Compression of amplified CPM dye laser pulses with 50 fs duration produced the long-standing world record of 6 fs (Fork, Cruz, Becker and Shank [1987]). Similar concepts have been recently used for external pulse compression of 13-fs pulses from a cavity dumped Ti:sapphire laser (Baltuska, Wei, Pshenichnikov, Wiersma and Szipöcs [1997]) and of 20-fs pulses from a Ti:sapphire laser amplifier (Nisoli, Stagira, Silvestri, Svelto, Sartania, Cheng, Lenzner, Spielmann and Krausz [1997]) resulting, in both cases, in approximately 4.5-fs pulses. In the latter case, the use of a noble-gas-filled hollow fiber resulted in pulse energies of about 0.5 mJ with 5.2 fs pulses and a peak power of 0.1 TW (Sartania, Cheng, Lenzner, Tempea, Spielmann, Krausz and Ferencz [1997]). More recently, adaptive pulse compression of a super-continuum produced in two cascaded hollow fibers generated 3.8 fs pulses with energies of up to 15 µJ at 1 kHz (Schenkel, Biegert, Keller, Vozzi, Nisoli, Sansone, Stagira, De Silvestri and Svelto [2003]). The pulse duration was fully characterized by spectral-phase interferometry for direct electric-field reconstruction (SPIDER). This is currently the world record in the visible and near-infrared spectral regime. The extra bandwidth obtained with SPM can be extremely large producing a white-light continuum (Fork, Shank, Hirlimann, Yen and Tomlinson [1983]), which can be used as a seed for broadband parametric amplification. Parametric processes can provide amplification with an even broader bandwidth than
90
Ultrafast solid-state lasers
[1, § 7
can typically be achieved in laser amplifiers. Noncollinear phase-matching at a crossing angle of 3.8◦ in Barium beta borate (BBO) provides more than 150 THz amplification bandwidth (Gale, Cavallari, Driscoll and Hache [1995]). With this type of set-up, parametric amplification has been successfully demonstrated with pulse durations of less than 5 fs (Shirakawa, Sakane, Takasaka and Kobayashi [1999], Zavelani-Rossi, Polli, Cerullo, De Silvestri, Gallmann, Steinmeyer and Keller [2002]). Much higher average power in compressed pulses can be achieved in combination with a passively mode-locked Yb:YAG thin-disk laser (Innerhofer, Südmeyer, Brunner, Häring, Aschwanden, Paschotta, Hönninger, Kumkar and Keller [2003]). Pulse compression after this laser resulted in 33-fs pulses with an average power of 18 W at a pulse repetition rate of 34 MHz. This gives a peak power of 12 MW (Südmeyer, Brunner, Innerhofer, Paschotta, Furusawa, Baggett, Monro, Richardson and Keller [2003]).
§ 7. Pulse characterization 7.1. Electronic techniques Electronic techniques for pulse-width measurements are limited to the picosecond regime. Today, photo detectors and sampling heads with bandwidths up to 60 GHz are commercially available. This means that the measured pulse duration is limited to about 7 ps. This follows from simple linear system analysis for which the impulse response of a photo detector or a sampling head can be approximated by a Gauss-function. The impulse response for a given system bandwidth B has a FWHM τFWHM in units of ps: τFWHM [ps] ≈
312 GHz . B[GHz]
(7.1)
The impulse response of a measurement system can be determined from the impulse response of each element in the detection chain: 2 τFWHM (7.2) = τ12 + τ22 + τ32 + · · ·, where, for example, τ1 is the FWHM of the impulse response of the photo detector, τ2 of the sampling head and so on. Thus, with a 40 GHz sampling head and a 60 GHz photo detector we only measure a impulse response with FWHM of 9.4 ps.
1, § 7]
Pulse characterization
91
7.2. Optical autocorrelation Optical autocorrelation techniques with second harmonic generation of two identical pulses that are delayed with respect to each other are typically used to measure shorter pulses (Sala, Kenney-Wallace and Hall [1980], Weiner [1983]). We distinguish between collinear and non-collinear intensity autocorrelation measurements for which the two pulse beams in the nonlinear crystal are either collinear or non-collinear. In the non-collinear case the second harmonic signal only depends on the pulse overlap and is given by I2ω (t) ∝ I (t)I (t − t) dt. (7.3) The FWHM of the autocorrelation signal I2ω (t) is given by τAu and determines the FWHM pulse duration τp of the incoming pulse I (t). However, τAu depends on the specific pulse shape (Table 10) and any phase information is lost in this measurement. So normally, transform-limited pulses are assumed. This assumption is only justified as long as the measured spectrum also agrees with the assumption of pulse shape and constant phase (i.e. the time bandwidth product corresponds to a transform-limited pulse, Table 10). For passively mode-locked lasers, for example, that allow for parabolic approximation of pulse formation mechanisms one expects a sech2 temporal and spectral pulse shape (Section 6). Passively mode-locked solid-state lasers with pulse durations well above 10 fs normally generate pulses close to this ideal sech2 -shape. Therefore, this noncollinear autocorrelation technique is a good standard diagnostic for such laser sources. For ultrashort pulse generation in the sub-10-fs regime this is generally not the case anymore. Experimentally, this is clearly indicated by more complex pulse spectra. Interferometric autocorrelation (IAC) techniques (Mindl, Hefferle, Schneider and Dörr [1983]) have been used to get more information. In IAC a collinear inTable 10 Optical pulses: defining equations for Gaussian and soliton pulse shapes, FWHM (full width half maximum) intensity pulse duration τp , time-bandwidth products νp τp for which νp is the FWHM of the spectral intensity, FWHM intensity autocorrelation pulse duration τAu τp /τ
νp τp
τp /τAu
2 I (t) ∝ exp t 2
√ 2 ln 2
0.4413
0.7071
Soliton I (t) ∝ sech2 τt
1.7627
0.3148
0.6482
Pulse shape Gauss
τ
92
Ultrafast solid-state lasers
[1, § 7
tensity autocorrelation is fringe resolved and gives some indication how well the pulse is transform-limited. However, we still do not obtain full phase information about the pulse. The temporal parameters have usually been obtained by fitting an analytical pulse shape with constant phase to the autocorrelation measurement. Theoretical models of the pulse formation process motivate the particular fitting function. For lasers obeying such a model the a priori assumption of a theoretically predicted pulse shape is well-motivated and leads to good estimates of the pulse duration as long as the measured spectrum also agrees with the theoretical prediction. However, we have seen that fits to an IAC trace with a more complex pulse spectrum tend to underestimate the true duration of the pulses.
7.3. FROG and SPIDER For a more precise measurement, a variety of methods have been proposed to fully reconstruct both pulse amplitude and phase from measured data only (Chilla and Martinez [1991], Kane and Trebino [1993], Chu, Heritage, Grant, Liu, Dienes, White and Sullivan [1995], Rhee, Sosnowski, Tien and Norris [1996], Iaconis and Walmsley [1998]). Of these methods, especially frequency-resolved optical gating (FROG, Kane and Trebino [1993] and Trebino, DeLong, Fittinghoff, Sweetser, Krumbügel and Richman [1997]) and spectral phase interferometry for direct electric-field reconstruction (SPIDER, Iaconis and Walmsley [1998] and Iaconis and Walmsley [1999]) have found widespread use. FROG is a characterization method based on the measurement of a spectrally resolved autocorrelation signal followed by an iterative phase-retrieval algorithm to extract the intensity and phase of the laser pulse. SPIDER is a self-referencing variant of spectral interferometry. Conventional spectral interferometry measures the spectral phase differences between two pulses. To access the spectral phase of a single pulse, SPIDER generates a spectral shear between the carrier frequencies of two replicas of this pulse. The spectral shear is generated by upconversion of the two replicas with a strongly chirped pulse using sum-frequency generation in a nonlinear optical crystal. The phase information of the resulting interferogram allows the direct reconstruction of the spectral phase of the input pulse. In the sub-10-femtosecond regime non-collinear FROG (Baltuska, Pshenichnikov and Wiersma [1998], Cheng, Fürbach, Sartania, Lenzner, Spielmann and Krausz [1999], Durfee, Backus, Kapteyn and Murnane [1999]), collinear FROG (Gallmann, Steinmeyer, Sutter, Matuschek and Keller [2000]), and SPIDER (Gallmann, Sutter, Matuschek, Steinmeyer, Keller, Iaconis and Walmsley [1999]) have been used. The applicability of a particular measure-
1, § 8]
Carrier envelope offset (CEO)
93
ment technique in this pulse duration regime is not solely determined by possible bandwidth limitations but also by the requirements on the nonlinear optical process or the beam geometry involved (Gallmann, Sutter, Matuschek, Steinmeyer and Keller [2000]). SPIDER offers more bandwidth than any other technique and still gives valid results even if the beam profile is not spatially coherent anymore (Gallmann, Steinmeyer, Sutter, Rupp, Iaconis, Walmsley and Keller [2001]). Thus, we can measure pulses even in the single-cycle regime (Schenkel, Biegert, Keller, Vozzi, Nisoli, Sansone, Stagira, De Silvestri and Svelto [2003]). SPIDER even allows for kHz single shot characterization (Kornelis, Biegert, Tisch, Nisoli, Sansone, De Silvestri and Keller [2003]). Only fully characterized pulses in phase and amplitude will provide reliable information about pulse shape and pulse duration – this becomes even more important for pulses with very broad and complex spectra. Any other technique, such as fitting attempts to IAC traces, is very erroneous and generally underestimates the pulse duration.
§ 8. Carrier envelope offset (CEO) Progress in ultrashort pulse generation (Steinmeyer, Sutter, Gallmann, Matuschek and Keller [1999]) has reached a level where the slowly varying envelope approximation starts to fail. Pulse durations of Ti:sapphire laser oscillators have reached around 5 fs which is so short that only about two optical cycles of the underlying electric field fit into the full-width half maximum of the pulse envelope. More recently, even shorter pulses of 3.8 fs have been achieved with adaptive pulse compression of a cascaded hollow fiber supercontinuum (Schenkel, Biegert, Keller, Vozzi, Nisoli, Sansone, Stagira, De Silvestri and Svelto [2003]). These are currently the shortest pulses generated in the visible and near infrared spectral regime. For such short pulses the maximum electric field strength depends strongly on the exact position of the electric field with regards to the pulse envelope (fig. 11), i.e. the carrier envelope offset (CEO) (Telle, Steinmeyer, Dunlop, Stenger, Sutter and Keller [1999]). In passively mode-locked lasers this carrier envelope offset is a freely varying parameter because the steady-state boundary condition only requires that the pulse envelope is the same after one round-trip. Therefore the CEO phase may exhibit large fluctuations, even when all other laser parameters are stabilized. We have discussed the physical origin of these fluctuations before (Helbing, Steinmeyer, Keller, Windeler, Stenger and Telle [2001], Helbing, Steinmeyer, Stenger, Telle and Keller [2002]). Because nonlinear laser– matter interaction depends strongly on the strength of the electric field, these
94
Ultrafast solid-state lasers
[1, § 8
Fig. 11. Electric field of a few-cycle pulse for different values of the carrier-envelope-offset (CEO) phase of 0 (solid line) and π/2 (dashed line).
CEO fluctuations cause strong signal fluctuations in nonlinear experiments such as high harmonic generation (Brabec and Krausz [2000]), attosecond pulse generation (Drescher, Hentschel, Kienberger, Tempea, Spielmann, Reider, Corkum and Krausz [2001]), photoelectron emission (Paulus, Grasborn, Walther, Villoresi, Nisoli, Stagira, Priori and Silvestri [2001]), etc. Different techniques have been proposed to stabilize these CEO fluctuations in the time domain (Xu, Spielmann, Poppe, Brabec, Krausz and Haensch [1996]) and in the frequency domain (Telle, Steinmeyer, Dunlop, Stenger, Sutter and Keller [1999]). The frequency domain technique is much more sensitive and is the technique that is being used today. Using the f -to-2f heterodyne technique with additional external spectral broadening (Jones, Diddams, Ranka, Stentz, Windeler, Hall and Cundiff [2000], Apolonski, Poppe, Tempea, Spielmann, Udem, Holzwarth, Hänsch and Krausz [2000]) we achieved a long-term CEO stabilization with residual 10 attosecond timing jitter which corresponds to 0.025 rad rms CEO phase noise in a (0.01 Hz–100 kHz) bandwidth (Helbing, Steinmeyer, Stenger, Telle and Keller [2002]). The f -to-2f interference technique requires an octave-spanning spectrum. So far, all attempts to generate this spectrum directly from a laser source have only reached unsatisfactory control of the CEO frequency, which was mainly caused by poor signal strength. Therefore most experiments required additional spectral broadening, e.g., in an external microstructure fiber. The continuum generation process with its strong nonlinearity, however, introduces additional CEO noise. It is important to note that CEO stabilization is achieved for the pulses after the microstructure fiber, which means that the pulses directly from the Ti:sapphire laser will exhibit excess CEO phase noise even with a perfectly working CEO stabilization. In our experiments we observed
1, § 9]
Outlook
95
typical phase drift values of about 1.2 rad/mW oscillator power fluctuations (Helbing, Steinmeyer and Keller [2003]). The relative phase between the carrier and the envelope of an optical pulse is the key parameter linking the fields of precision frequency metrology and ultrafast laser physics (Udem, Holzwarth and Hänsch [2002]). As we have discussed, the spectrum of a mode-locked laser consists of a comb of precisely equally spaced frequencies. The uniformity of this frequency comb has been demonstrated to a relative uncertainty below 10−15 (Udem, Reichert, Holzwarth and Hänsch [1999]). Knowledge of only two parameters, the comb spacing and a common offset frequency of all modes, provides one with a set of reference frequencies, similar to the tick marks on a ruler. The beat (difference-frequency) signal of this unknown frequency with the closest frequency in the ruler always gives a beat signal smaller than half the comb period. Thus, an optical frequency in the 100 THz regime can be translated down to a microwave frequency in the range of 100 MHz, which can then be measured very accurately. This can be used for an all-optical atomic clock that is expected to outperform today’s state-of-the-art caesium clocks (Udem, Holzwarth and Hänsch [2002]). The mode-locked laser provides a phase-locked clockwork mediating between radio frequencies and the terahertz frequencies of the lines in the optical comb, effectively rendering optical frequencies countable. Details on precision frequency measurements with mode-locked laser combs can be found in Reichert, Holzwarth, Udem and Hänsch [1999], Holzwarth, Zimmermann, Udem and Hänsch [2001], Cundiff, Ye and Hall [2001] and Bauch and Telle [2002].
§ 9. Outlook We have shown that the technology of ultrafast lasers has become very refined and is now suitable for application in many areas. Points of particular importance in this respect are: • The transition from dye lasers to solid-state lasers, which can be compact, powerful, efficient and long-lived. It has been shown that solid-state lasers can generate pulses which are even shorter than those generated in dye lasers with much more average output power. • The development of diode lasers for direct pumping of solid-state lasers. This has lead not only to very efficient and compact lasers, but also to mode-locked lasers with tens of watts of output power. • The development of novel saturable absorbers such as KLM which has pushed the frontier in ultrashort pulse duration into the two-optical-cycle regime.
96
Ultrafast solid-state lasers
[1, § 9
• The development of semiconductor saturable absorber mirrors (SESAMs) which passively mode-lock all kind of diode-pumped solid-state lasers without Q-switching instabilities and can be optimized for operation in very different parameter regimes such as laser wavelength, pulse duration, pulse repetition rate and power levels. For the next few years we expect many new exciting developments in the field of diode-pumped solid-state lasers: • For example, new solid-state gain materials, for example Cr2+ :ZnSe which appears to be suitable for the generation of pulses with 20 fs duration or less in a new spectral region around 2.7 µm. • Very high power levels (> 100 W of average power) should become possible with passively mode-locked thin disk lasers. Pulse durations just below 1 ps are feasible and with new materials the regime of 200 fs or even below should become accessible with similarly high powers. • Nonlinear frequency conversion stages (based on second-harmonic generation, sum frequency mixing, or parametric oscillation) will be pumped with high-power mode-locked lasers to generate short and powerful pulses at other wavelengths. The high power makes the nonlinear conversion efficiencies very high with very simple arrangements (Südmeyer, Aus der Au, Paschotta, Keller, Smith, Ross and Hanna [2001]). This will be of interest for application in largescreen RGB display systems, for example. • Simple external pulse compression combined with novel high-average-power solid-state lasers now allows for peak intensities as high as 12 MW with 33fs pulses at the full pulse repetition rate of the laser oscillator (Südmeyer, Brunner, Innerhofer, Paschotta, Furusawa, Baggett, Monro, Richardson and Keller [2003]). This could be focused to a peak intensity of 1014 W/cm2 , a regime where high field laser physics such as high harmonic generation (Ferray, L’Huillier, Li, Lompré, Mainfray and Manus [1988], Lewenstein, Balcou, Ivanov, L’Huillier and Corkum [1994], Salieres, L’Huillier, Antoine and Lewenstein [1999]) and laser plasma generated X-rays (Murnane, Kapteyn, Rosen and Falcone [1991]) are possible at more than 10 MHz pulse repetition rate. This improves signal-to-noise ratios in measurements by 4 orders of magnitude compared to the standard sources at kHz repetition rates. This would be important for low-power applications such as X-ray imaging and microscopy (Schmidt, Bauer, Wiemann, Porath, Scharte, Andreyev, Schönhense and Aeschlimann [2002]), femtosecond EUV and soft-X-ray photoelectron spectroscopy (Bauer, Lei, Read, Tobey, Gland, Murnane and Kapteyn [2001]) and ultrafast X-ray diffraction (Siders, Cavalleri, Sokolowski-Tinten, Toth, Guo, Kammler, Hoegen, Wilson, Linde and Barty [1999], Rousse, Rischel,
1, § 9]
Outlook
97
Fourmaux, Uschmann, Sebban, Grillon, Balcou, Förster, Geindre, Audebert, Gauthier and Hulin [2001]). • Very high pulse repetition rates with > 100 GHz should be possible from passively mode-locked diode-pumped solid-state lasers at different wavelength, even in the wavelength range around 1.5 µm and 1.3 µm for telecom applications. • As an alternative to ion-doped gain media, optically pumped semiconductor lasers with an external cavity and a SESAM for passive mode locking (Hoogland, Dhanjal, Roberts, Tropper, Häring, Paschotta, Morier-Genoud and Keller [2000]) will generate higher output powers and shorter pulses in the regime of multi-GHz pulse repetition rates (Häring, Paschotta, Gini, Morier-Genoud, Melchior, Martin and Keller [2001], Häring, Paschotta, Aschwanden, Gini, Morier-Genoud and Keller [2002]). Thus, all these examples show that the development of ultrafast diode-pumped sources has not come to an end but will continue to deliver superior performances for many established and new applications. In addition, research to produce pulses of even shorter duration is underway. Currently, the most promising path to attosecond pulse generation and attosecond spectroscopy is high-harmonic generation (recent reviews are given in Salieres, L’Huillier, Antoine and Lewenstein [1999] and DiMauro and Agostini [1995]). Nearly single-cycle-excitation optical pulses can produce harmonic frequencies in gases extending even into the water window (2.3–4.4 nm), and a continuum broad enough to support attosecond pulses. Key to this harmonic generation is intense optical pulses of a few cycles, since even slightly longer pulses result in reduced harmonic generation and a lower cut-off energy. It has been speculated early on that pulses from existing sources of high-harmonic generation exhibit attosecond time signature (Corkum [1995], Antoine, L’Huillier and Lewenstein [1996]). First measurements indicate single attosecond pulse duration (Hentschel, Kienberger, Spielmann, Reider, Milosevic, Brabec, Corkum, Heinzmann, Drescher and Krausz [2001]), but also demonstrate that measurement of these short pulses can be even a bigger challenge than the generation of the pulses themselves. We would expect that with attosecond time resolution we open up a new world of physics with as much impact as has been demonstrated in the 70s and 80s with the transition from picosecond to femtosecond time resolution. First time resolved measurements have been done (Drescher, Hentschel, Kienberger, Uiberacker, Yakovlev, Scrinzi, Westerwalbesloh, Kleineberg, Heinzmann and Krausz [2002]). At this point, however, we still have some significant challenges to tackle before we demonstrate standard attosecond pulse generation and attosecond spec-
98
Ultrafast solid-state lasers
[1
troscopy. Solving all of these challenges will make the research in ultrashort pulse generation very exciting and rewarding for many years to come.
References Agnesi, A., Pennacchio, C., Reali, G.C., Kubecek, V., 1997, High-power diode-pumped picosecond Nd:YVO4 laser, Opt. Lett. 22, 1645–1647. Agrawal, G.P., Olsson, N.A., 1989, Self-phase modulation and spectral broadening of optical pulses in semiconductor laser amplifiers, IEEE J. Quantum Electron. 25, 2297–2306. Angert, N.B., Borodin, N.I., Garmash, V.M., Zhitnyuk, V.A., Okhrimchuk, V.G., Siyuchenko, O.G., Shestakov, A.V., 1988, Lasing due to impurity color centers in yttrium aluminum garnet crystals at wavelengths in the range 1.35–1.45 µm, Sov. J. Quantum Electron. 18, 73–74. Antoine, P., L’Huillier, A., Lewenstein, M., 1996, Attosecond pulse trains using high-order harmonics, Phys. Rev. Lett. 77, 1234–1237. Apolonski, A., Poppe, A., Tempea, G., Spielmann, C., Udem, T., Holzwarth, R., Hänsch, T.W., Krausz, F., 2000, Controlling the phase evolution of few-cycle light pulses, Phys. Rev. Lett. 85, 740. Asaki, M.T., Huang, C.-P., Garvey, D., Zhou, J., Kapteyn, H.C., Murnane, M.M., 1993, Generation of 11-fs pulses from a self-mode-locked Ti:sapphire laser, Opt. Lett. 18, 977–979. Aus der Au, J., Kopf, D., Morier-Genoud, F., Moser, M., Keller, U., 1997, 60-fs pulses from a diodepumped Nd:glass laser, Opt. Lett. 22, 307–309. Aus der Au, J., Loesel, F.H., Morier-Genoud, F., Moser, M., Keller, U., 1998, Femtosecond diodepumped Nd:glass laser with more than 1-W average output power, Opt. Lett. 23, 271–273. Aus der Au, J., Schaer, S.F., Paschotta, R., Hönninger, C., Keller, U., Moser, M., 1999, High-power diode-pumped passively modelocked Yb:YAG lasers, Opt. Lett. 24, 1281–1283. Aus der Au, J., Spühler, G.J., Südmeyer, T., Paschotta, R., Hövel, R., Moser, M., Erhard, S., Karszewski, M., Giesen, A., Keller, U., 2000, 16.2 W average power from a diode-pumped femtosecond Yb:YAG thin disk laser, Opt. Lett. 25, 859. Bado, P., Bouvier, M., Coe, J.S., 1987, Nd:YLF mode-locked oscillator and regenerative amplifier, Opt. Lett. 12, 319–321. Baltuska, A., Pshenichnikov, M.S., Wiersma, D.A., 1998, Amplitude and phase characterization of 4.5-fs pulses by frequency-resolved optical gating, Opt. Lett. 23, 1474–1476. Baltuska, A., Wei, Z., Pshenichnikov, M.S., Wiersma, D.A., Szipöcs, R., 1997, All-solid-state cavity dumped sub-5-fs laser, Appl. Phys. B 65, 175–188. Bartels, A., Dekorsky, T., Kurz, H., 1999, Femtosecond Ti:sapphire ring laser with 2-GHz repetition rate and its application in time-resolved spectroscopy, Opt. Lett. 24, 996–998. Basu, S., Byer, R.L., 1988, Continuous-wave mode-locked Nd:glass laser pumped by a laser diode, Opt. Lett. 13, 458–460. Bauch, A., Telle, H.R., 2002, Frequency standards and frequency measurement, Rep. Prog. Phys. 65, 789–843. Bauer, M., Lei, C., Read, K., Tobey, R., Gland, J., Murnane, M.M., Kapteyn, H.C., 2001, Direct observation of surface chemistry using ultrafast soft-X-ray pulses, Phys. Rev. Lett. 87, 025501. Beaudoin, Y., Chien, C.Y., Coe, J.S., Tapié, J.-L., Mourou, G., 1992, Ultrahigh-contrast Ti:sapphire/Nd:glass terawatt laser system, Opt. Lett. 17, 865–867. Becker, M.F., Kuizenga, K.J., Siegman, A.E., 1972, Harmonic mode locking of the Nd:YAG laser, IEEE J. Quantum Electron. QE-8, 687–693. Beddard, T., Sibbett, W., Reid, D.T., Garduno-Mejia, J., Jamasbi, N., Mohebi, M., 1999, High-averagepower, 1-MW peak-power self-mode-locked Ti:sapphire oscillator, Opt. Lett. 24, 163–165.
1]
References
99
Bilinsky, I.P., Fujimoto, J.G., Walpole, J.N., Missaggia, L.J., 1999, InAs-doped silica films for saturable absorber applications, Appl. Phys. Lett. 74, 2411–2413. Brabec, T., Krausz, F., 2000, Intense few-cycle laser fields: Frontiers of nonlinear optics, Rev. Mod. Phys. 72, 545–591. Brabec, T., Spielmann, C., Krausz, F., 1991, Mode locking in solitary lasers, Opt. Lett. 16, 1961–1963. Braun, B., Hönninger, C., Zhang, G., Keller, U., Heine, F., Kellner, T., Huber, G., 1996, Efficient intracavity frequency doubling of a passively modelocked diode-pumped Nd:LSB laser, Opt. Lett. 21, 1567–1569. Braun, B., Kärtner, F.X., Keller, U., Meyn, J.-P., Huber, G., 1996, Passively Q-switched 180 ps Nd:LSB microchip laser, Opt. Lett. 21, 405–407. Braun, B., Keller, U., 1995, Single frequency Q-switched ring laser with an A-FPSA, Opt. Lett. 20, 1020–1022. Braun, B., Weingarten, K.J., Kärtner, F.X., Keller, U., 1995, Continuous-wave mode-locked solid-state lasers with enhanced spatial hole-burning. Part I: Experiments, Appl. Phys. B 61, 429–437. Brovelli, L.R., Jung, I.D., Kopf, D., Kamp, M., Moser, M., Kärtner, F.X., Keller, U., 1995, Selfstarting soliton modelocked Ti:sapphire laser using a thin semiconductor saturable absorber, Electron. Lett. 31, 287–289. Brovelli, L.R., Keller, U., Chiu, T.H., 1995, Design and operation of antiresonant Fabry–Perot saturable semiconductor absorbers for mode-locked solid-state lasers, J. Opt. Soc. Amer. B 12, 311– 322. Brovelli, L.R., Lanker, M., Keller, U., Goossen, K.W., Walker, J.A., Cunningham, J.E., 1995, Antiresonant Fabry–Perot quantum well modulator to actively modelock and synchronize solid-state lasers, Electron. Lett. 31, 381–382. Brunner, F., Paschotta, R., Aus der Au, J., Spühler, G.J., Morier-Genoud, F., Hövel, R., Moser, M., Erhard, S., Karszewski, M., Giesen, A., Keller, U., 2001, Widely tunable pulse durations from a passively mode-locked thin-disk Yb:YAG laser, Opt. Lett. 26, 379–381. Brunner, F., Spühler, G.J., Aus der Au, J., Krainer, L., Morier-Genoud, F., Paschotta, R., Lichtenstein, N., Weiss, S., Harder, C., Lagatsky, A.A., Abdolvand, A., Kuleshov, N.V., Keller, U., 2000, Diodepumped femtosecond Yb:KGd(WO4 )2 laser with 1.1-W average power, Opt. Lett. 25, 1119–1121. Brunner, F., Südmeyer, T., Innerhofer, E., Morier-Genoud, F., Paschotta, R., Kisel, V.E., Shcherbitsky, V.G., Kuleshov, N.V., Gao, J., Contag, K., Giesen, A., Keller, U., 2002, 240-fs pulses with 22-W average power from a mode-locked thin-disk Yb:KY(WO4 )2 laser, Opt. Lett. 27, 1162–1164. Burns, D., Hetterich, M., Ferguson, A.I., Bente, E., Dawson, M.D., Davies, J.I., Bland, S.W., 2000, High-average-power (> 20 W) Nd:YVO4 lasers mode locked by strain-compensated saturable Bragg reflectors, J. Opt. Soc. Amer. B 17, 919–926. Carrig, T.J., Pollock, C.R., 1993, Performance of a cw forsterite laser with krypton ion, Ti:sapphire, and Nd:YAG pump lasers, IEEE J. Quantum Electron. 29, 2835–2844. Cerullo, G., De Silvestri, S., Laporta, P., Longhi, S., Magni, V., Taccheo, S., Svelto, O., 1994, Continuous-wave mode locking of a bulk erbium-ytterbium glass laser, Opt. Lett. 19, 272–274. Chai, B.H.T., Lefaucheur, J.-L., Stalder, M., Bass, M., 1992, Cr:LiSr0.8 Ca0.2 AlF6 tunable laser, Opt. Lett. 17, 1584–1586. Chang, Y., Maciejko, R., Leonelli, R., Thorpe, A., 1998, Self-starting passively mode-locked tunable Cr4+ :yttrium-aluminium-garnet laser with a single prism for dispersion compensation, Appl. Phys. Lett. 73, 2098–2100. Chen, Y.-K., Wu, M.C., 1992, Monolithic colliding pulse modelocked quantum well lasers, IEEE J. Quantum Electron. 28, 2176–2185. Chen, Y.K., Wu, M.C., Tanbun-Ek, T., Logan, R.A., Chin, M.A., 1991, Subpicosecond monolithic colliding-pulse modelocked multiple quantum well lasers, Appl. Phys. Lett. 58, 1253–1255. Cheng, Z., Fürbach, A., Sartania, S., Lenzner, M., Spielmann, C., Krausz, F., 1999, Amplitude and chirp characterization of high-power laser pulses in the 5-fs regime, Opt. Lett. 24, 247–249.
100
Ultrafast solid-state lasers
[1
Chilla, J.L.A., Martinez, O.E., 1991, Direct determination of the amplitude and the phase of femtosecond light pulses, Opt. Lett. 16, 39–41. Cho, S.H., Bouma, B.E., Ippen, E.P., Fujimoto, J.G., 1999, Low-repetition-rate high-peak-power Kerrlens mode-locked Ti:Al2 O3 laser with a multiple-pass cell, Opt. Lett. 24, 417–419. Cho, S.H., Kärtner, F.X., Morgner, U., Ippen, E.P., Fujimoto, J.G., Cunningham, J.E., Knox, W.H., 2001, Generation of 90-nJ pulses with a 4-MHz repetition-rate Kerr-lens mode-locked Ti:Al2 O3 laser operating with net positive and negative intracavity dispersion, Opt. Lett. 26, 560–562. Chu, K.C., Heritage, J.P., Grant, R.S., Liu, K.X., Dienes, A., White, W.E., Sullivan, A., 1995, Direct measurement of the spectral phase of femtosecond pulses, Opt. Lett. 20, 904–906. Chudoba, C., Fujimoto, J.G., Ippen, E.P., Haus, H.A., Morgner, U., Kärtner, F.X., Scheuer, V., Angelow, G., Tschudi, T., 2001, All-solid-state Cr:forsterite laser generating 14-fs pulses at 1.3 µm, Opt. Lett. 26, 292–294. Chung, J., Siegman, A.E., 1995, Optical-Kerr-enhanced mode locking of a lamp-pumped Nd:YAG laser, IEEE J. Quantum Electron. 31, 582–590. Collings, B.C., Bergman, K., Knox, W.H., 1997, True fundamental solitons in a passively mode-locked short cavity Cr4+ :YAG laser, Opt. Lett. 22, 1098–1100. Collings, B.C., Stark, J.B., Tsuda, S., Knox, W.H., Cunningham, J.E., Jan, W.Y., Pathak, R., Bergman, K., 1996, Saturable Bragg reflector self-starting passive mode locking of a Cr4+ :YAG laser pumped with a diode-pumped Nd:YVO4 laser, Opt. Lett. 21, 1171–1173. Collins, R.J., Kisliuk, P., 1962, Control of population inversion in pulsed optical masers by feedback modulation, J. Appl. Phys. 33, 2009–2011. Conlon, P.J., Tong, Y.P., French, P.M.W., Taylor, J.R., Shestakov, A.V., 1994, Passive modelocking and dispersion measurement of a sub-100-fs Cr4+ :YAG laser, Opt. Lett. 19, 1468–1470. Corkum, P., 1995, Breaking the attosecond barrier, Opt. Photon., 18–22, May 1995. Couderc, V., Louradour, F., Barthelemy, A., 1999, 2.8 ps pulses from a modelocked diode pumped Nd:YVO4 laser using quadratic polarization switching, Opt. Commun. 166, 103–111. Cundiff, S.T., Ye, J., Hall, J.L., 2001, Optical frequency synthesis based on mode-locked lasers, Rev. Sci. Instrum. 27, 3750–3771. Curley, P.F., Ferguson, A.I., 1991, Actively modelocked Ti:sapphire laser producing transform-limited pulses of 150 fs duration, Opt. Lett. 16, 1016–1018. Curley, P.F., Spielmann, C., Brabec, T., Krausz, F., Wintner, E., Schmidt, A.J., 1993, Operation of a femtosecond Ti:sapphire solitary laser in the vicinity of zero group-delay dispersion, Opt. Lett. 18, 54–56. De Maria, A.J., Stetser, D.A., Heynau, H., 1966, Self mode-locking of lasers with saturable absorbers, Appl. Phys. Lett. 8, 174–176. Delfyett, P.J., 1995, High power ultrafast semiconductor injection diode lasers, in: Duling, I.I.N. (Ed.), Compact Sources of Ultrashort Pulses, Cambridge University Press, New York, pp. 274–328. Delfyett, P.J., 2003, Ultrafast single- and multiwavelength modelocked semiconductor lasers: physics and applications, in: Fermann, M.E., Galvanauskas, A., Sucha, G. (Eds.), Ultrafast Lasers: Technology and Applications, Marcel Dekker Inc., New York, pp. 219–321. Delfyett, P.J., Florez, L.T., Stoffel, N., Gmitter, T., Andreadakis, N.C., Silberberg, Y., Heritage, J.P., Alphonse, G.A., 1992, High-power ultrafast laser diodes, IEEE J. Quantum Electron. 28, 2203– 2219. Demidovich, A.A., Kuzmin, A.N., Ryabtsev, G.I., Danailov, M.B., Strek, W., Titov, A.N., 2000, Influence of Yb concentration on Yb:KYW laser properties, J. Alloys Compounds 300–301, 238–241. Diels, J.-C., 1990, Femtosecond dye lasers, in: Duarte, F., Hillman, L. (Eds.), Dye Lasers Principles: with Applications, Academic Press, Boston, pp. 41–132. Diels, J.-C., Stryland, E.W.V., Benedict, G., 1978, Generation and measurement of pulses of 0.2 ps duration, Opt. Commun. 25, 93.
