EDITORIAL ADVISORY BOARD
G.S. Agarwal
Stillwater, USA
T. Asakura
Sapporo, Japan
M.V. Berry
Bristol, England
C. Brosseau
Brest, France
A.T. Friberg
Stockholm, Sweden
F. Gori
Rome, Italy
D.F.V. James
Toronto, Canada
P. Knight
London, England
G. Leuchs
Erlangen, Germany
P. Milonni
Los Alamos, NM, USA
J.B. Pendry
London, England
J. Peˇrina
Olomouc, Czech Republic
J. Pu
Quanzhou, PR China
W. Schleich
Ulm, Germany
Elsevier Radarweg 29, PO Box 211, 1000 AE Amsterdam, The Netherlands The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, UK
First edition 2009 c 2009 Elsevier B.V. All rights reserved Copyright No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without the prior written permission of the publisher Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone (+44) (0) 1865 843830; fax (+44) (0) 1865 853333; e-mail:
[email protected]. Alternatively you can submit your request online by visiting the Elsevier web site at http://elsevier.com/locate/ permissions, and selecting Obtaining permission to use Elsevier material Notice No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made Library of Congress Catalog Card number: 61-19297 ISBN: 978-0-444-53528-3 ISSN: 0079-6638 For information on all Elsevier publications visit our web site at elsevierdirect.com 09 10 11 12
10 9 8 7 6 5 4 3 2 1
Printed and bound in Great Britain
PREFACE
This volume continues the tradition of its 53 predecessors of presenting accounts of developments in different areas of optics and related subjects. The first article by J. Turunen and A. Friberg describes basic concepts relating to optical fields, which possess various invariant features on propagation. The article presents an account of the generation of such beams and of their uses, for example for wave modulation, and in metrology. Their uses for particle guiding and trapping, and in nonlinear optics are also discussed. The second article by D. Mogilevtsev and S. Kilin reviews modern methods used in the analysis of quantum systems that interact with reservoirs. Particular attention is paid to electromagnetic fields in structured media such as photonic crystals. The article, which follows, by C. Brosseau presents a thorough account of a rather fascinating historical development of polarisation and coherence optics. The author argues that geometric algebra provides a unifying conceptual framework from which to view different descriptions of polarisation states. In the fourth article, T.C. Ralph and C.J. Pryde review the field of optical quantum computing. The article considers various implementations that have been proposed, as well as the experimental progress that has been made towards realising them. Both linear and nonlinear approaches are discussed, and particle and field encodings are examined. The prospects for large-scale optical quantum computing in terms of the most promising architectures are considered, and also the technical requirements needed for realising them. The next article by K. Sakoda and J.W. Haus presents a review of research on photonic crystals. The subject originated at microwave frequencies in the 1990s. After much effort, successful results were obtained at optical frequencies, leading to the creation of highly localised waves in the optical range of the electromagnetic spectrum. The article describes the history and the state-of-art of the subject. The concluding article by L.A. Romero and F.M. Dickey deals with a relatively recent development in the field of diffraction gratings; namely, the production of so-called beam splitting gratings. Such gratings, v
vi
Preface
one-dimensional as well as two-dimensional, are used to divide a laser beam into many beams. Gratings of these kinds are used for various industrial and scientific applications, for example in parallel processing, and as elements in laser machine sensors. Emil Wolf Department of Physics and Astronomy and The Institute of Optics, University of Rochester, Rochester, New York 14627, USA October 2009
CONTENTS
Preface
v
1. Propagation-Invariant Optical Fields Jari Turunen and Ari T. Friberg 1. Introduction 1.1. Diffraction-free Beams 1.2. Historical Overview 2. Scalar Fields 2.1. Durnin’s General Solution 2.2. Bessel Field Modes 2.3. Angular Spectra of Propagation-invariant Fields 2.4. Solutions in Elliptic and Parabolic Coordinate Systems 2.5. Custom-designed Propagation-invariant Fields 2.6. Relation to Self-imaging, Rotating, Frozen, and Airy Fields 2.7. Finite-aperture Approximations 2.8. Finite-angular-spectrum Approximations 3. Electromagnetic Fields 3.1. Vectorial Extension of Scalar Results 3.2. Particular Polarization Configurations 3.3. Solutions with Two Montgomery’s Rings 3.4. Finite Approximations 4. Partially Coherent Fields 4.1. General Field Representation 4.2. Specific Examples 4.3. Finite Approximations 4.4. Electromagnetic Solutions 5. Polychromatic and Pulsed Localized Waves 5.1. Field Representations 5.2. Dispersion Relations and Spectra 5.3. Examples of Stationary Fields 5.4. Examples of Pulsed Localized Waves 5.5. Partially Coherent and Electromagnetic Solutions 5.6. Finite Approximations 6. Propagation Through Lens Systems 6.1. Monochromatic Fields in Paraxial Systems 6.2. Finite-aperture Fields in Paraxial Systems 6.3. Polychromatic Fields in Paraxial Systems 7. Generation Methods 7.1. Spatial Wavefront Modulation 7.2. Spatial-frequency Selection and Modulation 7.3. Axicons and Diffractive Elements
1 2 2 3 7 7 8 10 13 13 16 18 22 24 25 26 30 31 32 32 33 35 38 40 41 42 47 51 57 59 60 60 63 65 66 66 67 68 vii
Contents
viii
7.4. Tandem Optical Systems 7.5. Optical Resonators 7.6. Reconfigurable Devices 7.7. Generation of Polychromatic and Pulsed Fields 8. Applications 8.1. Precision Instrumentation and Metrology 8.2. Manipulation of Atoms and Particles 8.3. Nonlinear Optics 9. Conclusions Acknowledgments References
70 71 73 73 76 77 78 79 80 81 81
2. Theoretical Tools for Quantum Optics in Structured Media Dmitri Mogilevtsev and Sergei Kilin 1. Introduction 2. Exact and Approximate Solutions for System Plus Reservoir 2.1. Spontaneous Emission into a Bosonic Reservoir: The Problem 2.2. The Density-of-States and Photonic Band-Gaps 2.3. Spontaneous Dynamics and the Markovian Approximation 2.4. The Cluster Expansion and Collective Operators 2.5. The Pseudo-Modes Method 2.6. Dephasing into a Bosonic Reservoir 3. Master Equations: Deterministic Approaches 3.1. Projection Operator Method and the Cumulant Expansion 3.2. Validity of Perturbative Approximations: Spontaneous Emission 3.3. A System-Reservoir Border 3.4. Separating the Reservoir 3.5. Solving Master equations I 4. Master Equations: Stochastic Approaches 4.1. Quantum Monte-Carlo Procedure 4.2. Non-Lindblad Quantum Monte-Carlo Procedure 4.3. Pseudomodes and Quantum Monte-Carlo Procedure 4.4. Quantum Diffusion Procedure 4.5. Non-Markovian Quantum Diffusion Procedure 5. Concluding Remarks Acknowledgements References
89 90 92 93 95 97 99 103 105 109 111 114 117 120 123 127 129 133 136 138 141 144 145 145
3. Polarization and Coherence Optics: Historical Perspective, Status, and Future Directions Christian Brosseau
149
1. 2. 3. 4. 5. 6.
149 151 163 172 188 197 199 199
Introduction From Bartholinus (1669) to Stokes (1852) From Stokes (1852) to Poincar´e (1892) From Poincar´e (1892) to Wolf (1954) From Wolf (1954) to Wolf (2007) Summary and Prospects Acknowledgments References
Contents
ix
4. Optical Quantum Computation Tim C. Ralph and Geoff J. Pryde
209
1. Introduction 1.1. Quantum Computation 1.2. Quantum Optics 2. Optical Qubits 2.1. Dual-Rail Encoding 2.2. Single-Rail Encoding 3. Universal Optical Quantum Gate Sets 3.1. Nonlinear Gates 3.2. Linear Optics Gates 3.3. Coherent State Gates 4. Cluster States 5. Fault Tolerance 5.1. Loss Tolerance 5.2. Telecorrection 5.3. Thresholds and Resource Counts for Optical Schemes 6. Experimental Demonstrations 6.1. Linear Optics, Particle Encodings 6.2. Circuits and Protocols Using Linear Optics Gates 6.3. Linear Optics, Field Encodings 6.4. Nonlinear Optical Quantum Computing Experiments 7. Auxiliary Components 7.1. Detectors 7.2. Sources 7.3. Memory 7.4. Integrated Optics 8. Conclusion and Summary Acknowledgments References
210 210 213 215 215 217 218 219 223 229 234 236 237 238 239 242 242 249 250 251 252 253 255 259 261 262 263 263
5. Science and Engineering of Photonic Crystals Kazuaki Sakoda and Joseph W. Haus
271
1. Introduction 2. Pursuit of the 3D PBG 2.1. Photonic Band Calculations 2.2. Realization of the 3D PBG 3. Eigenmode Symmetry and Transmittance 4. Defect Modes and Waveguides 4.1. Defect Modes 4.2. PC Waveguides 5. PC Slab 5.1. Fabrication of PC Slabs 5.2. Dispersion Relation 5.3. Diffraction by Periodic Structures 5.4. Air-Bridge PC Slab
271 273 273 279 284 288 288 290 290 290 292 296 299
x
Contents
6. Cavity QED with PC 7. Band Edge Enhancement 8. Concluding Remarks References
303 306 310 313
6. The Mathematical Theory of Laser Beam-Splitting Gratings Louis A. Romero and Fred M. Dickey
319
1. Introduction 2. The Basic Mathematical Problem 2.1. Reduction to a Problem in Fourier Analysis 2.2. Three Optimization Problems 2.3. Problems with an Even Number of Modes 2.4. Two Dimensional Gratings 3. An Upper Bound on the Efficiency 3.1. Some Basic Theorems 3.2. Numerical Results 4. Two Beam Splitting 5. Dammann Gratings 5.1. Introduction 5.2. The Role of α 5.3. Some Properties of Dammann Transmission Functions 5.4. Formulation Using Lagrange Multipliers 5.5. The Effect of Symmetry 5.6. Problems with an Even Number of Modes 6. One Dimensional Continuous Gratings 6.1. The Uniformity Optimization Problem 6.2. Solutions Using the Calculus of Variations 6.3. Numerical Calculations 7. Two Dimensional Fourier Series 7.1. Two Dimensional Fourier Series on General Lattices 7.2. Comments on Two Dimensional Symmetry 8. Two Dimensional Beam Splitting 8.1. The General Theory 8.2. Symmetry and Symmetry Breaking 8.3. Examples Using Square Gratings 8.4. Examples Using Hexagonal Gratings 9. Summary Acknowledgements Appendix A Appendix B References
320 326 326 327 331 333 333 333 336 338 341 341 347 348 350 353 354 354 356 358 359 361 363 366 368 368 370 371 375 378 380 380 381 384
Author Index for Volume 54 Subject Index for Volume 54 Contents of Previous Volumes Cumulative Index - Volumes 1–54
387 407 411 423
CHAPTER
1 Propagation-Invariant Optical Fields Jari Turunen a and Ari T. Friberg a,b,c a University of Joensuu, Department of Physics and Mathematics,
P.O. Box 111, FI-80101 Joensuu, Finland b Helsinki University of Technology, Department of Applied Physics, P.O. Box 3500, FI-02015 TKK, Finland c Royal Institute of Technology, Department of Microelectronics and Applied Physics, Electrum 229, SE-164 40 Kista, Sweden
Contents
1.
2.
3.
4.
Introduction 1.1. Diffraction-free Beams 1.2. Historical Overview Scalar Fields 2.1. Durnin’s General Solution 2.2. Bessel Field Modes 2.3. Angular Spectra of Propagation-invariant Fields 2.4. Solutions in Elliptic and Parabolic Coordinate Systems 2.5. Custom-designed Propagation-invariant Fields 2.6. Relation to Self-imaging, Rotating, Frozen, and Airy Fields 2.7. Finite-aperture Approximations 2.8. Finite-angular-spectrum Approximations Electromagnetic Fields 3.1. Vectorial Extension of Scalar Results 3.2. Particular Polarization Configurations 3.3. Solutions with Two Montgomery’s Rings 3.4. Finite Approximations Partially Coherent Fields 4.1. General Field Representation 4.2. Specific Examples 4.3. Finite Approximations 4.4. Electromagnetic Solutions
2 2 3 7 7 8 10 13 13 16 18 22 24 25 26 30 31 32 32 33 35 38
c 2009 Elsevier B.V. Progress in Optics, Volume 54 ISSN 0079-6638, DOI 10.1016/S0079-6638(10)05406-5 All rights reserved.
1
2
Propagation-Invariant Optical Fields
5.
Polychromatic and Pulsed Localized Waves 5.1. Field Representations 5.2. Dispersion Relations and Spectra 5.3. Examples of Stationary Fields 5.4. Examples of Pulsed Localized Waves 5.5. Partially Coherent and Electromagnetic Solutions 5.6. Finite Approximations 6. Propagation Through Lens Systems 6.1. Monochromatic Fields in Paraxial Systems 6.2. Finite-aperture Fields in Paraxial Systems 6.3. Polychromatic Fields in Paraxial Systems 7. Generation Methods Spatial Wavefront Modulation 7.1. 7.2. Spatial-frequency Selection and Modulation 7.3. Axicons and Diffractive Elements 7.4. Tandem Optical Systems 7.5. Optical Resonators 7.6. Reconfigurable Devices 7.7. Generation of Polychromatic and Pulsed Fields 8. Applications 8.1. Precision Instrumentation and Metrology 8.2. Manipulation of Atoms and Particles 8.3. Nonlinear Optics 9. Conclusions Acknowledgments References
40 41 42 47 51 57 59 60 60 63 65 66 66 67 68 70 71 73 73 76 77 78 79 80 81 81
1. INTRODUCTION 1.1. Diffraction-free Beams Diffraction by circular and annular apertures in classical optical instruments, including microscopes, photographic lenses, and astronomical telescopes, imposes fundamental resolution limits, which cannot be overcome by the introduction of any aberrations or apodization of the aperture. A circular aperture is ideal, in the sense that it produces the best confinement of optical energy around the optical axis. If a central obstruction with an increasing radius is introduced, the side lobes of the diffraction pattern grow in relation to the axial peak in the focal plane, but the focal depth increases simultaneously. Ultimately, with a thin annular aperture, the focal-plane diffraction pattern approaches the distribution J02 (αρ), where J0 (x) is the zeroth-order Bessel function of the first kind, α is a scale factor, and ρ is the radial coordinate. If the thin annular aperture is located in the front focal plane of an ideal positive lens, the transverse J02 (αρ) profile is observed over a long
Introduction
3
range behind the lens. Remarkably, the scale factor α is invariant within this range, as demonstrated by Durnin, Miceli Jr., and Eberly (1987). With an aberration-free lens of unlimited aperture this propagationinvariant range would actually extend to infinity, leading to a socalled ‘diffraction-free beam’ introduced by Durnin (1987) as an exact solution to the Helmholtz equation in free space. Durnin’s papers soon attracted both enthusiasm and criticism (Durnin, Miceli Jr., & Eberly, 1991; Sprangle & Hafizi, 1991). A flood of contributions followed, in which the theory, production methods, and applications of these wave fields were investigated. This work is reviewed here. We emphasize that ‘diffraction-free beams’ (or ‘non-diffracting beams’ as they are sometimes also called) are not beams in the conventional sense. Beams are usually understood as spatially well-confined pencils of light with narrow far-zone diffraction patterns. However, the envelope of the transverse intensity distribution of a diffraction-free beam falls off as 1/ρ when ρ → ∞, which implies that the field contains an infinite amount of energy. As with a plane wave, which is indeed the simplest (though trivial) example of a propagation-invariant field, the effects of diffraction become evident as soon as the field is truncated. In addition, the far-zone diffraction patterns of fields introduced by Durnin are delta-function rings rather than being concentrated around the optical axis. Hence we use the more general term ‘propagation-invariant field’ instead of ‘diffraction-free beam’.
1.2. Historical Overview The history of propagation-invariant wave fields can be traced back at least to Stratton (1941) and van Nie (1964), who considered elementary non-diffracting solutions, now known as Bessel beams, in electromagnetism. Bessel-like beams were produced more than half a century ago by McLeod (1954) as axial line images formed by conical axicons (Jaroszewicz, Burvall, & Friberg, 2005). Axicons were also employed in the 1980’s in Moscow to generate ‘continuous laser sparks’ from high-power lasers (Bunkin, Korobkin, Kurinyi, Polonskii, & Pyatnitskii, 1983; Polonskiy, 2005). Diffractive optical elements were employed to produce non-diffracting Bessel beams of any order by Vasara, Turunen, and Friberg (1989b). Subsequently diffraction-free beams have been generated by a wide variety of methods, including laser resonators, spherically aberrated lenses, dual-element optical systems, interferometers, spatial light modulators, and nonlinear liquid crystals. These will be discussed in more detail later. For general overviews of axicon fields and Bessel beams, we refer the reader to Soroko (1996), Jaroszewicz (1997), and McGloin and Dholakia (2005).