1]
References
101
DiMauro, L.F., Agostini, P., 1995, Ionization dynamics in strong laser fields, in: Advances in Atomic, Molecular and Optical Physics, vol. 35, Academics Press, p. 79. Drescher, M., Hentschel, M., Kienberger, R., Tempea, G., Spielmann, C., Reider, G.A., Corkum, P.B., Krausz, F., 2001, X-ray pulses approaching the attosecond frontier, Science 291, 1923–1927. Drescher, M., Hentschel, M., Kienberger, R., Uiberacker, M., Yakovlev, V., Scrinzi, A., Westerwalbesloh, Th., Kleineberg, U., Heinzmann, U., Krausz, F., 2002, Time-resolved atomic inner-shell spectroscopy, Nature 419, 803–807. Druon, F., Balembois, F., Georges, P., Brun, A., Courjaud, A., Hönninger, C., Salin, F., Aron, A., Mougel, F., Aka, G., Vivien, D., 2000, Generation of 90-fs pulses from a modelocked diodepumped Yb:Ca4 Gdo(Bo3 )3 laser, Opt. Lett. 25, 423–425. Druon, F., Chenais, S., Raybaut, P., Balembois, F., Georges, P., Gaume, R., Aka, G., Viana, B., Mohr, S., Kopf, D., 2002, Diode-pumped Yb:Sr3 Y(BO3 )3 femtosecond laser, Opt. Lett. 27, 197–199. Duling III, I.N., 1995, Modelocking of all-fiber lasers, in: Duling III, I.N. (Ed.), Compact Sources of Ultrashort Pulses, Cambridge University Press, New York, pp. 140–178. Durfee III, C.G., Backus, S., Kapteyn, H.C., Murnane, M.M., 1999, Intense 8-fs pulse generation in the deep ultraviolet, Opt. Lett. 24, 697–699. Dymott, M.J.P., Ferguson, A.I., 1994, Self-mode-locked diode-pumped Cr:LiSAF laser, Opt. Lett. 19, 1988–1990. Dymott, M.J.P., Ferguson, A.I., 1995, Self-mode-locked diode-pumped Cr:LiSAF laser producing 34fs pulses at 42 mW average output power, Opt. Lett. 20, 1157–1159. Dymott, M.J.P., Ferguson, A.I., 1997, Pulse duration limitations in a diode-pumped femtosecond Kerrlens modelocked Cr:LiSAF laser, Appl. Phys. B 65, 227–234. Ell, R., Morgner, U., Kärtner, F.X., Fujimoto, J.G., Ippen, E.P., Scheuer, V., Angelow, G., Tschudi, T., Lederer, M.J., Boiko, A., Luther-Davis, B., 2001, Generation of 5-fs pulses and octave-spanning spectra directly from a Ti:sapphire laser, Opt. Lett. 26, 373–375. Evans, J.M., Spence, D.E., Sibbett, W., Chai, B.H.T., Miller, A., 1992, 50-fs pulse generation from a self-mode-locked Cr:LiSrAlF6 laser, Opt. Lett. 17, 1447–1449. Falcoz, F., Balembois, F., Georges, P., Brun, A., 1995, Self-starting self-mode-locked femtosecond diode-pumped Cr:LiSAF laser, Opt. Lett. 20, 1874–1876. Fan, T.Y., 1993, Heat generation in Nd:YAG and Yb:YAG, IEEE J. Quantum Electron. 29, 1457–1459. Fan, T.Y., Sanchez, A., 1990, Pump source requirements for end-pumped lasers, IEEE J. Quantum Electron. 26, 311–316. Fermann, M.E., 1994, Ultrashort-pulse sources based on single-mode rare-earth-doped fibers, Appl. Phys. B 58, 197–209. Fermann, M.E., 1995, Nonlinear polarization evolution in passively modelocked fiber lasers, in: Compact Sources of Ultrashort Pulses, Cambridge University Press, New York, pp. 179–207. Fermann, M.E., 2003, Ultrafast fiber oscillators, in: Fermann, M.E., Galvanauskas, A., Sucha, G. (Eds.), Ultrafast Lasers: Technology and Applications, Marcel Dekker Inc., New York, pp. 89– 154. Fermann, M.E., Andrejco, M.J., Stock, M.L., Silberberg, Y., Weiner, A.M., 1993, Passive modelocking in erbium fiber lasers with negative group delay, Appl. Phys. Lett. 62, 910–912. Fermann, M.E., Galvanauskas, A., Sucha, G., Harter, D., 1997, Fiber-lasers for ultrafast optics, Appl. Phys. B 65, 259–275. Ferray, M., L’Huillier, A., Li, X.F., Lompré, L.A., Mainfray, G., Manus, C., 1988, Multiple-harmonic conversion of 1064 nm radiation in rare gases, J. Phys. B: At. Mol. Opt. Phys. 21, L31–L35. Flood, C.J., Walker, D.R., van Driel, H.M., 1995, CW diode-pumped and FM modelocking of a Nd:KGW laser, Appl. Phys. B 60, 309–312. Fluck, R., Häring, R., Paschotta, R., Gini, E., Melchior, H., Keller, U., 1998, Eyesafe pulsed microchip laser using semiconductor saturable absorber mirrors, Appl. Phys. Lett. 72, 3273–3275.
102
Ultrafast solid-state lasers
[1
Fluck, R., Jung, I.D., Zhang, G., Kärtner, F.X., Keller, U., 1996, Broadband saturable absorber for 10-fs pulse generation, Opt. Lett. 21, 743–745. Fluck, R., Zhang, G., Keller, U., Weingarten, K.J., Moser, M., 1996, Diode-pumped passively modelocked 1.3 µm Nd:YVO4 and Nd:YLF lasers by use of semiconductor saturable absorbers, Opt. Lett. 21, 1378–1380. Fork, R.L., Cruz, C.H.B., Becker, P.C., Shank, C.V., 1987, Compression of optical pulses to six femtoseconds by using cubic phase compensation, Opt. Lett. 12, 483–485. Fork, R.L., Greene, B.I., Shank, C.V., 1981, Generation of optical pulses shorter than 0.1 ps by colliding pulse modelocking, Appl. Phys. Lett. 38, 617–619. Fork, R.L., Martinez, O.E., Gordon, J.P., 1984, Negative dispersion using pairs of prisms, Opt. Lett. 9, 150–152. Fork, R.L., Shank, C.V., Hirlimann, C., Yen, R., Tomlinson, W.J., 1983, Femtosecond white-light continuum pulses, Opt. Lett. 8, 1–3. French, P.M.W., Mellish, R., Taylor, J.R., Delfyett, P.J., Florez, L.T., 1993, Modelocked all-solid-state diode-pumped Cr:LiSAF laser, Opt. Lett. 18, 1934–1936. Gäbel, K.M., Rußbüldt, P., Lebert, R., Valster, A., 1998, Diode pumped Cr3+ :LiCAF fs-laser, Opt. Commun. 157, 327–334. Gale, G.M., Cavallari, M., Driscoll, T.J., Hache, F., 1995, Sub-20 fs tunable pulses in the visible from an 82 MHz optical parametric oscillator, Opt. Lett. 20, 1562–1564. Gallmann, L., Steinmeyer, G., Sutter, D.H., Matuschek, N., Keller, U., 2000, Collinear type-II secondharmonic-generation frequency-resolved optical gating for the characterization of sub-10-fs optical pulses, Opt. Lett. 25, 269–271. Gallmann, L., Steinmeyer, G., Sutter, D.H., Rupp, T., Iaconis, C., Walmsley, I.A., Keller, U., 2001, Spatially resolved amplitude and phase characterization of femtosecond optical pulses, Opt. Lett. 26, 96–98. Gallmann, L., Sutter, D.H., Matuschek, N., Steinmeyer, G., Keller, U., 2000, Techniques for the characterization of sub-10-fs optical pulses: a comparison, Appl. Phys. B 70, S67–S75. Gallmann, L., Sutter, D.H., Matuschek, N., Steinmeyer, G., Keller, U., Iaconis, C., Walmsley, I.A., 1999, Characterization of sub-6-fs optical pulses with spectral phase interferometry for direct field reconstruction, Opt. Lett 24, 1314–1316. Geusic, J.E., Marcos, H.M., Uitert, L.G.V., 1964, Laser oscillations in Nd-doped yttrium aluminum, yttrium gallium, and gadolinium garnets, Appl. Phys. Lett. 4, 182–184. Gibson, A.F., Kimmitt, M.F., Norris, B., 1974, Generation of bandwidth-limited pulses from a TEA CO2 laser using p-type germanium, Appl. Phys. Lett. 24, 306–307. Gires, F., Tournois, P., 1964, Interferometre utilisable pour la compression d’impulsions lumineuses modulees en frequence, C. R. Acad. Sci. Paris 258, 6112–6115. Godil, A.A., Hou, A.S., Auld, B.A., Bloom, D.M., 1991, Harmonic mode locking of a Nd:BEL laser using a 20-GHz dielectric resonator/optical modulator, Opt. Lett. 16, 1765–1767. Goodberlet, J., Jacobson, J., Fujimoto, J.G., Schulz, P.A., Fan, T.Y., 1990, Self-starting additive pulse modelocked diode-pumped Nd:YAG laser, Opt. Lett. 15, 504–506. Goodberlet, J., Wang, J., Fujimoto, J.G., Schulz, P.A., 1989, Femtosecond passively mode-locked Ti:sapphire laser with a nonlinear external cavity, Opt. Lett. 14, 1125–1127. Graf, Th., Ferguson, A.I., Bente, E., Burns, D., Dawson, M.D., 1999, Multi-Watt Nd:YVO4 laser, mode locked by a semiconductor saturable absorber mirror and side-pumped by a diode-laser bar, Opt. Commun. 159, 84–87. Haiml, M., Siegner, U., Morier-Genoud, F., Keller, U., Luysberg, M., Lutz, R.C., Specht, P., Weber, E.R., 1999, Optical nonlinearity in low-temperature-grown Ga:As: Microscopic limitations and optimization strategies, Appl. Phys. Lett. 74, 3134–3136. Hargrove, L.E., Fork, R.L., Pollack, M.A., 1964, Locking of HeNe laser modes induced by synchronous intracavity modulation, Appl. Phys. Lett. 5, 4.
1]
References
103
Häring, R., Paschotta, R., Aschwanden, A., Gini, E., Morier-Genoud, F., Keller, U., 2002, High power passively modelocked semiconductor laser, IEEE J. Quantum Electron. 38, 1268–1275. Häring, R., Paschotta, R., Fluck, R., Gini, E., Melchior, H., Keller, U., 2001, Passively Q-switched microchip laser at 1.5 µm, J. Opt. Soc. Amer. B 18, 1805–1812. Häring, R., Paschotta, R., Gini, E., Morier-Genoud, F., Melchior, H., Martin, D., Keller, U., 2001, Picosecond surface-emitting semiconductor laser with > 200 mW average power, Electron. Lett. 37, 766–767. Harmer, A.L., Linz, A., Gabbe, D.R., 1969, Fluorescence of Nd3+ in lithium yttrium fluoride, J. Phys. Chem. Solids 30, 1483–1491. Haus, H.A., 1975a, Theory of modelocking with a fast saturable absorber, J. Appl. Phys. 46, 3049– 3058. Haus, H.A., 1975b, A theory of forced mode locking, IEEE J. Quantum Electron. 11, 323–330. Haus, H.A., 1975c, Theory of mode locking with a slow saturable absorber, IEEE J. Quantum Electron. 11, 736–746. Haus, H.A., 1995, Short pulse generation, in: Duling, I.I.N. (Ed.), Compact Sources of Ultrashort Pulses, Cambridge University Press, New York, pp. 1–56. Haus, H.A., 2000, Mode-locking of lasers, IEEE J. Sel. Top. Quantum Electron. 6, 1173–1185. Haus, H.A., Fujimoto, J.G., Ippen, E.P., 1991, Structures for additive pulse modelocking, J. Opt. Soc. Amer. B 8, 2068–2076. Haus, H.A., Fujimoto, J.G., Ippen, E.P., 1992, Analytic theory of additive pulse and Kerr lens mode locking, IEEE J. Quantum Electron. 28, 2086–2096. Haus, H.A., Silberberg, Y., 1986, Laser mode locking with addition of nonlinear index, IEEE J. Quantum Electron. 22, 325–331. Haus, H.A., Tamura, K., Nelson, L.E., Ippen, E.P., 1995, Strechted-pulse additive pulse mode-locking in fiber ring lasers: Theory and experiment, IEEE J. Quantum Electron. 31, 591–598. Helbing, F.W., Steinmeyer, G., Keller, U., 2003, Carrier-envelope offset phase-locking with attosecond timing jitter, Special issue on Stabilization of Modelocked Lasers, IEEE J. Sel. Top. Quantum Electron. 9, 1030–1040. Helbing, F.W., Steinmeyer, G., Keller, U., Windeler, R.S., Stenger, J., Telle, H.R., 2001, Carrierenvelope offset dynamics of mode-locked lasers, Opt. Lett. 27, 194–196. Helbing, F.W., Steinmeyer, G., Stenger, J., Telle, H.R., Keller, U., 2002, Carrier-envelope offset dynamics and stabilization of femtosecond pulses, Appl. Phys. B 74, S35–S42. Hellwarth, R.W. (Ed.), 1961, Advances in Quantum Electronics, Columbia Univ. Press, New York. Henrich, B., Beigang, R., 1997, Self-starting Kerr-lens mode locking of a Nd:YAG-laser, Opt. Commun. 135, 300–304. Hentschel, M., Kienberger, R., Spielmann, Ch., Reider, G.A., Milosevic, N., Brabec, T., Corkum, P., Heinzmann, U., Drescher, M., Krausz, F., 2001, Attosecond metrology, Nature 414, 511–515. Hofer, M., Ober, M.H., Haberl, F., Fermann, M.E., 1992, Characterization of ultrashort pulse formation in passively mode-locked fiber lasers, IEEE J. Quantum Electron. 28, 720–728. Holzwarth, R., Zimmermann, M., Udem, T., Hänsch, T.W., 2001, Optical clockworks and the measurement of laser frequencies with a mode-locked frequency comb, IEEE J. Quantum Electron. 37, 1493–1501. Hönninger, C., Morier-Genoud, F., Moser, M., Keller, U., Brovelli, L.R., Harder, C., 1998, Efficient and tunable diode-pumped femtosecond Yb:glass lasers, Opt. Lett. 23, 126–128. Hönninger, C., Paschotta, R., Graf, M., Morier-Genoud, F., Zhang, G., Moser, M., Biswal, S., Nees, J., Braun, A., Mourou, G.A., Johannsen, I., Giesen, A., Seeber, W., Keller, U., 1999, Ultrafast ytterbium-doped bulk lasers and laser amplifiers, Appl. Phys. B 69, 3–17. Hönninger, C., Paschotta, R., Morier-Genoud, F., Moser, M., Keller, U., 1999, Q-switching stability limits of cw passive modelocking, J. Opt. Soc. Amer. B 16, 46–56.
104
Ultrafast solid-state lasers
[1
Hönninger, C., Zhang, G., Keller, U., Giesen, A., 1995, Femtosecond Yb:YAG laser using semiconductor saturable absorbers, Opt. Lett. 20, 2402–2404. Hoogland, S., Dhanjal, S., Roberts, J.S., Tropper, A.C., Häring, R., Paschotta, R., Morier-Genoud, F., Keller, U., 2000, Passively modelocked diode-pumped surface-emitting semiconductor laser, IEEE Photon. Technol. Lett. 12, 1135–1137. Huang, C.-P., Asaki, M.T., Backus, S., Murnane, M.M., Kapteyn, H.C., Nathel, H., 1992, 17-fs pulses from a self-mode-locked Ti:sapphire laser, Opt. Lett. 17, 1289–1291. Huang, C.-P., Kapteyn, H.C., McIntosh, J.W., Murnane, M.M., 1992, Generation of transform-limited 32-fs pulses from a self-mode-locked Ti:sapphire laser, Opt. Lett. 17, 139–141. Hughes, D.W., Barr, J.R.M., Hanna, D.C., 1991, Mode locking of a diode-laser-pumped Nd:glass laser by frequency modulation, Opt. Lett. 16, 147–149. Hughes, D.W., Phillips, M.W., Barr, J.R.M., Hanna, D.C., 1992, A laser-diode-pumped Nd:glass laser: Mode-locked, high power, and single frequency performance, IEEE J. Quantum Electron. 28, 1010–1017. Iaconis, C., Walmsley, I.A., 1998, Spectral phase interferometry for direct electric field reconstruction of ultrashort optical pulses, Opt. Lett. 23, 792–794. Iaconis, C., Walmsley, I.A., 1999, Self-referencing spectral interferometry for measuring ultrashort optical pulses, IEEE J. Quantum Electron. 35, 501–509. Innerhofer, E., Südmeyer, T., Brunner, F., Häring, R., Aschwanden, A., Paschotta, R., Hönninger, C., Kumkar, M., Keller, U., 2003, 60-W average power in 810-fs pulses from a thin-disk Yb:YAG laser, Opt. Lett. 28, 367–369. Ippen, E.P., 1994, Principles of passive mode locking, Appl. Phys. B 58, 159–170. Ippen, E.P., Eichenberger, D.J., Dixon, R.W., 1980, Picosecond pulse generation by passive modelocking of diode lasers, Appl. Phys. Lett. 37, 267–269. Ippen, E.P., Haus, H.A., Liu, L.Y., 1989, Additive pulse modelocking, J. Opt. Soc. Amer. B 6, 1736– 1745. Ippen, E.P., Shank, C.V., Dienes, A., 1972, Passive modelocking of the cw dye laser, Appl. Phys. Lett. 21, 348–350. Ishida, Y., Naganuma, K., 1994, Characteristics of femtosecond pulses near 1.5 µm in a self-modelocked Cr4+ :YAG laser, Opt. Lett. 19, 2003–2005. Islam, M.N., Sunderman, E.R., Soccolich, C.E., Bar-Joseph, I., Sauer, N., Chang, T.Y., Miller, B.I., 1989, Color center lasers passively mode locked by quantum wells, IEEE J. Quantum Electron. 25, 2454–2463. Jeong, T.M., Kang, E.C., Nam, C.H., 1999, Temporal and spectral characteristics of an additive-pulse mode-locked Nd:YLF laser with Michelson-type configuration, Opt. Commun. 166, 95–102. Jiang, W., Bowers, J., 1995, Ultrafast vertical cavity semiconductor lasers, in: Duling, I.I.N. (Ed.), Compact Sources of Ultrashort Pulses, Cambridge University Press, New York, pp. 208–273. Johnson, A.M., Simpson, W.M., 1983, Continuous-wave modelocked Nd:YAG-pumped subpicosecond dye lasers, Opt. Lett. 8, 554–556. Johnson, A.M., Simpson, W.M., 1985, Tunable femtosecond dye laser synchronously pumped by the compressed second harmonic of Nd:YAG, J. Opt. Soc. Amer. B 2, 619–625. Jones, D.J., Diddams, S.A., Ranka, J.K., Stentz, A., Windeler, R.S., Hall, J.L., Cundiff, S.T., 2000, Carrier-envelope phase control of femtosecond mode-locked lasers and direct optical frequency synthesis, Science 288, 635–639. Juhasz, T., Lai, S.T., Pessot, M.A., 1990, Efficient short-pulse generation from a diode-pumped Nd:YLF laser with a piezoelectrically induced diffraction modulator, Opt. Lett. 15, 1458–1460. Jung, I.D., Brovelli, L.R., Kamp, M., Keller, U., Moser, M., 1995, Scaling of the A-FPSA design toward a thin saturable absorber, Opt. Lett. 20, 1559–1561. Jung, I.D., Kärtner, F.X., Brovelli, L.R., Kamp, M., Keller, U., 1995, Experimental verification of soliton modelocking using only a slow saturable absorber, Opt. Lett. 20, 1892–1894.
1]
References
105
Jung, I.D., Kärtner, F.X., Henkmann, J., Zhang, G., Keller, U., 1997, High-dynamic-range characterization of ultrashort pulses, Appl. Phys. B 65, 307–310. Jung, I.D., Kärtner, F.X., Matuschek, N., Sutter, D.H., Morier-Genoud, F., Shi, Z., Scheuer, V., Tilsch, M., Tschudi, T., Keller, U., 1997, Semiconductor saturable absorber mirrors supporting sub-10 fs pulses, Appl. Phys. B 65, 137–150. Jung, I.D., Kärtner, F.X., Matuschek, N., Sutter, D.H., Morier-Genoud, F., Zhang, G., Keller, U., Scheuer, V., Tilsch, M., Tschudi, T., 1997, Self-starting 6.5 fs pulses from a Ti:sapphire laser, Opt. Lett. 22, 1009–1011. Kafka, J.D., Watts, M.L., Pieterse, J.-W.J., 1992, Picosecond and femtosecond pulse generation in a regeneratively modelocked Ti:sapphire laser, IEEE J. Quantum Electron. 28, 2151–2162. Kane, D.J., Trebino, R., 1993, Characterization of arbitrary femtosecond pulses using frequencyresolved optical gating, IEEE J. Quantum Electron. 29, 571–578. Kärtner, F.X., Aus der Au, J., Keller, U., 1998, Modelocking with slow and fast saturable absorbers – What’s the difference?, IEEE J. Sel. Top. Quantum Electron. 4, 159–168. Kärtner, F.X., Jung, I.D., Keller, U., 1996, Soliton modelocking with saturable absorbers, Special Issue on Ultrafast Electronics, Photonics and Optoelectronics, IEEE J. Sel. Top. Quantum Electron. (JSTQE) 2, 540–556. Kärtner, F.X., Keller, U., 1995, Stabilization of soliton-like pulses with a slow saturable absorber, Opt. Lett. 20, 16–18. Kärtner, F.X., Kopf, D., Keller, U., 1995, Solitary pulse stabilization and shortening in actively modelocked lasers, JOSA B 12, 486–496. Kärtner, F.X., Matuschek, N., Schibli, T., Keller, U., Haus, H.A., Heine, C., Morf, R., Scheuer, V., Tilsch, M., Tschudi, T., 1997, Design and fabrication of double-chirped mirrors, Opt. Lett. 22, 831–833. Kean, P.N., Zhu, X., Crust, D.W., Grant, R.S., Landford, N., Sibbett, W., 1989, Enhanced modelocking of color center lasers, Opt. Lett. 14, 39–41. Keller, U., 1994, Ultrafast all-solid-state laser technology, Appl. Phys. B 58, 347–363. Keller, U., 1999, Semiconductor nonlinearities for solid-state laser modelocking and Q-switching, in: Garmire, E., Kost, A. (Eds.), Nonlinear Optics in Semiconductors, vol. 59, Academic Press, Inc., Boston, pp. 211–286, Chapter 4. Keller, U., 2003, Recent developments in compact ultrafast lasers, Nature 424, 831–838. Keller, U., Chiu, T.H., 1992, Resonant passive modelocked Nd:YLF laser, IEEE J. Quantum Electron. 28, 1710–1721. Keller, U., Chiu, T.H., Ferguson, J.F., 1993, Self-starting femtosecond mode-locked Nd:glass laser that uses intracavity saturable absorbers, Opt. Lett. 18, 1077–1079. Keller, U., Kopf, D., Optical component for generating pulsed laser radiation, US Patent No. 6,466,604, priority date 19 May 1995. Keller, U., Knox, W.H., Roskos, H., 1990, Coupled-cavity resonant passive modelocked Ti:sapphire laser, Opt. Lett. 15, 1377–1379. Keller, U., Li, K.D., Khuri-Yakub, P.T., Bloom, D.M., Weingarten, K.J., Gerstenberger, D.C., 1990, High-frequency acousto-optic modelocker for picosecond pulse generation, Opt. Lett. 15, 45–47. Keller, U., Miller, D.A.B., Boyd, G.D., Chiu, T.H., Ferguson, J.F., Asom, M.T., 1992, Solid-state lowloss intracavity saturable absorber for Nd:YLF lasers: an antiresonant semiconductor Fabry–Perot saturable absorber, Opt. Lett. 17, 505–507. Keller, U., ’t Hooft, G.W., Knox, W.H., Cunningham, J.E., 1991, Femtosecond pulses from a continuously self-starting passively mode-locked Ti:sapphire laser, Opt. Lett. 16, 1022–1024. Keller, U., Valdmanis, J.A., Nuss, M.C., Johnson, A.M., 1988, 53 ps pulses at 1.32 µm from a harmonic mode-locked Nd:YAG laser, IEEE J. Quantum Electron. 24, 427–430. Keller, U., Weingarten, K.J., Kärtner, F.X., Kopf, D., Braun, B., Jung, I.D., Fluck, R., Hönninger, C., Matuschek, N., Aus der Au, J., 1996, Semiconductor saturable absorber mirrors (SESAMs)
106
Ultrafast solid-state lasers
[1
for femtosecond to nanosecond pulse generation in solid-state lasers, IEEE J. Sel. Top. Quantum Electron. 2, 435–453. Kellner, T., Heine, F., Huber, G., Hönninger, C., Braun, B., Morier-Genoud, F., Keller, U., 1998, Soliton mode-locked Nd:YAlO3 laser at 930 nm, J. Opt. Soc. Amer. B 15, 1663–1666. Kemp, A.J., Stormont, B., Agate, B., Brown, C.T.A., Keller, U., Sibbett, W., 2001, Gigahertz repetition-rates from a directly diode-pumped femtosecond Cr:LiSAF laser, Electron. Lett. 37, 1457–1458. Kigre Inc., QX Laser Glasses, Data Sheet, 1996. Kim, B.G., Garmire, E., Hummel, S.G., Dapkus, P.D., 1989, Nonlinear Bragg reflector based on saturable absorption, Appl. Phys. Lett. 54, 1095–1097. Knox, W.H., Saturable Bragg Reflector, US Patent Nr. 5,627,854, priority date 15 March 1995. Koechner, W., 1996, Solid-State Laser Engineering, in: Schawlow, A.L., Siegman, A.E., Tamir, T., Lotsch, H.K.V. (Eds.), Springer Series in Optical Sciences, vol. 1, Springer-Verlag, Heidelberg, Germany. Kopf, D., Aus der Au, J., Keller, U., Bona, G.L., Roentgen, P., 1995, 400-mW continuous-wave diodepumped Cr:LiSAF laser based on a power-scalable concept, Opt. Lett. 20, 1782–1784. Kopf, D., Kärtner, F.X., Keller, U., Weingarten, K.J., 1995, Diode-pumped modelocked Nd:glass lasers using an A-FPSA, Opt. Lett. 20, 1169–1171. Kopf, D., Kärtner, F., Weingarten, K.J., Keller, U., 1994, Pulse shortening in a Nd:glass laser by gain reshaping and soliton formation, Opt. Lett. 19, 2146–2148. Kopf, D., Keller, U., Emanuel, M.A., Beach, R.J., Skidmore, J.A., 1997, 1.1-W cw Cr:LiSAF laser pumped by a 1-cm diode-array, Opt. Lett. 22, 99–101. Kopf, D., Prasad, A., Zhang, G., Moser, M., Keller, U., 1997, Broadly tunable femtosecond Cr:LiSAF laser, Opt. Lett. 22, 621–623. Kopf, D., Spühler, G.J., Weingarten, K.J., Keller, U., 1996, Mode-locked laser cavities with a single prism for dispersion compensation, Appl. Opt. 35, 912–915. Kopf, D., Weingarten, K.J., Brovelli, L.R., Kamp, M., Keller, U., 1994, Diode-pumped 100-fs passively mode-locked Cr:LiSAF laser with an antiresonant Fabry–Perot saturable absorber, Opt. Lett. 19, 2143–2145. Kopf, D., Weingarten, K.J., Brovelli, L.R., Kamp, M., Keller, U., 1995, Sub-50-fs diode-pumped mode-locked Cr:LiSAF with an A-FPSA, in: Conference on Lasers and Electro-Optics (CLEO), paper CWM2. Kopf, D., Weingarten, K.J., Zhang, G., Moser, M., Emanuel, M.A., Beach, R.J., Skidmore, J.A., Keller, U., 1997, High-average-power diode-pumped femtosecond Cr:LiSAF lasers, Appl. Phys. B 65, 235–243. Kopf, D., Zhang, G., Fluck, R., Moser, M., Keller, U., 1996, All-in-one dispersion-compensating saturable absorber mirror for compact femtosecond laser sources, Opt. Lett. 21, 486–488. Kornelis, W., Biegert, J., Tisch, J.W.G., Nisoli, M., Sansone, G., De Silvestri, S., Keller, U., 2003, Single-shot kilohertz characterization of ultrashort pulses by spectral phase interferometry for direct electric-field reconstruction, Opt. Lett. 28 (4), 281–283. Krainer, L., Paschotta, R., Aus der Au, J., Hönninger, C., Keller, U., Moser, M., Kopf, D., Weingarten, K.J., 1999, Passively mode-locked Nd:YVO4 laser with up to 13 GHz repetition rate, Appl. Phys. B 69, 245–247. Krainer, L., Paschotta, R., Lecomte, S., Moser, M., Weingarten, K.J., Keller, U., 2002, Compact Nd:YVO4 lasers with pulse repetition rates up to 160 GHz, IEEE J. Quantum Electron. 38, 1331– 1338. Krainer, L., Paschotta, R., Moser, M., Keller, U., 2000a, Passively modelocked picosecond lasers with up to 59 GHz repetition rate, Appl. Phys. Lett. 77, 2104–2105. Krainer, L., Paschotta, R., Moser, M., Keller, U., 2000b, 77 GHz soliton modelocked Nd:YVO4 laser, Electron. Lett. 36, 1846–1848.
1]
References
107
Krainer, L., Paschotta, R., Spühler, G., Moser, M., Keller, U., 1999, 29 GHz modelocked miniature Nd:YVO4 laser, Electron. Lett. 35, 1160–1161. Krainer, L., Paschotta, R., Spühler, G.J., Klimov, I., Teisset, C., Weingarten, K.J., Keller, U., 2002, Tunable picosecond pulse-generating laser with a repetition rate exceeding 10 GHz, Electron. Lett. 38, 225–226. Krausz, F., Fermann, M.E., Brabec, T., Curley, P.F., Hofer, M., Ober, M.H., Spielmann, C., Wintner, E., Schmidt, A.J., 1992, Femtosecond solid-state lasers, IEEE J. Quantum Electron. 28, 2097–2122. Krausz, F., Spielmann, C., Brabec, T., Wintner, E., Schmidt, A.J., 1992, Generation of 33 fs optical pulses from a solid-state laser, Opt. Lett. 17, 204–206. Kubecek, V., Couderc, V., Bourliaguet, B., Louradour, F., Barthelemy, A., 1999, 4-W and 23-ps pulses from a lamp-pumped Nd:YAG laser passively mode-locked by polarization switching in a KTP crystal, Appl. Phys. B 69, 99–102. Kück, S., Petermann, K., Pohlmann, U., Huber, G., 1995, Near-infrared emission of Cr4+ -doped garnets: Lifetimes, quantum efficiencies, and emission cross sections, Phys. Rev. B 51, 17323– 17331. Kuizenga, D.J., Siegman, A.E., 1970a, FM and AM mode locking of the homogeneous laser. Part I: Theory, Part II: Experimental results, IEEE J. Quantum Electron. 6, 694–715. Kuizenga, D.J., Siegman, A.E., 1970b, FM and AM mode locking of the homogeneous laser. Part II: Experimental results in a Nd:YAG laser with internal FM modulation, IEEE J. Quantum Electron. 6, 709–715. Kuleshov, N.V., 2003, Private communication. Kuleshov, N.V., Lagatsky, A.A., Podlipensky, A.V., Mikhailov, V.P., Huber, G., 1997, Pulsed laser operation of Yb-doped KY(WO4 )2 and KGd(WO4 )2 , Opt. Lett. 22, 1317–1319. Kuleshov, N.V., Lagatsky, A.A., Shcherbitsky, V.G., Mikhailov, V.P., Heumann, E., Jensen, T., Diening, A., Huber, G., 1997, CW laser performance of Yb and Er, Yb doped tungstates, Appl. Phys. B 64, 409–413. Kutovoi, S.A., Laptev, V.V., Matsnev, S.Y., 1991, Lanthanum scandoborate as a new highly efficient active medium of solid state lasers, Sov. J. Quantum Electron. 21, 131–132. Laporta, P., Longhi, S., Marchesi, M., Taccheo, S., Svelto, O., 1995, 2.5 GHz and 5 GHz harmonic mode-locking of a diode-pumped bulk erbium-ytterbium glass laser at 1.5 microns, IEEE Photon. Technol. Lett. 7, 155–157. Lefort, L., Puech, K., Butterworth, S.D., Svirko, Y.P., Hanna, D.C., 1999, Generation of femtosecond pulses from order-of-magnitude pulse compression in a synchronously pumped optical parametric oscillator based on periodically poled lithium niobate, Opt. Lett. 24, 28–30. Lemoff, B.E., Barty, C.P.J., 1992, Generation of high-peak-power 20-fs pulses from a regeneratively initiated, self-mode-locked Ti:sapphire laser, Opt. Lett. 17, 1367–1369. Lewenstein, M., Balcou, P., Ivanov, M.Y., L’Huillier, A., Corkum, P.B., 1994, Theory of highharmonic generation by low-frequency laser fields, Phys. Rev. A 49, 2117–2132. Li, K.D., Sheridan, J.A., Bloom, D.M., 1991, Picosecond pulse generation in Nd:BEL with a highfrequency acousto-optic mode locker, Opt. Lett. 16, 1505–1507. LiKamWa, P., Chai, B.H.T., Miller, A., 1992, Self-mode-locked Cr3+ :LiCaAlF6 laser, Opt. Lett. 17, 1438–1440. Lincoln, J.R., Dymott, M.J.P., Ferguson, A.I., 1994, Femtosecond pulses from an all-solid-state Kerrlens mode-locked Cr:LiSAF laser, Opt. Lett. 19, 1210–1212. Lincoln, J.R., Ferguson, A.I., 1994, All-solid-state self-mode locking of a Nd:YLF laser, Opt. Lett. 19, 2119–2121. Liu, H., Nees, J., Mourou, G., 2001, Diode-pumped Kerr-lens mode-locked Yb:KY(WO4 )2 laser, Opt. Lett. 26, 1723–1725. Liu, J.M., Chee, J.K., 1990, Passive modelocking of a cw Nd:YLF laser with a nonlinear external coupled cavity, Opt. Lett. 15, 685–687.
108
Ultrafast solid-state lasers
[1
Liu, K.X., Flood, C.J., Walker, D.R., van Driel, H.M., 1992, Kerr lens mode locking of a diodepumped Nd:YAG laser, Opt. Lett. 17, 1361–1363. Liu, L.Y., Huxley, J.M., Ippen, E.P., Haus, H.A., 1990, Self-starting additive pulse modelocking of a Nd:YAG laser, Opt. Lett. 15, 553–555. Liu, X., Qian, L., Wise, F., Zhang, Z., Itatani, T., Sugaya, T., Nakagawa, T., Torizuka, K., 1998, Femtosecond Cr:forsterite laser diode pumped by a double-clad fiber, Opt. Lett. 23, 129–131. Loesel, F.H., Horvath, C., Grasbon, F., Jost, M., Niemz, M.H., 1997, Selfstarting femtosecond operation and transient dynamics of a diode-pumped Cr:LiSGaF laser with a semiconductor saturable absorber mirror, Appl. Phys. B 65, 783–787. Longhi, S., Laporta, P., Taccheo, S., Svelto, O., 1994, Third-order-harmonic mode locking of a bulk erbium:ytterbium:glass at a 2.5-GHz repetition rate, Opt. Lett. 19, 1985–1987. Longhi, S., Sorbello, G., Taccheo, S., Laporta, P., 1998, 5-GHz repetition-rate dual-wavelength pulsetrain generation from an intracavity frequency-modulated Er-Yb:glass laser, Opt. Lett. 23, 1547– 1549. Magni, V., Cerullo, G., de Silvestri, S., Monguzzi, A., 1995, Astigmatism in Gaussian-beam selffocussing and in resonators for Kerr-lens mode-locking, JOSA B 12, 476–485. Major, A., Langford, N., Graf, T., Burns, D., Ferguson, A.I., 2002, Diode-pumped passively modelocked Nd:KGd(WO4 )2 laser with 1-W average output power, Opt. Lett. 27, 1478–1480. Major, A., Langford, N., Graf, T., Ferguson, A.I., 2002, Additive-pulse mode locking of a diodepumped Nd:KGd(WO4 )2 laser, Appl. Phys. B 75, 467–469. Maker, G.T., Ferguson, A.I., 1989a, Modelocking and Q-switching of a diode laser pumped neodymium-doped yttrium lithium fluoride laser, Appl. Phys. Lett. 54, 403–405. Maker, G.T., Ferguson, A.I., 1989b, Frequency-modulation mode locking of diode-pumped Nd:YAG laser, Opt. Lett. 14, 788–790. Malcolm, G.P.A., Curley, P.F., Ferguson, A.I., 1990, Additive pulse modelocking of a diode-pumped Nd:YLF laser, Opt. Lett. 15, 1303–1305. Martinez, O.E., Fork, R.L., Gordon, J.P., 1984, Theory of passively modelocked lasers including selfphase modulation and group-velocity dispersion, Opt. Lett. 9, 156–158. Martinez, O.E., Fork, R.L., Gordon, J.P., 1985, Theory of passively modelocked lasers for the case of a nonlinear complex propagation coefficient, J. Opt. Soc. Amer. B 2, 753. Matuschek, N., Gallmann, L., Sutter, D.H., Steinmeyer, G., Keller, U., 2000, Back-side coated chirped mirror with ultra-smooth broadband dispersion characteristics, Appl. Phys. B 71, 509–522. Matuschek, N., Kärtner, F.X., Keller, U., 1997, Exact coupled-mode theories for multilayer interference coatings with arbitrary strong index modulations, IEEE J. Quantum Electron. 33, 295–302. Matuschek, N., Kärtner, F.X., Keller, U., 1999, Analytical design of double-chirped mirrors with custom-tailored dispersion characteristics, IEEE J. Quantum Electron. 35, 129–137. McCarthy, M.J., Maker, G.T., Hanna, D.C., 1991, Efficient frequency doubling of a self-starting additive-pulse mode-locked diode-pumped Nd:YAG laser, Opt. Commun. 82, 327–332. McClung, F.J., Hellwarth, R.W., 1962, Giant optical pulsations from ruby, J. Appl. Phys. 33, 828–829. McConnell, G., Ferguson, A.I., Langford, N., 2002, Additive-pulse modelocking of a diode-pumped Nd3+ :YVO4 laser, Appl. Phys. B 74, 7–9. Mellish, P.M., French, P.M.W., Taylor, J.R., Delfyett, P.J., Florez, L.T., 1994, All-solid-state femtosecond diode-pumped Cr:LiSAF laser, Electron. Lett. 30, 223–224. Mellish, R., Barry, N.P., Hyde, S.C.W., Jones, R., French, P.M.W., Taylor, J.R., van der Poel, C.J., Valster, A., 1995, Diode-pumped Cr:LiSAF all-solid-state femtosecond oscillator and regenerative amplifier, Opt. Lett. 20, 2312–2314. Mellish, R., Chernikov, S.V., French, P.M.W., Taylor, J.R., 1998, All-solid-state compact high repetition rate modelocked Cr4+ :YAG laser, Electron. Lett. 34, 552–553.
1]
References
109
Meyn, J.-P., Jensen, T., Huber, G., 1994, Spectroscopic properties and efficient diode-pumped laser operation of neodymium doped lanthanum scandium borate, IEEE J. Quantum Electron. 30, 913– 917. Miller, A., LiKamWa, P., Chai, B.H.T., Van Stryland, E.W., 1992, Generation of 150-fs tunable pulses in Cr:LiSAF, Opt. Lett. 17, 195–197. Mindl, T., Hefferle, P., Schneider, S., Dörr, F., 1983, Characterisation of a train of subpicosecond laser pulses by fringe resolved autocorrelation measurements, Appl. Phys. B 31, 201–207. Mitschke, F.M., Mollenauer, L.F., 1987, Ultrashort pulses from the soliton laser, Opt. Lett. 12, 407– 409. Mocker, H.W., Collins, R.J., 1965, Mode competition and self-locking effects in a Q-switched ruby laser, Appl. Phys. Lett. 7, 270–273. Mollenauer, L.F., Stolen, R.H., 1984, The soliton laser, Opt. Lett. 9, 13–15. Mollenauer, L.F., Stolen, R.H., Gordon, J.P., 1980, Experimental observation of picosecond pulse narrowing and solitons in optical fibers, Phys. Rev. Lett. 45, 1095–1098. Morgner, U., Kärtner, F.X., Cho, S.H., Chen, Y., Haus, H.A., Fujimoto, J.G., Ippen, E.P., Scheuer, V., Angelow, G., Tschudi, T., 1999, Sub-two-cycle pulses from a Kerr-lens mode-locked Ti:sapphire laser: addenda, Opt. Lett. 24, 920. Mougel, F., Dardenne, K., Aka, G., Kahn-Harari, A., Vivien, D., 1999, Ytterbium-doped Ca4 GdO(BO3 )3 : an efficient infrared laser and self-frequency doubling crystal, J. Opt. Soc. Amer. B 16, 164–172. Moulton, P.F., 1986, Spectroscopic and laser characteristics of Ti:Al2 O3 , J. Opt. Soc. Amer. B 3, 125–132. Murnane, M.M., Kapteyn, H.C., Rosen, M.D., Falcone, R.W., 1991, Ultrafast X-ray pulses from laserproduced plasmas, Science 251, 531–536. Negus, D.K., Spinelli, L., Goldblatt, N., Feugnet, G., 1991, Sub-100 femtosecond pulse generation by Kerr lens modelocking in Ti:sapphire, in: Dubé, G., Chase, L. (Eds.), Advanced Solid-State Lasers, vol. 10, Optical Society of America, Washington, DC, pp. 120–124. Nelson, L.E., Fleischer, S.B., Lenz, G., Ippen, E.P., 1996, Efficient frequency doubling of a femtosecond fiber laser, Opt. Lett. 21, 1759–1761. New, G.H.C., 1972, Modelocking of quasi-continuous lasers, Opt. Commun. 6, 188–192. New, G.H.C., 1974, Pulse evolution in mode-locked quasi-continuous lasers, IEEE J. Quantum Electron. 10, 115–124. Nisoli, M., Stagira, S., Silvestri, S.D., Svelto, O., Sartania, S., Cheng, Z., Lenzner, M., Spielmann, C., Krausz, F., 1997, A novel high-energy pulse compression system: generation of multigigawatt sub-5-fs pulses, Appl. Phys. B 65, 189–196. Ober, M.H., Hofer, M., Keller, U., Chiu, T.H., 1993, Self-starting, diode-pumped femtosecond Nd:fiber laser, Opt. Lett. 18, 1532–1534. Paschotta, R., Aus der Au, J., Spühler, G.J., Morier-Genoud, F., Hövel, R., Moser, M., Erhard, S., Karszewski, M., Giesen, A., Keller, U., 2000, Diode-pumped passively mode-locked lasers with high average power, Appl. Phys. B 70, S25–S31. Paschotta, R., Keller, U., 2001, Passive mode locking with slow saturable absorbers, Appl. Phys. B 73, 653–662. Paschotta, R., Keller, U., 2003, Ultrafast solid-state lasers, in: Fermann, M.E., Galvanauskas, A., Sucha, G. (Eds.), Ultrafast Lasers: Technology and Applications, Marcel Dekker Inc., New York, pp. 1–60. Paschotta, R., Spühler, G.J., Sutter, D.H., Matuschek, N., Keller, U., Moser, M., Hövel, R., Scheuer, V., Angelow, G., Tschudi, T., 1999, Double-chirped semiconductor mirror for dispersion compensation in femtosecond lasers, Appl. Phys. Lett. 75, 2166–2168.
110
Ultrafast solid-state lasers
[1
Paulus, G.G., Grasborn, F., Walther, H., Villoresi, P., Nisoli, M., Stagira, S., Priori, E., Silvestri, S.D., 2001, Absolute-phase phenomena in photoionization with few-cycle laser pulses, Nature 414, 182– 184. Payne, S.A., Chase, L.L., Newkirk, H.W., Smith, L.K., Krupke, W.F., 1988, Cr:LiCAF: a promising new solid-state laser material, IEEE J. Quantum Electron. 24, 2243–2252. Payne, S.A., Chase, L.L., Smith, L.K., Kway, W.L., Newkirk, H., 1989, Laser performance of LiSrAlF6 :Cr3+ , J. Appl. Phys. 66, 1051–1056. Pestryakov, E.V., Trunov, V.I., Matrosov, V.N., Razvalyaev, V.N., 1984, Generation of ultrashort pulses that are tunable in the 0.7–0.8 µm range in a laser based on alexandrite, Bull. Acad. Sci. USSR 48, 94–98. Petricevic, V., Gayen, S.K., Alfano, R.R., 1988, Laser action in chromium-doped forsterite, Appl. Phys. Lett. 52, 1040–1042. Petrov, V., Shcheslavskiy, V., Mirtchev, T., Noack, F., Itatani, T., Sugaya, T., Nakagawa, T., 1998, High-power self-starting femtosecond Cr:forsterite laser, Electron. Lett. 34, 559–561. Piché, M., Salin, F., 1993, Self-mode locking of solid-state lasers without apertures, Opt. Lett. 18, 1041–1043. Prasankumar, R.P., Chudoba, C., Fujimoto, J.G., Mak, P., Ruane, M.F., 2002, Self-starting mode locking in a Cr:forsterite laser by use of non-epitaxially-grown semiconductor-doped silica films, Opt. Lett. 27, 1564–1566. Pujol, M.C., Bursukova, M.A., Güell, F., Mateos, X., Sole, R., Gavalda, Jna., Aguilo, M., Massons, J., Diaz, F., Klopp, P., Griebner, U., Petrov, V., 2002, Growth, optical characterization, and laser operation of a stoichiometric crystal KYb(WO4 )2 , Phys. Rev. B 65, 165121. Ramaswamy, M., Gouveia-Neto, A.S., Negus, D.K., Izatt, J.A., Fujimoto, J.G., 1993, 2.3-ps pulses from a Kerr-lens mode-locked lamp-pumped Nd:YLF laser with a microdot mirror, Opt. Lett. 18, 1825–1827. Reichert, J., Holzwarth, R., Udem, T., Hänsch, T.W., 1999, Measuring the frequency of light with mode-locked lasers, Opt. Commun. 172, 59–68. Rhee, J.-K., Sosnowski, T.S., Tien, A.-C., Norris, T.B., 1996, Real-time dispersion analyzer of femtosecond laser pulses with use of a spectrally and temporally resolved upconversion technique, J. Opt. Soc. Amer. B 13, 1780–1785. Ripin, D.J., Chudoba, C., Gopinath, J.T., Fujimoto, J.G., Ippen, E.P., Morgner, U., Kärtner, F.X., Scheuer, V., Angelow, G., Tschudi, T., 2002, Generation of 20-fs pulses by a prismless Cr:YAG laser, Opt. Lett. 27, 61–63. Rizvi, N.H., French, P.M.W., Taylor, J.R., 1992a, Continuously self-mode-locked Ti:sapphire laser that produces sub-50-fs pulses, Opt. Lett. 17, 279–281. Rizvi, N.H., French, P.M.W., Taylor, J.R., 1992b, 50-fs pulse generation from a self-starting cw passively mode-locked Cr:LiSrAlF6 laser, Opt. Lett. 17, 877–879. Rizvi, N.H., French, P.M.W., Taylor, J.R., 1992c, Generation of 33-fs pulses from a passively modelocked Cr3+ :LiSrAlF6 laser, Opt. Lett. 17, 1605–1607. Rizvi, N.H., French, P.M.W., Taylor, J.R., Delfyett, P.J., Florez, L.T., 1993, Generation of pulses as short as 93 fs from a self-starting femtosecond Cr:LiSrAlF6 lasers by exploiting multiple-quantumwell absorbers, Opt. Lett. 18, 983–985. Roskos, H., Robl, T., Seilmeier, A., 1986, Pulse shortening to 25 ps in a cw mode-locked Nd:YAG laser by introducing an intracavity etalon, Appl. Phys. B 40, 59–65. Rousse, A., Rischel, C., Fourmaux, S., Uschmann, I., Sebban, S., Grillon, G., Balcou, P., Förster, E., Geindre, J.P., Audebert, P., Gauthier, J.C., Hulin, D., 2001, Non-termal melting in semiconductors measured at femtosecond resolution, Nature 410, 65–68. Ruddock, I.S., Bradley, D.J., 1976, Bandwidth-limited subpicosecond pulse generation in modelocked cw dye lasers, Appl. Phys. Lett. 29, 296–297.
1]
References
111
Ruffing, B., Nebel, A., Wallenstein, R., 1997, A 20-W KTA-OPO synchronously pumped by a cw modelocked Nd:YVO4 oscillator-amplifier system, in: Conference on Lasers and Electro-Optics, p. 199, paper CWB2. Sala, K.L., Kenney-Wallace, G.A., Hall, G.E., 1980, CW autocorrelation measurements of picosecond laser pulses, IEEE J. Quantum Electron. 16, 990–996. Saleh, B.E.A., Teich, M.C., 1991, Fundamentals of Photonics, John Wiley & Sons, Inc. Salieres, P., L’Huillier, A., Antoine, P., Lewenstein, M., 1999, Study of the spatial and temporal coherence of high-order harmonics, in: ISBN 0-12-003841-2, Academic Press, p. 83. Salin, F., Squier, J., Piché, M., 1991, Modelocking of Ti:sapphire lasers and self-focusing: a Gaussian approximation, Opt. Lett. 16, 1674–1676. Sam, R.C., 1989, Alexandrite lasers, in: Cheo, P.K. (Ed.), Handbook of Solid-State Lasers, Marcel Dekker, New York, pp. 349–455. Sartania, S., Cheng, Z., Lenzner, M., Tempea, G., Spielmann, C., Krausz, F., Ferencz, K., 1997, Generation of 0.1-TW 5-fs optical pulses at a 1-kHz repetition rate, Opt. Lett. 22, 1562–1564. Sarukura, N., Ishida, Y., Nakano, H., 1991, Generation of 50 fs pulses from a pulse compressed, cw, passively modelocked Ti:sapphire laser, Opt. Lett. 16, 153–155. Sasnett, M.W., 1989, Propagation of multimode laser beams – The M 2 factor, in: Hall, D.R., Jackson, P.E. (Eds.), The Physics and Technology of Laser Resonators, Adam Hilger, New York, pp. 132– 142. Schenkel, B., Biegert, J., Keller, U., Vozzi, C., Nisoli, M., Sansone, G., Stagira, S., De Silvestri, S., Svelto, O., 2003, Generation of 3.8-fs pulses from adaptive compression of a cascaded hollow fiber supercontinuum, Opt. Lett. 28, 1987–1989. Scheps, R., Myers, J.F., Serreze, H.B., Rosenberg, A., Morris, R.C., Long, M., 1991, Diode-pumped Cr:LiSrAlF6 laser, Opt. Lett. 16, 820–822. Schmidt, O., Bauer, M., Wiemann, C., Porath, R., Scharte, M., Andreyev, O., Schönhense, G., Aeschlimann, M., 2002, Time-resolved two-photon photoemission electron microscopy, Appl. Phys. B 74, 223–227. Schön, S., Haiml, M., Gallmann, L., Keller, U., 2002, Fluoride semiconductor saturable-absorber mirror for ultrashort pulse generation, Opt. Lett. 27, 1845–1847. Schön, S., Haiml, M., Keller, U., 2000, Ultrabroadband AlGaAs/CaF2 semiconductor saturable absorber mirrors, Appl. Phys. Lett. 77, 782. Schott Glass Technologies, 1978, Glass for Laser Applications, Data Sheets. Schott Glass Technologies, 1999, Glass for Laser Applications, Data Sheets. Seas, A., Petricevic, V., Alfano, R.R., 1993, Self-mode-locked chromium-doped forsterite laser generates 50 fs pulses, Opt. Lett. 18, 891–893. Sennaroglu, A., Pollock, C.R., Nathel, H., 1993, Generation of 48-fs pulses and measurement of crystal dispersion by using a regeneratively initiated self-mode-locked chromium-doped forsterite laser, Opt. Lett. 18, 826–828. Sennaroglu, A., Pollock, C.R., Nathel, H., 1994, Continuous-wave self-mode-locked operation of a femtosecond Cr4+ :YAG laser, Opt. Lett. 19, 390–392. Shah, J., 1996, Ultrafast spectroscopy of semiconductors and semiconductor nanostructures, in: Springer Series in Solid-State Sciences, vol. 115. Shank, C.V., 1988, Generation of ultrashort optical pulses, in: Kaiser, W. (Ed.), Ultrashort Laser Pulses and Applications, Springer-Verlag, Heidelberg, Chapter 2. Shank, C.V., Ippen, E.P., 1974, Subpicosecond kilowatt pulses from a modelocked cw dye laser, Appl. Phys. Lett. 24, 373–375. Sharp, R.C., Spock, D.E., Pan, N., Elliot, J., 1996, Opt. Lett. 21, 881. Shirakawa, A., Sakane, I., Takasaka, M., Kobayashi, T., 1999, Sub-5-fs visible pulse generation by pulse-front-matched noncollinear optical parametric amplification, Appl. Phys. Lett. 74, 2268– 2270.
112
Ultrafast solid-state lasers
[1
Siders, C.W., Cavalleri, A., Sokolowski-Tinten, K., Toth, C., Guo, T., Kammler, M., von Hoegen, M.H., Wilson, K.R., von der Linde, D., Barty, C.P.J., 1999, Detection of nonthermal melting by ultrafast X-ray diffraction, Science 286, 1340–1342. Siegman, A.E., 1986, Lasers, University Science Books, Mill Valley, CA. Siegner, U., Keller, U., 2000, Nonlinear optical processes for ultrashort pulse generation, in: Bass, M., Stryland, E.W., Williams, D.R., Wolfe, W.L. (Eds.), Handbook of Optics, vol. III, McGraw–Hill, Inc., New York. Skidmore, J.A., Emanuel, M.A., Beach, R.J., Benett, W.J., Freitas, B.L., Carlson, N.W., Solarz, R.W., 1995, High-power continuous wave 690 nm AlGaInP laser-diode arrays, Appl. Phys. Lett. 66, 1163–1165. Smith, L.K., Payne, S.A., Kway, W.L., Chase, L.L., Chai, B.H.T., 1992, Investigation of the laser properties of Cr3+ :LiSrGaF6 , IEEE J. Quantum Electron. 28, 2612–2618. Sorokina, I.T., Sorokin, E., Wintner, E., Cassanho, A., Jenssen, H.P., Noginov, M.A., 1996, Efficient cw TEM00 and femtosecond Kerr-lens modelocked Cr:LiSrGaF laser, Opt. Lett. 21, 204–206. Sorokina, I.T., Sorokin, E., Wintner, E., Cassanho, A., Jenssen, H.P., Szipöcs, R., 1996, Prismless passively mode-locked femtosecond Cr:LiSGaF laser, Opt. Lett. 21, 1165–1167. Sorokina, I.T., Sorokin, E., Wintner, E., Cassanho, A., Jenssen, H.P., Szipöcs, R., 1997, 14-fs pulse generation in Kerr-lens modelocked prismless Cr:LiSGaF and Cr:LiSAF lasers observation of pulse self-frequency shift, Opt. Lett. 22, 1716–1718. Spälter, S., Böhm, M., Burk, M., Mikulla, B., Fluck, R., Jung, I.D., Zhang, G., Keller, U., Sizmann, A., Leuchs, G., 1997, Self-starting soliton-modelocked femtosecond Cr(4+):YAG laser using an antiresonant Fabry–Perot saturable absorber, Appl. Phys. B 65, 335–338. Spence, D.E., Kean, P.N., Sibbett, W., 1991, 60-fsec pulse generation from a self-mode-locked Ti:sapphire laser, Opt. Lett. 16, 42–44. Spielmann, C., Krausz, F., Brabec, T., Wintner, E., Schmidt, A.J., 1991, Femtosecond passive modelocking of a solid-state laser by dispersively balanced nonlinear interferometer, Appl. Phys. Lett. 58, 2470–2472. Spühler, G.J., Dymott, M., Klimov, I., Luntz, G., Baraldi, L., Kilburn, I., Crosby, P., Thomas, S., Zehnder, O., Teisset, C.Y., Brownell, M., Weingarten, K.J., Dangel, R., Offrein, B.J., Bona, G.L., Buccafusca, O., Kaneko, Y., Krainer, L., Paschotta, R., Keller, U., 2002, 40 GHz pulse generating source with sub-picosecond timing jitter, Electron. Lett. 38, 1031–1033. Spühler, G.J., Gallmann, L., Fluck, R., Zhang, G., Brovelli, L.R., Harder, C., Laporta, P., Keller, U., 1999, Passively modelocked diode-pumped erbium-ytterbium glass laser using a semiconductor saturable absorber mirror, Electron. Lett. 35, 567–568. Spühler, G.J., Golding, P.S., Krainer, L., Kilburn, I.J., Crosby, P.A., Brownell, M., Weingarten, K.J., Paschotta, R., Haiml, M., Grange, R., Keller, U., 2003, Novel multi-wavelength source with 25GHz channel spacing tunable over the C-band, Electron. Lett. 39, 778–780. Spühler, G.J., Paschotta, R., Fluck, R., Braun, B., Moser, M., Zhang, G., Gini, E., Keller, U., 1999, Experimentally confirmed design guidelines for passively Q-switched microchip lasers using semiconductor saturable absorbers, J. Opt. Soc. Amer. B 16, 376–388; Errata, J. Opt. Soc. Amer. B 18 (2001), 886. Spühler, G.J., Paschotta, R., Keller, U., Moser, M., Dymott, M.J.P., Kopf, D., Meyer, J., Weingarten, K.J., Kmetec, J.D., Alexander, J., Truong, G., 1999, Diode-pumped passively mode-locked Nd:YAG laser with 10-W average power in diffraction-limited beam, Opt. Lett. 24, 528–530. Spühler, G.J., Reffert, S., Haiml, M., Moser, M., Keller, U., 2001, Output-coupling semiconductor saturable absorber mirror, Appl. Phys. Lett. 78, 2733–2735. Spühler, G.J., Südmeyer, T., Paschotta, R., Moser, M., Weingarten, K.J., Keller, U., 2000, Passively mode-locked high-power Nd:YAG lasers with multiple laser heads, Appl. Phys. B 71, 19–25. Steinmeyer, G., Sutter, D.H., Gallmann, L., Matuschek, N., Keller, U., 1999, Frontiers in ultrashort pulse generation: Pushing the limits in linear and nonlinear optics, Science 286, 1507–1512.
1]
References
113
Stingl, A., Lenzner, M., Spielmann, Ch., Krausz, F., Szipöcs, R., 1995, Sub-10-fs mirror-dispersioncontrolled Ti:sapphire laser, Opt. Lett. 20, 602–604. Stingl, A., Spielmann, C., Krausz, F., Szipöcs, R., 1994, Generation of 11-fs pulses from a Ti:sapphire laser without the use of prisms, Opt. Lett. 19, 204–206. Südmeyer, T., Aus der Au, J., Paschotta, R., Keller, U., Smith, P.G.R., Ross, G.W., Hanna, D.C., 2001, Novel ultrafast parametric systems: high repetition rate single-pass OPG and fiber-feedback OPO, J. Phys. D: Appl. Phys. 34, 2433–2439. Südmeyer, T., Brunner, F., Innerhofer, E., Paschotta, R., Furusawa, K., Baggett, J.C., Monro, T.M., Richardson, D.J., Keller, U., 2003, Nonlinear femtosecond pulse compression at high average power levels using a large mode area holey fiber, Opt. Lett. 28, 1951–1953. Sutter, D.H., Jung, I.D., Kärtner, F.X., Matuschek, N., Morier-Genoud, F., Scheuer, V., Tilsch, M., Tschudi, T., Keller, U., 1998, Self-starting 6.5-fs pulses from a Ti:sapphire laser using a semiconductor saturable absorber and double-chirped mirrors, IEEE J. Sel. Top. Quantum Electron. 4, 169–178. Sutter, D.H., Steinmeyer, G., Gallmann, L., Matuschek, N., Morier-Genoud, F., Keller, U., Scheuer, V., Angelow, G., Tschudi, T., 1999, Semiconductor saturable-absorber mirror-assisted Kerr-lens mode-locked Ti:sapphire laser producing pulses in the two-cycle regime, Opt. Lett. 24, 631–633. Svelto, O., 1998, Principles of Lasers, fourth ed., Plenum Press, New York. Szipöcs, R., Ferencz, K., Spielmann, C., Krausz, F., 1994, Chirped multilayer coatings for broadband dispersion control in femtosecond lasers, Opt. Lett. 19, 201–203. Tamura, K., Ippen, E.P., Haus, H.A., 1995, Pulse dynamics in stretched-pulse fiber lasers, Appl. Phys. Lett. 67, 158–160. Tapié, J.-L., Mourou, G., 1992, Shaping of clean, femtosecond pulses at 1.053 µm for chirped-pulse amplification, Opt. Lett. 17, 136–138. Telle, H.R., Steinmeyer, G., Dunlop, A.E., Stenger, J., Sutter, D.H., Keller, U., 1999, Carrier-envelope offset phase control: A novel concept for absolute optical frequency measurement and ultrashort pulse generation, Appl. Phys. B 69, 327–332. Thoen, E.R., Koontz, E.M., Joschko, M., Langlois, P., Schibli, T.R., Kärtner, F.X., Ippen, E.P., Kolodziejski, L.A., 1999, Two-photon absorption in semiconductor saturable absorber mirrors, Appl. Phys. Lett. 74, 3927–3929. Tomaru, T., 2001, Two-element-cavity femtosecond Cr:YAG laser operating at a 2.6-GHz repetition rate, Opt. Lett. 26, 1439–1441. Tomaru, T., Petek, H., 2000, Femtosecond Cr4+ :YAG laser with an L-fold cavity operating at a 1.2 GHz repetition rate, Opt. Lett. 25, 584–586. Tomlinson, W.J., Stolen, R.H., Shank, C.V., 1984, Compression of optical pulses chirped by self-phase modulation in fibers, J. Opt. Soc. Amer. B 1, 139–149. Tong, Y.P., French, P.M.W., Taylor, J.R., Fujimoto, J.O., 1997, All-solid-state femtosecond sources in the near infrared, Opt. Commun. 136, 235–238. Tong, Y.P., Sutherland, J.M., French, P.M.W., Taylor, J.R., Shestakov, A.V., Chai, B.H.T., 1996, Selfstarting Kerr-lens mode-locked femtosecond Cr4+ :YAG and picosecond Pr3+ :YLF solid-state lasers, Opt. Lett. 21, 644–646. Treacy, E.B., 1969, Optical pulse compression with diffraction gratings, IEEE J. Quantum Electron. 5, 454–458. Trebino, R., DeLong, K.W., Fittinghoff, D.N., Sweetser, J., Krumbügel, M.A., Richman, B., 1997, Measuring ultrashort laser pulses in the time-frequency domain using frequency-resolved optical gating, Rev. Sci. Instrum. 68, 1–19. Tsuda, S., Knox, W.H., Cundiff, S.T., 1996, High efficiency diode pumping of a saturable Bragg reflector-mode-locked Cr:LiSAF femtosecond laser, Appl. Phys. Lett. 69, 1538–1540.
114
Ultrafast solid-state lasers
[1
Tsuda, S., Knox, W.H., de Souza, E.A., Jan, W.Y., Cunningham, J.E., 1995, Low-loss intracavity AlAs/AlGaAs saturable Bragg reflector for femtosecond mode locking in solid-state lasers, Opt. Lett. 20, 1406–1408. Udem, T., Holzwarth, R., Hänsch, T.W., 2002, Optical frequency metrology, Nature 416, 233–237. Udem, T., Reichert, J., Holzwarth, R., Hänsch, T.W., 1999, Accurate measurement of large optical frequency differences with a mode-locked laser, Opt. Lett. 24, 881–883. Uemura, S., Torizuka, K., 1999, Generation of 12-fs pulses from a diode-pumped Kerr-lens modelocked Cr:LiSAF laser, Opt. Lett. 24, 780–782. Uemura, S., Torizuka, K., 2003, Development of a diode-pumped Kerr-lens mode-locked Cr:LiSAF laser, IEEE J. Quantum Electron. 39, 68–73. Vail’ev, P., 1995, Ultrafast Diode Lasers, Artech House, Inc., Boston. Valdmanis, J.A., Fork, R.L., 1986, Design considerations for a femtosecond pulse laser balancing self phase modulation, group velocity dispersion, saturable absorption, and saturable gain, IEEE J. Quantum Electron. 22, 112–118. Valdmanis, J.A., Fork, R.L., Gordon, J.P., 1985, Generation of optical pulses as short as 27 fs directly from a laser balancing self-phase modulation, group-velocity dispersion, saturable absorption, and saturable gain, Opt. Lett. 10, 131–133. van der Ziel, J.P., Logan, R.A., Mikulyak, R.M., 1981, Generation of subpicosecond pulses from an actively modelocked GaAs laser in an external ring cavity, Appl. Phys. Lett. 39, 867–869. Vasil’ev, P.P., Morzov, V.N., Popov, Y.M., Sergeev, A.B., 1986, Subpicosecond pulse generation by tandem-type AlGaAs DH laser with colliding pulse modelocking, IEEE J. Quantum Electron. 22, 149–151. Wagenblast, P.C., Morgner, U., Grawert, F., Schibli, T.R., Kärtner, F.X., Scheuer, V., Angelow, G., Lederer, M.J., 2002, Generation of sub-10-fs pulses from a Kerr-lens mode-locked Cr:LiCAF laser oscillator by use of third-order dispersion-compensating double-chirped mirrors, Opt. Lett. 27, 1726–1728. Walker, A.C., Kar, A.K., Ji, W., Keller, U., Smith, S.D., 1986, All-optical power limiting of CO2 laser pulses using cascaded optical bistable elements, Appl. Phys. Lett. 48, 683–685. Walker, S.J., Avramopoulos, H., Sizer II, T., 1990, Compact mode-locked solid-state lasers at 0.5- and 1-GHz repetition rates, Opt. Lett. 15, 1070–1072. Walling, J.C., Jenssen, H.P., Morris, R.C., O’Dell, E.W., Peterson, O.G., 1979, Tunable laser performance in BeAl2 O4 Cr3+ , Opt. Lett. 4, 182–183. Wang, H.S., Kam Wa, P.L., Lefaucheur, J.L., Chai, B.H.T., Miller, A., 1994, Cw and self-modelocking performance of a red pumped Cr:LiSCAF laser, Opt. Commun. 110, 679–688. Wasik, G., Helbing, F.W., König, F., Sizmann, A., Leuchs, G., 2001, Bulk Er:Yb:glass soliton femtosecond laser, in: Conference on Lasers and Electro-Optics (CLEO), paper CMA4. Weiner, A.M., 1983, Effect of group velocity mismatch on the measurement of ultrashort optical pulses via second harmonic generation, IEEE J. Quantum Electron. 19, 1276–1283. Weingarten, K.J., Braun, B., Keller, U., 1994, In-situ small-signal gain of solid-state lasers determined from relaxation oscillation frequency measurements, Opt. Lett. 19, 1140–1142. Weingarten, K.J., Godil, A.A., Gifford, M., 1992, FM modelocking at 2.85 GHz using a microwave resonant optical modulator, IEEE Photon. Technol. Lett. 4, 1106–1109. Weingarten, K.J., Keller, U., Chiu, T.H., Ferguson, J.F., 1993, Passively mode-locked diode-pumped solid-state lasers using an antiresonant Fabry–Perot saturable absorber, Opt. Lett. 18, 640–642. Weingarten, K.J., Shannon, D.C., Wallace, R.W., Keller, U., 1990, Two gigahertz repetition rate, diode-pumped, mode-locked, Nd:yttrium lithium fluoride (YLF) laser, Opt. Lett. 15, 962–964. Weingarten, K.J., Spühler, G.J., Keller, U., Krainer, L., Semiconductor saturable absorber mirror, US Patent No. 6,538,298, priority date Dec. 10, 2001.