4
Propagation-Invariant Optical Fields
Ideal propagation-invariant waves are characterized by the fact that the field factors in transversally and longitudinally dependent parts behave in such a way that the intensity of the longitudinal part is independent of propagation distance. The Bessel beams and other, more general rotating and spiraling fields containing e.g. Neumann and Hankel functions, are examples of this factorization in circular cylindrical coordinates (Ch´avez-Cerda, McDonald, & New, 1996). An almost limitless variety of non-diffracting patterns and arrays, whether random or periodic, can be produced in the Cartesian coordinate system (Bouchal, 2002; Cottrell, Craven, & Davis, 2007; Turunen, Vasara, & Friberg, 1991). The Helmholtz equation, which governs the propagation of monochromatic scalar waves in optics, also separates into appropriate transversal and longitudinal parts in both elliptic and parabolic cylindrical coordinates, leading to propagation-invariant Mathieu beams (Dartora, Zamboni´ Rached, Nobrega, Recami, & Hern´andez-Figueroa, 2003; Guti´errezVega, Iturbe-Castillo, & Ch´avez-Cerda, 2000; Guti´errez-Vega, IturbeCastillo, Ramir´ez, Tepich´ın, Rodr´ıguez-Dagnino, Ch´avez-Cerda, & New, 2001) and propagation-invariant parabolic beams (Bandres, Guti´errez´ Vega, & Ch´avez-Cerda, 2004; Lopez-Mariscal, Bandres, Ch´avez-Cerda, & Guti´errez-Vega, 2005), respectively. The higher-order circularly symmetric non-diffracting waves are doughnut (dark, hollow) beams that possess helical wavefronts corresponding to optical vortices (phase singularities). Since diffraction-free vortex beams exert radiation pressure and carry orbital angular momentum, they can be used for micro-particle trapping in optical tweezers (He, Heckenberg, & Rubinstein-Dunlop, 1995) and rotation in optical wrenches (Ch´avez-Cerda, Padgett, Allison, New, Guti´errez-Vega, O’Neil, MacVicar, & Courtial, 2002; Volke-Sepulveda, Garc´es-Ch´avez, Ch´avez-Cerda, Artl, & Dholakia, 2002). The theory of coherent, non-diffracting scalar waves was early on generalized into the domain of (spatially) partially coherent, propagationinvariant fields (Friberg, Vasara, & Turunen, 1991; Turunen, Vasara, & Friberg, 1991) and also extended to electromagnetic fields that satisfy Maxwell’s equations rigorously (Bouchal & Oliv´ık, 1995; Mishra, 1991; Turunen & Friberg, 1993a). A classification of the propagation-invariant vector fields based on the complex Poynting theorem, as well as experimental demonstrations, were presented by Bouchal, Hor´ak, and Wagner (1996) and some special cases were investigated by Hor´ak, Bouchal, and Bajer (1997). Nonstationary electromagnetic fields whose monochromatic components are non-diffracting waves were analyzed by Bouchal, Bajer, and Bertolotti (1998) with the help of the Hertz vectors. Also, white-light propagation-invariant beams were demonstrated ´ recently (Fischer, Brown, Morris, Lopez-Mariscal, Wright, Sibbett, & Dholakia, 2005). While partially coherent, rigorously diffraction-free
Introduction
5
vector fields have received little attention so far, the generation and characteristics of various propagation-invariant electromagnetic waves have been studied; these works will be discussed more later. A valuable general review of non-diffracting scalar and electromagnetic fields, including their physical properties, the effects of partial coherence, experiments, and applications, is that by Bouchal (2003). As well as continuous-wave non-diffracting beams, pulsed fields that propagate in vacuum or in uniform medium fully undistorted, without changing the shape of their intensity profile in space or time, have also attracted considerable interest. These fields are called ‘localized waves’ and their peaks may propagate also at superluminal or subluminal speeds. Luminal non-diffractive and non-dispersive solutions to the wave equation and the Maxwell equations have been known to exist for a long time. Brittingham (1983) discovered the so-called focus wave modes, which are soliton-like linear waves that travel in three-dimensional space at the speed of light. The theory was generalized by Sezginer (1985), who also showed how to generate finite-energy approximations of such focused pulses. Aperture production methods of focus wave modes and other localized fields were analyzed by Shaarawi (1997). Focus wave modes were demonstrated experimentally at optical frequencies by Reivelt and Saari (2002b). Although an analytical expression was already given by Mackinnon (1978), subluminal localized waves have posed fundamental problems. Numerical simulations and approximative solutions have been attempted (Longhi, 2004; Salo & Salomaa, 2001a,b). Expanding the spectrum in a Fourier series, exact localized subluminal solutions of the Mackinnontype, valid for arbitrarily chosen frequency bandwidths, were recently put forward by Zamboni-Rached and Recami (2008). In principle, the refractive-diffractive setup proposed by Reivelt and Saari (2002a) should allow the demonstration of optical subluminal waves with velocities at least slightly below the speed of light. Strongly subluminal localized fields can be obtained only for narrow-band radiation. The so-called ‘frozen waves’ were introduced by Zamboni-Rached (2004) as ‘zero-velocity’ solutions allowing one to generate highly transversally localized intensity ´ distributions with tailored longitudinal profiles (Dartora, Nobrega, Dartora, Viana, & Filho, 2006; Zamboni-Rached, 2006b; Zamboni-Rached & Recami, 2008; Zamboni-Rached, Recami, & Hernand´ez-Figueroa, 2005). Although static in the sense that the intensity distribution is independent of time, such solutions are monochromatic and so the question of group velocity does not arise. The best known and most extensively studied among all localized pulsed fields are the ones that propagate at superluminal speeds.
6
Propagation-Invariant Optical Fields
A cornerstone in this research is the work of Lu and Greenleaf (1992a,b) in which they generalized the Bessel beams into broadband non-diffractive and non-dispersive pulses, known as the X waves. The name stems from the characteristic X-shaped transverse intensity profiles of these fields. Despite the simple property that all frequency components propagate at the same angle to the beam axis, as is typical of ‘conical waves’ (Fagerholm, Friberg, Huttunen, Morgan, & Salomaa, 1996), X waves are difficult to realize in optics since their spectral content has its maximum at zero frequency. This obstacle was, however, overcome in a seminal experiment by Saari and Reivelt (1997) (see also Valtna, Reivelt, and Saari (2007)). The superluminal speed of Bessel-type X waves has been observed experimentally (Alexeev, Kim, & Milchberg, 2002). Various extensions of X waves have been introduced by Salo, Fagerholm, Friberg, and Salomaa (2000) and the propagation of localized pulses has also been considered in ˜ dispersive media (Porras, 2001; Sonajalg & Saari, 1997; Zamboni-Rached, ´ Nobrega, Hern´andez-Figueroa, & Recami, 2003). The localized fields were of particular interest in acoustics. The first (narrowband) ultrasonic Bessel-beam transducer was fabricated by Hsu, Margetan, and Thompson (1989) and the potential of non-diffracting waves in high-resolution medical imaging was soon recognized (Lu & Greenleaf, 1990). Biomedical imaging with ‘limited diffraction’ beams is now a major research field. Localized X wave solutions have also been formulated for full vector fields satisfying Maxwell’s ´ equations (Recami, 1998; Recami, Zamboni-Rached, Nobrega, Dartora, & Hernand´ez-Figueroa, 2003). In the microwave regime, pulsed X waves were experimentally observed by Mugnai, Ranfagni, and Ruggeri (2000). Besides optics, acoustics, and electromagnetism, localized waves are relevant in any area of physics and engineering where a wave equation plays a role. Non-diffracting fields exhibit a self-reconstruction property both in the linear (Bouchal, Wagner, & Chlup, 1998) and nonlinear regimes (Butkus, Gadonas, Januˇsonis, Piskarskas, Regelskis, Smilgeviˇcius, & Stabinis, 2002). After the ground-breaking work of Wulle and Herminghaus (1983), spatio-temporally localized optical waves have been studied and applied extensively in nonlinear optics. As examples, we cite second-harmonic generation of X waves (Conti, 2003) and spontaneous formation of Xshaped light bullets (DiTrapani, Valiulis, Piskarskas, Jedrkiewicz, Trull, Conti, & Trillo, 2003). As useful reviews on the various aspects of localized fields we mention Saari and Reivelt (2004), Kiselev (2007), and the recent book edited by Hern´andez-Figueroa, Zamboni-Rached, and Recami (2008).
Scalar Fields
7
2. SCALAR FIELDS The defining property of an ideal propagation-invariant optical field is that its intensity distribution must be exactly the same across any transverse plane z = constant. Let us begin the discussion of such fields in the framework of scalar theory.
2.1. Durnin’s General Solution Assuming a monochromatic wave field of frequency ω and time dependence exp (−iωt), the complex amplitude U (x, y, z) of the field must satisfy the scalar Helmholtz equation ∇ 2 + k 2 U (x, y, z) = 0, (2.1) where k = ω/c = 2π/λ, c is the speed of light, and λ is the vacuum wavelength of the field. We look for solutions of Equation (2.1) in the form U (x, y, z) = V (x, y) exp (iβz) ,
(2.2)
where β is a constant and V (x, y) is an as-yet unknown function. Solutions of this type are exactly propagation-invariant in the sense that I (x, y, z) = |U (x, y, z)|2 = |V (x, y)|2 = |U (x, y, 0)|2 = I (x, y, 0) (2.3) for all z ≥ 0. Here I (x, y, z) denotes the optical intensity. The separation of variables expressed in Equation (2.2) is familiar from waveguide theory. Thus any propagation-invariant field may be interpreted as an eigenmode of free space, with propagation constant β. We note, in passing, that one might be tempted to search for more general solutions with βz replaced by an arbitrary real function 8(x, y, z), but the discussion of Turunen, Vasara, and Friberg (1991) indicates that 8(x, y, z) = βz is indeed the most general form for propagation-invariant fields. This is shown more explicitly by Belskii (1992). Substitution of Equation (2.2) into Equation (2.1) leads to the basic differential equation for propagation-invariant scalar waves:
2 ∇⊥ + α 2 V (x, y) = 0,
(2.4)
2 = ∂ 2 /∂ x 2 + ∂ 2 /∂ y 2 denotes the transverse Laplace operator and where ∇⊥ we have the dispersion relation
α2 + β 2 = k 2.
(2.5)
8
Propagation-Invariant Optical Fields
The general solution of Equation (2.4) may be expressed in the form first presented by Whittaker (1902) and Havelock (1904): V (x, y) =
Z
2π
A(φ) exp [iα (x cos φ + y sin φ)] dφ,
(2.6)
0
where A(φ) is an arbitrary complex function. Formally, Equation (2.6) represents V (x, y) as a superposition of homogeneous plane waves, of wave number α and amplitude A(φ), which propagate in all directions in the x y plane. Equation (2.6) may also be written in polar transverse coordinates (ρ, ψ), with x = ρ cos ψ and y = ρ sin ψ, as V (ρ, ψ) =
Z
2π
A(φ) exp [iαρ cos (ψ − φ)] dφ.
(2.7)
0
In combination with Equations (2.2) and (2.5), and with time dependence exp (−iωt), Equation (2.6) is precisely identical to the expression presented by Durnin (1987) for propagation-invariant scalar fields. Vicari (1989) and Ruschin (1994) have pointed out that mathematically valid solutions are obtained also for imaginary and complex values of α in Equation (2.6), but these are of lesser interest since such fields decay exponentially in the z direction.
2.2. Bessel Field Modes The general propagation-invariant field considered above, contains certain interesting special cases. To find these we insert in Equation (2.6) a Fourier expansion ∞ X
A(φ) =
am exp (imφ) ,
(2.8)
A(φ) exp (−imφ) dφ.
(2.9)
m=−∞
where the Fourier coefficients are am =
1 2π
2π
Z 0
Let us use the polar coordinates (ρ, ψ) and employ the Bessel function identity i−m Jm (x) = 2π
2π
Z 0
exp [i (mφ + x cos φ)] dφ,
(2.10)
Scalar Fields
9
where Jm is a Bessel function of the first kind and order m. On writing, for simplicity of notation, cm = 2π im am ,
(2.11)
we obtain U (ρ, ψ, z) = exp (iβz)
∞ X
cm Jm (αρ) exp (imψ) .
(2.12)
m=−∞
This is still a completely general scalar expression for propagationinvariant fields as a superposition of Bessel functions. In the simplest special case one can write cm = 0 for all but one value of m. This leads to the concept of Bessel field modes with rotationally symmetric transverse intensity distributions of the form I (ρ) ∝ Jm2 (αρ). The lowest-order (fundamental or J0 ) Bessel field mode, with I (ρ) ∝ J02 (αρ) and a binary-valued (zero or π radians, see Figure 1b) phase, exhibits a narrow axial peak surrounded by a system of concentric side lobes, as illustrated in Figure 1a. The full width at half maximum (FWHM) of the central peak is ρ ≈ 2.25/α and the first zero is located at ρ ≈ 2.405/α. The maximum of the first side lobe, which is 0.162 times the peak intensity, is located at ρ ≈ 3.832/α. The first-order J1 Bessel mode (Figure 1c) exhibits an axial zero surrounded by a strong circular ring and a system of weaker concentric side lobes. The maximum of the first ring is located at a distance ρ ≈ 1.841/α from the axis. The (inner) radius, where the intensity is one half of the peak intensity of the first ring, is ρ ≈ 1.83/α. If this is compared to the FWHM of the J0 mode, it may be concluded that the axis of the J1 mode is defined more precisely than that of the J0 mode. Higher-order Jm Bessel modes feature a wider axial dark spot (Figure 1e). All field modes with m 6= 0 exhibit an azimuthally spiral-like phase profile (Figure 1d and 1f) with a vortex at the z-axis, and may thus be said to possess angular momentum. Sufficiently far away from the optical axis we may use the asymptotic expression for Bessel functions to obtain Jm2 (αρ)
2 π 1 2 m− . ∼ cos αρ − π αρ 2 2
(2.13)
Thus the ring system associated with the Bessel mode becomes radially periodic, of period π/α, far away from the axis. In fact, the local period converges to this value quite rapidly, as shown in Table 1. In view of Equation (2.13), the envelope of the ring maxima decays as 1/ρ. Since the
Propagation-Invariant Optical Fields
(b) 20
10
10
αy
(a) 20
αy
10
0
–10
–10 –20
0
–20 –20 –10
0
10
20
–20 –10
0 αx
10
20
–20
0 αx
10
20
αx
10
10
αy
(d) 20
αy
(c) 20
0 –10
0 –10 –20
–20 –20 –10
0
10
20
–10
αx
(e) 20
(f)
10
αy
αy
10 0 –10 –20
20
0 –10
–20
–10
0
αx
10
20
–20 –20
–10
0 αx
10
20
FIGURE 1 Transverse field distributions of Bessel fields of different orders. Top: J0 field. Middle: J1 field. Bottom: J2 field. Left column: amplitude |V (x, y)|; white indicates the maximum of the J0 field, to which the other scales are normalized. Right column: phase; white and black indicate extreme values of −π and π , respectively
ring circumference grows linearly with ρ it is clear that, asymptotically, each ring contains an equal amount of energy (see Table 1). Unfortunately, this also means that the energy content of the entire Bessel beam is infinite and the beam is therefore physically unrealizable in the strict sense. However, the same is, of course, true for a plane wave. Note finally from Table 1 that the bright central lobe of the fundamental Bessel field in fact contains less energy than any of the surrounding rings.
2.3. Angular Spectra of Propagation-invariant Fields All propagation-invariant fields described by Equations (2.2) and (2.6) may be characterized by a simple property associated with the
Scalar Fields
11
TABLE 1 Some properties of the ring system associated with the J0 Bessel field lobe n:o
1
2
3
4
5
q1
outer radius/α width/α energy
2.405 – 4.897
5.520 3.115 6.187
8.654 3.134 6.252
11.791 3.137 6.268
14.931 3.140 6.274
π(q − 1/4) π 2π
distribution of wave vectors k in the plane-wave representation of the field. To see this, we first introduce a polar-coordinate notation for wave vectors: k = k x , k y , k z = ( f cos ϕ, f sin ϕ, w) ,
(2.14)
1/2 where w = k 2 − f 2 = k cos θ , k = |k|, and θ is the angle between the wave vector k and the z-axis. The properties of scalar fields in the wave-vector space are investigated most naturally by using the angular spectrum representation (Mandel and Wolf (1995), sect. 3.2). Ignoring evanescent waves, which, by definition, can not exist in a propagation-invariant field, we have1 U (x, y, z) =
2π
Z kZ 0
f T ( f, ϕ) exp [i f (x cos ϕ + y sin ϕ) + iwz] d f dϕ, (2.15)
0
where the function T ( f, ϕ) =
1 (2π )2
ZZ
∞
U (x, y, 0) exp [−i f (x cos ϕ + y sin ϕ)] dxdy (2.16)
−∞
is known as the angular spectrum of plane waves associated with U (x, y, 0). Comparing Equations (2.2) and (2.6) with Equation (2.15), we obtain w = β and T ( f, ϕ) = α −1 δ( f − α)A(ϕ),
(2.17)
where δ( f ) denotes the Dirac delta function. In view of Equation (2.17), the angular spectrum of a propagationinvariant field is non-zero only at f = α. Therefore the field is a conical 1 The general representation for propagation-invariant fields, Equations (2.2) and (2.6), can be derived alternatively, by inserting Equation (2.15) into the second and fourth expressions in Equation (2.3), i.e. |U (x, y, z)| = |U (x, y, 0)|.
12
Propagation-Invariant Optical Fields
wave with cone angle θ . According to Equation (2.5), sin θ = α/k and cos θ = β/k. The fact that all wave vectors must lie on the surface of a single cone may actually be viewed as an alternative defining property of (scalar) propagation-invariant fields. Clearly the angle φ in Equation (2.6) may be interpreted as the azimuthal angle ϕ in the spatial-frequency domain and A(φ) is a weight function of the plane-wave components along the surface of the cone. A choice of A(φ) = δ(φ − φ0 ) reduces the field to an oblique plane wave. With A(φ) = δ(φ) + δ(φ − π ) a sinusoidal interference pattern of two intersecting plane waves is obtained, and the choice of A(φ) = δ(φ) + δ(φ − π/2) + δ(φ − π ) + δ(φ − 3π/2) leads to a crossed sinusoidal pattern, as considered by Hafizi and Sprangle (1991). The cone angle θ is limited to the range 0 ≤ θ ≤ π/2 if we wish to obtain the desired exp (iβz) dependence. The limit θ = 0 corresponds to a plane wave that propagates in the direction of the positive z-axis. In the paraxial limit we assume θ 1 rad. The Taylor approximation of β then gives U (x, y, z) = V (x, y) exp (ikz) exp −iα 2 z/2k ,
(2.18)
where V (x, y) is given by Equation (2.6). Thus the paraxial approximation has no effect in the functional form of the transverse intensity distribution. In the opposite limit θ = π/2 we have β = 0, α = k, and U (x, y, z) =
Z
2π
A(φ) exp [ik (x cos φ + y sin φ)] dφ.