1]
References
115
Wu, M.C., Chen, Y.K., Tanbun-Ek, T., Logan, R.A., Chin, M.A., Raybon, G., 1990, Transform-limited 1.4 ps optical pulses from a monolithic colliding-pulse modelocked quantum well laser, Appl. Phys. Lett. 57, 759–761. Xu, L., Spielmann, C., Krausz, F., Szipöcs, R., 1996, Ultrabroadband ring oscillator for sub-10-fs pulse generation, Opt. Lett. 21, 1259–1261. Xu, L., Spielmann, C., Poppe, A., Brabec, T., Krausz, F., Haensch, T.W., 1996, Route to phase control of ultrashort light pulses, Opt. Lett. 21, 2008–2010. Xu, L., Tempea, G., Spielmann, C., Krausz, F., Stingl, A., Ferencz, K., Takano, S., 1998, Continuouswave mode-locked Ti:sapphire laser focusable to 5 × 1013 W/cm2 , Opt. Lett. 23, 789–791. Yan, L., Ling, J.D., Ho, P.-T., Lee, C.H., 1986, Picosecond-pulse generation from a continuous-wave neodymium:phosphate glass laser, Opt. Lett. 11, 502–503. Yanovsky, V., Pang, Y., Wise, F., Minkov, B.I., 1993, Generation of 25-fs from a self-modelocked Cr:forsterite laser with optimized group-delay dispersion, Opt. Lett. 18, 1541–1543. Yanovsky, V.P., Wise, F.W., Cassanho, A., Jenssen, H.P., 1995, Kerr-lens mode-locked diode-pumped Cr:LiSGAF laser, Opt. Lett. 20, 1304–1306. Yariv, A., Yeh, P., 1984, Optical Waves in Crystals, John Wiley & Sons, New York. Zavelani-Rossi, M., Polli, D., Cerullo, G., De Silvestri, S., Gallmann, L., Steinmeyer, G., Keller, U., 2002, Few-optical-cycle laser pulses by OPA: broadband chirped mirror compression and SPIDER characterization, Appl. Phys. B 74, S245–S251. Zayhowski, J.J., Dill III, C., 1995, Coupled cavity electro-optically Q-switched Nd:YVO4 microchip lasers, Opt. Lett. 20, 716–718. Zayhowski, J.J., Harrison, J., 1997, Miniature solid-state lasers, in: Gupta, M.C. (Ed.), Handbook of Photonics, CRC Press, New York, pp. 326–392. Zayhowski, J.J., Mooradian, A., 1989, Single-frequency microchip Nd lasers, Opt. Lett. 14, 24–26. Zeller, S.C., Krainer, L., Spühler, G.J., Weingarten, K.J., Paschotta, R., Keller, U., 2003, Passively modelocked 40-GHz Er:Yb:glass laser, Appl. Phys. B 76, 787–788. Zhang, Z., Nakagawa, T., Torizuka, K., Sugaya, T., Kobayashi, K., 1999, Self-starting modelocked Cr4+ :YAG laser with a low-loss broadband semiconductor saturable-absorber mirror, Opt. Lett. 24, 1768–1770. Zhang, Z., Torizuka, K., Itatani, T., Kobayashi, K., Sugaya, T., Nakagawa, T., 1997a, Self-starting mode-locked femtosecond forsterite laser with a semiconductor saturable-absorber mirror, Opt. Lett. 22, 1006–1008. Zhang, Z., Torizuka, K., Itatani, T., Kobayashi, K., Sugaya, T., Nakagawa, T., 1997b, Femtosecond Cr:forsterite laser with modelocking initiated by a quantum well saturable absorber, IEEE J. Quantum Electron. 33, 1975–1981. Zhou, F., Malcolm, G.P.A., Ferguson, A.I., 1991, 1-GHz repetition-rate frequency-modulation modelocked neodymium lasers at 1.3 µm, Opt. Lett. 16, 1101–1103. Zhou, J., Taft, G., Huang, C.-P., Murnane, M.M., Kapteyn, H.C., Christov, I.P., 1994, Pulse evolution in a broad-bandwidth Ti:sapphire laser, Opt. Lett. 19, 1149–1151. Zirngibl, M., Stulz, L.W., Stone, J., Hugi, J., DiGiovanni, D., Hansen, P.B., 1991, 1.2 ps pulses from passively modelocked laser diode pumped Er-doped fiber ring lasers, Electron. Lett. 27, 1734– 1735.
E. Wolf, Progress in Optics 46 © 2004 Elsevier B.V. All rights reserved
Chapter 2
Multiple scattering of light from randomly rough surfaces by
Andrei V. Shchegrov Novalux, Inc., Corporate Headquarters, 1220 Midas Way, Sunnyvale, CA 94085, USA
Alexei A. Maradudin Department of Physics and Astronomy and Institute for Surface and Interface Science, University of California, Irvine, CA 92697-4575, USA
and
Eugenio R. Méndez División de Física Aplicada, Centro de Investigación Científica y de Educación, Superior de Ensenada, Apartado Postal 2732, Ensenada, Baja California 22800, Mexico
ISSN: 0079-6638
DOI: 10.1016/S0079-6638(03)46002-2 117
Contents
Page § 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
119
§ 2. Characterization of randomly rough surfaces . . . . . . . . . . . . . .
121
§ 3. Equations for electromagnetic fields and observable quantities . . . .
126
§ 4. Weak localization effects in the multiple scattering of light from randomly rough surfaces. Enhanced backscattering . . . . . . . . . . . .
133
§ 5. Angular intensity correlation functions . . . . . . . . . . . . . . . . .
180
§ 6. Multiple-scattering effects in the scattering of light from complex media bounded by a rough surface . . . . . . . . . . . . . . . . . . . . .
189
§ 7. Near-field effects: localization phenomena for surface waves . . . . .
217
§ 8. Spectral changes induced by multiple scattering . . . . . . . . . . . .
221
§ 9. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
229
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
233
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
233
118
§ 1. Introduction One of the first concepts studied in any course in optics or electromagnetic theory is the reflection and transmission of light on a planar interface between two different materials. The importance of this phenomenon is evident from our everyday experience since all material objects are bounded by surfaces. However, any realistic surface is rough to some degree, and most of us are used to making a distinction between the specular reflection caused by a planar surface component and the diffuse reflection, or scattering, caused by the deviation of the surface from planarity. The systematic investigation of the scattering of light from rough surfaces started in the beginning of the 20th century and for the most part of this century was focused on single-scattering phenomena and their theoretical description. Theoretically, the majority of the problems were treated by either a firstorder perturbation theory (small-height expansion) or the Kirchhoff approximation assuming smooth surface roughness with large horizontal scales. The reader interested in the single-scattering of light from rough surfaces is referred to numerous books and review articles addressing this mature subject (Beckmann and Spizzichino [1963], Shmelev [1972], Bass and Fuks [1980], DeSanto and Brown [1986], Ogilvy [1991], Nieto-Vesperinas [1991], Maradudin, Nieto-Vesperinas and Thorsos [1994a, 1994b], Voronovich [1994]). This review is devoted to the topic of multiple scattering of light from randomly rough surfaces. Some foundations of the multiple-scattering theory were generally laid out by earlier workers (see, e.g., Bass and Fuks [1980]) but the mathematical complexities requiring the evaluation of multi-dimensional integrals or the solution of large matrix equations did not allow achieving quantitative results and rendered the investigation of multiple-scattering phenomena impossible. In many instances, the evaluation of light scattering from very rough surfaces (large amplitude and/or large slope) had to be done by unjustifiably pushing the limits of single-scattering theories that allowed some quantitative analysis. The same modeling convenience reasons led to the predominant use of impenetrable surfaces. The situation began to change rapidly in the 1980s when powerful computers became available. 119
120
Multiple scattering of light from randomly rough surfaces
[2, § 1
In 1985, it was predicted theoretically (McGurn, Maradudin and Celli [1985]) that the light scattered from a randomly rough surface exhibits a peak in the retroreflection direction. The peak was attributed to a coherent interference multiple-scattering phenomenon, that is similar to the weak localization effect occurring in the electron transport through disordered conductors and, generally, in the propagation of any wave through a disordered volume (Kuga and Ishimaru [1984], Tsang and Ishimaru [1984], van Albada and Lagendijk [1985], Wolf and Maret [1985], Sheng [1995], Barabanenkov, Kravtsov, Ozrin and Saichev [1991]). The experimental confirmation of the enhanced backscattering effect for random surfaces followed shortly thereafter (Méndez and O’Donnell [1987], O’Donnell and Méndez [1987]). In the 1990s, the effect was investigated in more complex systems involving rough surfaces, such as layered media, nonlinear media, materials supporting surface and guided waves, surface systems involving random and deterministic scatterers, etc. It was found that coherent multiple-scattering enhancement can occur in transmission through randomly rough surfaces of penetrable media, and that light scattered diffusely from some surface systems can exhibit satellite peaks that appear in directions other than the retroreflection direction. Coherent signatures were found and analyzed in the angular intensity correlation functions. More recent work done in the second half of 1990s addressed coherent phenomena in the near field and spectral changes in the scattering of polychromatic light induced by multiple scattering. All this effort helped to create adequate theoretical and computational methodologies that are now widely accepted and used by most workers in the field. Progress in the studies of multiple scattering of light from rough surfaces makes it timely to address the key achievements in this review article and stimulate further investigations. Applied science and technology is becoming increasingly interdisciplinary, and many topics addressed in this review have implications in such areas as radar technology, the laser and optics industry, microscopy, control of surface quality in various industries, etc. The structure of this review is as follows. In Section 2, we introduce a statistical formalism for characterizing randomly rough surfaces. In Section 3, we discuss the equations for the electromagnetic field that need to be solved to obtain experimentally observable characteristics of the scattered light. These characteristics are also introduced in Section 3. Section 4 introduces the enhanced backscattering effect, and discusses its origin and the common techniques for its experimental and theoretical investigations. These techniques are also relevant in the study of other multiple-scattering effects discussed in later sections. The coherent multiple-scattering phenomena occurring in the higher-order moments of the intensity of the scattered field are reviewed in Section 5. Section 6 ad-
2, § 2]
Characterization of randomly rough surfaces
121
dresses the studies of multiple-scattering phenomena in rough surface systems, more complex than a simple interface between two linear media. In Section 7, we discuss multiple-scattering phenomena for surface electromagnetic waves and related near-field optics effects. The subject of Section 8 is spectral changes induced by the multiple scattering of polychromatic light from randomly rough surfaces. Finally, in Section 9, we review the up-to-date achievements in the field and discuss some directions for future research. § 2. Characterization of randomly rough surfaces In this section we summarize some of the geometrical and statistical properties of randomly rough surfaces that will be used in the rest of this article. We will restrict our attention to surfaces that are planar in the absence of roughness. Such surfaces can be classified as two-dimensional and one-dimensional. In the former case the departure of the surface from the plane (assumed to be the plane x3 = 0) depends on both of the coordinates x1 and x2 in that plane. In the latter case it depends on only one of the coordinates in that plane, say x1 . Since a one-dimensional random surface can be regarded as a special case of a two-dimensional random surface, we consider the latter type of surface first, and then specialize its properties to the one-dimensional case. The reader interested in the scattering of light from randomly rough surfaces that are not planar in the absence of roughness can find appropriate discussions in the studies by Schiffer [1990], Farias, Vasconcelos, Cesar and Maradudin [1994], and Fuks and Voronovich [1999]. 2.1. Two-dimensional random surfaces The equation defining a two-dimensional surface can be written in the form x3 = ζ (x ),
(2.1)
where the real function ζ (x ) is called the surface profile function. It is a singlevalued function of the two-dimensional position vector in the plane x3 = 0 given by x = xˆ 1 x1 + xˆ 2 x2 , where xˆ 1 and xˆ 2 are unit vectors along the x1 - and x2 -axes, respectively. The region ζ (x )min < x3 < ζ (x )max is called the selvedge region. We will assume that ζ (x ) is differentiable with respect to x1 and x2 as many times as is necessary. In practice this usually means that its first and second partial derivatives exist. The interesting class of continuous, but non-differentiable functions ζ (x ) representing fractal surfaces (Mandelbrot [1977], Rothrock and Thorndike [1980]) is beyond the scope of this review because our central topic
122
Multiple scattering of light from randomly rough surfaces
[2, § 2
– multiple-scattering phenomena – has not yet been studied in enough detail for fractal surfaces. The interested reader is referred to the investigations of the singlescattering phenomena for fractal surfaces by Berry [1979], Jakeman [1986], Jordan, Hollins, Jakeman and Prewitt [1988], Jaggard and Sun [1990], Lin, Lee, Lim and Lee [1995], and Sheppard [1996]. The fact that we are dealing with a randomly rough surface forces us to characterize it by certain statistical properties. Underlying this characterization is the assumption that there is not a single function ζ(x ). Instead there is an ensemble of realizations of this function. Physical properties associated with a randomly rough surface are to be averaged over this ensemble, and it is assumed that this ensemble average does not differ from the spatial average over a single, sufficiently large, surface. 2.1.1. Gaussian random surfaces Most of the existing studies of the scattering of light from a random surface are based on the assumption that the surface profile function ζ(x ) is a zero-mean, strictly stationary, isotropic, Gaussian random process. The characteristic functional of a random process ζ (x ) is defined in general by the formula Φ[u] = exp i d2 x u(x )ζ (x ) , (2.2) where the angle brackets denote an average over the ensemble of realizations of ζ (x ) and u(x ) is an arbitrary function. Assuming that (j ) u(x ) = (2.3) λj δ x − x , j (j )
where δ(x ) is the Dirac delta-function, and that the {x } constitute an arbitrary array of points in the plane x3 = 0, we obtain for Φ[u] the expression (j ) λj ζ x , Φ[u] = exp i (2.4) j
so that multipoint moments can be obtained by differentiation, (1) (2) (k) 1 ∂ k Φ[u] ζ x ζ x · · · ζ x = k . i ∂λ1 ∂λ2 · · · ∂λk {λj }=0
(2.5)
In the case that ζ (x ) is a zero-mean Gaussian random process, the functional Φ simplifies to (Rytov, Kravtsov and Tatarskii [1989], p. 29)
1 d2 x d2 x u(x ) ζ(x )ζ(x ) u(x ) . Φ[u] = exp − (2.6) 2
2, § 2]
Characterization of randomly rough surfaces
123
When eq. (2.6) is combined with eq. (2.3), and eq. (2.5) is used, it is readily found that multipoint moments are given by (1) (α1 ) (α2 ) (αk−1 ) (αk )
ζ x · · · ζ x(k) (2.7) = ζ x ζ x · · · ζ x ζ x d.p.
for even k and vanish for odd k, where d.p. denotes the sum over different permutations of pairs from the sequence 1, 2, . . . , k. If the surface profile function is a strictly stationary isotropic Gaussian random process, the two-point moment can be written in the form
ζ (x )ζ (x ) = δ 2 W |x − x | , (2.8) where
δ 2 = ζ 2 (x ) ,
(2.9)
so that δ is the rms height of the surface. The fact that the surface height autocorrelation function W (|x − x |) depends on the coordinates x and x only through their difference is a reflection of the assumed stationarity of ζ(x ), while its dependence only on the modulus of x − x is a reflection of the assumed isotropy of the random process ζ (x ). The surface height autocorrelation function W (|x |) possesses several important general properties. It follows from eqs. (2.8) and (2.9) that W (0) = 1. It is also not difficult to show that −1 W |x | 1,
(2.10)
(2.11)
by making use of the result [ζ (x ) ± ζ (x )]2 0. In addition, W (|x |) tends to zero as |x | → ∞. This property follows from the fact that on a randomly rough surface the heights of the surface at two widely separated points are uncorrelated. The transverse correlation length of the surface roughness, a, is defined as the distance over which the function W (|x |) decreases significantly from its value of unity at the origin. For calculational purposes it is convenient, often even necessary, to introduce the Fourier integral representation of the surface profile function, 2 d k ik ·x e ζˆ (k ), ζ (x ) = (2.12) (2π)2 where k is a two-dimensional wave vector, k = xˆ 1 k1 + xˆ 2 k2 . In the case that ζ (x ) is a zero-mean, stationary, isotropic, Gaussian random process, the Fourier
124
Multiple scattering of light from randomly rough surfaces
[2, § 2
coefficient ζˆ (k ) is also a zero-mean, stationary, isotropic, Gaussian random process. Its statistical properties are defined completely by the two-point moment
(2.13) ζˆ (k )ζˆ (k ) = (2π)2 δ k + k δ 2 g |k | . The function g(|k |) appearing in eq. (2.13) is called the power spectrum of the surface roughness, and is defined by g |k | = d2 x e−ik ·x W |x | . (2.14) The power spectrum g(|k |) is a non-negative function of |k |, and is normalized according to d2 k (2.15) g |k | = 1. 2 (2π) Although several analytic expressions have been assumed for W (|x |) (see, for example, Ogilvy [1991], pp. 13–14), perhaps the most commonly used representation is the Gaussian expression W |x | = exp −x2 /a 2 . (2.16) The 1/e decay length a entering this expression is the transverse correlation length for this model of surface roughness. The corresponding power spectrum is given by g |k | = πa 2 exp −a 2 k2 /4 . (2.17) 2.1.2. Non-Gaussian random surfaces Although the discussion of the statistical properties of random surfaces up to this point has been based on the assumption that the surface profile function ζ(x ) is a Gaussian random process, many surfaces of practical importance are defined by surface profile functions that are not Gaussian random processes. Well-known examples of such surfaces are a very rough sea (Beckmann [1973]), and terrain with sharp ridges and round valleys (Beckmann [1973]). Nevertheless, the scattering of light from non-Gaussian surfaces, especially multiple scattering effects from such surfaces, has received less attention than the scattering of light from Gaussian surfaces (Kivelson and Moszkowski [1963], Beckmann [1973], Berry [1973], Brown [1982], Jakeman and Hoenders [1982], Henyey [1983], Kim, Méndez and O’Donnell [1987], Wu, Chen and Fung [1988a, 1988b], Henyey, Creamer, Dysthe, Schult and Wright [1988], Michel, Maradudin and Méndez [1989], Creamer, Henyey, Schult and Wright [1989], Jakeman [1991a, 1991b], Tatarskii and Voronovich [1994], Tatarskii [1995a], Tatarskii and Tatarskii [1996]).
2, § 2]
Characterization of randomly rough surfaces
125
One of the chief difficulties in theoretical studies of non-Gaussian random surfaces is that there are not many random functions ζ(x ) for which the characteristic functional Φ[u] is known. In recent work Tatarskii [1995a] has shown how to generate a random function ζ (x ) whose characteristic functional has a non-Gaussian form, and has used the result to calculate an average of the form ζ (x ) exp[iαζ (x )], which arises in the small-slope approximation of the theory of rough surface scattering (Tatarskii and Tatarskii [1994]). Subsequently, Tatarskii and Tatarskii [1996] generalized the latter result and showed how this approach can be used to generate a two-dimensional non-Gaussian random surface for use in computer simulation studies of rough surface scattering. To our knowledge, such computer simulation calculations based on this model have been carried out yet. 2.2. One-dimensional random surfaces Not all random surfaces of interest in theoretical and experimental studies of rough surface scattering are two-dimensional. In recent years methods for fabricating highly one-dimensional random surfaces in the laboratory have been developed (Knotts and O’Donnell [1990], Méndez, Ponce, Ruiz-Cortés and Gu [1991]). This has made possible detailed comparisons between experimental results obtained for well-characterized one-dimensional random surfaces and results of theoretical calculations for the same surfaces, which are computationally much easier to perform than they are for two-dimensional random surfaces (Knotts, Michel and O’Donnell [1993], West and O’Donnell [1995], Maradudin, McGurn and Méndez [1995]). The height of a one-dimensional random surface above the plane x3 = 0 is given by the equation x3 = ζ (x1 ). The geometrical and statistical properties of the surface profile function ζ (x1 ) follow directly from those of the surface profile function ζ (x ) of a two-dimensional random surface on suppressing all references k2 in the Fourier to x2 in the coordinate representation, and to the wavenumber representation. In the latter representation all integrals d2 k f (k )/(2π)2 must ∞ be replaced by −∞ dk1 f (k1 )/(2π). On the assumption that ζ (x1 ) is a zero-mean stationary Gaussian random process, two forms of the surface height autocorrelation function W (|x1 |) will be considered in this article. These are the Gaussian form, W |x1 | = exp −x12 /a 2 , (2.18) and the West–O’Donnell [1995] form W |x1 | = sin(kmax x1 ) − sin(kmin x1 ) / (kmax − kmin )x1 ,
(2.19)
126
Multiple scattering of light from randomly rough surfaces
where kmax > kmin . The corresponding power spectra are √ g |k| = πa exp −a 2 k 2 /4
[2, § 3
(2.20)
and g |k| =
π θ |k| − kmin θ kmax − |k| , kmax − kmin
(2.21)
where θ (x) is the Heaviside unit step function. The manner in which kmin and kmax in eqs. (2.19) and (2.21) are defined will be discussed in Section 4.
§ 3. Equations for electromagnetic fields and observable quantities 3.1. Physical quantities studied in rough surface scattering problems In this section, we discuss scattering geometries commonly studied in the literature, and present basic equations and definitions for random fields and their statistical moments. The majority of existing theoretical calculations, studying multiple scattering of light from rough surfaces, has been done within the model of one-dimensional roughness. Such a model reduces the scattering problem to the solution of scalar, one-dimensional equations. As compared to the general twodimensional case, such a simplification reduces both computational cost and effort needed to understand the physics behind the mathematical formalism. Apart from the change of polarization in the scattering process, all multiple-scattering phenomena that exist in the two-dimensional case are present in the one-dimensional case. Thus, we will follow the trend of the existing literature and center our presentation on the one-dimensional case. The studies of scattering from twodimensional rough surfaces will be referenced in appropriate sections, and mathematical formalisms for the two-dimensional case will be introduced when necessary. Figure 1 illustrates two geometries that are most commonly considered in rough surface scattering investigations. A beam of light of frequency ω is incident from the vacuum side (x3 > ζ (x1 )) onto the randomly rough surface of a homogeneous dielectric medium (fig. 1a) or a film structure (fig. 1b). In linear problems, scattering materials are characterized by complex, frequency-dependent dielectric functions ε(ω) and εd (ω), with the case of a perfectly conducting substrate given by the limit ε(ω) → −∞. Assuming that the surface is one-dimensionally rough and that the plane of incidence is the x1 x3 -plane, one can decouple the scattering problems for p-polarized light, characterized by the magnetic field vector H(x1 , x3 ; t) = (0, H2 (x1 , x3 |ω), 0) e−iωt , and for s-polarized light, character-
2, § 3]
Equations for electromagnetic fields and observable quantities
127
Fig. 1. Scattering geometries commonly considered in rough surface scattering investigations.
ized by the electric field vector E(x1 , x3 ; t) = (0, E2 (x1 , x3 |ω), 0) e−iωt . Following the first theoretical study of coherent phenomena in the multiple scattering of light from rough surfaces (McGurn, Maradudin and Celli [1985]), we discuss the mathematical formulation of the rough surface scattering problem in the geometry shown in fig. 1a. The solution of Maxwell’s equations in the region x3 > ζ(x1 )max that represents the sum of an incident plane wave and a scattered field can be written in the form ∞ dq > inc R(q|k) eiqx1+α0 (q,ω)x3 , U (x1 , x3 |ω) = U (x1 , x3 |ω) + (3.1) −∞ 2π where U > (x1 , x3 |ω) = H2> (x1 , x3 |ω) for p-polarized light and U > (x1 , x3 |ω) = E2> (x1 , x3 |ω) for s-polarized light, U inc (x1 , x3 |ω) = eikx1 −iα0 (k,ω)x3 , 1/2 , |q| < ω/c, α0 (q, ω) = (ω/c)2 − q 2 1/2 = i q 2 − (ω/c)2 ,
|q| > ω/c,
(3.2) (3.3a) (3.3b)
and R(q|k) determines the amplitude of the wave scattered from the state with wavenumber k = (ω/c) sin θ0 into the state with wavenumber −∞ < q < ∞. The states with |q| < ω/c represent propagating waves that are characterized by the scattering angle θs (see fig. 1) with q = (ω/c) sin θs . In many cases, it is useful to introduce the scattering S-matrix (Brown, Celli, Coopersmith and Haller [1983], Arsenieva and Feng [1993], Sánchez-Gil, Maradudin and Méndez [1995]) 1/2 R(q|k), S(q|k) = α0 (q, ω)/α0 (k, ω)
(3.4)
which is useful in the analysis of such properties of the scattering theory as unitarity and reciprocity.
128
Multiple scattering of light from randomly rough surfaces
[2, § 3
The scattered intensity I (q|k) is defined in terms of R(q|k) or S(q|k) as (Malyshkin, McGurn, Leskova, Maradudin and Nieto-Vesperinas [1997a, 1997b]) I (q|k) =
2 2 1 ω 1 ω α0 (q, ω) R(q|k) = S(q|k) , L1 c α0 (k, ω) L1 c
(3.5)
where L1 denotes the length of the x1 -axis covered by the random surface. Another widely used characteristic of the angular distribution of the intensity of scattered light is the so-called differential reflection coefficient (DRC), ∂R/∂θs , which is defined such that (∂R/∂θs ) dθs is the fraction of the total time-averaged incident flux that is scattered into the angular interval (θs , θs + dθs ). In scattering from a random surface it is the average of the differential reflection coefficient over the ensemble of realizations of the surface profile function ∂R/∂θs that must be calculated, where here and in all that follows the angle brackets denote this ensemble average. Since specular scattering is not of interest in theoretical studies of enhanced backscattering, even at normal incidence, what is usually calculated is the contribution to the mean DRC from the light that has been scattered diffusely. This is given by 2
2 1 ω cos2 θs ∂R R(q|k) − R(q|k) , = (3.6) ∂θs diff L1 2πc cos θ0 where the contribution from the mean, specularly reflected field is subtracted from the total intensity. In addition to the first-order statistical moment (average) of the intensity of the scattered light, it is often useful to study its higher-order moments. The secondorder moment is known as the angular intensity correlation function and is defined by
C(q, k|q , k ) = I (q|k)I (q |k ) − I (q|k) I (q |k ) . (3.7) In some cases (Simonsen, Maradudin and Leskova [1999]), it is convenient to use the normalized angular intensity correlation function, which is equal to C(q, k|q , k ) divided by I (q|k)I (q |k ). Some authors also define and analyze the properties of unnormalized or normalized amplitude correlation functions, with I (q|k) in eq. (3.7) being replaced by the scattering amplitude R(q|k). Thus, both the DRC and correlation functions are expressed in terms of the scattering amplitude R(q|k). The calculation of R(q|k) is the central problem of rough surface scattering theory. We next outline two alternative methods which are most commonly used to obtain the equations for R(q|k) and note that these equations had been derived long before it became possible to solve them in the multiple-scattering regime.