(2.19)
0
Therefore all wave vectors of the total field are confined in the x y plane. The plane-wave description of propagation-invariant fields helps to provide physical insight into the spatial resolution of Bessel field modes. The transverse scale of any Bessel mode is inversely proportional to α. Since α = k sin θ , the scale shrinks when θ is increased: the full width at half maximum of the central lobe of the fundamental Bessel field is given by FWHM ≈ 0.358
λ . sin θ
(2.20)
The smallest possible transverse scale for a propagating field is achieved in the limit θ = π/2. Ruschin and Leizer (1998) noted that there is no reason why we could not reduce the transverse scale further by using purely imaginary values of β. However, this leads to an evanescent field with an exponential decay of the axial intensity, which no longer satisfies Equation (2.3).
Scalar Fields
13
2.4. Solutions in Elliptic and Parabolic Coordinate Systems Propagation-invariant solutions can be expressed in all coordinate systems in which the transverse coordinates and the z coordinate are separable. In addition to the Cartesian and circular cylindrical systems, there are only two other such systems, namely the elliptic and parabolic cylindrical coordinate systems. Indeed, Guti´errez-Vega, Iturbe-Castillo, and Ch´avez-Cerda (2000) have shown that in the elliptic cylindrical coordinates the propagation-invariant solutions can be expressed in terms of radial and azimuthal Matthieu functions. Correspondingly, Bandres, Guti´errez-Vega, and Ch´avez-Cerda (2004) showed that in the parabolic cylindrical coordinates the solutions can be expressed in a series form. The resulting Matthieu and parabolic propagation-invariant fields, of course, have angular spectra confined on a single ring (in circular cylindrical coordinates) since the change of the coordinate system does not affect this defining property. Thus these solutions are merely different representations of the superpositions of Bessel fields discussed above. Nevertheless, they feature quite interesting visual appearances.
2.5. Custom-designed Propagation-invariant Fields Since the azimuthal variation A(φ) in Equation (2.6) is arbitrary, the transverse intensity distribution I (x, y) = |V (x, y)|2 can assume an infinite variety of different forms. The question naturally arises as to what propagation-invariant distributions I (x, y) are realizable under the constraint that the angular spectrum must be confined onto a ring. A complete answer is not known, but a number of interesting particular solutions exist. Kettunen and Turunen (1998) introduced a scheme for generating arrays of Bessel field modes centred at prescribed positions, which was demonstrated subsequently by Lautanen, Kettunen, Laakkonen, and Turunen (2000). If we shift the origin of a propagation-invariant field from (0, 0) to (x0 , y0 ) by substitutions x → x − x0 , y → y − y0 in Equation (2.6), the field remains as a conical wave with the original cone angle θ but A(φ) → A(φ) exp [−iα (x0 cos φ + y0 sin φ)] .
(2.21)
Thus, to generate an array of N propagation-invariant fields centred at (xn , yn ), with angular spectra An (φ), we choose in Equation (2.6)
A(φ) =
N X n=1
cn An (φ) exp [−iα (xn cos φ + yn sin φ)] ,
(2.22)
14
Propagation-Invariant Optical Fields
where cn are complex numbers. Consider, as an example, the generation of two parallel J0 field modes: N = 2, (x1 , y1 ) = (−x0 , 0), (x2 , y2 ) = (x0 , 0), c1 = c2 = 1/2, and A1 (φ) = A2 (φ) = 1. In view of Equation (2.22) A(φ) = cos (αx0 cos φ) .
(2.23)
For sufficiently large x0 , two well-resolved Bessel fields are obtained. Another interesting class of fields is obtained if the only non-vanishing terms in Equation (2.12) are m = p and m = − p. If we assume c p = c− p = 1/2 for even p, or c p = −c− p = 1/2 for odd p, we obtain U (ρ, φ, z) = J p (αρ) cos ( pφ) exp (iβz) .
(2.24)
The intensity distribution therefore exhibits 2 p bright spots surrounded by weaker spots around a dark central area (Figure 2a), while the phase is binary-valued (Figure 2b). Yet another interesting class of propagation-invariant fields was considered by Gu´erineau and Primot (1999) and by Gu´erineau, Harchaoui, Primot, and Heggarty (2001). They pointed out that any field with its angular spectrum consisting of points that reside simultaneously at a single ring and on a regular Cartesian grid must be propagationinvariant and transversally periodic. The resulting field distribution can assume a variety of interesting forms, such as that illustrated in Figure 2c and 2d. Here we have chosen A(φ) =
6 1X δ [φ − (n − 1)π/3] , 6 n=1
(2.25)
which yields U (x, y, z) =
n hα √ i 1 exp (iβz) cos(αx) + cos x + 3y 3 2 hα √ io + cos x − 3y . 2
(2.26)
Thus a hexagonally periodic propagation-invariant field, exhibiting a regular array of bright spots on a weak background and a binary phase distribution, is obtained. More examples of this type of field were presented by Bajer and Hor´ak (1996), who considered different polarization configurations among the plane waves constituting the field. Clearly, propagation-invariant fields are obtained for any function A(φ), including those with ‘rough’ amplitude and phase distributions,
Scalar Fields
10
10
αy
(b) 20
αy
(a) 20
0
0 –10
–10 –20 –20
–10
0 αx
10
–20 –20
20
10
10
αy
(d) 20
αy
(c) 20
0
–10
0 αx
10
–20 –20
20
(f) 20
10
10
0
αy
αy
0 αx
10
20
–10
0 αx
10
20
–10
0 αx
10
20
0
(e) 20
–10 –20 –20
–10
–10
–10 –20 –20
15
0 –10
–10
0 αx
10
20
–20 –20
FIGURE 2 Top: transverse amplitude distribution and phase of a propagation-invariant field in Equation (2.24) with p = 3. Middle: transversally hexagonally periodic propagation-invariant field. Bottom: propagation-invariant speckle-like pattern. Left column: amplitude, right column: phase
which lead to speckle-like transversal spatial profiles. Such fields, with apparently random but propagation-invariant intensity distributions, were first demonstrated by Turunen, Vasara, and Friberg (1991) and reconsidered recently, with a similar setup, by Cottrell, Craven, and Davis (2007). A field of this type is illustrated in Figure 2e and 2f. The pattern is generated by a simulated diffuser with complex amplitude A(φ) of the form of Equation (2.8) with 11 randomly chosen non-zero complex coefficients.
16
Propagation-Invariant Optical Fields
The use of an incoherent imaging setup with an annular Fourierplane filter to generate custom-structured propagation-invariant fields has been considered by Bouchal (2002). In such a setup, the annular filter transforms each object point into an independent J0 Bessel field, which is centred in a position defined by the object point location and the magnification of the system. In the case of continuous objects, the method works only for (nearly) incoherent illumination, for which the final intensity distribution is the sum of the intensities of the individual Bessel fields. In the coherent case a quasi-random propagation-invariant field results, because the intensity distribution is the squared absolute value of the sum of the complex amplitudes of the individual Bessel fields. In general, the custom design of propagation-invariant fields can be considered a Fourier-domain beam shaping problem in diffractive optics, with an annular (delta-function ring) support for the angular spectrum. Such beam shaping problems can be solved efficiently using Iterative Fourier-Transform Algorithms (IFTAs) described, e.g. by Bryngdahl and Wyrowski (1990) and Aagedal, Wyrowski, and Schmid (1997, chap. 6). The application of the basic IFTA (the Gerchberg-Saxton algorithm) to the synthesis of propagation-invariant fields has been discussed by Courtial, Whyte, Bouchal, and Wagner (2006).
2.6. Relation to Self-imaging, Rotating, Frozen, and Airy Fields Indebetouw (1989) was the first (see also Szwaykowski and Ojeda˜ Castaneda (1991) and Belskii (1992)) to point out the connection between propagation-invariant fields and self-imaging or longitudinally periodic fields reviewed by Patorski (1989). The latter satisfy the condition U (x, y, z + z T ) = exp (iξ ) U (x, y, z),
(2.27)
where ξ is an arbitrary phase factor. The longitudinal period z T is often called the Talbot distance because the phenomenon of self-imaging was first observed by Talbot (1836) in his studies on diffraction gratings. Montgomery (1967) showed that the general expression for fields that satisfy Equation (2.27) is of the form2 U (x, y, z) =
X q∈Q
exp iβq z
Z
2π
Aq (φ) exp iαq (x cos φ + y sin φ) dφ, (2.28)
0
where Aq (φ) are arbitrary functions, αq2 + βq2 = k 2 , βq = (ξ + 2πq) /z T , 2 Laterally periodic fields generated by regular gratings observed by Talbot (1836) are longitudinally periodic only in the paraxial approximation. See Indebetouw (1992) for generalizations of the self-imaging effect.
Scalar Fields
17
and q is an integer. The terms with different values of q represent conical waves with cone angles θq = arccos [(q + ξ/2π ) λ/z T ]
(2.29)
and the set Q contains only those values of q for which 0 ≤ θq ≤ π/2. If we choose ξ ≈ 2π z T /λ (and if z T λ), small values of |q| represent paraxial conical fields propagating in the positive z direction. Clearly, Equation (2.28) is a superposition of propagation-invariant fields descibed by Equations (2.2) and (2.6). Thus the latter form a subset of self-imaging fields: while the angular spectrum of any propagationinvariant field is confined to a single ring, the angular spectrum of a selfimaging field may contain several rings with different radii, known as Montgomery’s rings. Fourier-series expansions of the functions Aq (φ) and the use of Equation (2.10) yield a generalization of Equation (2.12): U (ρ, ψ, z) = exp (iξ z/z T )
∞ X X
cqm Jm (αq ρ)
q∈Q m=−∞
× exp [i (2πqz/z T + mφ)] ,
(2.30)
where cqm are arbitrary constants. If we retain only the terms with q ± m = constant, the intensity distribution rotates (without radial expansion) by 2π radians about the optical axis upon propagation over the distance z T . Such rotating scalar fields, which employ two Montgomery’s rings with different radii, have been discussed by Ch´avez-Cerda, McDonald, and New (1996), Paterson and Smith (1996b), Schechner, Piestun, and ¨ Shamir (1996), Kotlyar, Soifer, and Khonina (1997), P¨aa¨ kkonen, Lautanen, Honkanen, Kuittinen, Turunen, Khonina, Kotlyar, Soifer, and Friberg (1998), and Khonina, Kotlyar, Soifer, Lautanen, Honkanen, and Turunen (1999). A method to construct fields with sharply defined transversal profiles and controlled longitudinal intensity distributions as superpositions of Bessel fields with different propagation constants was introduced by Zamboni-Rached (2004) and Zamboni-Rached, Recami, and Hernand´ezFigueroa (2005), who called such fields ‘frozen waves’. They consider rotationally symmetric (monochromatic) superpositions of J0 Bessel fields with different propagation constants U (ρ, z) =
M X m=−M
Am J0 (αm ρ) exp (iβm z) ,
(2.31)
18
Propagation-Invariant Optical Fields
2 +β 2 = k 2 and A are complex constants. The aim is to obtain an where αm m m axial intensity distribution I (0, z) = |F(z)|2 in the range 0 ≤ z ≤ L. Taking βm = Q + 2π m/L and choosing Q such that the βm are positive and real for all −M ≤ m ≤ M, the axial distribution is well approximated by
1 Am = L
L
Z
F(z) exp (i2π mz/L) dz.
(2.32)
0
The transverse distribution is fixed by the constants Am through Equation (2.31), but there is some ‘design freedom’ in the choice of Q and the phase of F(z). Another topic that has attracted considerable recent attention is the concept of ‘accelerating’ Airy fields, originally introduced by Berry and Balazs (1979) in quantum mechanics and demonstrated by Siviloglou, Broky, Dogariu, and Christodoulides (2007) in optics. Pulsed versions of such Airy waves were discussed by Saari (2008). The Airy beams are not propagation-invariant fields in the sense of the definition (2.3) adopted here, but rather interference patterns in which the sharply modulated peak region appears to shift sideways at an accelerating rate (along a parabolic curve) as the field propagates. However, the transverse shape and scale of the field remains propagation-invariant.
2.7. Finite-aperture Approximations Good finite-energy approximations of propagation-invariant fields exist, which cover substantial distances without essential modification of the transverse scale. We proceed to consider several such approximations, referring only occasionally to techniques for producing them. A more systematic review of the production methods will be the subject of Section 7. Conceptually, the most straightforward approach to generate finiteenergy approximations of propagation-invariant fields, is to assume that the ideal field is truncated by a radially symmetric aperture with a realvalued amplitude transmission function P(ρ). Flattened Gaussian profiles have been discussed by Borghi, Santarsiero, and Gori (1997), while Jiang (1996) considered super-Gaussian functions P(ρ) = exp −(ρ/R)q ,
(2.33)
where R is the transverse scale factor of the aperture and q ≥ 2 is the super-Gaussian parameter. With q = 2 the assumed aperture is Gaussian, and in the limit q → ∞ one obtains a hard-edged aperture. Figure 3a illustrates axial profiles for three different values of q in Equation (2.33), namely q = 2, q = 8, and q = 32. Here we assume
Scalar Fields
(a) 1.4
(b) 1 |U (ρρ, 0.78L)|
1.2 I (0, z)
1 0.8 0.6 0.4 0.2 0
19
0.2
0.4
0.6
0.8 z/L
1
1.2
0.8 0.6 0.4 0.2
1.4
0
20
40
60
80 100 120 140
FIGURE 3 (a) Axial intensity profiles of super-Gaussian apertured Bessel fields. (b) Transverse amplitude profiles of the same fields at a distance z = 0.78L. Dotted lines: q = 2. Dashed lines: q = 8. Solid lines: q = 32
that the field is paraxial with cone angle θ = 1◦ and that the 10th zero of the incident J0 Bessel field is located at ρ = R. Hence R ≈ 280λ. With large values of q, strong axial fluctuations appear in the profile; the extreme case of a hard-edged aperture has been examined in detail, e.g. by Durnin (1987) and DeNicola (1991). The amplitude of these fluctuations increases upon propagation distance until the field experiences a rapid exponential-like decay towards zero. This decay begins near the end of the geometrical-optics prediction of the propagation-invariant range L = R/ tan θ,
(2.34)
given by Durnin, Miceli Jr., and Eberly (1987), which follows directly from the conical-wave nature of the field. In Figure 3a the propagation distance z has been normalized to L ≈ 16000λ. Figure 3b illustrates the radial amplitude distributions of the three fields considered in Figure 3a at a distance z = 0.78L, close to the end of the geometrical propagationinvariant range. It is seen in all cases that the minima are no longer zero as for an ideal Bessel field. At larger propagation distances, energy starts to drift increasingly away from the centre region until finally an approximation of an annulus with a linearly expanding radius is formed. The amplitude of the axial intensity fluctuations decreases when the super-Gaussian parameter q is reduced, i.e. the aperture is apodized. Simultaneously the average axial intensity of the propagation-invariant field begins to decrease upon propagation. With q = 2 no fluctuations are seen, but at z = L the axial intensity is less than 15% of its value at z = 0. Thus a compromise can be found between the axial fluctuations and the decay of axial intensity by choosing the super-Gaussian parameter q appropriately. The dashed line in Figure 3 with q = 8 represents such a compromise: no appreciable fluctuations are seen (these start to appear as soon as q > 8) and the axial intensity remains nearly constant until
20
Propagation-Invariant Optical Fields
z ≈ 0.7L. Up to this distance, the centre region of the transverse profile remains virtually ideal, with the first few minima close to zero. The fluctuations observed in the case of a hard-edged aperture may be interpreted as resulting from the interference of the converging conical wave and a boundary diffraction wave (Born and Wolf (1999), sect. 8.9, Rubinowicz (1965)) generated at the sharp edge of the aperture. Hence it is clear that the fluctuations can be suppressed by apodization of the aperture. This approach was proposed independently by Cox and D’Anna (1992) and by Herman and Wiggins (1992). Jiang, Lu, and Liu (1995) noted that the (envelope of the) longitudinal profile resembles the radial illumination distribution in the aperture, a fact that can be understood qualitatively on the basis of a geometrical interpretation of the conical wave. A more quantitative explanation was provided by Borghi, Santarsiero, and Gori (1997). The propagation characteristics of apertured Bessel field modes, Bessel–Gauss fields with q = 2 in Equation (2.33), and Gaussian beams have been compared rather extensively by Overfelt and Kenney (1991). We proceed to examine in more detail the case q = 2 in Equation (2.33), i.e, the Bessel–Gauss beam introduced by Gori, Guattari, and Padovani (1987b). The Bessel–Gauss source is described by the field distribution at z = 0 of the form U (ρ, 0) = J0 (αρ) exp −ρ 2 /w02 ,
(2.35)
where α and w0 are the defining source parameters; we have replaced R in Equation (2.33) by w0 to conform with common notation and dropped the explicit ψ dependence. Thus, if we let α → ∞ with fixed w0 , Equation (2.35) describes the waist of a conventional Gaussian beam. If, on the other hand, we fix α and let w0 → ∞, a fundamental J0 Bessel field is obtained. The field radiated by the source in Equation (2.35) may be interpreted as a superposition of Gaussian beams that propagate along the surface of the cone associated with the fundamental Bessel field mode (see, e.g. Palma (1997)). If, therefore, the divergence angle θG = λ/π w0 of a Gaussian beam is substantially smaller than the cone angle θ , the farfield diffraction pattern is essentially a Gaussian ring.3 More precisely, substitution of Equation (2.35) into Equation (2.16) gives the rotationally symmetric angular spectrum of the Bessel–Gauss beam: T( f ) =
1 2 1 w0 I0 (w02 α f /2) exp − w02 α 2 + f 2 , 2 4
(2.36)
3 Such rings appear in studies on far fields of axicons and circular gratings; see Fedotowski and Lehovec (1974) and B´elanger and Rioux (1978).
Scalar Fields
21
where I0 is the modified Bessel function of the first kind and zero order. Thus the far-field radiation pattern is an annulus of finite width, provided that at least a few zeros of the Bessel function are within the radius w0 . An analytical expression for the evolution of Bessel–Gauss beams is obtained by inserting Equation (2.35) into the Fresnel propagation formula. The result is ! # " αρ α2 w0 z J0 exp [i8(z)] exp i k − U (ρ, z) = w(z) 2k 1 + iz/z R ik 1 2 2 2 2 − × exp − ρ + z α /k , (2.37) w2 (z) 2R(z) where 8(z) = − arctan (z/z R )
(2.38)
is the Gouy phase shift and the quantities h i1/2 w(z) = w0 1 + (z/z R )2
(2.39)
R(z) = z + z R2 /z
(2.40)
and
represent the usual propagation parameters (the 1/e2 beam width and the radius of wavefront curvature, respectively) of a conventional Gaussian beam with Rayleigh range z R = π w02 /λ. Because of the Gaussian aperture at z = 0, the axial intensity profile h i I (0, z) = [w0 /w(z)]2 exp −2z 2 α 2 /k 2 w2 (z)
(2.41)
does not feature rapid oscillations, a fact we already saw in Figure 3b. As pointed out by Gori, Guattari, and Padovani (1987a), mth-order Bessel–Gauss beams radiated by sources of the form U (ρ, ψ, 0) = Jm (αρ) exp (imψ) exp −ρ 2 /w02 ,
(2.42)
are also solutions of the paraxial Helmholtz equation. Such higher-order Bessel–Gauss fields have been investigated in detail by Bagini, Frezza,
22
Propagation-Invariant Optical Fields Q
L
P
z
f0
z = –2F
z=0
FIGURE 4 Conceptual optical system for generation of finite-energy approximations of propagation-invariant fields by spatial-frequency selection. Here Q and P are complex-amplitude transmission filters, while L is a Fourier-transforming lens of focal length F: the fields immediately behind z = −2F and immediately in front of z = 0 are thus Fourier transforms of each other. The shaded area represents the region illuminated by one point at the filter Q
Santarsiero, Schettini, and Schirripa Spagnolo (1996). These fields may be interpreted as a superposition of Gaussian beams with wave vectors located on a cone defined by the angle θ , and complex amplitudes weighted by the azimuthal angular spectrum A(φ) = exp (imφ). There are many possible extensions and variations of the Bessel–Gauss beams described above. Gaussian apodization of modified Im Bessel fields has been studied by Ruschin (1994). Bagini, Frezza, Santarsiero, Schettini, and Schirripa Spagnolo (1996) view such fields as superpositions of Gaussian beams with their wave vectors on a cylindrical surface. As a further extension, these authors consider Bessel–Gauss fields of complex argument, interpreting them as superpositions of Gaussian beams with their waists centred on a circle and wave vectors lying on the surface of a cone (see also Palma (1997)). Non-paraxial propagation of radiation from Bessel–Gauss sources has been analyzed by Borghi, Santarsiero, and Porras (2001).
2.8. Finite-angular-spectrum Approximations An alternative approach to produce a nearly propagation-invariant field is suggested by the experimental configuration of Durnin, Miceli Jr., and Eberly (1987), who used a thin annular aperture in the front focal plane of a Fourier-transform lens. This experiment simulates a field with a ring-like angular spectrum of finite width. Figure 4 is a conceptual illustration of the general geometry, in which L represents the Fourier-transforming lens and two filters with complex-amplitude transmission functions Q( f, φ) and P(ρ) are located at the planes z = −2F and z = 0. When illuminated by a plane wave, the filter Q( f, φ) realizes an approximation of a deltafunction ring in the radial variable and also produces the distribution A(φ) in the azimuthal direction. The Fourier-transformed field incident on the
Scalar Fields
23
second filter P(ρ) at the plane z = 0 is an approximation of a conical wave with cone angle θ = arctan ( f 0 /F), where f 0 is the centre radius of the aperture of filter Q( f, φ). The nature of the field in the region z > 0 depends critically on the functional forms of the filters Q( f, φ) and P(ρ). If the annulus of Q( f, φ) is sufficiently wide, the aperture P(ρ) may not truncate the field incident onto it at all. If, on the other hand, the function Q( f, φ) has a sufficiently narrow radial transmittance around f = f 0 , the entire aperture P(ρ) is illuminated essentially uniformly. Let us first consider fields truncated purely in the spatial-frequency domain, i.e. ignore the filter P(ρ). Writing Q( f, φ) = R( f )A(φ), where the function R( f ) is centred at f = α, and taking A(φ) = 1/2π for simplicity, we discuss approximations of the J0 Bessel field mode. Interpreting Q( f, φ) as the angular spectrum T ( f, φ) defined in Equation (2.16), and expressing Equation (2.15) in polar coordinates, leads to a rotationally symmetric field formula U (ρ, z) =
Z
k
f R( f )J0 ( fρ) exp [iβ( f )z] d f,
(2.43)
0
1/2 with β( f ) = k 2 − f 2 . This expression can be evaluated numerically for any given R( f ) and propagation distance z. At z = 0 an ideal fundamental Bessel field is generated if R( f ) = δ( f − α). For other functional forms of R( f ), the field at z = 0 is a coherent superposition of ideal fundamental Bessel fields with different scaling factors f . Thus the result is a transversely blurred version of the J0 field. Since the Bessel contributions with different scales have different propagation constants β( f ), the degree of blurring grows with increasing z. Following the approach of McCutchen (1964), Arsenault and Boivin (1967), and Colombeau, Froehly, and Vampoille (1979), we assume ρ = 0, make the paraxial approximation, and write f 2 = q in Equation (2.43). This yields an expression for the axial field distribution in the form U (0, z) =
1 exp (ikz) 2
Z
k2
R
√ q exp (iqz/2k) dq.
(2.44)
0
This result reveals that the on-axis profile of the field is essentially the Fourier transform of its radial angular spectrum, but expressed as a function of f 2 rather than f . In the geometry of Figure 4 this distribution is centred at the plane z = 0, where the aperture P(ρ) may modulate the field further by selecting only the central part. This, of course, affects the axial intensity distribution in z > 0.
24
Propagation-Invariant Optical Fields
3. ELECTROMAGNETIC FIELDS The scalar analysis presented above is applicable to optical fields only if the cone angle θ is paraxial — otherwise the longitudinal components of the electromagnetic field become important and the concept of propagation invariance no longer has a unique definition. The straightforward extension of the scalar definition is to demand that each scalar component of the field satisfies, separately, a condition of the form (2.3). Since the time-averaged Poynting vector hS(x, y, z, t)i is often considered as a measure of the intensity of an electromagnetic field, it appears natural to demand that hSz (x, y, z, t)i = hSz (x, y, 0, t)i
(3.1)
for all z > 0. This can actually be shown to be equivalent to the definition introduced by Bajer and Hor´ak (1996), in which the divergence of the transverse part of hS(x, y, z, t)i is set equal to zero. On the other hand, the time-averaged electric energy density hwe (x, y, z, t)i is frequently taken as the quantity observable with usual square-law detectors. Hence one could promote the definition hwe (x, y, z, t)i = hwe (x, y, 0, t)i.
(3.2)
A corresponding definition for the total energy density hw(x, y, z, t)i = hwe (x, y, z, t)i + hwm (x, y, z, t)i
(3.3)
is also possible; note that it is not equivalent to Equation (3.2) because, in general, the magnetic energy density hwm (x, y, z, t)i 6= hwe (x, y, z, t)i. We note that, in a certain sense, a definition based on electric energy density appears most natural. Not only do square-law detectors respond to this quantity, but the (locally interpreted) time-averaged Poynting vector can assume negative values; this phenomenon is well known in focusing problems (Richards & Wolf, 1959) and it has also been predicted for propagation-invariant fields by Turunen and Friberg (1993a). The first rigorous propagation-invariant solution of Maxwell’s equations was presented by Mishra (1991). More general treatments, in which different definitions of propagation invariance are employed, have been given by Turunen and Friberg (1993a), Bouchal and Oliv´ık (1995), Bouchal, Hor´ak, and Wagner (1996), Bajer and Hor´ak (1996), Hor´ak, Bouchal, and Bajer (1997), and Tervo, Vahimaa, and Turunen (2002). We begin by extending the scalar results into the vectorial domain and then
Electromagnetic Fields
25
present some examples of electromagnetic propagation-invariant fields, which sometimes have no counterparts in scalar theory.
3.1. Vectorial Extension of Scalar Results Let us consider a class of electromagnetic propagation-invariant fields that is obtained if expressions of the type of Equations (2.2) and (2.6) are introduced for two independent components of the electromagnetic field and the remaining components are solved from Maxwell’s equations. In general, we write the electric and magnetic field vectors in the form E(x, y, z) = exp (iβz)
2π
Z
A(φ) exp [iα (x cos φ + y sin φ)] dφ,
(3.4)
B(φ) exp [iα (x cos φ + y sin φ)] dφ.
(3.5)
0
H(x, y, z) = exp (iβz)
2π
Z 0
If the x and y components of the electric field are chosen as the independent functions, A x (φ) and A y (φ) are arbitrary and the remaining components of the vectors A(φ) and B(φ) are obtained from Maxwell’s equations: α A z (φ) = − cos φ A x (φ) + sin φ A y (φ) , β ! # " β α2 α2 2 Bx (φ) = − sin φ cos φ A x (φ) + 1 + 2 sin φ A y (φ) , k Z β2 β " ! # β α2 α2 2 B y (φ) = 1 + 2 cos φ A x (φ) + 2 sin φ cos φ A y (φ) , kZ β β α sin φ A x (φ) − cos φ A y (φ) , Bz (φ) = − kZ
(3.6) (3.7) (3.8) (3.9)
where Z = (µ0 /0 )1/2 is the impedance, 0 the permittivity, and µ0 the permeability of vacuum. These expressions agree, apart from slight notational rearrangements, with the results of Turunen and Friberg (1993a). In the paraxial limit (α → 0 and β → k) the longitudinal field components E z and Hz vanish. In the opposite limit (α → k and β → 0) one obtains two-dimensional (z-independent) fields, known as reflaxicon fields considered by Turunen and Friberg (1993b), which are the electromagnetic counterparts of Equation (2.19). The electromagnetic propagation-invariant field solutions may also be expressed in terms of Bessel functions, in analogy with the scalar case. Let us write the electric and magnetic fields in forms analogous to
26
Propagation-Invariant Optical Fields
Equation (2.12), i.e. E(ρ, ψ, z) = exp (iβz) H(ρ, ψ, z) = exp (iβz)
∞ X m=−∞ ∞ X
em Jm (αρ) exp (imψ) ,
(3.10)
hm Jm (αρ) exp (imψ) ,
(3.11)
m=−∞
and choose ex,m and e y,m as the independent coefficients. The rest of the coefficients are obtained by considering Equations (2.9) and (3.6)–(3.9): α iex,m+1 − iex,m−1 − e y,m+1 − e y,m−1 , (3.12) 2β β α2 h x,m = − e y,m + iex,m+2 − iex,m−2 − e y,m+2 − e y,m−2 − 2e y,m , kZ 4kβ Z (3.13) β α2 h y,m = ex,m+2 + ex,m−2 + ie y,m+2 − ie y,m−2 − 2ex,m , ex,m − kZ 4kβ Z (3.14) α h z,m = − ex,m+1 + ex,m−1 + ie y,m+1 − ie y,m−1 . (3.15) 2k Z ez,m =
The expressions for ez,m and hm depend on ex,m±1 , e y,m±1 , ex,m±2 , and e y,m±2 , in addition to ex,m and e y,m , i.e. there is coupling between different Bessel orders. Thus, in the electromagnetic case, there are no pure mthorder Bessel field modes: if only one value of m is retained in the expansion (3.10), the electric energy density depends not only on Jm (αρ) but also on Jm±1 (αρ), and the magnetic energy density depends, in addition, on Jm±2 (αρ).
3.2. Particular Polarization Configurations We now focus our attention on certain simple special cases of electromagnetic propagation-invariant fields, with different polarization properties. In the paraxial approximation, the state of polarization of the field can be defined analogously with a plane wave, and one may speak of elliptically, circularly, and linearly polarized fields without ambiguity. However, in case of non-paraxial fields, the z component is usually present in addition to the x and y components. Instead of dealing with the field directly, we consider the transverse component of the electric angular spectrum, A⊥ (φ) = A(φ) − zˆ A z (φ) = xˆ A x (φ) + yˆ A y (φ) = ρˆ Aρ (φ) + φˆ Aφ (φ), (3.16)
Electromagnetic Fields
27
where x, ˆ yˆ , zˆ are the Cartesian, and ρ, ˆ φˆ the polar unit vectors. The latter are of interest since they permit the definition of radially and azimuthally polarized fields; see Tidwell, Ford, and Kimura (1992) and Jordan and Hall (1994), respectively. In a radially/azimuthally polarized field the polarization vector always points in the radial/azimuthal direction and hence, if the field is incident on a transverse planar interface, it is purely TM/TE polarized, as pointed out by Ruschin and Leizer (1998). Let us first investigate the choice A⊥ (φ) = (2π )−1 x, ˆ which implies that the transverse angular spectrum is linearly polarized in the x direction. Then the electric field vector takes the form E(ρ, ψ, z) = xˆ J0 (αρ) − iˆz J1 (αρ) tan θ sin ψ exp (iβz) ,
(3.17)
where tan θ = α/β and θ is the cone angle, which approaches zero in the paraxial limit. Hence we see that the field is linearly polarized, in the usual sense, only in the paraxial limit, while a finite z component emerges in non-paraxial cases. The time-averaged electric energy density, hwe (ρ, ψ, z, t)i =
i 1 h 2 0 J0 (αρ) + J12 (αρ) tan2 θ sin2 ψ , 2
(3.18)
is clearly non-rotationally-symmetric in non-paraxial cases. The magnetic field and the magnetic energy density can also be calculated straightforwardly. While the expressions are somewhat more complicated, they are also rotationally non-symmetric, but the field remains propagationinvariant also, if the magnetic energy density or the Poynting vector is considered. We next examine the case of a right/left (±) circularly polarized transverse angular spectrum defined as √ A⊥ (φ) = (2 2π )−1 xˆ ± i yˆ .
(3.19)
Now 1 xˆ ± i yˆ J0 (αρ) − zˆ J1 (αρ) tan θ exp (±iψ) exp(iβz), (3.20) E(ρ, ψ, z) = √ 2 and hwe (ρ, ψ, z, t)i =
1 1 0 J02 (αρ) + J12 (αρ) tan2 θ . 2 2
(3.21)
28
Propagation-Invariant Optical Fields
Hence the electric energy density is rotationally symmetric also in the nonparaxial domain, but the radial distribution depends significantly on the cone angle θ . In the paraxial case the distribution of the electric energy density is the same as in the ‘linearly polarized’ case. Finally, we discuss fields with azimuthally and radially polarized transverse angular spectra A⊥ (φ) = (2π )−1 xˆ cos φ + yˆ sin φ
(3.22)
and A⊥ (φ) = (2π )−1 −xˆ sin φ + yˆ cos φ ,
(3.23)
respectively. In the former case Aρ (φ) = 0 and in the latter case Aφ (φ) = 0 in Equation (3.16). The electric field distribution is E(ρ, ψ, z) = iφˆ J1 (αρ) exp (iβz)
(3.24)
in the azimuthally polarized case, and E(ρ, ψ, z) = iρˆ J1 (αρ) − zˆ J0 (αρ) tan θ exp (iβz)
(3.25)
in the radially polarized case. Both field distributions are rotationally symmetric and the azimuthally polarized field has no z component, thus being azimuthally polarized in the exact sense also in the non-paraxial domain. The distributions of the electric energy density are 1 0 J12 (αρ) 2
(3.26)
i 1 h 2 0 J1 (αρ) + J02 (αρ) tan2 θ 2
(3.27)
hwe (ρ, ψ, z, t)i = in the azimuthally polarized case, and hwe (ρ, ψ, z, t)i =
in the radially polarized case. In both cases the paraxial contribution has a zero at the origin, instead of an axial maximum. It is worthwhile noticing that the two solutions just considered, are the only possible azimuthally and radially polarized propagation-invariant field modes, as shown by Tervo, Vahimaa, and Turunen (2002). Some of the particular field solutions considered above are illustrated in Figure 5. Here the cone angle is assumed to be either θ = 30◦ or θ = 45◦
Electromagnetic Fields
(b)
20
20
10
10
0
0
αy
αy
(a)
–10
–10
–20 –20
–10
0
10
–20 –20
20
–10
0
10
20
αx
αx
(d)
(c)
10
10
0
0
αy
20
αy
20
–10
–10
–20 –20
–10
0 αx
10
–20 –20
20
(e)
–10
0 αx
10
20
–10
0 αx
10
20
(f) 20
20
10
10
0
0
αy
αy
29
–10
–20 –20
–10
–10
0 αx
10
20
–20 –20
FIGURE 5 Distributions of the square root of the electric energy density for some electromagnetic propagation-invariant fields with cone angles θ = 30◦ (left column) and θ = 45◦ (right column). Top: linear polarization. Middle: circular polarization. Bottom: radial polarization. Black denotes zero energy density
to demonstrate the significant changes in transverse profiles that take place in the non-paraxial domain. The square root of the electric energy density is plotted to display the transversal structure more clearly, and to simplify comparison with the amplitude profiles of scalar fields. When moving into the non-paraxial domain, the linearly polarized field shown
30
Propagation-Invariant Optical Fields
in Figure 5a and 5b becomes rotationally non-symmetric, as pointed out by Mishra (1991). The circularly polarized field in Figure 5c and 5d retains its symmetry, as does the radially polarized field in Figure 5e and 5f. In the former case, however, the increasing contribution of the J1 term reduces the axial intensity and decreases the fringe contrast. In the radially polarized case, the paraxial field has an axial zero, which is filled in the non-paraxial domain, where the fringe contrast decreases more rapidly than in the circularly polarized case. In the azimuthally polarized case, full contrast is retained also in the non-paraxial domain and the result is always similar to Figure 1c.