2, § 3]
Equations for electromagnetic fields and observable quantities
129
3.2. Rayleigh method The Rayleigh method, which we discuss first, is based on the so-called Rayleigh hypothesis which states that the expressions for the fields in the regions x3 > ζ (x1 )max and x3 < ζ (x1 )min can be continued on to the surface itself (Lord Rayleigh [1896, 1907]). Thus, the expression (3.1) and an analogous expression for the field transmitted into the medium can be used in satisfying the boundary conditions at x3 = ζ (x1 ). Toigo, Marvin, Celli and Hill [1977] employed this hypothesis together with Green’s integral theorem and the extinction theorem to derive a single integral equation for the scattering amplitude R(q|k) in the case of p polarization: ∞ dq (+) M (p|q)R(q|k) = −M (−)(p|k), (3.8) −∞ 2π where M (±) (p|q) =
[pq ± α(p, ω)α0 (q, ω)]σ I α(p, ω) ∓ α0 (q, ω)|p − q , (3.9) α(p, ω) ∓ α0 (q, ω)
∞ I (γ |Q) = −∞ dx1 e−iQx1 e−iγ ζ(x1 ) , and α(q, ω) = [ε(ω)(ω/c)2 − q 2 ]1/2 with Re α(q, ω) > 0, Im α(q, ω) > 0. The exponent σ is unity for p-polarization and zero for s-polarization (the latter case was considered by Brown, Celli, Coopersmith and Haller [1983]). Equation (3.8) is called the reduced Rayleigh equation and it is exact within the limits of validity of the Rayleigh hypothesis. The limits of validity of the Rayleigh hypothesis have been explored by several researchers, primarily in the context of electromagnetic and acoustic scattering from deterministic surfaces. Petit and Cadilhac [1966], Millar [1971], Hill and Celli [1978], and Maystre [1984] established that the Rayleigh method is rigorous for sinusoidal surfaces with height to period ratios smaller than 0.07. Since the √ rms height of the sinusoidal surface is 2 times smaller than its amplitude, the range of validity of the Rayleigh method for randomly rough surfaces can be estimated as δ/D < 0.1,
(3.10)
where δ is the rms height and D is the mean distance between consecutive peaks and valleys of the√surface profile ζ (x1 ) (for the Gaussian correlation function (2.20), D = πa/ 6). However, even the early studies we have just mentioned suggest that the practical utility of the Rayleigh method can be extended well beyond the inequality (3.10). Several recent investigations (Soto-Crespo, NietoVesperinas and Friberg [1990], Madrazo and Maradudin [1997]) support this suggestion by comparing the predictions of the Rayleigh method with the exact nu-
130
Multiple scattering of light from randomly rough surfaces
[2, § 3
merical solution and show that the Rayleigh equations can provide accurate results for surfaces with δ/D as large as 0.47. Thus, the reduced Rayleigh equation gives a convenient mathematical formulation of the scattering problem for smooth surfaces. As we will see in Section 4.2, it provides a good starting point for both analytical (perturbative) and numerical calculations. One of the particular advantages of the reduced Rayleigh equation is that its structure (3.8) is generic to many geometries, and not only to the case of the scattering of light from a single, one-dimensional rough interface. In the general case of electromagnetic scattering from a two-dimensional rough surface x3 = ζ (x ), the electric vector is expressed in terms of the 2 × 2 matrix of scattering amplitudes, whose elements Rij (q |k ) describe scattering of a j -polarized wave with wavenumber k into an i-polarized wave with wavenumber q (j = p, s). The reduced Rayleigh equation (3.8) in this case becomes a (±) matrix integral equation. The matrix elements Mij (p |q ) which appear in this equation were derived by Brown, Celli, Haller and Marvin [1984]. A 2 × 2 matrix reduced Rayleigh equation for the amplitudes of the p- and s-polarized components of the transmitted field have been obtained by Greffet [1988] for the case in which light incident in one dielectric medium enters a second dielectric medium through a two-dimensional random interface. Furthermore, eq. (3.8) was shown to describe the scattering of light from a one-dimensional rough surface of a film in the geometry shown in fig. 1b for both p (Sánchez-Gil, Maradudin, Lu, Freilikher, Pustilnik and Yurkevich [1996]) and s (Sánchez-Gil, Maradudin, Lu, Freilikher, Pustilnik and Yurkevich [1994]) polarizations, when the film is free-standing or deposited on a perfectly conducting substrate. A reduced Rayleigh equation has been derived for the amplitude of an s-polarized electromagnetic field transmitted through a dielectric film whose illuminated face is a one-dimensional rough surface (Leskova, Maradudin, Méndez and Simonsen [2000]). The film is deposited on a planar surface of a substrate of a different dielectric material. A reduced Rayleigh equation for the amplitude of an s-polarized electromagnetic field transmitted through a dielectric film whose back surface is a one-dimensional rough surface while its illuminated surface is planar and which is sandwiched between two different dielectric media has been derived by Maradudin, Méndez, Leskova and Simonsen [2001]. In the case that the latter structure is illuminated by a p-polarized electromagnetic field, the reduced Rayleigh equation for the amplitude of the scattered field has been obtained by Leskova, Maradudin, Shchegrov and Méndez [1998]. It has also been proved to be possible to derive reduced Rayleigh equations for scattering of electromagnetic waves from a dielectric film whose illuminated surface is a two-dimensional random surface while its back surface is planar, and for scattering from a dielectric film whose illu-
2, § 3]
Equations for electromagnetic fields and observable quantities
131
minated surface is planar but whose back surface is a two-dimensional random surface (Soubret, Berginc and Bourrely [2001a]). In the former case the dielectric film was deposited on a perfectly conducting or a metallic substrate. In the latter case it was deposited on a metallic substrate. A set of reduced Rayleigh equations for the scattering of electromagnetic waves from a dielectric film, both of whose surfaces are two-dimensional randomly rough surfaces, has been obtained by Soubret, Berginc and Bourrely [2001b] in the case where the film is deposited on a metallic substrate. The derivation of reduced Rayleigh equations for the scattering of light from, and its transmission through, other multilayered planar structures with interfaces perturbed by one- or two-dimensional roughness appears to be a feasible task if the procedures developed for simpler structures are followed. 3.3. Surface integral equations for electromagnetic fields The mathematical formulation of the scattering problem provided by the Rayleigh method cannot be used for rough surfaces with large slopes. In this case, one has to consider the exact equations for the electromagnetic fields. We present here the exact formulation based on the use of Green’s second integral theorem (Huygens’ principle) in the half-spaces x3 > ζ (x1 ) and x3 < ζ (x1 ) (again, we focus on the geometry of fig. 1 in the one-dimensional roughness case). The scattering amplitude R(q|k) introduced in eq. (3.1) is expressed in terms of the surface values of the total field U(x1 ) ≡ U > (x1 , ζ(x1 )|ω) and its normal derivative V(x1 ) = (∂/∂N )U > (x1 , x3 |ω)|x3 =ζ(x1 ) through (Maradudin, Michel, McGurn and Méndez [1990]) ∞ i R(q|k) = dx1 e−iqx1 −iα0 (q,ω)ζ(x1 ) 2α0 (q, ω) −∞ × i qζ (x1 ) − α0 (q, ω) U(x1 ) − V(x1 ) . (3.11) The so-called source functions U(x1 ) and V(x1 ) obey a pair of coupled integral equations ∞ inc dx1 H0 (x1 |x1 )U(x1 ) − L0 (x1 |x1 )V(x1 ) , (3.12a) U(x1 ) = U (x1 ) + 0=
−∞
∞ −∞
dx1 Hε (x1 |x1 )U(x1 ) − νLε (x1 |x1 )V(x1 ) ,
where Hε (x1 |x1 ) = (i/4)(∂/∂N )H(1) 0 (1)
Lε (x1 |x1 ) = (i/4)H0
(3.12b)
ε(ω) (ω/c)ξ x =ζ(x ) ,
ε(ω) (ω/c)ξ x =ζ(x ) , 3
1
3
(3.13a)
1
(3.13b)
132
Multiple scattering of light from randomly rough surfaces
[2, § 3
∂/∂N = −ζ (x1 )(∂/∂x1 ) + (∂/∂x3 ), H0 (z) is the Hankel function of the first 1/2 , and η is a positive infinitesikind, ξ = (x1 − x1 )2 + (ζ (x1 ) − x3 + η)2 mal. The kernels H0 (x1 |x1 ) and L0 (x1 |x1 ) are obtained by setting ε(ω) = 1 in Hε (x1 |x1 ) and Lε (x1 |x1 ), respectively. Equations (3.11)–(3.13) are exact equations for both p and s polarizations, with ν in eq. (3.12b) equal to ε(ω) for ppolarization and 1 for s-polarization. The inhomogeneous term in eq. (3.12a) is given by U inc (x1 ) ≡ U inc (x1 , ζ(x1 )|ω). Since a numerical solution of eqs. (3.12) is especially difficult due to the infinite limits of integration, it is often useful to replace a plane incident wave (3.2) by a beam of finite width: π/2 2 2 wω − w ω (θ−θ0 )2 −i ω (x1 sin θ−x3 cos θ) inc dθ e 4c2 e c , (3.14) U (x1 , x3 |ω) = √ 2 πc −π/2 (1)
where w is the half-width of the beam. The half-width of the intercept of the beam with the plane x3 = 0 is g = w/ cos θ0 . The contribution from the diffusely scattered light to the mean DRC has to be modified as compared to eq. (3.6) 2 cos2 θs ω |R b (q|k)|2 − |R b (q|k)|2 ∂R , = (3.15) ∂θs diff (2π)3/2 wc pinc (θ0 ) wω π due to the change in incident flux. Here pinc (θ0 ) = 12 [erf( √ ( 2 − θ0 )) + 2c
erf( √wω ( π2 + θ0 ))] (Leskova, Maradudin and Shchegrov [1997a]), and R b (q|k) 2c is still given by eq. (3.11) – the superscript b is used to distinguish the scattering amplitude due to the incident beam (3.14) from the scattering amplitude R(q|k) due to plane wave illumination. A detailed derivation of eqs. (3.11)–(3.13) is given by Maradudin, Michel, McGurn and Méndez [1990]. Similar analysis can be implemented for a layered system shown in fig. 1b (Lu, Maradudin and Michel [1991]). In the onedimensional roughness case, each interface adds two source functions (the field and its normal derivative), therefore it also adds two integral equations. When one of the interfaces is perfectly conducting, either the field or its normal derivative vanish identically at that interface, thereby reducing the number of equations by one. Furthermore, for a planar perfectly conducting interface (as in fig. 1b) the method of images allows eliminating the integral equations at that interface altogether (Shchegrov and Maradudin [1999]). The scattering of light from a two-dimensional rough surface x3 = ζ(x ) has to be considered in its full vectorial formulation. This increases the number of unknown source functions and integral equations (Kong [1990], Tran and Maradudin [1994]). Due to its complexity scattering from two-dimensional rough surfaces is
2, § 4]
Weak localization effects in the multiple scattering of light
133
often studied in the cases of scattering of a scalar plane wave from Dirichlet (Tran and Maradudin [1992, 1993], Macaskill and Kachoyan [1993]) or Neumann (Tran and Maradudin [1993]) surfaces. § 4. Weak localization effects in the multiple scattering of light from randomly rough surfaces. Enhanced backscattering 4.1. The nature of enhanced backscattering effect We start the review of coherent multiple-scattering effects with a qualitative description of the most widely known of these phenomena – enhanced backscattering. The incorporation of multiple scattering into the theories of rough surface scattering began in the 1980s. In 1985, McGurn, Maradudin and Celli [1985] studied the scattering of p-polarized (transverse magnetic) light from a onedimensional randomly rough metal surface, and calculated the mean intensity of the diffusely scattered light by evaluating an infinite sum in powers of the surface profile function. The calculated angular dependence (3.6) of the mean intensity of the light that has been scattered diffusely revealed a well-defined peak in the retroreflection direction. The origin of this novel phenomenon of enhanced backscattering was linked to the coherent interaction of the multiply-scattered, p-polarized, surface electromagnetic waves (surface plasmon polaritons), supported by the vacuum–metal interface, with their reciprocal partners, in which the light and the surface plasmon polaritons interact with the roughness at the same points on the surface, but in the reverse order. This mechanism is illustrated in fig. 2. The partial waves, resulting from such reciprocal scattering events, have the same amplitude and phase if the wave vectors of the incident and scattered waves
Fig. 2. Illustration of the surface plasmon-polariton mechanism for the enhanced backscattering of light. The double-scattered reciprocal waves shown in the retroreflection direction interfere constructively since they have the same phase and the same amplitude. The excitation of intermediate surface waves (surface plasmon polaritons) is essential for backscattering enhancement.
134
Multiple scattering of light from randomly rough surfaces
[2, § 4
Fig. 3. The intensity of light scattered in plane from a considerably rough two-dimensional random surface measured as a function of the polar scattering angle. The incident light is s polarized, its wavelength is λ = 633 nm, and the polar angle of incidence is θ0 = −20◦ . (0) s → s scattering data; (+) s → p scattering data (O’Donnell and Méndez [1987]).
are oppositely directed, i.e. these partial waves interfere constructively. For scattering into directions other than the retroreflection direction, the different partial waves become dephased, so that only their intensities add. Thus, the intensity of scattering into the retroreflection direction is a factor of two larger than the intensity of scattering into those other directions, because of the interference terms that contribute to the intensity in the former case. The contribution of single-scattering processes usually significantly reduces this factor of two enhancement, since it is not subject to the coherent backscattering mechanism illustrated in fig. 2. The treatment by McGurn, Maradudin and Celli [1985] relied on the assumption of weakly rough surfaces (with typical amplitudes small compared to the wavelength) that support surface plasmon polaritons. However, the first experimental observation of the enhanced backscattering effect (Méndez and O’Donnell [1987], O’Donnell and Méndez [1987]) was done for surfaces with large roughness amplitudes. These experimental results (fig. 3) stimulated the development of non-perturbative, computational techniques for calculating the multiple scattering of light from very rough surfaces (see Section 4.2.2). This experimental and theoretical work led to the conclusion that the formation of enhanced backscattering from very rough surfaces does not require surface waves and can be interpreted as shown in fig. 4. The large amplitudes and slopes of the surface cause the multiple scattering of light. The intermediate waves traveling from one point of scattering to another are not necessarily surface waves.
2, § 4]
Weak localization effects in the multiple scattering of light
135
Fig. 4. A double-scattering sequence and its reciprocal partner that contribute to the enhanced backscattering of light incident normally on a very rough random surface. D is the distance between the first and last point on the surface struck by the light, and a is the average distance between consecutive maxima and minima on the surface (a can be approximately identified with the transverse correlation length for the Gaussian power spectrum (2.16)).
Fig. 5. The same geometry as in fig. 4, but showing the partial wave scattered into a direction described by the scattering angle θs .
Figures 2 and 4 illustrate the two main mechanisms for the enhanced backscattering of light from a randomly rough surface. The height of the enhanced backscattering peak on the background of the usually dominant single-scattering contribution is determined by the efficiency of one of these multiple-scattering mechanisms. The angular width δθ of the enhanced backscattering peak can be estimated by considering the phase difference φ between two wave paths shown in fig. 5 (where we assume the geometry of fig. 4). For normal incidence, φ = (2π/λ)θs D, where D is the distance between the first and the last points of the scattering sequence. The angular width θ of the main interference maximum centered at θs = 0 is given by twice the angular position θs = λ/2D of the first interference minimum, i.e. θ = λ/D.
(4.1)
136
Multiple scattering of light from randomly rough surfaces
[2, § 4
For the enhanced backscattering mechanisms depicted in figs. 4 and 5, the value of D can be estimated by the mean distance between consecutive hills and valleys on the surface, which is approximately equal to the roughness correlation length a (Maradudin and Michel [1990], Simeonov, McGurn and Maradudin [1997]). For the enhanced backscattering induced by surface waves (the case of fig. 2), D can be approximated by the mean free path l of a surface wave on a random surface. As we will see in the following sections, these estimates agree well with both theory and experiment. To complete our qualitative introduction to the enhanced backscattering effect, we stress that the averaging of the intensity of the scattered light over the ensemble of realizations of the surface profile function is essential for the observation of the effect. The angular dependence of the intensity of light scattered diffusely from any finite area of a rough surface represents a noisy sequence of bright and dark spots – a speckle pattern. In contrast to most peaks that disappear after taking the average over many realizations of the surface profile, the peak in the retroreflection direction remains. The nature of the enhanced backscattering from random surfaces allows relating this effect to a broader class of multiple-scattering phenomena called weak localization. Weak localization was originally studied in the context of electronic properties of disordered conductors (Abrahams, Anderson, Licciardello and Ramakrishnan [1979]). Electrons, which represent quantum wave packets, undergo multiple scattering from randomly distributed impurities. The coherent interference of partial waves due to reciprocal electron trajectories leads to the enhanced probability for an electron to return to its origin. Such weak localization of electrons results in the decrease of the electrical conductivity of a disordered sample. During the 1980s it was realized that weak localization is a general property of all waves in random media (a good discussion on the subject is presented in Sheng [1995]). In particular, the first experimental reports of the enhanced backscattering of light in media with volume disorder appeared in 1984–85 (Kuga and Ishimaru [1984], van Albada and Lagendijk [1985], Wolf and Maret [1985]), and were accompanied by theoretical interpretations describing coherent interference between reciprocal scattered waves (Tsang and Ishimaru [1984, 1985]) and its connection to weak localization (van Albada and Lagendijk [1985], Wolf and Maret [1985]). These investigations triggered the studies of multiple-scattering phenomena in rough surface scattering. Despite the weak localization origin of the enhanced backscattering effect from both random volumes and random surfaces, there is an essential difference. In volumes, a well-defined enhanced backscattering peak is formed after several hundreds or thousands of scattering events. In contrast, for surfaces, double-scattering processes are not
2, § 4]
Weak localization effects in the multiple scattering of light
137
only sufficient, but usually dominate the contribution of all multiple-scattering processes. The qualitative picture of the enhanced backscattering from randomly rough surfaces, presented in this section, is based on extensive theoretical and experimental work that followed the first efforts of the mid-1980s. In Section 4.2, devoted to theoretical techniques employed in the study of multiple scattering of light from randomly rough surfaces, we will gain deeper insight into the enhanced backscattering effect under different experimental conditions (discussed in Section 4.3). In particular, we will analyze the surface plasmon polariton mechanism for enhanced backscattering in Section 4.2.1 which reviews perturbative techniques for weakly rough surfaces. The other mechanism for enhanced backscattering (fig. 4) will be discussed in the context of numerical simulations used to study scattering from very rough surfaces. 4.2. Theoretical methods employed in the study of multiple-scattering phenomena We next review theoretical methods used in the study of enhanced backscattering of light from a randomly rough surface. The discussion will center on the geometry of a one-dimensional randomly rough surface of a homogeneous medium. However, the methods we will discuss are generic and are also used in investigating other multiple-scattering phenomena in systems with one-dimensional or two-dimensional rough surfaces (see Sections 5–8) with varying complexity of the computational effort required. 4.2.1. Weakly rough surfaces: perturbative techniques 4.2.1.1. Small-amplitude perturbation theory. The weak roughness limit has been at the focus of rough surface scattering theories since the beginning of the 20th century. In this limit, the amplitude of the scattered light is expanded in powers of the surface profile function ζ . The first-order term of this expansion describes single-scattering processes, and from the pioneering work of Mandel’shtam [1913] to the mid-1980s this approximation dominated theoretical investigations of rough surface scattering (Rice [1951], Brekhovskikh [1952], Bass and Fuks [1980]). Since the mid-1980s, several papers have considered the small-amplitude expansion of the scattering amplitude beyond the single-scattering approximation (see, e.g., Jackson, Winebrenner and Ishimaru [1988], Soto-Crespo, Nieto-Vesperinas and Friberg [1990], Maradudin and Méndez [1993], Shchegrov [1998b], O’Donnell, West and Méndez [1998], O’Donnell [2001]). Here we out-
138
Multiple scattering of light from randomly rough surfaces
[2, § 4
line the method developed by Shchegrov [1998b] that allows calculating the scattering amplitude and intensity recursively to any order in the surface profile function. We start from the reduced Rayleigh equation (3.8) and expand the functions M ± (p|q) in powers of the surface profile function, M (±) (p|q) = where
∞
ˆ (n) (p − q), µ(±) n (p|q)ζ
n=0 ∞ ζˆ (n) (Q) = −∞ dx1 e−iQx1 ζ n (x1 )
(4.2)
and
n
µ(±) n (p|q) = (−i) pq ± α(p, ω)α0 (q, ω)
σ
α(p, ω) ∓ α0 (q, ω)
n−1
,
(4.3) with the same convention for σ as in eq. (3.9). Then, we seek the scattering amplitude in the form of an integral series expansion in powers of ζ , R(q|k) = 2πδ(q − k)R (0) (k) + χ (1) (q|k)ζˆ (q − k) ∞ ∞ 1 ∞ dp1 dpn−1 (n) + ··· χ (q|p1 , p2 , . . . , pn−1 |k) n! −∞ 2π −∞ 2π n=2
× ζˆ (q − p1 )ζˆ (p1 − p2 ) · · · ζˆ (pn−1 − k),
(4.4)
(−) (+) where R (0) (k) = −µ0 (p|q)/µ0 (p|q)
is the amplitude Fresnel reflection coefficient for a planar surface. Substituting eqs. (4.3)–(4.4) into eq. (3.8) and equating the terms that have the same powers of ζ , we find that the functions χ (n) can be calculated recursively, χ (n) (q|p1 , p2 , . . . , pn |k) n−1 n (n−m) = ξ (n) (q|k) − (q|pn−m )χ (m) (pn−m |pn−m+1 , . . . , pn−1 |k), η m m=1 (4.5) (−)
(+)
(+)
where ξ (n) (q|k) = −[µn (q|k) + µn (q|k)R0(k)]/µ0 (q|q), η(n) (q|k) = n (+) µ(+) n (q|k)/µ0 (q|q), and m is the binomial coefficient. It follows from eq. (3.6) that ∂R/∂θs diff is given by 2 ω cos2 θs 2 (1) ∂R = δ χ (q|k) g |q − k| ∂θs diff 2πc cos θ0 ∞ ∞ ∞ dp1 dpn−1 (2n) 2n ··· Λ (q|p1 , . . . , pn−1 |k) . δ + −∞ 2π −∞ 2π n=2
(4.6)
2, § 4]
Weak localization effects in the multiple scattering of light
139
Fig. 6. A plot of ∂R/∂θs diff as a function of θs for p-polarized light of wavelength λ = 457.9 nm incident normally on a one-dimensional random silver surface (ε(ω) = −7.5 + i0.24) characterized by the Gaussian power spectrum (2.20) with δ = 5 nm and a = 100 nm. The calculations are based on eq. (4.6) (solid curve) and on eq. (4.37) (dashed curve). The inset shows ∂R/∂θs diff in the vicinity of the retroreflection direction (Maradudin and Méndez [1993]).
The analytical calculation of the functions Λ(2n) is done directly from eqs. (3.6) and (4.4) using the averaging procedure described in Section 2. This task is rather straightforward but increasingly tedious for higher-order terms. The singlescattering term, of O(δ 2 ) is written explicitly in eq. (4.6), and the next, multiplescattering term O(δ 4 ) can be found in the works by Leskova, Maradudin and Shchegrov [1997b] and Shchegrov [1998b]. The contributions to ∂R/∂θs diff from a selected group of terms of O(δ 6 ) has been obtained by O’Donnell, West and Méndez [1998]. In something of a tour-de-force, O’Donnell [2001] has calculated all the terms of O(δ 6 ) and of O(δ 8 ). In the scattering of p-polarized light from a weakly rough one-dimensional random metal surface that supports surface plasmon polaritons, the enhanced backscattering peak first arises in the contribution to ∂R/∂θs diff of the fourth order in δ, as shown in fig. 6 for the case that the surface roughness is characterized by a Gaussian power spectrum (2.20). The peak arises from the term in Λ(4) (q|p1|k) in the integrand in eq. (4.6) that contains the product χ (2) (q|p1 |k)χ (2)∗(q|q + k − p1 |k), and is associated with the coherent interfer-
140
Multiple scattering of light from randomly rough surfaces
[2, § 4
Fig. 7. A plot of ∂R/∂θs diff as a function of θs when p-polarized light of wavelength λ = 612 nm is incident normally on a one-dimensional random gold surface (ε(ω) = −9.00 + i1.29) characterized by the West–O’Donnell power spectrum (2.21) with δ = 10.9 nm, kmin = 8.4815 × 10−3 nm−1 and kmax = 13.260 × 10−3 nm−1 . The calculation of ∂R/∂θs diff was carried out on the basis of eq. (4.6) where terms through 0(δ 4 ) were retained (after Maradudin, McGurn and Méndez [1995]).
ence of a doubly-scattered surface plasmon polariton and its reciprocal partner (fig. 2). Enhanced backscattering of light from a one-dimensional random metal surface characterized by a Gaussian power spectrum has yet to be observed experimentally. The visibility of the enhanced backscattering peak is especially high for the West–O’Donnell surfaces, described in Section 2. The power spectrum of these surfaces (eq. (2.21)) forces the incident wave to be scattered only into states with wavenumbers kmin < q < kmax (for normal incidence). By choosing kmin and kmax to lie close to the left and to the right of the wavenumber ksp (ω) of a surface plasmon polariton of frequency ω on a planar surface, we force the system into the multiple-scattering regime. Figure 7 illustrates the angular dependence of the intensity of diffusely scattered p-polarized light calculated to fourth order in δ. The single-scattering contribution ∼ δ 2 appears as two lobes at large scattering angles, while the central part is completely dominated by multiple-scattering processes. Since the single-scattering contribution is absent in the retroreflection direction, the height of the enhanced backscattering peak in this case is twice that of the background. In experiments in which p-polarized light was scattered from a weakly rough one-dimensional random silver surface characterized by a West–O’Donnell power spectrum that is significant at twice the surface plasmon polariton wave number, 2ksp(ω), O’Donnell, West and Méndez [1998] observed an enhanced backscattering peak at large angles of incidence (60◦ –76◦ ) (fig. 8).
2, § 4]
Weak localization effects in the multiple scattering of light
141
Fig. 8. Measurements at three angles of incidence of the mean diffuse intensity Ip (θs ) in the scattering of p-polarized light of wavelength 1152 nm (ε(ω) = −61.0 + i6.2) from a one-dimensional random silver surface characterized by the West–O’Donnell power spectrum (2.21) with kmin = 9.93 × 10−3 nm−1 , kmax = 1.49 × 10−2 nm−1 , and δ = 11.1 nm (after O’Donnell, West and Méndez [1998]).
This peak is not predicted by the small-amplitude perturbation theory when the expansion of ∂R/∂θs diff in powers of δ 2 is terminated at the fourth-order term (Maradudin and Méndez [1993]). However, the calculation of the contribution to ∂R/∂θs diff from a selected set of terms of sixth order in δ by these authors predicts this peak. In his determination of ∂R/∂θs diff through terms of O(δ 8 ), O’Donnell [2001] observed an enhanced specular peak in the contribution of O(δ 8 ) when the weakly rough one-dimensional random surface was characterized by a West–O’Donnell power spectrum (2.21) and a sufficiently large value of δ was employed (fig. 9). This feature is not present in the contributions to ∂R/∂θs diff of lower orders in δ. This peak has not been observed experimentally up to the present time. These results illustrate the point that interesting new effects can be obtained when small-amplitude perturbation theory is extended to higher and higher orders. The small-amplitude perturbation expansion applied to the scattering of s-polarized light in the same geometries that exhibit the enhanced backscattering effect for p-polarized light show no enhanced backscattering. This is because s-polarized surface plasmon polaritons do not exist on a vacuum–metal interface (see, e.g., Maradudin [1982]). Thus, the small-amplitude perturbation theory provides a simple and fast approach to investigating multiple-scattering phenomena. It allows decoupling and analyzing contributions from single, double, and higher order scatter-
142
Multiple scattering of light from randomly rough surfaces
[2, § 4
(8)
Fig. 9. The calculated enhanced specular peak in the contribution Ip (θs |θ0 ) of eighth order in δ to the mean diffuse intensity in the scattering of p-polarized light of wavelength 457.9 nm (estimated ε(ω) = −9.5 + i0.24) from a random metal surface characterized by the West–O’Donnell power spectrum (2.21) with kmin = 1.70(ω/c), kmax = 2.25(ω/c), and δ/λ = 3 × 10−3 . The angle of incidence is θ0 = 52◦ (after O’Donnell [2001]).
ing processes to the scattered field and intensity – and such an advantage is especially important since in most practical cases the contribution from double-scattering processes greatly exceeds the contribution from other multiplescattering processes (Maradudin and Méndez [1993]). A rather subtle issue concerning the small amplitude perturbation theory is its limit of validity. The first and obvious requirement is the smallness of the rms roughness amplitude δ compared to the wavelength λ of the incident light: δ/λ 1. However, the derivation of the small-amplitude perturbation expansion is typically based on the Rayleigh equations (see Eqs. (4.2)–(4.6)). Therefore, this expansion is often believed to be valid only as long as the Rayleigh hypothesis is valid, i.e. the rms slope of the surface should also be small (Ogilvy [1991]). However, recent investigations (Jackson, Winebrenner and Ishimaru [1988], Tatarskii [1995b]) suggest that the small-amplitude perturbation series derived from the Rayleigh equations and from the exact surface integral equations based on Green’s theorem, coincide. Further, Tatarskii [1995b] argued that the only criterion for the validity of the small-amplitude series is the convergence of this series, which condition does not necessarily require the validity of the Rayleigh hypothesis. 4.2.1.2. Self-energy (many-body) perturbation theory. A somewhat deeper insight into weak localization processes on a weakly rough metal surface and into the nature of the perturbation parameter is provided by another perturbative ap-
2, § 4]
Weak localization effects in the multiple scattering of light
143
proach developed by McGurn, Maradudin and Celli [1985] where the first theoretical prediction of the enhanced backscattering from randomly rough surfaces was made. In this method, in the case of p-polarized incident light, one seeks the scattering amplitude in the form R(q|k) = 2πδ(q − k)R (0) (k) − 2iG(0)(q)T (q|k)G(0)(k)α0 (k),
(4.7)
where the functions α(k) and α0 (k) are the same as the functions α0 (k, ω) and α(k, ω) entering eq. (3.9) with the argument ω suppressed for brevity. The function G(0) (k) = iε(ω)/[ε(ω)α0 (k) + α(k)] is a Green’s function for surface plasmon polaritons at a planar surface, while the transition matrix T (q|k) is postulated to satisfy ∞ dq V (q|p)G(0)(p)T (p|k). T (q|k) = V (q|k) + (4.8) 2π −∞ Equations (3.8), (4.7), and (4.8) define the scattering potential V (q|k) which is the solution of the equation ∞ V (q|k) dp (+) M (q|p) − M (−) (q|p) 2iα0 (p) −∞ 2π (+) = M (q|k) ε(ω)α0 (k) − α(k) − M (−) (q|k) ε(ω)α0 (k) + α(k) ×
1 , 2ε(ω)α0 (k)
(4.9)
where M (±) (q|p) has been defined by eqs. (3.9). To first order in the surface profile function the solution of eq. (4.9) is V (1) (q|k) = i
ε(ω) − 1 ε(ω)qk − α(q)α(k) ζˆ (q − k). 2 ε (ω)
(4.10)
The use of the approximation to the scattering potential given by eq. (4.10) is called the small-roughness approximation (Toigo, Marvin, Celli and Hill [1977]). Equation (4.8) brings the scattering problem to the form known in the standard wave scattering theory (Sheng [1995]). We next introduce Green’s function for surface plasmon polaritons on the randomly rough metal surface as the solution of ∞ dp (0) (0) V (q|p)G(p|k) G(q|k) = 2πδ(q − k)G (k) + G (q) −∞ 2π = 2πδ(q − k)G(0) (k) + G(0) (q)T (q|k)G(0)(k).
(4.11)
Recalling eq. (4.7), we find that R(q|k) = −2πδ(q − k) − 2iG(q|k)α0(k).
(4.12)
144
Multiple scattering of light from randomly rough surfaces
[2, § 4
An important role in the theory being described here is played by the averaged Green’s function G(q|k). Due to the stationarity of the surface profile function G(q|k) is diagonal in q and k,
G(q|k) = 2πδ(q − k)G(k), (4.13) where G(k) is related to the unperturbed Green’s function G(0) (k) by G(k) =
1 G(0) (k)−1
− M(k)
.
(4.14)
The function M(k) in this expression is a self-energy that can be calculated as an expansion in powers of ζ (x1 ) as a solution of the equations
M(q|k) = 2πδ(q − k)M(k), (4.15a) ∞
dp M(q|k) = V (q|k) + M(q|p)G(0)(p) V (p|k) − V (p|k) . (4.15b) −∞ 2π Just as G(0) (k) has simple poles at the wave numbers ±ksp of the surface plasmon polaritons of frequency ω propagating along a planar vacuum–metal interface, G(k) has simple poles at the wavenumbers of the surface plasmon polaritons of frequency ω propagating along the randomly rough vacuum–metal interface. This result can be exploited to simplify calculations in which G(k) appears in the integrand of some integral, by the introduction of the pole approximation to G(k), G(k) ∼ =
C C − , k − ksp − i k + ksp + i
(4.16)
where C=
|ε1 (ω)|3/2 , ε12 (ω) − 1
(4.17)
ε1 (ω) = Re ε(ω), ε2 (ω) = Im ε(ω), and can be written as = ε + sp ,
(4.18)
with ε =
ε2 ksp , 2|ε1(ω)|(|ε1 | − 1)
sp = C Im M(ksp ).
(4.19a) (4.19b)
The function ε is the decay rate of the surface plasmon polariton as it propagates along the surface due to ohmic losses, i.e. due to the imaginary part of the dielectric constant, while sp is the decay rate of the surface plasmon polariton due to its roughness-induced conversion into volume electromagnetic waves
2, § 4]
Weak localization effects in the multiple scattering of light
145
in the vacuum and into other surface plasmon polaritons. In the small-roughness approximation sp is given by ∞
dp (1) 1 sp = C Im (4.20) V (k|p)G(p)V (1) (p|k) . L1 −∞ 2π In writing eq. (4.16) we have neglected the small renormalization of ksp arising from the real part of M(k), but have taken into account that sp can be comparable to ε . The use of eq. (4.12) in eq. (3.6) yields 2
2 1 2 ω3 ∂R = cos2 θs cos θ0 G(q|k) − G(q|k) . (4.21) 3 ∂θs diff L1 π c The two-particle Green’s function |G(q|k)|2 is given by the Bethe–Salpeter equation
G(q|k)G∗ (q|k) 2 2
= G(q|k) G∗ (q|k) + G(q) τ (q, q|k, k) G(k) , (4.22) where τ (q, q|k, k) is the reducible vertex function. The general reducible vertex function τ (q, q |k, k ) is related to the irreducible vertex function Γ (q, q |k, k ) by
τ (q, q |k, k ) = Γ (q, q |k, k ) ∞
dp + Γ (q, q |p, p ) G(p)G∗ (p ) τ (p, p |k, k ) . 2π −∞ (4.23) We approximate the irreducible vertex function by the sum of the contributions from all maximally-crossed diagrams, which describe the phase-coherent multiple-scattering processes that give rise to enhanced backscattering. It is given in the small roughness approximation by
Γ (q, q |k, k ) K(q + k) 2 2 φ(k) , = 2πδ(q − k − q + k ) W g |q − k| + W φ(q) 1 − K(q + k) (4.24) where
∞
dp 2 φ (p)G(p)G∗ (Q − p), 2π −∞ √ 1/2 φ(p) = π a exp −a 2 p2 /4 , K(q) = W
2
(4.25) (4.26)
146
Multiple scattering of light from randomly rough surfaces
[2, § 4
1 − ε(ω) ω2 W =δ . (4.27) ε(ω) c2 In obtaining the result given by eq. (4.24) several approximations were made in addition to the small-roughness approximation. The first is that the contribution to the scattering potential V (q|k) that is linear in the surface profile function ζ(x1 ), eq. (4.10), has been approximated by 1 − ε(ω) ω2 (4.28) ζˆ (q − k), ε(ω) c2 which is valid for small q and k. The second is that the imaginary part of the dielectric function ε(ω) has been neglected. The third is that the power spectrum of the surface roughness g(|q − k|) has been approximated by Antsygina, Freilikher, Gredeskul, Pastur and Slusarev [1991] as √ g |q − k| = π a exp −a 2 (q − k)2 /4 φ(q)φ(k). (4.29) V (1) (q|k)
This is a good approximation when g(|q − k|) is convolved with a function f (k) that is more sharply peaked than g(|k|) is. These approximations introduce small quantitative errors into the result for Γ (q, q |k, k ), but preserve its qualitative features while enabling an analytic result to be obtained. The result given by eq. (4.24) has to be substituted into eq. (4.23), which is then solved by iteration. However, in each of the integral terms of the iterative solution only the contribution associated with the first term on the right-hand side of eq. (4.24) is kept, and all contributions that contain the second term are omitted. The sum of the resulting integral terms is given by K(q − q ) φ(k). (4.30) 1 − K(q − q ) This is just the sum of the contributions associated with all the ladder diagrams. It equals the contribution of the second term on the right-hand side of eq. (4.24) when q = −k. Therefore, it cannot be neglected in comparison with the latter contribution. When this result is combined with the one given by eq. (4.24) we obtain as our approximation to the reducible vertex function
τ (q, q |k, k ) 2πδ(q − k − q + k )W 2 φ(q)
= 2πδ(q − k − q + k )
K(q − q ) K(q + k) + φ(k) . × W 2 g |q − k| + W 2 φ(q) 1 − K(q + k) 1 − K(q − q ) (4.31) The content of the result given by eq. (4.31) is made clearer if we evaluate the function K(Q) with the aid of the pole approximation (4.16) for the Green’s func-
2, § 4]
Weak localization effects in the multiple scattering of light
147
tion G(k). This approximation yields the result that G(p)G∗ (Q − p) = −
2i 2i C2 C2 + 2 2 Q − 2i (p − ksp ) + Q + 2i (p + ksp)2 + 2
2πC 2 2πiC 2 δ(p − ksp) + δ(p + ksp ), Q − 2i Q + 2i in the limit of small Q. With this result K(Q) becomes −
(4.32)
4sp (4.33) Q2 + 42 since, with the approximations made in the present calculation we find from eq. (4.20) that K(Q) =
sp = C 2 W 2 φ 2 (ksp ).