3.3. Solutions with Two Montgomery’s Rings We proceed to consider certain propagation-invariant solutions of Maxwell’s equations, which can not be anticipated in the framework of the scalar theory. Thus far we have only analyzed solutions with the spatial frequencies confined onto a single Montgomery’s ring, which are the only possible solutions within scalar theory. However, as shown by Turunen and Friberg (1993a) and Tervo, Vahimaa, and Turunen (2002), the situation is different with vector fields. A general solution for the class of electromagnetic propagation-invariant fields is yet to be found, but there are examples of fields with the spatial frequencies located on two different Montgomery’s rings. In such a case the different definitions of propagation-invariance, given in Equations (3.1)–(3.3), are no longer equivalent, as we will shortly see. Let us first investigate a combination of an axial plane wave, which is linearly polarized in y direction, and a Bessel field in Equation (3.17) that has only x and z components. According to Turunen and Friberg (1993a), the electric field then is E(ρ, ψ, z) = a yˆ exp (ikz) + b xˆ J0 (αρ) − iˆz J1 (αρ) tan θ sin ψ exp (iβz) ,
(3.28)
where a and b are constants. The electric energy density i h hwe (ρ, ψ, z, t)i = |a|2 + |b|2 J02 (αρ) + J12 (αρ) tan2 θ sin2 ψ
(3.29)
is clearly propagation-invariant. However, it is easy to establish that the magnetic energy density is not propagation-invariant: instead it selfimages, as does the z component of the time-averaged Poynting vector. As a second example, we consider the superposition of azimuthally and radially polarized fields with weight factors cφ and cρ , and the angular spectra confined onto two Montgomery’s rings with different radii αφ
Electromagnetic Fields
31
and αρ . These fields are orthogonal since the azimuthally polarized field in Equation (3.24) has only a φ component and the radially polarized field in Equation (3.25) has ρ and z components. The electric energy density of the superposition is readily found to be 0 hwe (r, t)i = 2
( " #) 2 2 α2 cφ J (αφ ρ) + cρ 2 J 2 (αρ ρ) + ρ J 2 (αρ ρ) , (3.30) 1 1 βρ2 0
1/2 where βρ = k 2 − αρ2 . It can be established that also the magnetic energy density and the z component of the time-averaged Poynting vector are propagation-invariant (see Tervo, Vahimaa, and Turunen (2002) for explicit expressions). Thus the field satisfies all the definitions (3.1)–(3.3).
3.4. Finite Approximations Finite-energy approximations of electromagnetic propagation-invariant fields can be constructed similarly to their scalar counterparts, either by assuming a finite aperture (as in Section 2.7), or by assuming that the angular spectrum of each field component is concentrated onto a ring (or two rings) of finite radial extent (as in Section 2.8). In the paraxial approximation this again leads to truly beam-like fields. Once the angular spectrum or the initial field distribution has been specified, the freespace propagation can be governed by applying the angular spectrum representation, or (in the paraxial case) the Fresnel diffraction formula, to each scalar component separately. Jordan and Hall (1994) considered an azimuthally polarized initial field of the form of Equation (3.24) but with a Gaussian apodizing envelope, i.e. E(ρ, ψ, 0) = iφˆ J1 (αρ) exp −ρ 2 /w02 .
(3.31)
They derived analytical expressions for the propagation of this azimuthally polarized Bessel–Gauss beam in free space by solving the Fresnel diffraction integral. They also considered the linearly polarized fundamental Bessel–Gauss beam and found that, apart from the polarization vector and a phase factor, the far-field distributions of the two are similar. A more detailed study of the subject was conducted by Greene and Hall (1996); see also Seshadri (2008). Hall (1996) extended the analysis to higher-order Bessel–Gauss fields in different polarization configurations. Greene and Hall (1998) provided illustrations of such beams and considered their diffraction from various objects.
32
Propagation-Invariant Optical Fields
4. PARTIALLY COHERENT FIELDS The property of propagation-invariance is not restricted to spatially fully coherent fields. The extension to spatially partially coherent fields presented by Turunen, Vasara, and Friberg (1991) is based on the coherence theory in the space-frequency domain developed by Wolf (1982).
4.1. General Field Representation Spatially partially coherent optical fields may be characterized in the space-frequency domain by the cross-spectral density function (Mandel and Wolf (1995), sect. 4.3.2) W (x1 , y1 , z 1 , x2 , y2 , z 2 ), which is a measure of the field correlations between two spatial points (x1 , y1 , z 1 ) and (x2 , y2 , z 2 ) at frequency ω. The condition for propagation-invariance in partially coherent optics is that W (x1 , y1 , z, x2 , y2 , z) = W (x1 , y1 , 0, x2 , y2 , 0)
(4.1)
for all z > 0. The general solution was derived by Turunen, Vasara, and Friberg (1991) using angular-spectrum techniques. The angular correlation function, defined as T ( f 1 , φ1 , f 2 , φ2 ) = (2π )
−4
ZZZZ
∞
W (x1 , y1 , 0, x2 , y2 , 0)
−∞
× exp [−i (x2 f 2 cos φ2 − x1 f 1 cos φ1 + y2 f 2 sin φ2 − y1 f 1 sin φ1 )] dx1 dx2 dy1 dy2 ,
(4.2)
of a propagation-invariant field must be of the form T ( f 1 , φ1 , f 2 , φ2 ) = S( f 1 , φ1 , φ2 )δ( f 1 − f 2 ),
(4.3)
where S( f, φ1 , φ2 ) is an arbitrary (real and positive) function. The corresponding expression for the cross-spectral density function then is Z k Z Z 2π W (x1 , y1 , z 1 , x2 , y2 , z 2 ) = f 2 S( f, φ1 , φ2 ) 0 0 1/2 2 2 × exp i (z 2 − z 1 ) k − f × exp [i f (x2 cos φ2 − x1 cos φ1 + y2 sin φ2 − y1 sin φ1 )] d f dφ1 dφ2 . (4.4)
Partially Coherent Fields
33
Friberg and Turunen (1994) have shown that fields defined by Equation (4.4) are eigenmodes of an infinite Fabry–Perot resonator in the sense of the coherence theory of resonator modes developed by Wolf and Agarwal (1984). In view of Equation (4.3), the characteristic property of all partially coherent propagation-invariant fields is that all plane-wave components are non-correlated in the radial direction, while the correlations may be arbitrary in the azimuthal direction. In other words, the general solution allows a continuum of Montgomery’s rings with different radii, provided that the fields on different rings are mutually uncorrelated. The correlations on each ring may, however, be arbitrary. Fields with wave vectors confined onto a single ring are a subset of the general solution. For such fields T ( f 1 , φ1 , f 2 , φ2 ) = α −2 A(φ1 , φ2 )δ( f 1 − α)δ( f 2 − α),
(4.5)
where A(φ1 , φ2 ) is an arbitrary complex function. The general solution (4.4) has a Bessel-function representation given by Turunen, Vasara, and Friberg (1991). On expanding the function S( f, φ1 , φ2 ) into a two-dimensional Fourier series with respect to the angular coordinates, we obtain W (x1 , y1 , z 1 , x2 , y2 , z 2 ) =
∞ X
∞ X
exp [i (mφ1 + nφ2 )]
m=−∞ n=−∞
Z ×
k
1/2 amn ( f )Jm ( fρ1 )Jn ( fρ2 ) exp i (z 2 − z 1 ) k 2 − f 2 d f, (4.6)
0
where amn ( f ) are related to the Fourier coefficients of S( f, φ1 , φ2 ). This result was also derived by Kowarz and Agarwal (1995) from their general Bessel-function representation of partially coherent fields. Some special cases of these fields have been discussed by Ostrovsky, Nartin´ez-Niconoff, and Martinez-San-Juan (2001, 2002).
4.2. Specific Examples Let us assume that the intensity distribution of a partially coherent field is uniform and the distribution of the complex degree of spatial coherence depends only on the transverse coordinate differences x1 − x2 and y1 − y2 , i.e. the field is of the Schell-model type (see Mandel and Wolf (1995), sects. 4.3.2 and 5.3.2). Approximations of such fields are generated in, e.g. acousto-optic devices (Ohtsuka, 1981; Turunen, Tervonen, & Friberg, 1990), and their invariant propagation properties have been investigated by Turunen (2002), who established that the spatial
Propagation-Invariant Optical Fields
34
frequencies of a uniform-intensity Schell-model field are uncorrelated and that, consequently, the complex degree of spatial coherence is propagation-invariant. One special case of such fields is the Gaussiancorrelated Schell-model field, and another example is obtained if we assume that the angular spectrum is confined on a single ring and choose A(φ1 , φ2 ) = δ(φ1 − φ2 ) in Equation (4.5). Now W (x1 , y1 , z 1 , x2 , y2 , z 2 ) = exp [iβ (z 2 − z 1 )] h i1/2 × J0 α (x2 − x1 )2 + (y2 − y1 )2 (4.7) and we have the fundamental Bessel-correlated field introduced by Gori, Guattari, and Padovani (1987b). Clearly, again, the intensity distribution is constant but the complex degree of spatial coherence is a sharply peaked function. The Bessel-correlated field is interesting also because it is one of the few fields for which the coherent modes introduced by Wolf (1982) have been evaluated explicitly. These modes, denoted by 9n (x, y, z), allow an expansion of the cross-spectral density function in the form W (x1 , y1 , z, x2 , y2 , z) =
∞ X
λn 9n∗ (x1 , y1 , z)9n (x2 , y2 , z),
(4.8)
n=0
where the functions 9n (x, y, z) and the weights λn satisfy the Fredholm integral equation (Mandel and Wolf (1995), sect. 4.7) ZZ
∞
W (x1 , y1 , z, x2 , y2 , z)9n (x1 , y1 , z)dx1 dy1 = λn 9n (x2 , y2 , z). (4.9)
−∞
Gori, Guattari, and Padovani (1987b) showed that the coherent modes of the Bessel-correlated field can be chosen as coherent, propagationinvariant Bessel field modes, i.e. 9n (ρ, ψ, z) = Jm (αρ) exp [i (mψ + βz)] ,
(4.10)
with equal weights. Gori, Santarsiero, and Borghi (2008) and Gori, Santarsiero, Borghi, and Ram´ırez-S´anchez (2008) pointed out that, because of this degeneracy, other choices are possible as well. They also presented the electromagnetic extension of the scalar results. Partially coherent propagation-invariant fields with sharply peaked intensity distributions exist as well. One example, due to Turunen, Vasara,
Partially Coherent Fields
35
and Friberg (1991), is h i1/2 2 2 W (x1 , y1 , z, x2 , y2 , z) = J0 α (x2 − x1 ) + (y2 − y1 ) h i1/2 2 2 + cos γ J0 α (x2 + x1 ) + (y2 + y1 ) , (4.11) where γ is an arbitrary real number. The intensity distribution is given by I (ρ, z) = 1 + J0 (2αρ) cos γ .
(4.12)
When γ = 0, an intensity peak on a luminous background is obtained, but if γ = π we have a field with a well-defined axial zero. Ponomarenko, Huang, and Cada (2007) named these fields dark and antidark diffractionfree beams (though they are not beam-like) and showed that the modal weights of such fields are given by λm = 1 + (−1)m cos γ . Examination of Equation (4.12) shows that the precision of definition of the optical axis in the case of a ‘dark’ partially coherent propagationinvariant field, pictured in Figure 6a, is comparable to that of the coherent J1 Bessel field. The amplitude and (binary) phase of the complex degree of spatial coherence at y1 = y2 = 0, W (x1 , 0, z, x2 , 0, z) , µ(x1 , 0, z, x2 , 0, z) = √ I (x1 , 0, z)I (x2 , 0, z)
(4.13)
of this wave field are displayed in Figure 6b and 6c, respectively. The function in Equation (4.13) could be measured by scanning the two pinholes in Young’s experiment along the x-axis. Additional examples of partially coherent propagation-invariant fields can easily be constructed. Bouchal and Peˇrina (2002) discussed a wide range of fields with the angular spectrum confined on a single ring, but with the degree of coherence modulated along the azimuthal direction. Gori, Santarsiero, Borghi, and Li (2007) provided numerous examples of fields generated by thin annular sources, and discussed methods to generate such fields, with tailored correlation properties, using primary incoherent sources and rotating diffusers with appropriately engineered statistics.
4.3. Finite Approximations In their original work on Bessel-correlated fields, Gori, Guattari, and Padovani (1987a) considered planar sources with finite, well-behaved
Propagation-Invariant Optical Fields
(b) 20
10
10
αx2
(a) 20
αy
36
0 –10 –20 –20
0 –10
–10
0 αx
10
–20 –20
20
–10
0 αx1
10
20
(c) 20
αx2
10 0 –10 –20 –20
–10
0 αx1
10
20
FIGURE 6 (a) Distribution of the square root of the intensity of a partially coherent propagation-invariant field given by Equation (4.12) with γ = −1. The absolute value of the complex degree of spatial coherence is shown in (b) and its phase in (c)
transverse intensity distributions I (x, y, 0) and a correlation function of the type in Equation (4.7), i.e. W (x1 , y1 , 0, x2 , y2 , 0) =
p
I (x1 , y1 , 0)I (x2 , y2 , 0) h i1/2 × J0 α (x2 − x1 )2 + (y2 − y1 )2 . (4.14)
They showed that if I (x, y, 0) depends only on the radial distance ρ from the origin, the coherent modes are of the form −1/2
9n (ρ, ψ, 0) = λn
p
I (ρ, 0) an Jn (αρ) exp (inψ) + bn J−n (αρ) exp (−inψ) ,
(4.15)
where an /bn is arbitrary. Moreover, the modal weights were shown to be λn = 2π
Z 0
∞
I (ρ, 0)Jn2 (αρ)ρdρ.
(4.16)
Partially Coherent Fields
37
Thus the coherent modes can be represented as superpositions of propagation-invariant Bessel fields, modulated by [I (ρ, 0)]1/2 . The effective number of non-zero eigenvalues increases with the width of I (ρ, 0) if α is constant, in agreement with the general property that the less coherent the field, the larger the number of coherent modes. Assuming that the intensity distribution of a Bessel-correlated Schellmodel source is Gaussian, i.e. I (ρ, 0) = I0 exp −2ρ 2 /w02 ,
(4.17)
with I0 a constant, the coherent modes at z = 0 represent sources of higherorder Bessel-Gauss beams. In this case the eigenvalues in the modal expansion are, according to Equation (4.16), λn =
π 2 w0 I0 exp −w02 α 2 /4 In w02 α 2 /4 , 2
(4.18)
where In is the modified Bessel function of order n. Using the propagation laws of higher-order Bessel-Gauss beams, Palma, Borghi, and Cincotti (1996) derived a closed-form expression for the partially coherent beam radiated by the J0 -correlated source. Explicitly, " # w02 ρ12 + ρ22 W (ρ1 , ψ1 , z, ρ2 , ψ2 , z) = I0 2 exp − 2 w (z) w (z) " # 2 z2 ik 2 2α w0 n 2 × exp ρ2 − ρ12 exp − 2 2 J0 α ρ1 exp [−i29(z)] 2R(z) w(z) k w (z) o1/2 + ρ22 exp [i29(z)] − 2ρ1 ρ2 cos (ψ1 − ψ2 ) , (4.19) where 9(z), w(z), and R(z) are given by Equations (2.38)–(2.40). The (paraxial) conical-wave nature of the field is expressed in the third exponential factor in Equation (4.19). Borghi (1999) has shown that J0 beams of the type of Equation (4.19) can be constructed using an incoherent superposition of suitably shifted and tilted Gaussian beams; see also Shchegrov and Wolf (2000). Such a decomposition of a partially coherent wave field into a set of coherent fields of identical basic form has its roots in the work of Gori and Palma (1978) and Gori (1980). Besides being illustrative, decompositions of this type are of significant engineering value: they allow one to propagate partially coherent fields (numerically) using standard coherent techniques
38
Propagation-Invariant Optical Fields
even in complex optical systems (Turunen & Vahimaa, 2008; Vahimaa & Turunen, 2006). Finally, propagation characteristics of partially coherent Bessel–Gauss fields with the transversal intensity distribution corresponding to Equation (2.35) and having a Gaussian Schell-model complex degree of spatial coherence, have been evaluated analytically by Zahid and Zubairy (1989).
4.4. Electromagnetic Solutions Vectorial partially coherent propagation-invariant fields have thus far received little attention in the literature. Such fields can be constructed by considering the 3 × 3 cross-spectral density tensor with components Wi j (r1 , r2 ), where (i, j) = (x, y, z) in Cartesian coordinates or (i, j) = (ρ, ψ, z) in circular cylindrical coordinates. To do this, one can assume that the appropriate scalar components Ti j ( f 1 , φ1 , f 2 , φ2 ) of the angular correlation tensor are of the form of Equation (4.3). Since Maxwell’s equations need to be satisfied and Wi j (r1 , r2 ) = W ji∗ (r2 , r1 ), only three components of the cross-spectral density function (such as Wx x , W yy , and Wx y ) are independent, but Wx y must be chosen such that the cross-spectral density tensor satisfies the appropriate realizability requirements related to non-negative definiteness (see Gori and Santarsiero (2007) and Gori, Santarsiero, Borghi, and Ram´ırez-S´anchez (2008) for more details of these conditions). The discussion of partially coherent (and partially polarized) electromagnetic fields is simplified in the paraxial approximation, since then the longitudinal components of the electric and magnetic fields can be neglected and a 2 × 2 matrix suffices to describe the polarization and coherence properties of the field, as discussed in general terms by Gori (1998), Gori, Santarsiero, Vicalvi, Borghi, and Guattari (1998), and Wolf (2003). In fact, the interesting example provided by Gori and Santarsiero (1998) could easily be adapted to Bessel fields. Tervo (2003) considered purely azimuthally polarized solutions for propagation-invariant fields, i.e. waves for which the only non-vanishing term of the cross-spectral density tensor is Wφφ . He showed that the only possible solutions are of the rotationally symmetric form Wφφ (ρ1 , z 1 , ρ2 , z 2 ) =
Z
k
f 2 S( f )J1 ( fρ1 )J1 ( fρ2 ) q 2 2 × exp i (z 2 − z 1 ) k − f d f, 0
(4.20)
with real and positive S( f ). Thus the angular spectrum of any azimuthally polarized solutions must be fully coherent and uniform in the azimuthal
Partially Coherent Fields
39
direction, but any incoherent radial superposition of rings of this type is allowed. Hence, if only a single ring is considered by choosing S( f ) = δ( f − α0 ), the field is necessarily fully coherent with Wφφ (ρ1 , z 1 , ρ2 , z 2 ) = α02 J1 (α0 ρ1 )J1 (α0 ρ2 ) exp [iβ0 (z 2 − z 1 )]
(4.21)
q and β0 = k 2 − α02 . Another closed-form solution, but in this case with partial spatial coherence properties, is obtained by assuming for z 1 = z 2 = z that S( f ) =
1/ f, if cα0 ≤ f ≤ α0 , 0, otherwise,
(4.22)
and 0 ≤ c ≤ 1 is real. Now, the (planar) cross-spectral density function takes on the form α0 {ρ1 [J0 (α0 ρ1 )J1 (α0 ρ2 ) − c J0 (cα0 ρ1 )J1 (cα0 ρ2 )] ρ22 − ρ12 − ρ2 [J0 (α0 ρ2 )J1 (α0 ρ1 ) − c J0 (cα0 ρ2 )J1 (cα0 ρ1 )]} (4.23)
Wφφ (ρ1 , z, ρ2 , z) =
and the electric energy density is proportional to Wφφ (ρ, z, ρ, z) =
1 2n 2 α J (α0 ρ) − J0 (α0 ρ)J2 (α0 ρ) 2 0 1h
io − c2 J12 (cα0 ρ) − J0 (cα0 ρ)J2 (cα0 ρ) .
(4.24)
At the limit c → 1, the fully coherent (single-ring) case is recovered. We stress that, for any value of c, the field is strictly propagation-invariant, not a finite-angular-spectrum approximation. Figure 7a, where we have taken c = 0, illustrates the electric energy density and shows that the axial intensity is zero. Because of the incoherent superposition of a continuum of coherent single-ring contributions, the other minima are largely smoothed out. The amplitude and phase of the complex degree of azimuthal coherence at y1 = y2 = 0, µφφ (x1 , 0, z, x2 , 0, z) = p
Wφφ (x1 , 0, z, x2 , 0, z) , (4.25) Wφφ (x1 , 0, z, x1 , 0, z)Wφφ (x2 , 0, z, x2 , 0, z)
are illustrated in Figure 7b and 7c, respectively, for c = 0.
Propagation-Invariant Optical Fields
(b) 20
10
10
α 0x 2
(a) 20
α 0y
40
0 10
0 10
20
20 20
10
0 α 0x
10
20
20
10
0 α 0x 1
10
20
(c) 20
α 0x 2
10 0 10 20 20
10
0 α 0x 1
10
20
FIGURE 7 (a) Distribution of the square root of the energy density of a partially coherent azimuthally polarized propagation-invariant field, given by Equation (4.24). The absolute value of the complex degree of azimuthal spatial coherence is shown in (b) and its phase in (c).
5. POLYCHROMATIC AND PULSED LOCALIZED WAVES Thus far we have assumed that the propagation-invariant field is either fully monochromatic or (in the case of partial coherence) we have considered a statistically stationary field at a single angular frequency ω. In this section we examine polychromatic fields that are propagationinvariant at each frequency separately, assuming complex-analytic-signal field representation. The nature of the field then depends critically on the correlations between these spectral components: in the case of complete non-correlation, a stationary polychromatic field ensues, while pulsed fields are obtained in the case of partial or full correlation between the frequency components. We concentrate on the two limiting cases of complete correlation and complete non-correlation since the intermediate range of partial correlation has not been discussed in the literature in the context of propagation-invariant fields. After some general considerations, we provide examples of both stationary and pulsed propagation-invariant solutions of the wave equation. For more details of coherent pulsed solutions, we refer to reviews by Kiselev (2007) and Hern´andez-Figueroa, Zamboni-Rached, and Recami (2008).
Polychromatic and Pulsed Localized Waves
41
5.1. Field Representations Showing the frequency dependence explicitly, we write Equations (2.2) and (2.6) in the form U (x, y, z; ω) = G(ω)V (x, y; ω) exp [iβ(ω)z]
(5.1)
and V (x, y; ω) =
2π
Z
A(φ) exp [iα(ω) (x cos φ + y sin φ)] dφ.
(5.2)
0
For simplicity, we assume the frequency-dependent azimuthal variation of the angular spectrum, G(ω)A(φ), to be separable in ω and φ, as is customary in the literature. Furthermore, α(ω) and β(ω) are related to ω through Equation (2.5): α 2 (ω) + β 2 (ω) = ω2 /c2 ,
(5.3)
where c is the speed of light in vacuum. The spectral density of the pulse, S(x, y, z; ω) = |U (x, y, z; ω)|2 = |G(ω)|2 |V (x, y; ω)|2 = S(x, y; ω) (5.4) is propagation-invariant at each frequency. If the frequency components U (x, y, z; ω) are fully correlated, the timedomain field representation is obtained by the inverse temporal Fourier transform U (x, y, z; t) =
Z
∞
U (x, y, z; ω) exp (−iωt) dω,
(5.5)
0
which gives a general expression U (x, y, z; t) =
Z
∞
G(ω)V (x, y; ω) exp {i [β(ω)z − ωt]} dω
(5.6)
0
for coherent pulsed solutions. The temporal intensity distribution of such solutions is I (x, y, z; t) = |U (x, y, z; t)|2 . An intuitively attractive (though not completely general) approach to discussing temporally partially coherent propagation-invariant fields is to consider them as successions of identical, mutually independent, fully coherent pulses of the form of Equation (5.6), originating at different
42
Propagation-Invariant Optical Fields
instants of time t 0 , and having a real and positive envelope function p(t 0 ). Then the mutual coherence function of the entire non-stationary field,4 describing correlations between two instants of time t1 and t2 at two planes z 1 and z 2 , can be expressed in the form 0(r1 , r2 ; t1 , t2 ) =
Z
∞
p(t 0 )U ∗ (r1 ; t1 − t 0 )U (r1 ; t2 − t 0 )dt 0 .
(5.7)
−∞
Note that, although we have assumed the field to be fully spatially coherent at each frequency ω, this is not necessarily true in the spacetime domain because, in general, the function 0 is not factorable in spatial coordinates. Obviously, the fully coherent pulse is obtained with the choice p(t 0 ) = δ(t 0 ) and the coherence of the field is reduced when the width of p(t 0 ) increases. A stationary field is obtained with the choice p(t 0 ) = 1 in Equation (5.7), which indicates an infinitely long train of independent identical pulses with a constant envelope. Use of the integral definition of the Dirac delta function then leads to a time-domain expression for spatially fully coherent stationary propagation-invariant fields in the form 0(x1 , y1 , x2 , y2 , 1z; τ ) =
Z
∞
|G(ω)|2 V ∗ (x1 , y1 ; ω)
0
× V (x2 , y2 ; ω) exp {i [β(ω)1z − ωτ ]} dω, (5.8) where 1z = z 2 − z 1 and τ = t2 − t1 . This expression allows one to calculate the coherence length and coherence time of the field, given by the characteristic decay lengths of 0(x, y, x, y; 1z, τ ) in 1z and τ , respectively. In general, these quantities depend on the position. Finally, I (x, y) = 0(x, y, x, y, 0; 0) =
∞
Z
S(x, y; ω)dω,
(5.9)
0
where S(x, y; ω) is given by Equation (5.4), represents the frequencyaveraged transverse intensity distribution of the stationary field.
5.2. Dispersion Relations and Spectra It is clear from the formulas presented above that the character of the polychromatic propagation-invariant field, whether pulsed or stationary, depends on the dispersion relation (5.3). In general, when considering 4 This function is discussed, e.g. by Bertolotti, Sereda, and Ferrari (1997) and the approach used here is an extension of that introduced for partially coherent plane-wave pulses by Vahimaa and Turunen (2006).
Polychromatic and Pulsed Localized Waves
43
pulsed fields, we may expand the expression for the propagation constant β(ω) in the usual form 1 ∂ 2 β(ω) ∂β(ω) β(ω) = β(ω0 ) + (ω − ω0 ) + ∂ω ω=ω0 2 ∂ω2
(ω − ω0 )2 + . . . ω=ω0
1 = β(ω0 ) + (ω − ω0 ) + Dω (ω − ω0 )2 + . . . vg
(5.10)
to define the group velocity vg and its dispersion Dω . If the coefficient Dω = 0, the field is dispersion-free around ω = ω0 , and if all higher-order coefficients vanish as well, the field is completely dispersion-free. We next proceed to consider some specific dispersion relations that have been studied in connection with localized waves. 5.2.1. Transversally Achromatic Waves As the first special case of interest, we assume that α(ω) = α is independent of ω, i.e. the transverse profile of the field is achromatic. In view of Equation (5.3), the propagation constant now takes the frequencydependent form β(ω) =
q
(ω/c)2 − α 2 .
(5.11)
The cone angle of the field becomes frequency-dependent as well, θ (ω) = arcsin(αc/ω),
(5.12)
and hence the angular spectrum is a coloured distribution with a finite radial extent. Porras (2001) and Hu and Guo (2002) considered the dispersive properties of achromatic waves, noting that q vg = c 1 − (αc/ω0 )2 = c cos θ(ω0 ),
(5.13)
with θ (ω0 ) given by Equation (5.12), and
Dω = −
α2c 2ω03
"
1−
αc ω0
2 #−3/2
.
(5.14)
Thus the group velocity is smaller than the speed of light c (subluminal field) and its dispersion is negative (anomalous dispersion) so that
44
Propagation-Invariant Optical Fields
the pulses always spread (see, e.g. Campbell and Soloway (1990)). Porras (2001) found that achromatic fields can propagate, to a good approximation, without dispersion. This can occur in a (normally) dispersive material, with the condition that the positive material dispersion cancels the negative dispersion of the free-space field. 5.2.2. Waves with Achromatic Angular Spectrum A second interesting type of propagation-invariant polychromatic field can be introduced by assuming that the cone angle θ is independent of frequency, i.e. θ (ω) = θ , which yields α(ω) = (ω/c) sin θ
(5.15)
β(ω) = (ω/c) cos θ.
(5.16)
and
Since all frequency components propagate with the same cone angle, the angular spectrum is a sharp achromatic ring. The transverse spatial intensity distribution exhibits colour variations, but they are the same at every transverse plane. The group velocity of a strictly conical polychromatic wave is, according to Equation (5.16), vg = c/ cos θ.
(5.17)
Hence the field is superluminal (vg > c), but only in the same sense as a light stripe on a screen illuminated by an obliquely incident planewave pulse, thus causing no physical problems as stressed by Saari and Reivelt (1997). The group velocity dispersion coefficient Dω = 0, as do all higher-order coefficients in Equation (5.10). Thus the field is now exactly dispersion-invariant, in addition to being propagation-invariant. 5.2.3. Fields with Achromatic Propagation Constant A third natural alternative is to assume that the propagation constant β is independent of ω (achromatic). In view of Equation (5.3), we now obtain α(ω) =
q
(ω/c)2 − β 2 .
(5.18)
For these fields, both the transverse profile and the angular spectrum exhibit colour variations since the cone angle θ depends on the frequency
Polychromatic and Pulsed Localized Waves
45
according to θ (ω) = arccos(βc/ω).
(5.19)
The group velocity is, in this case, infinite, but the wave field is again exactly dispersion-free. 5.2.4. General Dispersion-free Waves Reivelt and Saari (2000), Salo and Salomaa (2001a,b), Sheppard (2001, 2002), and others have considered propagation constants of the form β(ω) = γ ω/c − µ,
(5.20)
where γ ≥ 0 and µ are constants. Obviously, if µ = 0, the fields described by Equation (5.20) reduce to conical waves with achromatic angular spectrum and the cone angle given by cos θ = γ . In general, with Equation (5.20), θ (ω) = arccos (γ − cµ/ω)
(5.21)
r ω cµ 2 α(ω) = 1− γ − . c ω
(5.22)
and
Fields defined by Equation (5.20) are exactly dispersion-free with a constant group velocity vg = c/γ , thus being superluminal or subluminal, depending on whether γ < 1 or γ > 1. If γ = 1, vg = c and the group velocity is the same as the speed of light (luminal wave). If we allow bidirectional propagation of the monochromatic components, i.e. 0 ≤ θ (ω) ≤ π , the propagation constant is bound to the interval −ω/c ≤ β(ω) ≤ ω/c. Thus the subluminal condition γ > 1 yields µ > 0 and a limited range µc µc ≤ω≤ γ +1 γ −1
(5.23)
of allowed frequencies in G(ω). If only forward-propagating monochromatic components are accepted, i.e. we demand β(ω) > 0, an additional constraint ω > µc/γ = µvg
(5.24)
46
Propagation-Invariant Optical Fields
must be applied. This condition is of concern in practical aperture-based realizations of polychromatic propagation-invariant fields. Considering next the superluminal condition γ ≤ 1 for positive values of µ, we have the restriction ω≥
µc γ +1
(5.25)
for fields containing both backward- and forward-propagating monochromatic components, and the former are excluded by the constraint (5.24). Finally, with negative values of µ, the constraint is ω≥
−µc , 1−γ
(5.26)
and no backward-propagating waves can exist. 5.2.5. Spectra In order to provide numerical examples on polychromatic fields, we need to define the functional form of the frequency spectrum G(ω). Often one chooses a Gaussian spectrum # " (2/π )1/4 (ω − ω0 )2 G(ω) = √ exp − 2
(5.27)
R∞ of centre frequency ω0 and width , normalized such that −∞ |G(ω)|2 dω = 1. However, this spectrum is consistent with the analytic-signal representation, and convenient only if the negative frequencies are ignored, which can be done without truncating the spectrum significantly if ω0 , i.e. if the spectrum is sufficiently narrow-band. To be able to deal with wide-band sources satisfactorily, we employ (in suitable instances) a model frequency spectrum that vanishes if ω < ωmin , and equals (2a/c)n+1/2 G(ω) = √ (ω − ωmin )n exp [−a (ω − ωmin ) /c] 0(2n + 1)
(5.28)
R∞ if ω ≥ ωmin , and is normalized such that ωmin |G(ω)|2 dω = 1. Here n and a are real constants, which define (along with ωmin ≥ 0) the characteristics of the spectrum. The maximum is located at ω¯ = ωmin + nc/a. Thus, keeping n/a constant while increasing both n and a, narrows the spectrum around
Polychromatic and Pulsed Localized Waves
47
a constant centre frequency. The fraction of energy that falls in the region of backward wave propagation in the case of the dispersion relation (5.20) is Z
cµ/γ
|G(ω)|2 dω = 1 −
cµ/(γ +1)
0[2n + 1, 2aµ/γ (γ + 1)] , 0(2n + 1)
(5.29)
where 0(a, z) is the incomplete Gamma function. If we choose the peak frequency ω¯ to fall well into the region of forward wave propagation, i.e. ω¯ cµ/γ or n aµ/γ (γ + 1) = b, this fraction is negligible (less than 0.5% when n = 3b and less than 10−5 if n = 5b).