(4.34)
With the use of eq. (4.33) the reducible vertex function becomes
τ (q, q |k, k ) = 2πδ(q − k − q + k ) 4sp 4sp × W 2 g |q − k| + W 2 φ(q) + (q + k)2 + 4ε (q − q )2 + 4ε
× φ(k) . (4.35) When we set q = q and k = k in eq. (4.35) and combine the result with eqs. (4.21) and (4.22), we obtain finally for ∂R/∂θs diff 2 ω3 ∂R 2 G(q)2 = cos θ cos θ s 0 ∂θs diff π c3
sp 4sp 2 2 φ(k) × W g |q − k| + W φ(q) + ε (q + k)2 + 4ε 2 × G(k) . (4.36) The first term on the right-hand side of eq. (4.36) is the single-scattering contribution, the second is the contribution from the ladder diagrams, and the third is the contribution from the maximally-crossed diagrams. The latter two terms are equal for scattering into the retroreflection direction, i.e. when q = −k or θs = −θ0 . The prediction of enhanced backscattering is clearly seen through the presence of the denominator (q + k)2 + 4ε in the last term on the right-hand side of eq. (4.36), which is peaked at q = −k. This term, which is the contribution of the maximally crossed diagrams, expresses the consequence of the coherent interference of multiply-scattered surface plasmon polaritons with their reciprocal partners.
148
Multiple scattering of light from randomly rough surfaces
[2, § 4
We have presented the derivation of eq. (4.36) in some detail because the approach used and the approximations made will enable explicit results for angular intensity correlation functions to be obtained in Section 5. However, if the approximations made in obtaining the result given in eq. (4.36) are not made, in particular if the approximations given by eqs. (4.28) and (4.29) are not made, the contribution to the mean differential reflection coefficient from the diffuse component of the scattered light is still given by a very similar expression (McGurn, Maradudin and Celli [1985]), namely ∂R ∂θs diff 3 2 ω =4 cos2 θs cos θ0 G(q) c
A(q|k) A(−k|k) C 2 2 C2 , + × K(q|k) + 2 1 − (sp /)2 (q + k)2 + 42 1 − (sp /)2 (4.37) where K(q|k) = − − while A(q|k) is a smoothly varying function of q and k. An explicit expression for it is given by McGurn, Maradudin and Celli [1985]. Again, the first term on the right-hand side of eq. (4.37) is the single-scattering contribution, the second is the contribution from the ladder diagrams, and the third is the contribution from the maximally-crossed diagrams. The latter two terms are equal for scattering into the retroreflection direction q = −k. An important difference of the described approach, sometimes referred to as the self-energy perturbation theory (Sánchez-Gil, Maradudin and Méndez [1995]) from the small-amplitude perturbation theory is that the expansion of the scattering amplitude is made in powers of the scattering potential V (q|k) and not just in powers of the surface profile. The scattering potential is calculated to ensure the unitary and reciprocal character of the scattering formalism (Brown, Celli, Coopersmith and Haller [1983]) – these properties are not manifestly present in the small-amplitude expansion, although that theory is unitary and reciprocal. Further, a careful analysis shows that the infinite perturbation series summed to obtain eq. (4.31) has the form of perturbation series in powers of a small parameter sp /. Thus, the perturbative expansion assumes that the ohmic losses exceed the roughness-induced scattering losses for a surface plasmon polariton. When the condition sp / 1 is satisfied, the results expressed by eq. (4.6) through terms of fourth order in δ and eq. (4.37) practically overlap each other, as can be seen in fig. 6 (Maradudin and Méndez [1993]), indicating that the fourth-order δ 2 g(|q
k|)|[(ε(ω) − 1)/ε2 (ω)][ε(ω)qk
α(q)α(k)]|2 ,
2, § 4]
Weak localization effects in the multiple scattering of light
149
perturbation theory indeed captures the essential behavior predicted in the infinite order perturbation theory. The small differences can be attributed to two reasons. The first reason is the truncation of the scattering potential V (q|k) in eq. (4.10) at the linear term in ζ . The second reason is that the dispersion and decay of a surface plasmon polariton due to roughness effects are not renormalized in the small amplitude perturbation theory. This explains the observed difference in the widths of the enhanced backscattering peaks predicted by the two perturbation theories – as we have seen in Section 4.1, the width is determined by the mean free path of a surface plasmon polariton. It is worth pointing out that the result for ∂R/∂θs diff given by eq. (4.37) is able to predict an enhanced backscattering peak for large angles of incidence, which small-amplitude perturbation theory through terms of fourth order in δ is unable to do. The results of the two perturbation theories, eqs. (4.6) and (4.37), are also in good agreement with the experimental results of West and O’Donnell [1995] (fig. 10), which are the only experimental results to show the enhanced
Fig. 10. A plot of ∂R/∂θs diff as a function of θs for three angles of incidence when p-polarized light of wavelength λ = 612 nm is incident on a one-dimensional random gold surface (ε(ω) = −9.00 + i1.29) characterized by the West–O’Donnell power spectrum (2.24) with δ = 10.9 nm, kmin = 8.4815 × 10−3 nm−1 , and kmax = 13.200 × 10−3 nm−1 . (0) Experimental data of West and O’Donnell [1995]; (−) results based on eq. (4.6); () results based on eq. (4.37) (Maradudin, McGurn and Méndez [1995]).
150
Multiple scattering of light from randomly rough surfaces
[2, § 4
backscattering effect in scattering from weakly rough metal surfaces, where the mechanism responsible for the effect is the multiple scattering of the surface plasmon polaritons excited by the incident light. The limit sp / 1, which is not described by the presented theories, requires a fully self-consistent treatment of the scattering problem and is described by Celli, Maradudin, Marvin and McGurn [1985] for the one-dimensional case and by McGurn and Maradudin [1987] in the two-dimensional case. The enhanced backscattering effect was predicted for random surfaces of both dimensionalities. Other theoretical studies of the surface plasmon polariton-assisted enhanced backscattering from a weakly rough metal surfaces were done by Hanato, Ogura and Wang [1997] for the one-dimensional case, and McGurn and Maradudin [1996], Hanato, Ogura and Wang [1997], and Johnson [1999] for the two-dimensional case, with results showing agreement with the studies cited above. To conclude discussion of the enhanced backscattering effect from weakly rough surfaces, we point out that this effect is not specific only to metallic surfaces. In a theoretical work by Leskova, Maradudin and Shchegrov [1997b] it was shown that the effect can be observed for weakly rough surfaces of dipole-active media which support surface phonon polaritons or surface exciton polaritons. 4.2.2. Very rough surfaces: scattering simulations 4.2.2.1. Simulations based on the surface integral equations. The first experimental observations of the enhanced backscattering of light from random surfaces (Méndez and O’Donnell [1987], O’Donnell and Méndez [1987]) were carried out for very rough metal surfaces, for which the perturbation theories described in Section 4.2.1 are invalid. The theoretical approach that has become most commonly used in dealing with such surfaces is based on a numerical solution of the surface integral equations (3.12a), (3.12b), which are exact for surfaces with arbitrary geometrical properties. In this approach, the infinite range of integration in the integral terms of eqs. (3.12a), (3.12b) is replaced by a finite one (−L1 /2, L1 /2), since computers cannot deal with infinitely long surfaces. Then, a realization of a segment of a random surface profile x3 = ζ (x1 ) that covers the interval (−L1 /2, L1 /2) is generated numerically. In most cases, the surface is divided into N intervals of equal length x = L1 /N . Two algorithms for generating random realizations of x3 = ζ (x1 ) on a discrete mesh given the values of the rms roughness amplitude δ and the form of the correlation function can be found in the papers by Maradudin, Michel, McGurn and Méndez [1990] and by Freilikher, Kanzieper and Maradudin [1997] (the appendix of the latter article describes an effective
2, § 4]
Weak localization effects in the multiple scattering of light
151
but, to our knowledge, unpublished method suggested by V.I. Tatarskii). After the random surface is generated, the integrals in eqs. (3.12a), (3.12b) are approximated by finite sums of N terms each. Thus, the integral equations are replaced (n) with the matrix equations for the values of the field Un = U(x1 ) and its normal (n) derivative Vn = V(x1 ) at the midpoint x1(n) = −L + n − 12 x,
n = 1, . . . , N,
(4.38)
of each interval. To minimize the edge effects caused by the truncation of the integrals, the incident field is usually chosen to represent a beam of a finite width (eq. (3.14)). The resulting equations have the form Um = 2Uminc −
N
(0) Un − L(0) Hmn mn Vn ,
(4.39a)
n=1 N (ε) Hmn Un − νL(ε) Um = − mn Vn ,
(4.39b)
n=1
where the factor of two before the incident field term Uminc = U inc (x1(m) ) appears due to the contribution from the singular integration point x1 = x1 in eq. (3.12a). Expressions for the matrices H(0,ε) and L(0,ε) are obtained from eq. (3.13) and can be found in the paper by Maradudin, Michel, McGurn and Méndez [1990]. The system (4.39) of 2N equations is then solved numerically, and the source functions U and V obtained are substituted into the integral expression (3.11) for the scattering amplitude, which is also converted into a finite sum. The procedure must be repeated for many realizations of the surface profile function so that the average characteristics of the intensity of diffusely scattered light can be obtained. Due to the statistical nature of this computational method it is often referred to as scattering Monte Carlo simulations, while the general approach of discretizing the scattering system to convert the integral equations into matrix equations is known as the method of moments. This numerical simulation approach was first implemented for perfectly conducting surfaces by Nieto-Vesperinas and Soto-Crespo [1987]. Within the model of a perfect conductor, one has to solve only N coupled equations given by eq. (4.39a), since one of the source functions vanishes identically (due to the Dirichlet boundary condition U ≡ 0 for s-polarization or the Neumann boundary condition ∂U/∂n ≡ 0 for p-polarization). The enhanced backscattering effect was found for both s and p polarizations for surfaces with the Gaussian correlation function and typical values of the roughness rms amplitude δ and correlation
152
Multiple scattering of light from randomly rough surfaces
[2, § 4
length a of the order of the wavelength λ. As expected from the qualitative argument given in Section 4.1, the angular width of the enhanced backscattering peak turned out to be approximately equal to λ/a. The early work cited was followed by numerous subsequent investigations, including those done for one-dimensional surfaces of penetrable media (Tran and Celli [1988], Maradudin, Méndez and Michel [1989], and many others) and a few investigations of scattering from two-dimensional rough surfaces (Tran and Maradudin [1992, 1993]). These studies revealed several points that are essential in optimizing the speed and reliability of the numerical simulation approach as well as some criteria for the existence of pronounced coherent multiple-scattering phenomena. We summarize the most important of these findings below. The first issue concerns the optimal choice of the simulation parameters characterizing the surface (L1 , N , and x = L1 /N ) and the incident beam (half-width of the illuminated region g) for a given frequency ω (or wavelength λ = 2πc/ω), and the number of realizations Np of the surface profile. The discretization interval x must be small compared to λ to properly model wave propagation and scattering and small compared to the correlation length a to properly model the details of the surface profile. On the other hand, the length L1 needs to be large compared to a and to w (it is usually assumed that w λ). The achievement of the two obvious goals of minimizing x and maximizing L1 is limited by the finite number of points N = L1 /x that can be handled by a computer. Most investigations show that for comparable a and λ simulations of reasonable quality require x 0.2 min(a, λ),
L1 30λ,
g ∼ L1 /5.
(4.40)
These numbers, of course, have to be considered as only empirical estimates, based on the authors’ and other researchers’ experience with scattering simulations. The “marginally suitable” parameters x = 0.2λ and L1 = 30λ yield N = 150, thus requiring the solution of 2N = 300 coupled algebraic equations. This task can now be easily handled by any Pentium-class PC without special memory requirements. Although experiments usually deal with much longer surfaces and illuminated areas whose linear dimensions are larger than a few tens of a wavelength, scattering simulations appear to provide reasonably good modeling of experimental results when the statistical properties of a random surface are accurately known, and are introduced into a computer simulation of the scattering of light from that surface (see fig. 11 adapted from the paper by Knotts, Michel and O’Donnell [1993]). The second issue concerns the testing of the reliability of simulations that has to be done even for an intelligent choice of input parameters. Usually, it is wise to
2, § 4]
Weak localization effects in the multiple scattering of light
153
Fig. 11. The elements sij of the Stokes matrix and the +45◦ -polarized intensity, when light of wavelength λ = 3.392 µm is incident on a one-dimensional, randomly rough, gold surface with Gaussian power spectrum (2.21) characterized by δ = 1.73 µm and a = 3.43 µm. The angle of incidence is θ0 = 10◦ . Experimental results (solid curves) are compared with numerical results obtained by the use of an impedance boundary condition applied to a surface profile obtained directly by contact profilometry (Knotts, Michel and O’Donnell [1993]).
vary x, L1 , or N to check whether the result is stable against these variations. If not, a further decrease of x and/or increase of L1 is recommended. It is also useful to check the energy conservation requirement R + T = 1 (or R = 1 for perfectly conducting surfaces) for the total intensity reflection (R) and transmission (T ) coefficients. This criterion becomes especially important for surfaces supporting surface waves, which can propagate a distance of a few wavelengths even on rather rough surfaces. If the dissipative loss is expected to be low, then a bad energy conservation test means that the chosen surface is not long enough to ensure the total conversion of excited surface waves into the reflected/transmitted light. The typical energy conservation tests reported in the thoroughly tested numerical simulations in lossless geometries exceed 97%. While the two tests mentioned should be reasonably well satisfied for every realization of the surface profile, one must also obtain the convergence of the average intensity with an increase in the
154
Multiple scattering of light from randomly rough surfaces
[2, § 4
Fig. 12. A plot of ∂R/∂θs diff as a function of θs when light of wavelength λ = 632.8 nm is incident normally on a one-dimensional random surface of BaSO4 (nc = 1.628 + i0.0003) characterized by the Gaussian power spectrum (2.20) with δ = 1.2 µm and a = 2 µm. (a) p-polarized incident light (Maradudin, Méndez and Michel [1990]); (b) s-polarized incident light (Maradudin, Michel, McGurn and Méndez [1990]).
number Np of independent surface realizations. As usual in statistical sampling problems, the fluctuations around the average value are expected to decrease as 1/ Np . There has also been much evidence that large rms height, large rms slope surfaces that display a pronounced enhanced backscattering effect induced by the mechanism illustrated in fig. 4 have to be strongly reflective to ensure the efficacy of that mechanism. Such surfaces include, e.g., metallic and semiconductor surfaces in the frequency ranges where Re ε(ω) is negative, and perfectly conducting (model) surfaces. Highly reflective properties can also be expected from dielectrics with positive Re ε(ω) when Re ε(ω) is not very close to unity (high dielectric contrast). Figure 12 provides an illustration that the same one-dimensional randomly rough dielectric surface (ε = 2.25) can display the enhanced backscattering effect for s-polarized light (strongly reflected) and show no sign of the enhanced backscattering for p-polarized light (weakly reflected). The focus of current efforts in scattering simulations seems to be on improving the numerical algorithms to solve matrix equations for the source functions – such improvement is critically needed in dealing with two-dimensional rough surfaces. The abundance of proposed numerical schemes makes it impossible to describe them in detail within this article. In addition, none of these schemes appears to have proved itself as the best choice for all scattering configurations. We present only a brief outline of some of these algorithms. 4.2.2.2. Direct solution of matrix equation. The most straightforward approach, which was used, e.g., in the work of Maradudin, Michel, McGurn and Méndez [1990] cited, consists of the direct solution of the system by, e.g., the LU-
2, § 4]
Weak localization effects in the multiple scattering of light
155
factorization technique (Press, Teukolsky, Vetterling and Flannery [1992]). The speed of this approach scales as O(Nu3 ) with the number of unknowns Nu (equal to 2N for eq. (4.39a)) and requires a storage capacity of O(Nu2 ). While this exact approach can be safely chosen for Nu smaller than a few thousand (handled by ordinary computers), its utility rapidly decreases for the problems requiring large Nu , e.g., in the case of moderately rough metallic surfaces supporting surface plasmon polaritons that can propagate for considerable distances away from the illuminated area. Large values of Nu are also required in scattering problems with several interfaces (Lu, Maradudin and Michel [1991]). Finally, the necessary number of unknown Nu is especially large for the scattering from a twodimensional rough surface – for a vector field one has Nu = 4N 2 where N is the number of discretization points in the x1 - and x2 -directions, while 4 is the number of independent components of the vector field and its normal derivative (Ong, Celli and Maradudin [1993], Maradudin [1995]). Even a rather small value of N = 100 yields N = 40, 000 which exceeds both storage and speed limitations of best available computers by orders of magnitude. 4.2.2.3. Iterative techniques. A possible approach to circumvent this problem is to use an iterative scheme. For example, in a Neumann–Liouville iterative scheme one takes (2Uminc , 0) for (Um , Vm ) as an initial iteration and continues to carry out iterations without storing kernel matrices until convergence is reached – the implementation of this procedure in the one-dimensional roughness case is presented by Maradudin, Michel, McGurn and Méndez [1990]. The storage and speed of this approach scale as O(Nu2 ) and O(Nu ), respectively. The Neumann–Liouville scheme was used to simulate the enhanced backscattering of scalar waves from Neumann and Dirichlet surfaces with Nu = 2N 2 (Tran and Maradudin [1992, 1993]) and of vector electromagnetic waves (light) from metallic surfaces (Tran and Maradudin [1994]). An attractive feature of the Neumann–Liouville series is that the nth term describes an n-fold scattering process in which the field interacts with the surface n times. Strictly speaking, this identification is valid only in the high frequency or short wavelength limit (Liszka and McCoy [1982]). However, it remains qualitatively valid at lower frequencies as well. Simulations clearly show that the first-order (single-scattering) is normally dominant, but the enhanced backscattering peak appears only if the double-scattering processes are included (fig. 13). Several algorithms have been proposed to improve the convergence of iterations. For example, the method of ordered multiple interactions (MOMI), developed for one-dimensional perfectly conducting surfaces (Kapp and Brown [1996]) and later applied to one-dimensional rough surfaces of penetrable me-
156
Multiple scattering of light from randomly rough surfaces
[2, § 4
Fig. 13. Plot of the mean differential reflection coefficient and the contribution from the individual scattering processes. The computer simulations were performed for a two-dimensional rough Dirichlet surface with a Gaussian power spectrum characterized by δ = λ and a = 2λ, illuminated by a normally incident scalar beam. The n-fold scattering processes that are not labeled, decrease monotonically (Tran and Maradudin [1993]).
dia (Tran and Elson [1998a]), modifies the iteration series so that each term contains a large number of scattering events (predominantly scattering events that follow the same direction as the original scattering event). The resulting scheme still shows O(Nu2 ) and O(Nu ) storage and speed scaling, but its convergence is significantly better than that of the Neumann–Liouville iterations. Further improvements can be achieved by breaking up the matrices in the matrix equation into “weak” and “strong” parts (Tsang, Chan, Pak, Sangani, Ishimaru and Phu [1994], Tsang, Chan and Pak [1994]). Such breaking can be applied in both standard and MOMI-reconditioned matrix equations (Tran and Elson [1998b]). Improvements can also be achieved by combining the effective matrix-vector multiplication techniques and conjugate-gradient solvers of the matrix equations. Discussion on such improvements can be found in the papers by Chew and his collaborators developing the so-called FMM or fast multipole method (Michielssen, Boag and Chew [1996], Wagner, Song and Chew [1997]). In the cases considered, the FMM approaches were shown to improve the speed performance from O(Nu2 ) to O(Nu log Nu ), but the applicability of these methods (or necessity of their modification) to coherent multiple-scattering phenomena in rough surface scattering remains to be proven. Currently, the development of these efficient methods continues but none of them appears to have been accepted as a general and most efficient tool for all rough surface scattering simulations. A big part of the problem is that most of that work is done within the perfect conductor model. The matrices of equations for penetrable media are expected to have features (e.g., due to surface waves) that are poorly handled by iterative schemes or perfect-conductor-oriented matrix
2, § 4]
Weak localization effects in the multiple scattering of light
157
Fig. 14. A plot of the contribution to the mean differential reflection coefficient from the diffuse component of the scattered light when a beam of p-polarized light of wavelength λ = 457.9 nm is incident normally on a two-dimensional randomly rough silver surface (ε(ω) = −7.5 + i0.24) characterized by the Gaussian power spectrum (2.17), with δ = 1.0λ and a = 2.0λ. The solid curves give the contribution from in-plane scattering (φs = 0◦ ), and the dashed curves give the contribution from out-of-plane scattering (φs = 90◦ ) (Tran and Maradudin [1994]).
reconditioning. This is why the iterative series often rapidly diverges for surfaces that support surface waves (Tran [1998]). Another apparent problem with these approaches is difficulties with their extension to the two-dimensional case (e.g., MOMI) or lack of sufficient evidence that they can describe regimes when the enhanced backscattering peak is best visible (FMM). Despite several preliminary reports of the numerically simulated enhanced backscattering effect (Tran and Maradudin [1992, 1993, 1994], Macaskill and Kachoyan [1993], Tran, Celli and Maradudin [1994]) from two-dimensional rough surfaces, the numerical solution of the scattering from a two-dimensional randomly rough surface in the multiple-scattering regime remains the most serious challenge in the field. Current efforts remain centered on the perfect conductor and/or scalar models which reduce the number of equations. However, the effect that involves surface wave excitation and polarization change, observed in the very first experiments on enhanced backscattering, require the solution of the full vector problem for a penetrable medium. A powerful assistance in solving such problems is given by parallel computing, which is traditionally useful in Monte Carlo algorithms. For example, Tran and Maradudin [1994] were able to study a 128 × 128 points mesh modeling a metallic surface that exhibits enhanced backscattering (fig. 14) in this way. 4.2.2.4. Impedance approximation. A powerful technique to reduce the number of unknowns in the surface integral (or matrix) equations is provided by the method of impedance boundary conditions. In the general formulation of this method for a one-dimensional surface (Maradudin [1993]) the starting point is a
158
Multiple scattering of light from randomly rough surfaces
[2, § 4
linear relation between the values of the field and its unnormalized normal derivative on the surface x3 = ζ (x1 ) ∞ V(x1 ) = (4.41) dx1 K(x1 |x1 )U(x1 ) −∞
(the source functions U and V are introduced in Section 3.3). The non-local impedance K(x1 |x1 ) is expanded in powers of the optical skin depth d(ω), which is defined by −1/2 d(ω) = (c/ω) −ε(ω) (4.42) . In most practical cases the local impedance relation −σ (0) V(x1 ) = ε(ω) K (x1 )U(x1 )
(4.43)
(as before, σ = 1 for p-polarization and σ = 0 for s-polarization) was shown to yield results that agree with those obtained with a more accurate non-local relation (4.41) or with exact integral equations (3.12a) (Maradudin and Méndez [1996]). The accuracy of the impedance approximation was shown to improve for materials with small |d(ω)|. Such materials include most metals in the infrared and visible frequency ranges and dielectrics with Re ε(ω) considerably larger than unity (surprisingly, the values of Re ε(ω) > 2 can often be sufficient to make eq. (4.43) a good approximation). The local surface impedance K (0) (x1 ) in eq. (4.43) is given by
3 φ(x1 ) [ζ (x1 )]2 ζ (x1 ) (0) 2 +O d K (x1 ) = 1 + d(ω) − d (ω) , (4.44) d(ω) 2φ(x1) 8φ 4 (x1 ) where φ(x1 ) = 1 + [ζ (x1 )]2 . The term of O(1/d) was obtained by Rosich and Wait [1977], the term of O(1) was obtained by Depine [1987, 1988], the term of O(d) was calculated by Garcia-Molina, Maradudin and Leskova [1990], and the term of O(d 2 ) (not shown in eq. (4.44)) was derived by Maradudin [1993] and by Marvin and Celli [1994]. Since ζ2 (x1 ) is essentially the local curvature of the surface1 , it is therefore expected that the first few terms in the expansion (4.44) will provide a good approximation to K (0) (x1 ) when the skin depth is small compared to [ζ (x1 )]−1 , or to its rms value in the case of a random surface. Several recent papers discuss the improvements and generalizations of the local relation (4.44) – e.g., Maradudin and Méndez [1996] discuss the role and the magnitude of non-local corrections, while Mendoza-Suarez and Méndez [1997] generalize eq. (4.44) to the case of re-entrant surfaces. 1 Strictly speaking, the local curvature of the surface is ζ (x )/φ(x )3 . 1 1
2, § 4]
Weak localization effects in the multiple scattering of light
159
The use of eq. (4.41) or eq. (4.43) in eqs. (3.12) has the consequence that only the first of these two equations has to be solved, thereby dramatically reducing computational cost. This advantage becomes especially valuable for twodimensional rough surfaces, for which the vector analog of the impedance conditions (4.41) or (4.43) was derived by Ong, Celli and Maradudin [1993], Ong, Celli and Marvin [1994] and Maradudin [1995]. 4.2.2.5. Simulations based on the reduced Rayleigh equations. The scattering simulation approach we discussed in this subsection is based on the integral equations for the electromagnetic field in the coordinate space. Similar Monte Carlo simulations can be done in momentum (wavenumber) space for the reduced Rayleigh equation (3.8). Such simulations were carried out, e.g., by Michel [1994], and showed that the Rayleigh method can be used for fairly rough surfaces that do not allow a perturbative treatment. Madrazo and Maradudin [1997] and Simonsen and Maradudin [1999] performed such simulations for film systems and demonstrated a good overlap of their results with the predictions of the real-space simulations based on the exact surface integral equations. 4.2.3. Other approaches In the previous two subsections, we have described perturbation techniques for weakly rough surfaces and numerical simulation approaches that are usually used for large-amplitude, large-slope surfaces. These two approaches are the most commonly used since most multiple-scattering problems can be treated with one or the other of them. However, much work has been done since the mid-1980s in other directions. We next review these other approaches (without presenting their mathematical details) and separate them into two categories. The first category includes methods that appear to differ from the Rayleigh method (the starting point of perturbative expansions) or the exact surface integral equations (the starting point of numerical simulations) only in formalism but not in their range of validity or generality. The stochastic functional approach (Ogura, Kawanishi, Takahashi and Wang [1995], Ogura and Takahashi [1995], Ogura and Wang [1995], and Kawanishi, Ogura and Wang [1997]), utilizes the Wiener–Itô expansion of the random fields in the weak roughness limit. The stochastic functional approach generally yields results consistent with those obtained by the perturbative solution of the reduced Rayleigh equation and capture the coherent enhanced backscattering if the expansion is done beyond first order in the surface profile function. The phase perturbation theory, first suggested by Maradudin and Shen [1980], utilizes the reduced Rayleigh equation to develop an expression for the scat-
160
Multiple scattering of light from randomly rough surfaces
[2, § 4
tering S-matrix that contains a small-amplitude expansion series in the exponent: ∞ ∞ (−i)n Ψ (n) (q|x1|k) , (4.45) dx1 exp −i(q − k)x1 exp S(q|k) = n! −∞ n=1
where Ψ (n)
is O(ζ n )
and the relation of S(q|k) to the scattering amplitude R(q|k) is given by eq. (3.4). Sánchez-Gil, Maradudin and Méndez [1995] obtained the expansion (4.45) in a form that manifestly preserves the reciprocity of the scattering matrix: S(q|k) = S(−k| − q). They also investigated the ranges of applicability of the phase perturbation theory and the two perturbation theories described in Section 4.2.1 by comparing their results with the predictions of numerical simulations for a one-dimensional weakly rough surface of a dielectric with ε = 2.25. It was found that in both s- and p-polarizations the phase perturbation theory produced the best agreement with the simulation data for each angle of incidence considered, while the small amplitude perturbation theory was the least accurate of the three. This accuracy of the phase perturbation theory holds even for large angles of incidence for which both small-amplitude and self-energy perturbation theories do not work very well. The two-dimensional version of the phase perturbation theory was developed within a scalar model for the Dirichlet boundary condition, also in a reciprocal formulation (Fitzgerald and Maradudin [1994]). However, the multiple-scattering formalism provided by the phase perturbation theory has yet to prove its utility in the regime of coherent multiple-scattering phenomena, such as enhanced backscattering. The full-wave approach (Bahar [1987a, 1987b]) obtains approximate expressions for the scattered fields not by iterating the surface integral equations (3.12a) but by switching to local surface coordinates and obtaining approximate integral equations for the scattered field. The key double-scattering term that accounts for enhanced backscattering is written as a 6-dimensional integral even for a onedimensional surface (Bahar and El-Shenawee [1995]), which can be evaluated only when some simplifying assumptions are made (e.g., large surface rms curvatures compared to the wavelength). Unfortunately, the range of applicability for the full-wave approach appears to be unclear. The early theoretical work by Bahar and Fitzwater [1989] where enhanced backscattering was predicted, essentially relied on the approximation of mutual uncorrelation of heights and slopes. This approximation was shown to be invalid (Collin [1992]), even causing the artificial appearance of a retroreflection peak in the predominantly single-scattering regime. Although this contradiction was removed in more recent work (Bahar and El-Shenawee [1995]), the range of applicability of the full-wave method requires further examination.
2, § 4]
Weak localization effects in the multiple scattering of light
161
The second category includes approaches that are specifically tailored to the scattering geometry. The most famous of them is the Kirchhoff approximation, which has been used for decades (see, e.g., Beckmann and Spizzichino [1963], Bass and Fuks [1980]) to investigate single-scattering of waves from smooth surfaces with large-scale roughness by assuming that the scattered field is formed due to the specular reflection of light from the plane that is tangent to the surface at each point. After the discovery of enhanced backscattering from rough surfaces, the Kirchhoff approximation was extended to the second order, where the light is scattered twice from the surface, each time specularly from the plane tangent to the surface at the points struck by the light, with shadowing incorporated into the result (Ishimaru and Chen [1990], Chen and Ishimaru [1990]). The angular distribution of the intensity of the scattered light, obtained within the second-order Kirchhoff approximation displayed enhanced backscattering for one-dimensional perfectly conducting surfaces and was in agreement with the results of numerical simulations. The theory was later extended to the case of penetrable media (Ishimaru and Chen [1991]). In this context, we mention the paper by Li and Fung [1991] in which the authors reformulated the surface field integral equations for the scattering of light from a two-dimensional rough penetrable surface in such a way that the inhomogeneous terms represented the Kirchhoff approximation to the source functions. The iterative solutions of the reformulated equations were expected to converge rapidly in application to scattering from surfaces for which the Kirchhoff approximation is a good approximation to the exact result. However, no numerical results for the differential reflection coefficient were obtained on the basis of the reformulated equations. Subsequently, Sentenac and Maradudin [1992], reformulated the pair of surface integral equations (3.12) for scattering from a one-dimensional rough penetrable surface in such a way that the Neumann–Liouville series solution of the resulting pair of equations started with the Kirchhoff approximation to each of the source functions. Numerical results for the mean differential reflection coefficient obtained by an iterative solution of the reformulated equations were found to be in good qualitative and quantitative agreement with the results of a direct numerical solution of the original untransformed pair of equations. The same problem was studied by Stoddart [1992], by a different approach that also led to a pair of integral equations for the source functions that had the Kirchhoff approximation to these functions as the inhomogeneous terms. The convergence properties of the iterative solution of the resulting pair of integral equations were examined, but no numerical results were obtained from the transformed equations.