5.3. Examples of Stationary Fields Using Equation (5.8) as the starting point, we first consider the implications of some of the dispersion relations introduced above in the properties of stationary propagation-invariant fields, which are spatially fully coherent in the space-frequency domain. Some fields of this type have been studied experimentally, using Ti:Sapphire lasers, superluminescent diodes, halogen bulbs, and photonic-crystal fibres for ´ supercontinuum generation, by Fischer, Brown, Morris, Lopez-Mariscal, ´ Wright, Sibbett, and Dholakia (2005) and Fischer, Little, Smith, LopezMariscal, Brown, Sibbett, and Dholakia (2006). The examples provided below are adapted from Turunen (2008). 5.3.1. Achromatic Stationary Fields Let us first consider the dispersion relation (5.11), which implies that V (x, y; ω) is independent of ω, i.e. the transverse profile of the polychromatic field is achromatic. It immediately follows that the normalized mutual correlation function (complex degree of coherence), γ (x1 , y1 , x2 , y2 , 1z; τ ) 0(x1 , y1 , x2 , y2 , 1z; τ ) = , [0(x1 , y1 , x1 , y1 , 0; 0)0(x2 , y2 , x2 , y2 , 0; 0)]1/2
(5.30)
can be expressed in a factorable form γ (x1 , y1 , x2 , y2 , 1z; τ ) = γ (x1 , y1 , x2 , y2 , 0; 0)γ (x, y, x, y, 1z; τ ), (5.31) where the first factor γ (x1 , y1 , x2 , y2 , 0; 0) =
V ∗ (x1 , y1 )V (x2 , y2 ) |V (x1 , y1 )V (x2 , y2 )|
(5.32)
Propagation-Invariant Optical Fields
48
describes the transverse spatial coherence of the field in the space-time domain and the second part γ (x, y, x, y, 1z; τ ) io n hp R∞ 2 2 − α 2 1z − ωτ |G(ω)| (ω/c) dω exp i 0 = R∞ 2 0 |G(ω)| dω
(5.33)
describes the longitudinal and temporal coherence. In view of Equation (5.32), |γ (x1 , y1 , x2 , y2 , 0; 0)| = 1, i.e. the field is transversely fully spatially coherent also in the space-time domain. Furthermore, in view of Equation (5.33), the longitudinal and temporal coherence properties are independent of position and depend only on the chosen frequency spectrum G(ω), as in the case of a polychromatic plane wave. Note, however, that while the temporal coherence properties (1z = 0) of any achromatic propagation-invariant field are identical to those of a plane wave with the same frequency spectrum G(ω), this is not true for the longitudinal coherence properties (τ = 0) because of the dispersion relation (5.11). 5.3.2. Conical Stationary Fields In general, polychromatic propagation-invariant fields exhibit partial spatial and temporal coherence properties in the space-time domain even if they are fully spatially coherent at each frequency. This can be illustrated by considering polychromatic conical waves, i.e. expressions (5.15) and (5.16). Now one obtains 0(x1 , y1 , x2 , y2 , 1z; τ ) =
ZZ
2π
dφ dφ 0 A∗ (φ)A(φ 0 )F [τ − cos θ 1z/c 0 (5.34) + sin θ x1 sin φ − x2 sin φ 0 + y1 cos φ − y2 cos φ 0 /c ,
where ∞
Z
|G(ω)|2 exp (−iωx) dω
F(x) =
(5.35)
0
is the temporal coherence function of a polychromatic plane wave with power spectrum |G(ω)|2 . Note that, despite the complicated argument of this function in Equation (5.34), the complex degree of coherence on the
Polychromatic and Pulsed Localized Waves
49
optical axis takes on the simple form R∞ γ (0, 0, 0, 0, 1z; τ ) =
0
|G(ω)|2 exp [−iω (τ − cos θ 1z/c)] dω , (5.36) R∞ 2 0 |G(ω)| dω
i.e. the field behaves like a polychromatic plane wave propagating along the direction of the z-axis in a medium of effective (dispersion-free) refractive index N = cos θ. To illustrate more clearly the spatial coherence of polychromatic conical waves, we consider the fundamental case A(φ) = 1 as an example. Now V (x, y; ω) = V (ρ; ω) = J0 (sin θ ωρ/c)
(5.37)
and the mutual coherence function assumes the form 0(ρ1 , ρ2 , 1z; τ ) =
∞
Z
|G(ω)|2 J0 (sin θ ωρ1 /c) J0 (sin θ ωρ2 /c) 0
× exp [iω (cos θ 1z/c − τ )] dω,
(5.38)
where we have used Equation (5.8) directly. Taking ωmin = 0 in Equation (5.28) and denoting α(ω) ¯ = (ω/c) ¯ sin θ = sin θ/a, we can evaluate the radial intensity distribution 2 I (ρ) = 3 F2 1/2, n + 1/2, n + 1; 1, 1; − [α(ω)ρ/n] ¯
(5.39)
and the mutual coherence function between an arbitrary and an axial point, 2 ¯ , 0(ρ, 0, 0; 0) = 2 F1 n + 1/2, n + 1; 1; − [α(ω)ρ/2n]
(5.40)
in terms of hypergeometric functions P FQ . Figure 8a illustrates, for clarity, the square root of the intensity distribution for three different values of the spectral bandwidth, determined by the power n in Equation (5.28). Clearly, a wide spectrum (small n) wipes out the zeros of the monochromatic Bessel beam as different frequency components have different lateral scales, in agreement ´ with the experimental observations of Fischer, Brown, Morris, LopezMariscal, Wright, Sibbett, and Dholakia (2005). When the spectrum becomes narrower, the zeros begin to emerge, starting from those closest to the origin. Figure 8b shows the absolute value of the complex degree of
50
Propagation-Invariant Optical Fields
(a)
1
I (ρ)
0.8 0.6 0.4 0.2
0
2.5
5
7.5
10
12.5
15
17.5
10
12.5
15
17.5
α
(b) 1
|γγ (ρρ, 0, 0; 0)|
0.8 0.6 0.4 0.2
0
2.5
5
7.5
α FIGURE 8 Radial distributions of (a) the square root of intensity and (b) the absolute value of the complex degree of spatial coherence of polychromatic stationary J0 Bessel fields with n = 1 (solid line), n = 10 (dashed line) and n = 100 (dotted line) in Equation (5.28)
√ spatial coherence γ (ρ, 0, 0; 0) = 0(ρ, 0, 0; 0)/ I (ρ)I (0). With the widest spectral band (n = 1) the spatial coherence systematically reduces close to zero at distances of the order of the first minimum of the monochromatic Bessel field at frequency ω. ¯ Decreasing the spectral bandwidth increases the effective spatial coherence area, but one observes an intriguing quasiperiodic behaviour, with zeros5 close to the locations of the zeros of the monochromatic Bessel field. These minima persist however narrow the spectrum G(ω) is, but the envelope of |γ (ρ, 0, 0; 0)| approaches unity when n → ∞. 5 The function γ (ρ, 0, 0; 0) is real, and changes sign at these zero positions.
Polychromatic and Pulsed Localized Waves
51
5.4. Examples of Pulsed Localized Waves We next turn our attention to coherent pulsed fields, which are propagation-invariant at each angular frequency. The starting point is Equation (5.6), and we again study some of the dispersion relations introduced in Section 5.2. 5.4.1. Bessel Pulses Assuming dispersion relation (5.11) so that V (x, y) is independent of ω, the temporal wave form of the pulse is separable in coordinates (x, y) and (z; t): U (x, y, z; t) = V (x, y)
Z
∞
q 2 2 G(ω) exp (−iωt) exp iz (ω/c) − α dω. (5.41)
0
As already noted in Section 5.2, the pulse is dispersive (Liu & Fan, 1998). However, there is no coupling between the transverse-spatial and longitudinal/temporal contributions and, as a result, the pulse spread due to dispersion is independent of its spatial distribution. Hence the propagation characteristics are the same for all coherent propagationinvariant pulses with a given frequency spectrum G(ω). They can be determined from Equation (5.41) by simple numerical integration. With a sufficiently narrow-band spectrum G(ω), the propagation constant β(ω) can be approximated by the first three terms in Equation (5.10), with the group velocity and dispersion coefficient given by Equations (5.13) and (5.14), respectively. Expanding the square root in Equation (5.41) and rearranging terms, we have U (x, y, z; t) = V (x, y) exp {i [β(ω0 )z − ω0 t]} Z ∞ h i × G(ω) exp iDω z (ω − ω0 )2 exp −i t − z/vg (ω − ω0 ) dω. (5.42) 0
Hence the (z; t)-dependent part of the wave is like a plane wave of frequency ω0 propagating in the positive z direction (the first exponential factor), multiplied by an envelope function (the integral). This shows that propagation-invariant pulses in free space behave analogously to conventional plane-wave pulses in dispersive media, as noted by Hu and Guo (2002). The integration in Equation (5.42) can be performed analytically for the Gaussian frequency spectrum defined in Equation (5.27) assuming that ω0 so that the lower integration limit can be taken as −∞. Then, if
Propagation-Invariant Optical Fields
52
we denote by T = 2/ the pulse-envelope half-width at z = 0, and write zT =
1 T2 , = 4Dω 2 D ω
(5.43)
we obtain √
" # π 1 + iz/z T z 2 U (x, y, z; t) = V (x, y) √ exp − t− , (5.44) vg T 2 (z) 1 − iz/z T where q T (z) = T 1 + (z/z T )2 .
(5.45)
The time-dependent intensity distribution is given by " 2 # t − z/vg T exp − , I (x, y, z; t) = I0 |V (x, y)| T (z) T 2 (z) 2
(5.46)
with I0 = π 2 . Hence the function T (z) represents the temporal pulse √ width at distance z, which increases by a factor 2 as the pulse propagates from z = 0 to z = z T . Note that, in addition to temporal spreading, the (spatially) propagation-invariant pulse acquires a spherical phase (chirp) term upon propagation. For very short (few-cycle) pulses the temporal evolution becomes more complicated and the propagated pulse shape differs from Gaussian, as is obvious from Equation (5.42); see Hu and Guo (2002) for further details. It is instructive to consider some numerical values of the dispersion properties of Bessel pulses. Let us assume, for example, a central wavelength λ0 = 0.86 µm and a modest central cone angle θ (λ0 ) = 10◦ , so that the scalar field representation is adequate. Then α ≈ 1.269 µm−1 . For a T = 50 fs pulse we then have, from Equations (5.43) and (5.14), z T ≈ 20.5 mm. The radius of the finite aperture required to generate a propagation-invariant distance equal to z T would, according to Equation (2.34), be R = z T tan θ (λ0 ) ≈ 3.6 mm. 5.4.2. X Waves Consider next conical fields that satisfy relations (5.15) and (5.16), assuming a single Bessel field mode of order m with A(φ) = (2π )−1 i−m exp (imφ). In polar coordinates we now have V (ρ, ψ; ω) = Jm [(ω/c)ρ sin θ ] exp (imψ). Thus, from Equation (5.6), the space-time
Polychromatic and Pulsed Localized Waves
representation of the pulse takes the form Z ∞ ω ρ sin θ U (ρ, ψ, z; t) = exp (imψ) G(ω)Jm c h z 0 i × exp i cos θ − t ω dω. c
53
(5.47)
A particularly interesting class of fields, known as X waves, was introduced by Lu and Greenleaf (1992a). These waves are obtained if we assume the spectral distribution (5.28) with ωmin = 0. Equation (5.47) can now be integrated using the Laplace-transform formula Z
∞
x n Jm (ax) exp (−bx) dx 0 −(n+1)/2 b a 2 + b2 , = 0(m + n + 1)Pn−m √ a 2 + b2
(5.48)
where Pnm is the associated Legendre function. The result is Unm (ρ, ψ, z; t) =
0(m + n + 1) c (2a)n+1/2 exp (imψ) √ 0(2n + 1) 0 a + ict × Pn−m M(ρ, t 0 )−(n+1) , M(ρ, t 0 ) √
(5.49)
with t 0 = t − z cos θ/c and h 2 i1/2 . M(ρ, t 0 ) = ρ 2 sin2 θ + a + ict 0
(5.50)
The lowest-order solution with m = 0 is the original ‘broadband’ X wave introduced by Lu and Greenleaf (1992a) and the higher-order solutions discussed by Fagerholm, Friberg, Huttunen, Morgan, and Salomaa (1996) and Friberg, Fagerholm, and Salomaa (1997) are temporal derivatives of the fundamental solution, as one can easily verify. Some of the lowest-order solutions for X waves can readily be obtained by inserting explicit expressions for the Legendre functions. Thus, with (n, m) = (0, 0), we have √ −1 U00 (ρ, ψ, z; t) = 2ac M(ρ, t 0 ) , (5.51) and with (n, m) = (1, 0) we obtain √ −3 U10 (ρ, ψ, z; t) = a 2ac a + ict 0 M(ρ, t 0 ) .
(5.52)
Propagation-Invariant Optical Fields
54
5
5 x [ µ m]
(b) 10
x [ µ m]
(a) 10
0
0
–5
–5
–10 –10
–5
0 z [ µ m]
5
10
–10 –10
–5
0
5
10
z [ µm]
FIGURE 9 Cross-sectional amplitude plots (at t = 0) of X waves with (a) n = 1 and m = 0, and (b) n = m = 1. In both, the cone angle is θ = 20◦ and a = 1 µm. The shapes remain invariant at other values of t, moving at the speed vg given by Equation (5.17) in the positive z direction
Correspondingly, if m = 1, a trivial solution U01 (ρ, ψ, z; t) = 0 is obtained if n = 0, but a nontrivial solution √ −3 U11 (ρ, ψ, z; t) = a 2ac exp (iψ) ρ sin θ M(ρ, t 0 )
(5.53)
results if n = 1. Figure 9 illustrates cross sections of the amplitude profiles of the non-vanishing solutions given by Equations (5.52) and (5.53). Common to both is the striking X-shaped behaviour, which gives rise to the name X wave. An observer travelling with the wave at the cross of the X (i.e. t 0 = 0) sees a transversely localized wave. A stationary observer at the optical axis experiences a single temporally localized pulse, but an offaxis observer located in the region where the X arms have separated sees two separate pulses. Note that the field is not localized in the directions of the X arms, which correspond to conical wave fronts with the normal at an angle θ with respect to the z-axis. The pulse has a bright central peak around ρ = t 0 = 0 when m = 0, and a dark central region when m > 0, in analogy with monochromatic mth-order Bessel field modes. Though not shown here, side lobes begin to appear when n is increased: then the spectral bandwidth decreases and the transverse scale distribution of the monochromatic field components becomes narrower. For example, the first zero of the fundamental (m = 0) X wave solution then begins to form close to the position |x| ≈ 2.405a/n sin θ . 5.4.3. Focus Wave Modes Focus wave modes (FWMs) were introduced by Brittingham (1983) as exact dispersion-free solutions to the wave equation (and Maxwell’s
Polychromatic and Pulsed Localized Waves
55
equations). While first derived in a heuristic manner, it was soon noted by B´elanger (1984) and Sezginer (1985) that FWMs have the general (scalar) form h µ i U (x, y, z; t) = u(x, y; z − ct) exp −i (z + ct) , (5.54) 2 where u(x, y, z − ct) is any solution of the paraxial wave equation and µ is a constant. Hillion (1992) found that FWMs satisfy the dispersion relation (5.20) with γ = 1, implying that they are propagation-invariant fields at any frequency; they are also luminal dispersion-invariant fields. Considerable discussion and even controversy arose about the causality of focus wave mode solutions and their physical realizability; see, e.g. Heyman, Steinberg, and Felsen (1987), Hillion (1992), Ziolkowski, Besieris, and Shaarawi (1993), and Heyman and Felsen (1994). However, as we shall see, correct conclusions can easily be drawn in the frequency domain. Let us write γ = 1 in Equations (5.20) and (5.22), thus choosing luminal waves. Considering mth-order Bessel field modes, Equation (5.6) takes the form Z ∞ U (ρ, ψ, z; t) = exp [i (mψ − µz)] G(ω) 0 i h p × Jm ρ µ (2ω/c − µ) exp [i (z/c − t) ω] dω. (5.55) Let us further assume the spectrum of Equation (5.28). Changing variables x = a (ω − ωmin ) /c and denoting t 0 = t − z/c, we have √ 2n+1/2 c/a exp i mψ − µz − ωmin t 0 U (ρ, ψ, z; t) = √ 0(2n + 1) q Z ∞ n 0 2 × x exp − 1 + ict /a x Jm ρ 2µ (x/a + ωmin /c) − µ dx. (5.56) 0
For FWMs, ω has the absolute lower bound ωmin = cµ/2, as indicated by Equation (5.25). In this case the integral in Equation (5.56) can be evaluated in terms of hypergeometric functions: if m/2 + n + 1 > 0, we obtain U (ρ, ψ, z; t) =
√ 0(m/2 + n + 1) ac a n µm/2 2n+1/2−m/2 √ 0(2n + 1)0(m + 1) exp {i [mψ − µ(z + ct)/2]} × [a − i(z − ct)]m/2+n+1 " # µρ 2 /2 m × ρ 1 F1 m/2 + n + 1, m + 1, − . (5.57) a − i(z − ct)
56
Propagation-Invariant Optical Fields
With n = m = 0 we have the solution " # √ µρ 2 /2 ac exp − exp [−iµ(z + ct)/2] , (5.58) U (ρ, z; t) = a − i(z − ct) a − i(z − ct) which is precisely Brittingham’s original Gaussian focus wave mode. Other simple solutions also exist: if n = 1 and m = 2, we have a solution √ µa acρ 2
"
µρ 2 /2 exp − U (ρ, ψ, z; t) = a − i(z − ct) [a − i(z − ct)]3 × exp {i [2ψ − µ(z + ct)/2]}
#
(5.59)
with a zero on the optical axis, as one would expect since the J2 Bessel mode was used to construct it. The solutions (5.58) and (5.59) contain plane waves propagating towards negative half-space if cµ/2 ≤ ω < cµ. If only forwardpropagating plane-wave components are allowed, we must take ωmin = cµ in accordance with Equation (5.24). In this case no general analytic result can be obtained, but on axis (ρ = 0) we have, for m = n = 0, I (0, z; t) =
ac . a 2 + (z − ct)2
(5.60)
The same result is obtained from Equation (5.58). Thus the omission of backward-propagating waves does not influence the axial form of the fundamental FWM, although it has an effect in the off-axis field. Using large enough values of n in G(ω), the contribution of the backward-propagating components (with frequencies cµ/2 < ω < cµ) to the pulse amplitude can be made arbitrarily small as discussed by Shaarawi, Ziolkowski, and Besieris (1995), and as is evident from Equation (5.29). This reduces the spectral width of the field (towards optically realizable values) and facilitates single-aperture realizations of FWMs as discussed by Reivelt and Saari (2004). Figure 10 illustrates two FWM solutions with negligible backward-propagating contributions: m = 0, n = 5 and m = 1, n = 5. In both cases the crossed pulse shape already seen with X waves is apparent; for smaller values of n the X shape becomes increasingly blurred because a wider range of conical waves with different cone angles is effectively present. Some details of the side lobes of the basic J0 and J1 fields are also visible close to the optical axis. The√ first zero of the former is seen to form close to the position |x| ≈ 2.405 a/2µn, i.e. |x| ≈ 0.43 µm with the parameter choices made in Figure 10.