162
Multiple scattering of light from randomly rough surfaces
[2, § 4
The recently developed small-contrast perturbation theory (Leskova, Maradudin and Novikov [2000]) is built on the expansion of the scattering amplitude in powers of the small dielectric contrast η = ε0 − ε,
(4.46)
where ε0 is the dielectric constant of the medium of incidence and ε is the dielectric constant of the scattering medium. This approach can be applied, for example, to the scattering of X-rays from a randomly rough metal surface (Leskova and Maradudin [1997, 1999]), where ε0 = 1 and ε = ε(ω), the dielectric function of the metal. The term of nth order in the expansion of the scattered field in powers of η contains contributions of all orders in the surface profile function. This new expansion yields the enhanced backscattering peak when carried out through terms of fourth order in η(ω). There are also several novel theories that have been developed recently but were not shown to describe coherent multiple scattering phenomena such as enhanced backscattering. This should not be considered as a negative feature of these theories since they proved themselves very effective within their range of validity. These theories include, e.g., small-slope perturbation theory, the operator expansion formalism, and the coherent potential mean-field theory. The small-slope perturbation theory (Voronovich [1985, 1994, 1996]) is akin to the Kirchhoff series. This theory, developed for the scattering of scalar waves from perfectly conducting, smooth surfaces, represents the scattered field as an expansion in powers of small surface slopes. Further investigations including numerical testing against other methods’ predictions (Thorsos and Broschat [1995], Broschat and Thorsos [1997]), showed a robust performance of this approach, including the often problematic case of grazing angles of incidence. The operator expansion formalism is another approach that has been proposed to describe scattering from surfaces with small slopes, first for perfectly conducting surfaces (Milder [1996]) and then extended to surfaces of penetrable media (Smith [1996]). The expansion leads to efficient numerical schemes through the use of fast Fourier transforms. The small slopes assumed in both smallslope and operator expansions realistically model many physical situations such as the scattering of electromagnetic waves from the ocean surface. The coherent potential mean-field theory has been developed for both one-dimensional (Sentenac and Greffet [1998]) and two-dimensional dielectric surfaces (CalvoPerez, Sentenac and Greffet [1999]). In the version of this approach applicable to scattering from one-dimensional surfaces the dielectric constant of the system, ε(x1 , x3 ) = ε> θ (x3 − ζ (x1 )) + ε< θ (ζ (x1 ) − x3 ), where ε> is the dielectric constant of the region x3 > ζ (x1 ) while ε< is the dielectric constant of the region
2, § 4]
Weak localization effects in the multiple scattering of light
163
x3 < ζ (x1 ), is rewritten in the form ε(x1 , x3 ) = ε(x3 ) + [ε(x1, x3 ) − ε(x3 )], where ε(x3 ) is the average of ε(x1 , x3 ) with respect to x1 . The solution of the scattering problem is then obtained by regarding the fluctuation term [ε(x1, x3 ) − ε(x3 )] as a perturbation on the system described by ε(x3 ). An analogous procedure is followed in the studying of the scattering from a two-dimensional random interface between two dielectric media. This approach does not rely on the smallness of the roughness amplitudes or slopes and was shown to perform well for dielectric constants smaller than 5, but has yet to be extended to the multiple-scattering regime. The approaches reviewed in this subsection probably do not account for the full scope of ongoing theoretical research but certainly demonstrate the high level of activity and creativity of researchers in the very lively rough surface scattering field. 4.2.4. The polarization dependence of scattering from a one-dimensional rough surface In the preceding discussion of the scattering of light from a one-dimensional random surface, where it was assumed that the plane of incidence is normal to the generators of the surface, the plane of scattering coincides with the plane of incidence. We will call scattering in this geometry in-plane scattering. The incident electric field was linearly polarized either parallel or perpendicular to the grooves and ridges of the surface, i.e. it was s- or p-polarized, respectively. In this case the polarization of the scattered field was the same as that of the incident field. The diffuse scattering is therefore described by the two scattering cross sections ∂Rs /∂θs diff and ∂Rp /∂θs diff . However, it was shown by O’Donnell and Knotts [1991] and by Michel, Knotts and O’Donnell [1992] that the complete determination of the incoherent scattering properties (viz. those described by the second-order moments of the field) of a one-dimensional random surface, when the plane of incidence is normal to the generators of the latter, requires not just two but four scattering cross sections. Two of them are effectively ∂Rs /∂θs diff and ∂Rp /∂θs diff , but the other two must be determined by the use of an incident field that is a linear combination of p- and s-polarized fields. To obtain these scattering cross sections we begin by writing the electric vector of a beam of light incident from the vacuum side (x3 > ζ(x1 )) on a onedimensional surface defined by x3 = ζ (x1 ), with the plane of incidence the x1 x3 plane,
164
Multiple scattering of light from randomly rough surfaces
E> (x1 , x3 |ω)inc π/2 ωw dθ exp − ω2 w2 /4c2 (θ − θ0 )2 = √ 2 π c −π/2 × Ep eˆ p (θ )inc + Es eˆ s (θ )inc × exp i(ω/c)(x1 sin θ − x3 cos θ ) ,
[2, § 4
|θ0 | < π/2, (4.47)
where eˆ s (θ )inc = xˆ 2 ,
(4.48a)
eˆ p (θ )inc = −ˆx1 cos θ − xˆ 3 sin θ.
(4.48b)
The field given by eq. (4.47) is an exact solution of Maxwell’s equations in vacuum that in the limit (ωw/2c) 1 describes a Gaussian beam propagating in the direction defined by xˆ 1 sin θ0 − xˆ 3 cos θ0 . Its width at the origin of coordinates is 2w. The scattered electric field in the far field is then given by π/2 i dθs Ep rp (θs )ˆep (θs )sc + Es rs (θs )ˆes (θs )sc E(x1 , x3 |ω)sc = 4π −π/2 × exp i(ω/c)(x1 sin θs + x3 cos θs ) , (4.49) where eˆ s (θs )sc = xˆ 2 ,
(4.50a)
eˆ p (θs )sc = xˆ 1 cos θs − xˆ 3 sin θs
(4.50b)
and the scattering amplitudes rp (θs ) and rs (θs ) are given by ∞ rp (θs ) = dx1 exp −i(ω/c) x1 sin θs + ζ(x1 ) cos θs −∞
rs (θs ) =
∞ −∞
× i(ω/c) ζ (x1 ) sin θs − cos θs H (x1 |ω) − L(x1 |ω) , (4.51a) dx1 exp −i(ω/c) x1 sin θs + ζ(x1 ) cos θs
× i(ω/c) ζ (x1 ) sin θs − cos θs E(x1 |ω) − F (x1 |ω) . (4.51b) > The source functions H (x1 |ω) and E(x1 |ω) are given by U (x1 , ζ (x1 )), while the source functions L(x1 |ω) and F (x1 |ω) are given by (∂/∂N)U > (x1 , x3 |ω)|x3 =ζ(x1 ) ,
2, § 4]
Weak localization effects in the multiple scattering of light
165
where we recall that U > (x1 , x3 |ω) = H2> (x1 , x3 |ω)(E2> (x1 , x3 |ω)) for p- (s-)polarized light. It is desirable to define the polarizations of the incident and scattered fields in terms of directly measurable quantities. With this in mind it is necessary to calculate the power present in the projection of the incident and scattered fields on polarization states that are linear combinations of the p- and s-polarized states defined by the unit vectors (4.48) and (4.50), respectively. A useful description of polarized light in terms of these powers is through the Stokes vectors (Hecht [1970]), and Stokes matrices (Ishimaru [1978], van de Hulst [1981]). The following pairs of orthonormal vectors are needed to compute the Stokes vectors (Jackson [1975]): 1 eˆ + = √ (ˆep + eˆ s ), 2
1 eˆ − = √ (ˆep − eˆ s ) 2
(4.52)
1 eˆ R = √ (ˆep − iˆes ), 2
1 eˆ L = √ (ˆep + iˆes ), 2
(4.53)
and
where eˆ α with α = p, s, +, −, R, or L must be replaced by eˆ α (θ0 )inc in the case of the incident field, and by eˆ α (θs )sc in the case of the scattered field. The pairs of unit vectors (4.52) and (4.53) define the linearly polarized states at +45◦ , as well as right (R) and left (L) circularly polarized states. If (α, β) denotes any of the pairs (p, s), (+, −), or (R, L), the projection of a field E(x1 , x3 |ω) on eˆ γ (γ = α or β) is Eγ (x1 , x3 |ω) = [ˆe∗γ · E(x1 , x3 |ω)]ˆeγ , and we have that E(x1 , x3 |ω) = Eα (x1 , x3 |ω) + Eβ (x1 , x3 |ω). The elements of the Stokes vector V = (I, Q, U, V ) may be expressed as (Born and Wolf [1980]): I = Pp + Ps = P+ + P− = PR + PL ,
(4.54a)
Q = Pp − Ps ,
(4.54b)
U = P+ − P− ,
(4.54c)
V = PR − PL .
(4.54d)
In the case of the incident field Pα is the total power incident on the surface that is in the projection Eα (x1 , x3 |ω)inc of the incident field on eˆ α . To obtain this total incident power the magnitude of the 3-component of the time-averaged Poynting vector of Eα (x1 , x3 |ω)inc is integrated over the region of the x1 x2 -plane defined by −L1 /2 < x1 < L1 /2, −L2 /2 < x2 < L2 /2, where L1 and L2 are large lengths.
166
Multiple scattering of light from randomly rough surfaces
The Stokes vector of the incident field Vinc thus becomes |Ep |2 + |Es |2 Iinc |Ep |2 − |Es |2 Qinc Vinc = Uinc = Pinc 2 Re(Ep E ∗ ) , s Vinc 2 Im(Ep Es∗ )
[2, § 4
(4.55)
where Pinc is given by Pinc ∼ = L2 cw/[8(2π)1/2] when ωw/2c 1. The powers Pα required for the calculation of the Stokes vector of the scattered field given by eq. (4.49) are the differential powers per unit scattering angle Pα (θs )sc obtained from the projection of the plane wave scattered at an angle θs (the integrand on the right-hand side of eq. (4.49)) on the polarization state eˆ α (θs )sc . The function Pα (θs )sc is calculated by integrating the 3-component of the timeaveraged Poynting vector of this projection over the region −L1 /2 < x1 < L1 /2, −L2 /2 < x2 < L2 /2 of a plane in the far field that is parallel to the plane x3 = 0. The Stokes vector v(θs )sc calculated from the scattered field (4.49) with the use of eqs. (4.54) for a single realization of the rough surface is then |rp (θs )Ep |2 + |rs (θs )Es |2 2 |rp (θs )Ep |2 − |rs (θs )Es |2 c . v(θs )sc = L2 (4.56) 2 64π ω 2 Re[rp (θs )Ep r ∗ (θs )E ∗ ] s
s
2 Im[rp (θs )Ep rs∗ (θs )Es∗ ]
A matrix ↔ s (θs ) that relates the Stokes vector of the incident field given by eq. (4.55) to the Stokes vector of the scattered field given by eq. (4.56) through the equation v(θs )sc = ↔ s (θs )Vinc
(4.57)
can be obtained by using the fact that the amplitudes Ep and Es are arbitrary complex amplitudes. This matrix is found to have the form 0 0 s11 (θs ) s12 (θs ) s12 (θs ) s11 (θs ) 0 0 ↔ , s (θs ) = (4.58) 0 0 s33 (θs ) s34 (θs ) 0 0 −s34 (θs ) s33 (θs ) where
2 2 s11 (θs ) = C rp (θs ) + rs (θs ) , 2 2 s12 (θs ) = C rp (θs ) − rs (θs ) , s33 (θs ) = C 2 Re rp (θs )rs∗ (θs ) , s34 (θs ) = C −2 Im rp (θs )rs∗ (θs ) ,
(4.59a) (4.59b) (4.59c) (4.59d)
2, § 4]
Weak localization effects in the multiple scattering of light
167
with C=
L2 c2 . 32(2π)2ωPinc
(4.60)
To each realization of the surface profile function ζ(x1 ) corresponds a realizas (θs ). In the scattering of light from a tion of the vector v(θs )sc and of the matrix ↔ randomly rough surface it is the average of the vector v(θs )sc over the ensemble of realizations of the surface profile function,
V(θs )sc = v(θs )sc , (4.61) that is of interest. This equation defines the Stokes vector of the scattered light V(θs )sc . The average of the matrix ↔ s (θs ) over the ensemble of realizations of the surface profile function,
↔ S(θs ) = ↔ (4.62) s (θs ) , is the matrix that enables the Stokes vector of the scattered light to be calculated for any given polarization state of the incident light, ↔
V(θs )sc = S(θs )Vinc . ↔
(4.63)
The matrix S(θs ) is called the Stokes matrix of the randomly rough surface. Just as the mean differential reflection coefficients contain contributions from the specular and diffuse components of the scattered field, so do the elements of the Stokes matrix. Thus, if we replace the scattering amplitudes rp (θs ) and rs (θs ) in eqs. (4.59) by their ensemble averages rp (θs ) and rs (θs ), respectively, we obtain the contribution to the Stokes matrix from the specular component of the ↔ scattered field, S(θs )spec . Similarly, if we replace the scattering amplitudes rp (θs ) and rs (θs ) in eqs. (4.59) by their fluctuating parts rp (θs ) − rp (θs ) and rs (θs ) − rs (θs ), respectively, and then average the resulting matrix elements over the ensemble of realizations of the surface profile function, we obtain the contribution ↔ to the Stokes matrix from the diffuse component of the scattered field, S(θs )diff . It ↔ ↔ ↔ should be noted that S(θs ) = S(θs )spec + S(θs )diff . The elements of the Stokes matrix (4.58) provide a complete description of the polarization of the light scattered from a one-dimensional randomly rough surface when the plane of incidence is normal to the generators of the surface. We note for future reference that the elements of the Stokes matrix for the in-plane scattering of light from a one-dimensional perfectly conducting random surface are obtained by setting ε(ω) = −∞ everywhere in eqs. (4.59). Knotts, Michel and O’Donnell [1993] carried out a comparison of the results of a computer simulation calculation of the elements of the Stokes matrix for the
168
Multiple scattering of light from randomly rough surfaces
[2, § 4
in-plane scattering of light from a one-dimensional randomly rough gold surface with experimental results for these elements obtained in the same work. The frequency of the incident light was in the infrared region of the optical spectrum, so that the impedance boundary condition (4.43)–(4.44) was used in the solution of the integral equations of scattering theory, together with the actual surface profile, obtained by contact profilometry. The agreement between theory and experiment proved to be very good, both qualitatively and quantitatively (fig. 11). From results such as those presented in fig. 11 it can be said that the computational techniques developed for the study of the in-plane scattering of light from one-dimensional random surfaces are impressively accurate. The most detailed information about the polarization properties of light scattered from a one-dimensional random surface, however, is obtained from a study in which the plane of incidence is not perpendicular to the generators of the surface. In this scattering geometry the diffuse portion of the scattered radiation appears on the surface of a cone whose axis is parallel to the generators of the surface, rather than in a plane, because of the translational invariance of the scattering surface parallel to its generators. For this reason the scattering of light in this geometry is called conical scattering. In the earliest experimental studies of the conical scattering of light from a onedimensional random surface (Luna and Méndez [1995], Luna [1996]), and in the earliest theoretical studies of such scattering (Depine [1991, 1993], Li, Chang and Tsang [1994]), the dependence of the elements of the corresponding Stokes matrix on the angles of incidence and scattering was not obtained. The first theoretical determination of the elements of the Stokes matrix for the scattering of light from a one-dimensional random metal surface in a conical configuration, with multiplescattering processes of all orders taken into account was carried out by Novikov and Maradudin [1997]. This work was soon followed by analogous calculations for conical scattering from a perfectly conducting surface by Luna, Acosta-Ortiz and Zou [1998] and Novikov and Maradudin [1999], although in the former paper only single-scattering processes were taken into account. An interesting result of the theoretical study of the conical scattering of light from a one-dimensional perfectly conducting random surface (Novikov and Maradudin [1999]) is that the Stokes matrix in this case has exactly the same form as the Stokes matrix for in-plane scattering from the same surface, and the elements of the Stokes matrix in the former case can be obtained from the elements of the Stokes matrix in the latter case by replacing the wavelength λ of the incident field in the latter elements everywhere by an effective wavelength given by λeff = λ/ cos φ0 .
(4.64)
2, § 4]
Weak localization effects in the multiple scattering of light
169
The angle φ0 in this expression, called the conical angle, is the angle between the wave vector k, of the incident field and its projection on the x2 x3 plane. This result had been obtained earlier by Maystre [1984] in a study of conical scattering from a perfectly conducting classical grating. In the case of the conical scattering of light from a one-dimensional penetrable random surface, e.g., from a metal surface, all elements of the Stokes matrix are nonzero. The expressions for these elements are rather cumbersome, and the reader is referred to the paper by Novikov and Maradudin [1997] where they are presented in an explicit form. The first experimental results for several elements of the Stokes matrix in the conical scattering of light from a one-dimensional random metal surface were obtained by Luna [2002] and were in fairly good qualitative and quantitative agreement with theoretical results of Novikov and Maradudin [1997]. 4.3. Experimental techniques used in the study of multiple scattering effects, including the enhanced backscattering effect When discrepancies are found in the comparison of scattering theories with experimental data taken with commonly found surfaces, it is difficult to decide if they are due to some approximation made in the electromagnetic treatment of the problem, or to an incorrect choice of the statistical model. Much progress towards understanding the accuracy of various theoretical techniques and observing new phenomena has been made by the use of specially fabricated surfaces. In this subsection we first review the most relevant aspects of that work. We also examine some important issues concerning the relation between practical surface scattering measurements and theoretical work, and describe a typical experimental arrangement for measurements in the optical and infrared regions of the spectrum. Finally, we discuss some experimental results involving multiplescattering effects that are difficult to explore theoretically. 4.3.1. Fabrication of surfaces A simple assumption for theoretical work is that the scattering object consists of a single interface separating two homogeneous media, and that the curve defining the interface constitutes a realization of a stationary Gaussian random process with a Gaussian correlation function (see Section 2.1.1). Surfaces obtained, for example, by processes such as grinding, polishing, and sandblasting, depart from this idealized kind of random process and are difficult to model and characterize; they may present roughness over many scales, be non-stationary, and have illdefined statistics.
170
Multiple scattering of light from randomly rough surfaces
[2, § 4
For detailed comparisons between theory and experiment, surfaces with welldefined statistics are required, and methods of fabricating such surfaces have been devised. Moreover, multiple-scattering effects can be enhanced by the use of specially designed surfaces. We restrict our review to the fabrication of random surfaces on photoresist. Typically, the process starts by coating glass substrates with a thin layer of photoresist, which is a light-sensitive material in the blue and ultraviolet regions of the spectrum. The plates are then exposed to some specified light pattern, either sequentially or at once. The procedure followed to prepare and develop the plates varies depending on the kind of photoresist and the nature of the surface to be fabricated. Particular procedures have been described by several authors (e.g., Gray [1978], O’Donnell and Méndez [1987]). With the application of thick metal coatings by vacuum evaporation the sample can behave, in reflection, as the interface between two semi-infinite media. Other measures need to be taken to achieve a similar situation with dielectrics (Chaikina, Hernández-Walls and Méndez [2000], Chaikina, Negrete-Regagnon, Ruiz-Cortés and Méndez [2002]). The minimum standard deviation that can be produced in a controlled way is determined, mainly, by the quality of the substrate and the conditions of the clean room employed in the coating process. A fundamental requirement is that the natural roughness of the substrate and coating must be much smaller than the desired roughness. The maximum standard deviation depends on the thickness of the photoresist coating, and it is common to have samples that contain two or even three layers of photoresist. The most important and widely used method for the fabrication of Gaussiancorrelated Gaussian surfaces was devised by Gray [1978]. Surfaces fabricated with Gray’s method, or some variation of it, have been used extensively in studies involving enhanced backscattering and detailed comparisons with theory (e.g., Méndez and O’Donnell [1987], O’Donnell and Méndez [1987], Sant, Dainty and Kim [1989], Kim, Dainty, Friberg and Sant [1990], O’Donnell and Knotts [1991, 1992], Michel, Knotts and O’Donnell [1992], Knotts, Michel and O’Donnell [1993], Knotts and O’Donnell [1993, 1994], Luna, Méndez, Lu and Gu [1995], Luna and Méndez [1995], Méndez, Navarrete and Luna [1995], Chaikina, Navarrete, Méndez, Martínez and Maradudin [1998], Navarrete, Chaikina, Méndez and Leskova [2002], Chaikina, Hernández-Walls and Méndez [2000], Chaikina, Negrete-Regagnon, Ruiz-Cortés and Méndez [2002]). They have also been employed in studies of non-Gaussian speckle (Levine and Dainty [1982]), scattering by gamma-distributed surfaces (Kim, Méndez and O’Donnell [1987])
2, § 4]
Weak localization effects in the multiple scattering of light
171
and, more recently, in studies of second harmonic generation by random surfaces (O’Donnell, Torre and West [1996, 1997], O’Donnell and Torre [1997]). In this method, the central idea is to convert the random spatial fluctuations of a speckle pattern into a surface profile, by exposing the photoresist-coated plates to speckle patterns with the appropriate statistical properties. Consider the exposure of the plate to the intensity of a fully-developed speckle pattern, I (x ), produced by the scattering of coherent light from a diffuser. The probability density function associated with the intensity of the pattern is a negative exponential function (Goodman [1984], p. 16), and the intensity covariance function may be written in the form (Goodman [1984], p. 38) 2
I (x )I (x ) = C(x − x ) , (4.65) where I (x ) represents the fluctuations about the mean intensity, and ∞ q · |x − x | P (q )2 exp i ω dq . C(x − x ) = c z −∞
(4.66)
Here P (q ) represents the pupil function on the plane of the diffuser that produces the speckle, and z is the distance from the diffuser to the photoresist plate. We note that if the diffuser is illuminated with a Gaussian beam (Gaussian pupil function), C(x − x ) has also a Gaussian form. Assuming a linear response of the photoresist, the resulting profile is proportional to the exposure. That is ζ(x ) = αI (x )t,
(4.67)
where ζ(x ) denotes the fluctuations about the mean surface, α is a proportionality constant characterizing the material’s response to the exposing energy, and t is the exposure time. If the linearity assumption is satisfied, the normalized correlation function of the fabricated surface will also be Gaussian. To obtain an approximate Gaussian probability density function, the plate is exposed to several independent speckle patterns of equal mean intensity. The resulting intensity, I, is gamma-distributed (Goodman [1984], p. 23), I N−1 I , exp − PI (I) = (4.68) (N − 1)!I N I where N represents the number of exposures, I = Ij = I/N , and j = 1, 2, 3, . . . , N . As the number of exposures grows, the distribution of I approaches a Gaussian form, but the contrast of the total exposure decreases, reducing the roughness parameter δ of the fabricated surface. It is considered that a good compromise is obtained with between eight and ten exposures.
172
Multiple scattering of light from randomly rough surfaces
[2, § 4
Fig. 15. Electron micrographs of (a) a surface fabricated in photoresist, and (b) a ground glass surface. The areas shown are, approximately, 30 × 30 µm (O’Donnell and Méndez [1987]).
Surfaces with correlation lengths between a = 1 µm and 50 µm, and standard deviations of between δ = 0.1 µm and 2 µm, have been fabricated in this way. The difference between this kind of surface and others, such as ground glass surfaces is illustrated in fig. 15. The fabrication of what have been called quasi one-dimensional surfaces involves the use of nonisotropic speckle patterns (Sant, Dainty and Kim [1989], Kim, Dainty, Friberg and Sant [1990]). An improvement is possible by scanning the plate while exposing it in the direction of the elongation of the speckles (O’Donnell and Knotts [1991], Luna, Méndez, Lu and Gu [1995]). Through mechanical profilometry it has been verified that Gaussian-correlated surfaces with a Gaussian height distribution can be fabricated with this technique. However, detailed studies of the statistical properties of some surfaces has shown that, even when the height distribution is Gaussian, the higher order statistics of the profile may be incompatible with the assumption of a Gaussian random process (Knotts, Michel and O’Donnell [1993]). The other important technique for the fabrication of rough one-dimensional surfaces was introduced by West and O’Donnell [1995]. The surfaces fabricated with this technique are designed to be weakly rough and produce multiplescattering effects mediated by the excitation of surface waves. The surfaces have Gaussian statistics and a power spectrum g(k) that has constant height between the wavenumbers kmin and kmax , with little spectral power elsewhere (see Section 2.2). Mathematically speaking, the fabrication method is based on a means of generating Gaussian random processes discussed by Rice [1951]. The photoresist plate is exposed, sequentially, to sinusoidal intensity patterns produced by the interference of two intersecting light beams. Assuming a linear response of the
2, § 4]
Weak localization effects in the multiple scattering of light
173
material, the resulting developed profile ζ (x1 ) is ζ (x1 ) = α
N
In 1 + Mn cos(kn x1 − φn ) tn ,
(4.69)
n=1
where N represents the number of exposures, In is the intensity of the nth pattern, Mn is the fringe modulation, kn = nk is a wave number that depends on the angle of intersection of the beams, φn is a phase that determines the position of the fringe pattern, tn is the exposure time, and α is a proportionality constant characterizing the response of the material. The phases φn are chosen as independent random variables, uniformly distributed on the range (0, 2π). Then, ζ (x1 ) =
N
cn cos(kn x1 − φ),
(4.70)
n=1
where cn = αIn Mn tn . Taking an amplitude cn specified by 1/2 cn = 2g(kn )k ,
(4.71)
it may be shown that ζ (x1 ) has the required roughness spectrum g(k) and, as N increases and k decreases, the statistics converge to the Gaussian form. The method may thus be regarded as a means of constructing a realization of ζ (x1 ) from its randomly phased Fourier components, with appropriate amplitude weighting to produce g(k). Surfaces with standard deviations of heights of as little as δ = 10 nm have been successfully fabricated with this technique. 4.3.2. Scattering measurements We concentrate the discussion on the experimental estimation of the mean differential reflection coefficient. At first sight, this appears to be a simple task, but scattering measurements are full of subtleties (see, e.g., Stover [1995], Chapter 6). There are also some issues concerning the relation between theory and experiment that are worth discussing. The first problem encountered is that the mean differential reflection coefficient involves the estimation of an ensemble average, and averaging over many statistically equivalent experimental samples is not a practical proposition in most cases. Instead, to estimate the mean DRC, the averaging is performed by integration over a finite angular region with the detector. In the following lines, we try to justify this procedure taking, as an example, the scattering by one-dimensional stationary surfaces. For plane wave illumination, the DRC may be written in the form (see Section 3.3)
174
Multiple scattering of light from randomly rough surfaces
∂R(q) |r(q|k)|2 , = ∂θs F (k)L1
[2, § 4
(4.72)
where F (k) = 8πα0 (k, ω), and L/2 r(q|k) = F (q, x1) exp{−ıqx1} dx1.
(4.73)
Here, we have defined F (q, x1 ) = ıqζ (x1 ) − ıα0 (q, ω) U(x1 ) − V(x1 ) × exp −ıα0 (q, ω)ζ (x1 ) .
(4.74)
−L/2
As a function of x1 , the function F (q, x1) is a stationary random process that can be interpreted as the near-field complex amplitude reflected by the surface. The far-field scattering amplitude may be viewed as the Fourier transform of this quantity (the variation of F (q, x1 ) with q is slow), and the statistical estimation of the DRC amounts to the estimation of the spectral density function of the random process F (q, x1). The sample estimator of the DRC can be written in the form ∞ 1 ∂R(q) C(q, η) exp{ıqη} dη, = (4.75) ∂θs F (k) −∞ where we have defined ∞ η+ξ 1 ξ rect F (q, ξ )F ∗ (q, η + ξ ) dξ, (4.76) C(q, η) = rect L1 −∞ L1 L1 with rect(ξ ) representing the rectangle function (Goodman [1996], p. 13). The function C (q, η) is also a random function of η, with a slow dependence on q. In the limit of large L1 , it becomes the sample correlation function of the random process F (q, ξ ). Since the expectation value of the sample estimator yields the mean DRC, the estimator is not biased. However, the sample DRC does not converge to any function as the length of the sample is increased. It is known that the variance of the intensity fluctuations does not decrease as the length of the sample tends to infinity. In other words, we know that while the speckle size depends on the length of the surface, the speckle contrast is insensitive to L1 . Spectral estimators may be improved by smoothing the sample estimator (Jenkins and Watts [1968], Section 6.3). Convolution with an aperture function A(q) results in a smoothed estimator with reduced variance. We write ∞ ∂R(q) dq ∂R(q ) A(q − q ) = , (4.77) ∂θs ∂θs −∞ 2π
2, § 4]
Weak localization effects in the multiple scattering of light
where the aperture function has the property ∞ A(q) dq = 2π. −∞
175
(4.78)
Making use of the fact that C(q, η) is a slowly varying function of q, we can also write the smoothed estimator in the form ∞ ∂R(q) 1 = A(η)C(q, η) exp{ıqη} dη, (4.79) ∂θs F (k) −∞ where A(η) is the inverse Fourier transform of A(q). Expressions (4.77) and (4.79) represent standard forms for the estimation of spectral quantities (Jenkins and Watts [1968], p. 243). The expectation value of the smoothed estimator is ∞ ∂R(q) dq ∂R(q ) A(q − q ) = . (4.80) ∂θs ∂θs −∞ 2π We can see that, depending on the angular size and shape of the aperture function, the mean smoothed DRC can be a distorted version of the mean DRC. So, if the surface is statistically stationary (at least in the wide sense), the ensemble averaged DRC can be estimated through an angular average. To reduce the statistical fluctuations in the measurements it is necessary to average over an angle that is larger than the speckle size, but smaller than the characteristic features in the angular distribution of the mean intensity. The fulfillment of these conditions determines the minimum size of the sample (the speckle size is inversely proportional to L1 ). This is usually not a problem for scattering measurements in the visible, but can be an important consideration when working with one-dimensional surfaces in the infrared. Although there are instruments that can measure the full hemispherical distribution function, normally, angular light scattering measurements are only performed in the plane of incidence. The instruments employed for these measurements are called in-plane bidirectional scatterometers. A typical arrangement is shown schematically in fig. 16. In the figure, an expanded laser beam is sent through a periscope with a series of mirrors from which it exits horizontally toward the sample. Since the use of lock-in detection techniques is desirable, the illumination beam is usually chopped. For reflective measurements the last mirror of the periscope obstructs the detector in the vicinity of the backscattering direction and, in order to reduce this problem, the beam is focused on this last mirror; the obstruction can then be made narrower, permitting measurements close to the backscattering direction. With this arrangement, which is convenient for backscattering measurements with surfaces that do not present an appreciable coherent
176
Multiple scattering of light from randomly rough surfaces
[2, § 4
Fig. 16. Schematic diagram of a typical experimental setup for scattering measurements. The sample can be rotated about the vertical axis to set the angle of incidence. The detector arm rotates around the same axis and records the scattered intensity.
component, the surface is illuminated by a slightly divergent beam. For weakly rough surfaces that produce a strong coherent component, converging beam or collimated beam illumination can be more convenient. So, the choice of the form of the illumination beam depends on the nature of the surface and the kind of scattering measurement desired. The detection system usually consists of an aperture, a lens, and the detector. The lens is used to collect the scattered light over the region defined by the aperture and, at the same time, to form an image of the illuminated region of the surface. To ensure that the instrument does not distort the angular distribution of the intensity, it must be verified that the detector views the entire illuminated area of the surface, irrespective of the angle of detection. Aperture effects, represented by eq. (4.80), must also be minimized. Accurate scattering measurements are particularly difficult with weakly scattering surfaces and at large angles of incidence and scattering. When the scattering signals are weak in comparison with the strong coherent component, spurious reflections can easily overcome them, interfering with the signal and saturating the detector. The instrument signature (Stover [1995], Chapter 6) depends on the amount of stray light produced by the various elements of the scatterometer (mirrors, lenses, polarization elements, and apertures) and the aberrations of the system. 4.3.3. Observations of enhanced backscattering and other multiple scattering effects The first experimental works on enhanced backscattering with specially fabricated surfaces were reported by Méndez and O’Donnell [1987] (see also O’Donnell and Méndez [1987]). The motivation for that work was a critical evaluation of the
2, § 4]
Weak localization effects in the multiple scattering of light
177
Fig. 17. Photographs of the far-field scattering pattern in normal incidence from a surface similar to the one used for the results of fig. 3. The illumination wavelength is λ = 0.633 µm and the incident polarization is vertical. A and B correspond to orthogonally polarized components (O’Donnell and Méndez [1987]).
available theories. Instead, what was found were results that, quantitatively, are still difficult to reproduce theoretically. Figure 3 shows the angular distribution of the mean intensity for a two-dimensional surface with approximately Gaussian statistics, a Gaussian correlation function, and estimated parameters a = 1.4 µm and δ ≈ 1.0 µm. The parameters represent only estimates because a surface with such large slopes cannot be characterized by mechanical profilometry. There are two noteworthy features in the scattering distributions shown in fig. 3. The first one is the presence of an enhanced backscattering peak; the nature of this effect has already been discussed in Section 4.1. The second important feature is the presence of a relatively large cross-polarized component. Although these results were surprising at the time, their multiple scattering nature was quickly recognized.
178
Multiple scattering of light from randomly rough surfaces
[2, § 4
Fig. 18. Optical micrographs (NA = 0.8, 50×) of an area of a surface illuminated with linearly polarized light and viewed with a parallel (left) and perpendicular (right) analyzer. The areas where changes in the polarization are observed exhibit patterns that are consistent with the explanation of fig. 19 (O’Donnell and Méndez [1987]).
The far-field hemispherical scattering pattern produced by a surface with similar characteristics is shown in fig. 17. The photographs correspond to the co- and cross-polarized scattering patterns. Although the surface is isotropic, the scattering patterns are not. The symmetry is broken by the polarization vector of the incident field. The four-fold symmetry observed in the scattering patterns is also visible in the polarization components of the near-field intensity scattered by the surface, as shown in fig. 18. It is clear that important changes in the polarization take place in the valleys of the surface. These polarization changes may be understood with the help of fig. 19. Consider the incidence of vertically-polarized light upon a valley of a twodimensional perfectly conducting surface. Assume, for simplicity, that the surface can be considered locally flat. The light following the double scattering paths A and B will not lead to any changes in the polarization. However, for the diagonal cross paths C and D, the light will emerge polarized perpendicularly to the initial linear polarization state. The diagonal paths cause a rotation of the polarization vector by 90◦ , explaining the four-fold symmetry observed in the reflected near-field intensities of fig. 18. Measurements of the full hemispherical intensity distribution function of the scattered light from fabricated isotropic metallic surfaces that produce multiple scattering have been reported by Knotts and O’Donnell [1993]. Similar measurements for dielectric surfaces have been reported by Chaikina, Negrete-Regagnon, Ruiz-Cortés and Méndez [2002]. The results provide quantitative data for patterns like the ones shown in fig. 18. We end this section by discussing the problems that have prevented the observation of some interesting multiple-scattering effects.
2, § 4]
Weak localization effects in the multiple scattering of light
179
Fig. 19. Multiple scattering of polarized light from a valley in a two-dimensional perfectly conducting surface. Light scattered along the paths A and B preserve the incident polarization, but the light scattered along the paths C and D is cross-polarized (O’Donnell and Méndez [1987]).
The first theoretical prediction of enhanced backscattering from randomly rough surfaces appeared in 1985 (McGurn, Maradudin and Celli [1985]). The calculations, based on an infinite order perturbation theory, involved a silver Gaussian surface with Gaussian correlation function and parameters a = 100 nm and δ = 5 nm, illuminated with light of wavelength λ = 457.9 nm. Later calculations based on other perturbation theories (Maradudin and Méndez [1993]) and Monte Carlo methods (Michel [1994]) have confirmed the correctness of the prediction. Although the existence of the effect, and the mechanism responsible for its production have been confirmed quite convincingly by the experimental work of West and O’Donnell [1995], it has not been possible to observe enhanced backscattering with Gaussian-correlated weakly rough surfaces, and to obtain quantitative agreement with the theory. The main reason for this surprising fact is that the fabrication of surfaces with such characteristics is a difficult task. These difficulties are compounded by the usual problems associated with scattering measurements with weakly scattering surfaces. The satellite peaks, predicted to occur in the angular distribution of the scattered light by thin films with a random surface (see Section 6.1) have also proved
180
Multiple scattering of light from randomly rough surfaces
[2, § 5
difficult to observe. Here, it is not only the surface characteristics that must be controlled very precisely, but also the thickness of the film.
§ 5. Angular intensity correlation functions The mean differential reflection and transmission coefficients considered in the preceding sections of this review involve the first and second moments of the field scattered from or transmitted through a randomly rough surface. Recently, however, attention has begun to be directed toward the investigation of the conditions under which the amplitudes or intensities of light scattered or transmitted into the far field for different angles of incidence and scattering will be correlated, and the form the resulting correlation functions take. The study of these correlation functions requires the study of not only second moments of the scattered or transmitted fields, but of higher moments as well. Much of the work on multiple-scattering effects on angular correlation functions in rough surface scattering has been carried out for weakly rough random metal surfaces. Thus, Arsenieva and Feng [1993] studied perturbatively the angular correlation function of the scattering amplitude in the cross-polarized scattering of light from a two-dimensional randomly rough metal surface. The squared modulus of this correlation function gives the analog of the C (1) angular intensity correlation function studied earlier in the context of scattering from volume disordered systems (Feng, Kane, Lee and Stone [1988]). By keeping only the contribution from ladder diagrams they obtained what is now called the memory effect peak in the envelope of the C (1) angular intensity correlation function, and related this peak to the enhanced backscattering peak in the angular dependence of the intensity of the light scattered diffusely from the same surface. The same angular amplitude correlation function was also studied perturbatively by Freilikher and Yurkevich [1993]. By keeping only the contribution from maximally-crossed diagrams they obtained what is now called the reciprocal memory effect peak in the envelope of the C (1) angular intensity correlation function, and related it to the enhanced backscattering peak. The analogs of the C (2) and C (3) angular intensity correlation functions discussed earlier in the context of the scattering from volume disordered systems (Feng, Kane, Lee and Stone [1988]) were investigated theoretically in the scattering of p-polarized light from one-dimensional random metal surfaces by Malyshkin, McGurn, Leskova, Maradudin and Nieto-Vesperinas [1997a, 1997b] by small-amplitude perturbation theory. The surface roughness was assumed to be sufficiently weak that these correlations were caused primarily by the multiple
2, § 5]
Angular intensity correlation functions
181
scattering of the surface plasmon polaritons supported by the surface and excited by the incident field through the roughness of the surface. The work of Malyshkin, McGurn, Leskova, Maradudin and Nieto-Vesperinas [1997a, 1997b] also revealed that in rough surface scattering the angular intensity correlation function possesses a contribution that they called the C (10) correlation that is of the same order of magnitude as the C (1) correlation function. This correlation function was overlooked in the earlier studies of Arsenieva and Feng [1993] and of Freilikher and Yurkevich [1993], because the angular amplitude correlation function they investigated was equivalent to the use of the factorization approximation (Shapiro [1986]) in calculating the angular intensity correlation function. This approximation, which was not used in the work of Malyshkin, McGurn, Leskova, Maradudin and Nieto-Vesperinas [1997a, 1997b], is based on the assumption that the amplitude of the field scattered diffusely possesses a circular complex Gaussian joint probability density function (Goodman [1975, 1985]). The fact that it fails to predict the C (10) correlation function shows that the latter assumption is not universally valid. An indication that interesting correlation effects can occur when the factorization approximation is not used had been shown earlier by Nieto-Vesperinas and Sánchez-Gil [1992a, 1992b, 1993] in predicting what they called enhanced long-range correlation functions. In their work Malyshkin, McGurn, Leskova, Maradudin and Nieto-Vesperinas [1997a, 1997b] also showed that there is an additional class of correlations, intermediate between the C (1) and C (2) correlations in the order in the surface profile function in which they first appear. This class was therefore named the C (1.5) correlations. These correlations also arise when the factorization approximation is not made. They are intimately connected with the roughness-induced excitation of forward- and backward-propagating surface plasmon polaritons by the incident light, and introduce peaks into the angular intensity correlation function. Theoretical and experimental studies of angular intensity correlation functions represent a rather new sub-field within the larger field of multiple-scattering effects in the scattering of light from randomly rough surfaces. In this section we present an introduction to this field in the context of the scattering of light from one-dimensional, weakly rough, random surfaces. A study of angular intensity correlation functions in the scattering of light from two-dimensional, weakly rough, random metal surfaces that goes beyond the memory and reciprocal memory effects has been carried out by Malyshkin, McGurn, Leskova, Maradudin and Nieto-Vesperinas [1997b]. The reader interested in this topic is referred to this work.
182
Multiple scattering of light from randomly rough surfaces
[2, § 5
Thus, the general angular intensity correlation function C(q, k|q , k ) we consider here is defined by
C(q, k|q , k ) = I (q|k)I (q |k ) − I (q|k) I (q |k ) , (5.1) where the angle brackets denote an average over the ensemble of realizations of the surface profile function. The intensity I (q|k) entering this expression is defined in terms of the scattering matrix S(q|k) for the scattering of light of frequency ω from a one-dimensional random surface by (Arsenieva and Feng [1993]) 2 1 ω S(q|k) , I (q|k) = (5.2) L1 c where L1 is the length of the x1 -axis covered by the random surface, and the wave numbers k and q are related to the angles of incidence and scattering θ0 and θs , measured counterclockwise and clockwise from the normal to the mean scattering surface, respectively, by k = (ω/c) sin θ0 and q = (ω/c) sin θs . From eqs. (5.1) and (5.2) we can already obtain useful qualitative information about the correlation function C(q, k|q , k ). From the fact that the correlation of I (q|k) with itself should be stronger than the correlation of I (q|k) with I (q |k ) when q = q and k = k, a peak in C(q, k|q , k ) is expected when q = q and k = k. This peak has come to be called the memory effect peak. At the same time, since S(q|k) is reciprocal, S(q|k) = S(−k| − q), a peak in C(q, k|q , k ) is also expected when q = −k and k = −q. This peak is called the reciprocal memory effect peak. In terms of the scattering matrix S(q|k) the correlation function C(q, k|q , k ) takes the form
1 ω 2 C(q, k|q , k ) = 2 S(q|k)S ∗ (q|k)S(q |k )S ∗ (q |k ) L1 c
− S(q|k)S ∗ (q|k) S(q |k )S ∗ (q |k ) . (5.3) As it stands, the expression for C(q, k|q , k ) given by eq. (5.3) contains purely specular contributions, i.e. terms proportional to 2πδ(q − k) and/or 2πδ(q − k ). If we note that, due to the stationarity of the surface profile function, S(q|k) is diagonal in q and k, S(q|k) = 2πδ(q − k)S(k), we can eliminate these uninteresting specular contributions by rewriting C(q, k|q , k ) in terms of the diffuse part of the scattering matrix δS(q|k) = S(q|k) − S(q|k). In addition, if we use the relations between averages of the products of random functions and the corresponding cumulant averages (Kubo [1962]), we can finally write the contribution
2, § 5]
Angular intensity correlation functions
183
to C(q, k|q , k ) that is free from specular contributions as C(q, k|q , k )
2
2 1 ω 2 = 2 δS(q|k)δS ∗ (q |k ) + δS(q|k)δS(q |k ) L1 c
+ δS(q|k)δS ∗ (q|k)δS(q |k )δS ∗ (q |k ) c ,
(5.4)
where · · ·c denotes the cumulant average. The result given by eq. (5.4) is convenient for several reasons. Due to the stationarity of the surface profile function, the average δS(q|k)δS ∗ (q |k ) is proportional to 2πδ(q − k − q + k ). It therefore gives rise to the contribution to C(q, k|q , k ) called C (1) (q, k|q , k ), which contains the memory and reciprocal memory effects. Similarly, the average δS(q|k)δS(q |k ) is proportional to 2πδ(q − k + q − k ), and contributes the correlation function C (10) (q, k|q , k ) to C(q, k|q , k ). The third term on the right-hand side of eq. (5.4), δS(q|k)δS ∗ (q|k)δS(q |k )δS ∗ (q |k )c , is proportional to 2πδ(0), due to the stationarity of the surface profile function, and gives rise to the long-range and infinite range contributions to C(q, k|q , k ) given by the sum C (1.5) (q, k|q , k ) + C (2) (q, k|q , k ) + C (3) (q, k|q , k ). Thus, the contributions to C(q, k|q , k ) that have been named C (1) (q, k|q , k ) and C (10) (q, k|q , k ) are explicitly separated out in eq. (5.4). In addition, from eq. (5.4) we can easily estimate the relative magnitudes of the different contributions to C(q, k|q , k ). Since 2πδ(0) = L1 in onedimension, when the arguments of the delta-functions vanish the C (1) (q, k|q k ) and C (10) (q, k|q k ) correlation functions are independent of the length of the surface L1 , because they are proportional to [2πδ(0)]2 . At the same time the remaining term in eq. (5.4), that yields the sum C (1.5) + C (2) + C (3) , is inversely proportional to the length of the surface, since it is proportional to only the first power of 2πδ(0). Therefore, in the limit of a long surface or large illumination area the long-range and infinite-range correlation functions are small compared to the short-range correlation functions, and vanish in the limit of an infinitely long surface. Consequently, although they contain interesting multiple-scattering effects (Malyshkin, McGurn, Leskova, Maradudin and NietoVesperinas [1997a, 1997b]), they are weak, and therefore will not be considered further here. The preceding results are consistent with the assumptions and conclusions encountered in conventional speckle theory (Goodman [1975, 1985]). Thus, when the surface profile function is assumed to be a stationary random process, and the random surface is assumed to be infinitely long, the scattering matrix S(q|k)
184
Multiple scattering of light from randomly rough surfaces
[2, § 5
becomes the sum of a very large number of independent contributions from different points on the surface. On invoking the central limit theorem it is found that S(q|k) obeys complex Gaussian statistics. In this case eq. (5.4) becomes rigorously (Stoffregen [1979])
2
2 1 ω 2 δS(q|k)δS ∗ (q |k ) + δS(q|k)δS(q |k ) C(q, k|q , k ) = 2 L1 c (5.5a) ≡ C (1) (q, k|q , k ) + C (10) (q, k|q , k ),
(5.5b)
because all cumulant averages of products of more than two Gaussian random processes vanish. The third term on the right-hand side of eq. (5.4) therefore gives the correction to the prediction of the central limit theorem due to the finite length of the random surface. If it is further assumed, as is done in speckle theory, where the disorder is presumed to be strong, that δS(q|k) obeys circular complex Gaussian statistics (Goodman [1975, 1985]), then δS(q|k)δS(q |k ) vanishes, and the expression for C(q, k|q , k ) simplifies to
2 1 ω 2 C(q, k|q , k ) = 2 δS(q|k)δS ∗ (q |k ) L1 c ≡ C (1) (q, k|q , k ).
(5.6)
This result is often called the factorization approximation to C(q, k|q , k ) (Shapiro [1986]). The degree of surface roughness needed for δS(q|k) to change from a complex Gaussian random process to a circular complex Gaussian random process, and hence for C (10) (q, k|q , k ) to vanish, is discussed in Section 5.2.
5.1. Memory and reciprocal memory effects: theory and experiment We have seen from eq. (5.6) that the C (1) (q, k|q , k ) correlation function is given by C (1) (q, k|q , k ) =
2 1 ω2 δS(q|k)δS ∗ (q |k ) . 2 2 L1 c
From eq. (4.12) and the relation S(q|k) = (α0 (q)/α0 (k))1/2 R(q|k), we find that δS(q|k) can be written in terms of the Green’s function G(q|k) as
1/2 1/2 δS(q|k) = −2iα0 (q) G(q|k) − G(q|k) α0 (k). (5.7)
2, § 5]
Angular intensity correlation functions
185
It follows that the amplitude correlation function δS(q|k)δS ∗ (q |k ) takes the form
δS(q|k)δS ∗ (q |k ) = 4α0 (q)α0 (k)α0 (q )α0 (k )
× G(q|k)G∗ (q |k ) − G(q|k) G∗ (q |k ) . 1/2
1/2
1/2
1/2
(5.8)
G(q|k)G∗ (q |k )
is solved by the use of the If the Bethe–Salpeter equation for same approximations used in obtaining eq. (4.35), we obtain the result that C (1) (q, k|q , k ) 16 ω2 α0 (q)α0 (k)α0 (q ) L1 c2 2 2 2 2 × α0 (k ) G(q) G(q ) G(k) G(k ) W 2 g |q − k|
= 2πδ(q − k − q + k )
+ W 2 φ(q)
2 4sp 4sp φ(k) + 2 2 (q + k) + 4ε (q − q ) + 4ε
≡ 2πδ(q − k − q + k )
1 (1) C (q, k|q , q − q + k), L1 0
(5.9)
which defines the envelope function C0(1) (q, k|q , q − q + k). The term 4sp/[(q + k)2 + 4ε ] in this result, which arises from the sum of the contributions from the maximally-crossed diagrams, contributes a peak centered at q = −k, or θs = −θ0 , to the envelope function. This is the reciprocal memory effect peak. The term 4sp/[(q − q )2 + 4ε ], which arises from the sum of the contributions from the ladder diagrams, contributes a peak centered at q = q, or θs = θs , to the envelope function. This is the memory effect peak. Both of these peaks have now been observed experimentally in the scattering of ppolarized light from weakly rough, one-dimensional, random gold surfaces (West and O’Donnell [1999]) (fig. 20). From a comparison of the results given by eqs. (4.36) and (5.9), we see that an experimental determination of C (1) (q, k|q , k ) in the vicinity of either the memory effect peak or the reciprocal memory effect peak yields essentially the same information as a measurement of ∂R/∂θs diff in the vicinity of the enhanced backscattering peak, but without the experimental difficulties associated with placing a detector at the position of the source. It is this feature of the correlation function C (1) that emerged from the theoretical studies of it by Arsenieva and Feng [1993] and by Freilikher and Yurkevich [1993].
186
Multiple scattering of light from randomly rough surfaces
[2, § 5
Fig. 20. Experimental results for the correlation functions C(1) (solid curve) and C(10) (dotted curve) as functions of θ1 for θ0 = 6.3◦ and θs = 8.6◦ when p-polarized light of wavelength λ = 612.7 nm is incident on a one-dimensional randomly rough gold surface (ε(ω) = −9.00 + i1.29). The surface roughness is characterized by the West–O’Donnell power spectrum (2.24) with δ = 115.5 nm, kmin = 8.56 × 10−3 nm−1 , and kmax = 13.3 × 10−3 nm−1 . The peak in C(1) labeled ME is the memory effect peak; the peak labeled RME is the reciprocal memory effect peak (West and O’Donnell [1999]).
We conclude this discussion of the C (1) (q, k|q , k ) correlation function by noting that the property of a speckle pattern that is characterized by the presence of the factor 2πδ(q − k − q + k ) in C (1) (q, k|q , k ) is that if we change the angle of incidence in such a way that k goes into k = k + k, the entire speckle pattern shifts in such a way that any feature initially at q moves to q = q + k. This is why the C (1) correlation function was originally named the memory effect. In terms of the angles of incidence and scattering we have that if θ0 is changed to θ0 = θ0 + θ0 , any feature in the speckle pattern originally at θs is shifted to θs = θs + θs , where θs = (cos θ0 / cos θs )θ0 , to first order in θ0 . 5.2. The correlation function C (10)(q, k|q , k ): theory and experiment We see from eqs. (5.5) and (5.6) that the C (10)(q, k|q , k ) correlation function is given by C (10) (q, k|q , k ) =
2 1 ω2 δS(q|k)δS(q |k ) . 2 2 L1 c
(5.10)
With the use of eq. (5.7) the amplitude correlation function δS(q|k)δS(q |k ) becomes
1/2 1/2 1/2 1/2 S(q|k)δS(q |k ) = −4α0 (q)α0 (k)α0 (q )α0 (k )
× G(q|k)G(q |k ) − G(q|k) G(q |k ) . (5.11)
2, § 5]
Angular intensity correlation functions
187
If the same approximations are made in solving the Bethe–Salpeter equation for G(q|k)G(q |k ) as were made in obtaining G(q|k)G∗ (q |k ), we obtain for the correlation function C (10) (q, k|q , k ): C (10) (q, k|q , k ) 16 ω2 α0 (q)α0 (k)α0 (q )α0 (k ) L1 c2 2 2 2 2 × G(q) G(k) G(q ) G(k ) W 2 g |q − k|
= 2πδ(q − k + q − k )
2 L(q + q ) L(k − q ) + + W φ(q) φ(k) 1 − L(k − q ) 1 − L(q + q ) 2
≡ 2πδ(q − k + q − k ) where L(Q) = W 2
∞ −∞
1 (10) C (q, k|q , q + q − k), L1 0
dp φ(p)G(p)G(Q − p)φ(p), 2π
(5.12)
(5.13)
which defines the envelope function C0 (q, k|q , q + q − k). The function L(Q) displays no resonant behavior for values of Q within the range of variation of Q = k − q and Q = q + q . Hence the envelope function C (10) (q, k|q , q + q − k) is a structureless function of q (θs ) for fixed values of q and k, while k is determined from the condition k = q + q − k(sin θ0 = (10) sin θs + sin θs − sin θ0 ). The envelope function C0 (q, k|q , q + q − k) has been determined experimentally recently in the scattering of p-polarized light from weakly rough, one-dimensional gold surfaces (West and O’Donnell [1999]) (fig. 20), and is a structureless function of its arguments. The property of a speckle pattern that is characterized by the presence of the factor 2πδ(q − k + q − k ) in C (10)(q, k|q , k ) is that if we change the angle of incidence in such a way that k goes into k = k + k, a feature at q = k − q will be shifted to q = k + q, i.e. to a point as much to one side of the new specular direction as the original point was on the other side of the original specular direction. For one and the same incident beam (k = k), the C (10) correlation function therefore reflects the symmetry of the speckle pattern with respect to the specular direction (in wavenumber space). We have noted in Section 5 that in conventional speckle theory, where the surface roughness is assumed to be strong (Goodman [1975, 1985]), δS(q|k) is found to obey circular complex Gaussian statistics, in which case the correlation function C (10)(q, k|q , k ) must vanish. In this section, however, we have shown that (10)
188
Multiple scattering of light from randomly rough surfaces
[2, § 5
for weakly rough random surfaces C (10) (q, k|q , k ) is nonzero, which implies that for such surfaces δS(q|k) must obey only complex Gaussian statistics. How δS(q|k) transforms from a complex Gaussian random process into a circular complex Gaussian random process with increasing surface roughness has been analyzed by Leskova, Simonsen and Maradudin [2002], on the basis of the assumption that the surface profile function ζ (x1 ), and hence δS(q|k), is a stationary random process. The simplest version of phase perturbation theory (Sánchez-Gil, Maradudin and Méndez [1995]) for the scattering of a scalar plane wave from a one-dimensional random surface on which the Dirichlet boundary condition is satisfied was used to calculate δS(q|k). It was found that in the limit as δ/λ tends to zero the joint probability density of the {δS(q|k)} has the complex Gaussian form for a fixed value of the transverse correlation length a. However, as δ/λ increases, the joint probability density of the {δS(q|k)} approaches the circular complex Gaussian form. This can occur for values of δ/λ as small as about δ/λ = 0.6 for a fixed value of a.
5.3. Correlations in film systems Our discussion of angular intensity correlation functions up to now has been limited to those arising in the scattering of light from a one-dimensional random surface bounding a semi-infinite metal. Such surfaces can support at most a single surface electromagnetic mode, and the peaks occurring in the C (1) correlation function are associated with the excitation of this mode. If we turn to a structure that supports two or more surface or guided modes, such as a dielectric film on a reflecting substrate or a free-standing or supported film, it may be expected that, just as the contribution to the mean differential reflection coefficient from the diffuse component of the scattered field exhibits a richer structure – satellite peaks – than is the case for a structure supporting a single surface wave, so will the angular intensity correlation function. The possibility that this should be the case was first noted by Freilikher, Pustilnik and Yurkevich [1994], in the context of the angular intensity correlation function C (1) . In four subsequent theoretical studies the angular intensity correlations of light reflected from a randomly rough dielectric film on a perfectly conducting substrate (Sánchez-Gil [1997], McGurn and Maradudin [1997, 1998a]), or reflected from and transmitted through a randomly rough freestanding metal film (McGurn and Maradudin [1997, 1998b]), were calculated. The resulting correlation functions indeed have a richer structure than the correlation functions calculated for the light scattered from a semi-infinite metal with a randomly rough surface. In particular, the C (1) correlation function acquires additional peaks whose angular positions depend on differences of the wave numbers
2, § 6]
Multiple-scattering effects in the scattering
189
of the guided and surface waves. That is, there are now additional memory effect peaks, and additional reciprocal memory effect peaks. Such peaks, of course, have no counterparts in the scattering of light from a semi-infinite metal. Additional peaks also occur in the C (1.5) correlation function, but these are due simply to the multiplicity of guided and surface waves in the film structures, and their origin is the same as that of the peaks in C (1.5) occurring in the scattering from the surface of a semi-infinite metal.
§ 6. Multiple-scattering effects in the scattering of light from complex media bounded by a rough surface 6.1. Coherent effects associated with the interference of nonreciprocal optical paths Although enhanced backscattering is the best known coherent effect caused by the interference of multiply-scattered optical paths, it is not the only one. Like enhanced backscattering some of the other coherent effects arise from the interference of reciprocal optical paths. However, interesting effects also arise from the interference of nonreciprocal optical paths. In this section we survey these other coherent effects. The simplest and the earliest studied of the coherence effects arising from the interference of nonreciprocal optical paths was discovered in an investigation of the scattering of p-polarized light incident from vacuum on a one-dimensional random metal or n-type semiconductor surface in the presence of a static magnetic field oriented parallel to the generators of the surface (i.e. parallel to the x2 -axis), and hence normal to the plane of incidence, the x1 x3 -plane. The scattering in this case is co-polarized, and the plane of scattering coincides with the plane of incidence. In this geometry the p-polarized surface magnetoplasmon polariton at the frequency ω of the incident light supported by a planar interface between the vacuum and the metal or semiconductor becomes nonreciprocal. The wavenumber k+ (ω) of the surface magnetoplasmon polariton propagating in the +x1 -direction is different from that of the surface magnetoplasmon polariton propagating in the −x1 -direction, k− (ω). Both of these wavenumbers depend on the strength of the applied magnetic field. At each nonzero value of this field k+ (ω) is smaller than k− (ω) for all frequencies ω smaller than a lower critical frequency that depends on the field strength. At frequencies above this critical value only k+ (ω) exists: there is no backward propagating surface magnetoplasmon polariton. Then, at frequencies above a higher critical frequency k+ (ω) also no longer exists. Within the
190
Multiple scattering of light from randomly rough surfaces
[2, § 6
Fig. 21. Double-scattering sequences occurring in the scattering of p-polarized light from a semi-infinite metal or n-type semiconductor, bounded by a one-dimensional random surface, that supports a nonreciprocal surface plasmon polariton when an external dc magnetic field is directed into the plane of the figure.
frequency range where both k+ (ω) and k− (ω) exist, the difference k+ (ω) − k− (ω) increases in magnitude with increasing magnetic field strength. These nonreciprocal properties of the surface magnetoplasmon polariton give rise to an interesting multiple-scattering effect when p-polarized light is incident on the random surface of a semi-infinite metal or n-type semiconductor. Thus, let us consider the system depicted in fig. 21, and consider a doublescattering process in which the incident light of frequency ω excites a forward propagating surface magnetoplasmon polariton of wavenumber k+ (ω) through the breakdown of the infinitesimal translational invariance of the system caused by the roughness. The surface wave thus excited propagates along the surface, is scattered once more by the roughness, and is converted back into a volume wave in the vacuum that propagates away from the surface (fig. 21). All such doublescattering processes are assumed to be uncorrelated due to the random nature of the surface. However, to any given sequence there corresponds a partner, in which the light and surface wave are scattered from the same points, but in the reverse order. This means that the magnitude of the wavenumber of the backwardpropagating surface wave excited is k− (ω). The phase difference between the path (ABCD)+ and the path (A CBD )− , where the subscript denotes the surface wave excited by the incident light, is φ+− = rBC · (kin + ksc ) − |rBC | k+ (ω) − k− (ω) , (6.1) where kin and ksc are the wave vectors of the incident and scattered light, respectively, while rBC is the vector joining the points B and C. We see from eq. (6.1)
2, § 6]
Multiple-scattering effects in the scattering
191
that in the absence of the magnetic field, when k+ (ω) = k− (ω), constructive interference (φ+− = 0) occurs when ksc = −kin . This condition describes scattering into the retroreflection direction. The complex amplitudes of the reciprocal paths (ABCD)+ and (A CBD )− in this case are equal. They must be added, and the modulus of the sum squared, in calculating their contribution to the intensity of the scattered field. For scattering into directions other than the retroreflection direction the two waves have a nonzero phase difference, and very rapidly become incoherent, so that only their intensities add. Consequently, the intensity of scattering into the retroreflection direction is a factor of two larger than it is for scattering into other directions, due to the contribution from the cross terms in the squared-modulus. Of course, the contribution from the single-scattering processes has to be subtracted in obtaining this factor of two enhancement, since it is not subject to coherent backscattering. This increased intensity of scattering into a narrow angular interval about the retroreflection direction is called enhanced backscattering. However, in the presence of the magnetic field, when k+ (ω) is unequal to k− (ω), we see from eq. (6.1) that constructive interference occurs for scattering into a direction for which ksc = −kin . This gives rise to a peak in the angular distribution of the intensity of the diffuse component of the scattered light at the scattering angle corresponding to this value of ksc . Unlike the enhanced backscattering peak in zero applied field, however, the height of this peak has no simple relation to the height of the background at its position, when the contribution of the single-scattering processes is subtracted off, because the amplitudes of the two interfering waves are not equal in this case. If we note that the points B and C lie on the scattering surface, so that the vector rBC is essentially parallel to the mean scattering plane, we obtain from eq. (6.1) that the scattering angle at which this peak occurs is related to the angle of incidence θ0 by sin θs = − sin θ0 +
c k+ (ω) − k− (ω) . ω
(6.2)
Although the result given by eq. (6.2) was obtained on the basis of a doublescattering process, it remains unchanged when multiple-scattering processes of all orders are taken into account. Thus, since k+ (ω) is smaller than k− (ω), the coherent interference of the nonreciprocal optical paths shifts the position of the peak away from the retroreflection direction in the direction of larger scattering angles, and the magnitude of the shift is larger the larger is the strength of the applied magnetic field. These predictions have been verified by the results of perturbative (McGurn, Maradudin and Wallis [1991]) and computer simulation (Lu, Maradudin and Wallis [1991]) calculations of the scattering of p-polarized light
192
Multiple scattering of light from randomly rough surfaces
[2, § 6
Fig. 22. The contribution to the mean differential reflection coefficient from the diffuse component of the scattered light when a p-polarized beam of light of wavelength λ = 43.2 µm (ω/ωp = 0.68) is incident on a one-dimensional random surface of n-GaAs, characterized by the Gaussian power spectrum (2.23) with δ = 3.14 µm and a = 15.14 µm. The angle of incidence is θ0 = 5◦ . An external dc magnetic field H is applied parallel to the grooves and ridges of the surface. (a) ωc /ωp = 0 (H = 0); (b) ωc /ωp = 0.2 (H = 2.46 × 104 G); (c) ωc /ωp = 0.4 (H = 3.28 × 10−4 ); (d) ωc /ωp = 0.5 (H = 4.1 × 104 G), where ωc is the cyclotron frequency and ωp is the plasma frequency of the conduction electrons. The position of the enhanced backscattering peak is at θs = −5◦ , −6.8◦ , −7.2◦ , and −7.5◦ , respectively (Lu, Maradudin and Wallis [1991]).
from the structure depicted in fig. 21. In the results of these calculations, especially in those of the latter calculations depicted in fig. 22, it was found that the angle at which the peak in the contribution to the mean differential reflection coefficient occurred departed slightly from the value given by eq. (6.2) when the values of k+ (ω) and k− (ω) for a planar surface were used. This is due presumably to the fact that in the computer simulation calculations the roughness-induced corrections to the values of k+ (ω) and k− (ω) are taken into account. The coherent effect caused by the interference of nonreciprocal optical paths that we have just considered is present in the scattering of p-polarized light from a random surface that supports only a single surface electromagnetic wave, albeit a nonreciprocal one. If we now turn our attention to the scattering of light
2, § 6]
Multiple-scattering effects in the scattering
193
Fig. 23. Double-scattering sequences occurring in the scattering of electromagnetic waves from, and their transmission through, a bounded structure whose illuminated surface is a one-dimensional random surface, that supports several surface or guided waves and their reciprocal partners.
from, or its transmission through, a structure that supports two or more surface or guided waves, new features appear in the angular distribution of the intensity of the incoherent component of the reflected or transmitted light not present in the results described so far, that are also caused by the coherent interference of multiply-scattered nonreciprocal optical paths, even though the surface or guided waves themselves are now reciprocal. Thus, let us consider a free-standing dielectric or metallic film, characterized by an isotropic, complex, dielectric function ε(ω), in the region −d < x3 < ζ(x1 ), with vacuum in the regions x3 > ζ(x1 ) and x3 < −d (fig. 23). This structure is illuminated from the region x3 > ζ (x1 ) by a plane wave of p- or s-polarized light, whose plane of incidence is the x1 x3 -plane. The scattered and transmitted light in this geometry is co-polarized, and the plane of scattering and of transmission is also the x1 x3 -plane. We assume that this structure supports N ( 2) guided or surface electromagnetic waves at the frequency ω of the incident light (N = 2 in the case of a metal film illuminated by p-polarized light), whose wave numbers will be denoted by k1 (ω), k2 (ω), . . . kN (ω). Each of these guided or surface waves is assumed to be reciprocal. The existence of two or more guided or surface electromagnetic waves can add peaks to the angular distribution of the intensity of the light scattered dif-
194
Multiple scattering of light from randomly rough surfaces
[2, § 6
fusely from, or transmitted diffusely through, the film. To see how they arise, we examine double-scattering processes contributing to the scattered and transmitted fields. A typical double-scattering path ABCD and its partner A CBD (fig. 23) are each N -fold degenerate in the sense that along the segment BC there are N “channels” with different phase factors. The phase difference between the path (ABCD)m , and the path (A CBD )n , where the subscript gives the number of the waveguide mode excited by the incident light, is φnm = rBC · (kin + ksc ) + |rBC | kn (ω) − km (ω) . (6.3) We see from this expression that constructive interference (φnm = 0) can occur when n = m and ksc = −kin , i.e. due to the coherent interference of doubly-scattered reciprocal paths. Such scattering processes give rise to enhanced backscattering. However, we see from eq. (6.3) that constructive interference can also occur for scattering into other directions, ksc = −kin , for some n = m. The satisfaction of these conditions gives rise to satellite peaks in the angular distribution of the intensity of the light scattered diffusely. These are due to the coherent interference of doubly scattered nonreciprocal paths. The scattering angles at which these satellite peaks occur are related to the angle of incidence by sin θs = − sin θ0 ± (c/ω) kn (ω) − km (ω) , n = m. (6.4) In transmission, the phase difference between the double-scattering path (ABCE)m and its partner (A CBE )n (fig. 23) is φnm = rBC · (kin + k∗sc ) + |rBC | kn (ω) − km (ω) , (6.5) where the vector k∗sc is obtained from the vector ksc by reversing the sign of the 3-component of the latter. From eq. (6.5) it follows that constructive interference (φnm = 0) can occur when n = m and k∗sc = −kin . This is a new coherent effect arising from the interference of doubly-scattered reciprocal paths. The condition k∗sc = −kin translates into a peak in the angular distribution of the intensity of light transmitted diffusely through the film in the antispecular direction, i.e. in the direction opposite to that for specular reflection from the film. This peak is called enhanced transmission (McGurn and Maradudin [1989]). At the same time, we see from eq. (6.5) that constructive interference can also occur for transmission into other directions k∗sc = −kin for some n = m. The satisfaction of these conditions gives rise to satellite peaks in the angular distribution of the light transmitted diffusely. These peaks are due to the coherent interference of non-reciprocal optical paths. These satellite peaks occur at angles of transmission θt that are related
2, § 6]
Multiple-scattering effects in the scattering
195
to the angle of incidence by sin θt = − sin θ0 ± (c/ω) kn (ω) − km (ω) ,
n = m,
(6.6)
(Freilikher, Pustilnik, Yurkevich and Maradudin [1994], Sánchez-Gil, Maradudin, Lu and Freilikher [1995]). Although the preceding discussion has been based on the double-scattering processes depicted in fig. 23, the conclusions reached, in particular eqs. (6.4) and (6.6), remain valid when all higher-order multiple-scattering processes are taken into account. Of the N(N − 1) satellite peaks predicted to occur at the angles of scattering and transmission given by eqs. (6.4) and (6.6), respectively, not all may be observed, however. This will be the case when the magnitude of the right-hand side of eq. (6.4) or eq. (6.6) is larger than unity. Furthermore, among the real satellite peaks that should appear when the magnitude of the right-hand sides of eqs. (6.4) and (6.6) is smaller than unity, not all of them might be intense enough to be observable. We also note that the occurrence of satellite peaks in general appears to be limited to the case of the scattering of p- or s-polarized light from a one-dimensional random surface of a structure that supports two or more surface or guided waves. In the case of the scattering of light from an isotropic two-dimensional surface of such a structure the ensemble averaging of the scattered intensity restores isotropy in the mean scattering plane, and thereby eliminates the occurrence of special scattering angles at which satellite peaks could occur (Kaw