Polychromatic and Pulsed Localized Waves
(b) 4
4
2
2 x [ µm]
x [ µm]
(a)
57
0
0
–2
–2
–4
–4 –4
–2
0 z [ µm]
2
4
–4
–2
0 z [µm]
2
4
FIGURE 10 Cross-sectional amplitude plots (at t = 0) of focus wave modes with (a) m = 0, n = 5, and (b) m = 1, n = 5. In both cases a = 1 µm and µ = 3.1 µm−1 , implying a cone angle θ (ω) ¯ = 20◦ at the peak frequency ω¯ of the spectrum.
Since the frequency spectrum G(ω) of FWMs (as well as X waves) can assume any form allowed by the constraints discussed in Section 5.2.4, an unlimited variety of FWM solutions can be constructed. Sheppard and Saari (2008) presented several interesting band-limited solutions in terms of Lommel functions, familiar from the theory of waves in focal regions. In particular, they considered wave packets that propagate in the direction opposite to all of their monochromatic constituents.
5.5. Partially Coherent and Electromagnetic Solutions Pulsed, spectrally partially coherent propagation-invariant fields can be constructed from the coherent solutions, for example, using Equation (5.7). As soon as the function p(t 0 ) differs from the delta function, partial spatial coherence is introduced in the space-time domain. In other words, partial spectral coherence implies partial spatial coherence in the spacetime domain even if the field is spatially fully coherent in the spacefrequency domain. On the other hand, pulsed fields that are partially spatially coherent in the space-frequency domain can be constructed by assuming that the general solution (4.4) applies to each frequency component separately. In fact, it is not difficult to find a general condition for propagationinvariance of spectrally and spatially partially coherent (scalar) fields using the space-frequency-domain coherence theory of non-stationary fields discussed e.g. by Bertolotti, Sereda, and Ferrari (1997). The (twofrequency) cross-spectral density function W (x1 , y1 , z 1 , x2 , y2 , z 2 ; ω1 , ω2 ) is
58
Propagation-Invariant Optical Fields
propagation-invariant if W (x1 , y1 , z, x2 , y2 , z; ω1 , ω2 ) = W (x1 , y1 , 0, x2 , y2 , 0; ω1 , ω2 )
(5.61)
for all z. Using the angular spectrum representation and denoting β( f ; ω) =
q
(ω/c)2 − f 2 ,
(5.62)
the relevant condition is β( f 1 ; ω1 ) = β( f 2 ; ω2 ).
(5.63)
Thus the relationship between f 2 and f 1 must be such that the longitudinal components of the wave vectors at frequencies ω2 and ω1 are equal, i.e. the angular correlation function must be of the form T ( f 1 , φ1 , f 2 , φ2 ; ω1 , ω2 ) = S( f 1 , φ1 , φ2 ; ω1 , ω2 )δ
f2 −
q
f 12 − ω12 − ω22 /c2
(5.64)
in place of Equation (4.5), which applies to stationary fields. The general form of the cross-spectral density function then is, instead of (4.4), ω1 /c
2π q W (x1 , y1 , z 1 , x2 , y2 , z 2 ; ω1 , ω2 ) = f f 2 − ω12 − ω22 /c2 0 0 q 2 2 × S( f, φ1 , φ2 ; ω1 , ω2 ) exp i (z 2 − z 1 ) (ω1 /c) − f
Z
ZZ
× exp [−i f (x1 cos φ1 + y1 sin φ1 )] q 2 2 2 2 × exp i f − ω1 − ω2 /c (x2 cos φ2 + y2 sin φ2 ) d f dφ1 dφ2 . (5.65) The space-time properties of the field are governed by the (two-time) mutual correlation function ZZ ∞ 0(x1 , y1 , z 1 , x2 , y2 , z 2 ; t1 , t2 ) = W (x1 , y1 , z 2 , x2 , y2 , z 2 ; ω1 , ω2 ) 0
× exp [i (ω1 t1 − ω2 t2 )] dω1 dω2 .
(5.66)
Unfortunately, the expression obtained by inserting Equation (5.65) into Equation (5.66) is not very transparent. We point out, however, that Equation (5.61) is not the only possible definition of propagation-invariance of
Polychromatic and Pulsed Localized Waves
59
non-stationary fields; a less strict definitions that requires only the invariance of the spectral density leads to a larger class of solutions. In his original paper on focus wave modes, Brittingham (1983) considered localized solutions of Maxwell’s equations, i.e. vector fields. Vectorial solutions can be constructed from scalar ones in several ways, for example by assuming that two independent Cartesian components of the electromagnetic field are of the form of a scalar solution. The remaining components are then calculated from Maxwell’s equations in the same way that we did for monochromatic propagation-invariant fields. This approach is applicable to the general solution (5.65). Alternatively, electric and magnetic Hertz potentials of the form of scalar solutions can be used to construct TE and TM electromagnetic waves. This approach was used by Sezginer (1985) to generate electromagnetic focus wave modes and by Recami (1998) to construct vectorial X wave solutions.
5.6. Finite Approximations The transverse truncation of pulsed fields can be treated, in general, simply by considering each frequency component separately. Thus all the methods introduced in Sections 2.7 and 2.8 are readily available, provided that the resulting finite-energy monochromatic fields are summed with the weights given by the frequency spectrum. One version of the approach outlined above was applied by ZamboniRached (2006a) to incident ideal pulses, limited at z = 0 by a circular aperture of radius R. He assumed that each monochromatic component propagates like an ideal wave up to the geometrical range z = R/ tan θ(ω) (if t = 0), vanishing thereafter in the neighbourhood of the optical axis. Such an approximation proved reasonably accurate; considering several dispersion relations for θ (ω) and different frequency spectra, analytical expressions for axial intensity distributions were evaluated for a number of subluminal, luminal, and superluminal pulses including focus wave modes. The assumption of Gaussian instead of hard-edged apodization, and the dispersion relation (5.19), lead to the concept of Bessel–Gauss pulses. They were introduced by Overfelt (1991), who derived an analytical expression for their propagation and also showed, e.g. that the fundamental focus wave mode solution is a special case of these pulses. The angular spectrum of Bessel–Gauss pulses was given in closed form by Donnelly and Ziolkowski (1993). It was pointed out by Reivelt and Saari (2004) that, with a proper choice of the frequency spectrum (by decreasing the bandwidth), purely forward-propagating pulses with finite energy can be obtained, which are experimentally realizable, as discussed at the end of Section 5.4.3.
60
Propagation-Invariant Optical Fields
6. PROPAGATION THROUGH LENS SYSTEMS Lenses and lens systems are employed widely in the production of monochromatic propagation-invariant fields, and to transform their properties, such as the transverse scale, as we will see in Section 7. Chromatic effects in optical systems, which may also contain diffractive elements with strong dispersive characteristics, can be helpful in the production and tailoring of polychromatic propagation-invariant fields with the desired dispersion relation. We therefore begin this section by considering the propagation of initially propagation-invariant fields through paraxial optical systems described by the usual ABCD matrix (Kogelnik & Li, 1966), first addressed by Wolf (1988) in the coherent case, and by Friberg, Vasara, and Turunen (1991) in the partially coherent case. The discussion is then extended to fields of finite extent, and to the effects of chromatic variations in the propagation of polychromatic and pulsed fields.
6.1. Monochromatic Fields in Paraxial Systems The propagation of optical fields through an optical system characterized by its ABCD matrix is governed by the extended form of the Fresnel diffraction integral, presented by Collins (1970): Z Z ∞ ik D 2 ik exp (ik L) 2 U (x, y, z 0 ) = − exp x +y U (x 0 , y 0 , 0) 2π B 2B −∞ ik ik A 0 2 02 0 0 x +y exp − x x + yy dx 0 dy 0 , (6.1) × exp 2B B Rz where L = 0 0 n(z)dz, n(z) is the axial refractive-index profile of the optical system, A, B, C, and D are the elements of the 2 × 2 matrix obtained by multiplication of individual element matrices as explained in Kogelnik and Li (1966), and z = z 0 is the output plane of the system. It should be stressed that the expression (6.1) is based on the paraxial approximation and therefore only paraxial propagation-invariant fields in the form of Equation (2.18) can, strictly speaking, be treated. Inserting Equations (2.6) and (2.18) into Equation (6.1), performing the integrations with the aid of the formula Z
∞
2
exp −ax + bx dx = 0
r
π b2 exp a 4a
! ,
(6.2)
Propagation Through Lens Systems
61
and assuming that the refractive indices of the input and output spaces are equal (which implies that AD − BC = 1), we obtain U (x, y, z 0 ) =
x y 1 ikC 2 exp (i8) exp x + y2 V , , A 2A A A
(6.3)
where 8 = k L − α 2 /2k B/A and V (x, y) is given by Equation (2.6). Equation (6.3) provides the field at the output plane z = z 0 of a fixed optical system but it does not describe propagation behind that plane. To solve this propagation problem we consider a fixed ABCD system followed by propagation in free space over a distance 1z. The elements of the combined system are A0 = A + C1z, B 0 = B + D1z, C 0 = C, and D 0 = D. On substituting these into Equation (6.3), we find U (x, y, z 0 + 1z) =
ik exp [i8(1z)] exp x 2 + y2 m(1z) 2R(1z) Z 2π iα A(φ) exp × (x cos φ + y sin φ) dφ, (6.4) m(1z) 0
where 8(1z) = k (L + 1z) −
α 2 B + D1z 2k A + C1z
(6.5)
represents the on-axis phase delay up to the distance z = z 0 + 1z, m(1z) = A + C1z
(6.6)
can be interpreted as the transverse scale (magnification) of the field, and R(1z) = A/C + 1z
(6.7)
can be understood as the radius of wavefront curvature, as in the standard theory of Gaussian beams, or in the theory of Bessel–Gauss beams in Section 2.7. In view of Equation (6.6), paraxial optical systems produce, in general, shape-invariant fields with linearly expanding or contracting transverse scales. The linearity of the scale change distinguishes these fields from other shape-invariant fields, such as the usual Gaussian, Hermite–Gaussian, or Laguerre–Gaussian laser beams, or from Bessel–Gauss beams. In contrast with such beams, the evolution of R(z) in Equation (6.7) is also linear.
62
Propagation-Invariant Optical Fields
The passage of partially coherent propagation-invariant fields through paraxial optical systems can also be treated with the ABCD matrix formalism. Inserting the general expression (4.4) into the lens-system propagation formula for partially coherent light, Friberg, Vasara, and Turunen (1991) showed that W (x1 , y1 , z 0 , x2 , y2 , z 0 ) = Z
k
ZZ
× 0
2π
1 ikC 2 2 2 2 exp x − x + y − y 2 1 2 1 2A A2
f 2 S( f, φ1 , φ2 ) exp [i f (x2 cos φ2 − x1 cos φ1
0
+ y2 sin φ2 − y1 sin φ1 ) /A] d f dφ1 dφ2 .
(6.8)
Further propagation in free space across a distance 1z can be governed by replacing A and A/C in Equation (6.8) with m(1z) and R(1z), respectively. Let us next consider the transmission of propagation-invariant fields in some specific types of optical systems often used to form and modify propagation-invariant fields. 6.1.1. Afocal Systems Optical systems with matrix elements (A, B, C, D) = (A, B, 0, 1/A) are called afocal. Since C = 0, a strictly propagation-invariant field is obtained as the output: in view of Equations (6.6) and (6.7), R(1z) = ∞ for all 1z and m(1z) = A is constant. The on-axis phase takes the form 1 α 2 8(1z) = 8(z 0 ) + k − 1z, 2k A
(6.9)
where 8(z 0 ) = k/L − α 2 B/2k A and the factor in front of 1z is the paraxial approximation of the output propagation constant β of the field, modified by the system through scaling of α. An afocal system is useful in transforming, by a magnification factor m = A, the transversal scale and the cone angle of the propagation-invariant field. Thus it allows one, for example, to extend the propagation-invariant range of a finite-aperture field at the expense of expanding the transverse scale. 6.1.2. Imaging Systems Systems characterized by matrix elements (A, B, C, D) = (A, 0, C, 1/A) are known as imaging systems. They reproduce, at the output plane, the intensity distribution of any paraxial incident field. This is, naturally, the case also for propagation-invariant fields, but a quadratic phase term is generated unless the imaging system is also afocal (C = 0).
Propagation Through Lens Systems
63
6.1.3. Fourier-transforming Systems Systems with matrix elements (A, B, C, D) = (A, B, −1/B, 0) are collimators, while those with elements (A, B, C, D) = (0, B, −1/B, D) are focusing systems in paraxial ray optics: the former convert a point source into a bundle of parallel rays, while the latter operate in reverse. The element B is in both cases interpreted as the focal length F of the system. In the wave picture, both systems form an exact Fourier transform of the input field at z = 0 into the output plane z = z 0 , if A and D vanish. A system of this type was already considered in Figure 4 in connection with schematic generation of finite-energy approximations of propagation-invariant fields. The general solutions given by Equations (6.3) and (6.8) are not valid for exact Fourier-transforming systems. In this case, Equation (6.1) can be used directly to obtain U (ρ, ψ, z 0 ) = −
ik exp (ik L) A(ψ)δ(ρ − Fα/k). 2π F
(6.10)
Thus a delta-function ring of radius ρ = Fα/k is observed at the output plane, with an azimuthal complex amplitude proportional to A(ψ). Conversely, it follows from Equation (6.1) that, if a ring-like field with this radius illuminates a collimating system of focal length F, an ideal (paraxial) propagation-invariant field is obtained as proposed by Durnin, Miceli Jr., and Eberly (1987). Analogously, in the case of partially coherent incident fields of the form of Equation (4.4), one obtains a field with W (ρ1 , ψ1 , z 0 , ρ2 , ψ2 , z 0 ) = S(ρ1 , ψ1 , ψ2 )δ(ρ1 − ρ2 )
(6.11)
in the output plane of the focusing system. Finally, propagation-invariant fields of the form of Equation (4.4) can be generated by inserting a wave field with a cross-spectral density function (6.11) into a collimating system, as demonstrated by Turunen, Vasara, and Friberg (1991).
6.2. Finite-aperture Fields in Paraxial Systems Considering the focusing system in more detail, we note that Equations (6.4)–(6.7) with A = D = 0 are valid at any plane except for z = z 0 , where the delta ring is formed. Thus, at any plane other than this particular plane, the field has either a linearly contracting (z < z 0 ) or expanding (z > z 0 ) scale, with m(1z) → 0 when 1z → 0. Thus the field appears to ‘defy diffraction’ also in a focusing system, in the sense that its transverse scale can become arbitrarily small compared to the wavelength. Such ‘diffraction-free’ behaviour in focusing systems is, of course, again a consequence of our assumption of infinite fields, which is in
64
Propagation-Invariant Optical Fields
contradiction to the assumption of paraxial wave propagation in the system. As soon as any truncation in either the incident field or the aperture of the system is introduced (even if non-paraxial propagation were allowed), the dilemma disappears. The field close to the focal plane z = z 0 becomes an annulus with propagation-distance-dependent radial ¨ Huang, Zhang, Kong, and Zhai (1996), width and field distribution. Lu, Lu¨ and Huang (1996), and DeNicola (1996) have studied the focusing of such finite-aperture spatially coherent propagation-invariant fields by numerical methods. Such studies show, as expected from the qualitative considerations, a shallow increase in the axial intensity of a focused field towards the focal plane until a maximum is reached before the field begins to assume the annular focal-plane form. Focusing of partially spatially coherent Bessel-correlated fields, with a ring-like angular correlation function, was considered by Gbur and Visser (2003). They showed that an annulus (with a dark centre) is again formed in the focal plane, as expected because the finite aperture of the focusing lens only smears the Fourierring, irrespective of the intensity or coherence properties of the field. More insight into, e.g. the focusing problem can be achieved by considering the propagation of Bessel–Gauss beams in paraxial systems, since this allows one to derive closed-form expressions. The topic has been investigated thoroughly by Santarsiero (1996) and Palma (1997). One way to approach the problem is to insert the expression for a Bessel–Gauss field at z = 0 into Equation (6.1) and perform the integrations directly. Another method, which provides more physical insight, is to consider the Bessel–Gauss beam as a bundle of decentred Gaussian elementary beams, propagating in well-defined directions. In this way one can determine the propagation of the Bessel–Gauss beam in the system by tracing a meridional ray and a single, representative decentred Gaussian beam through the system. The result of both approaches is, of course, the same: an incident Bessel–Gauss beam remains a Bessel–Gauss beam in any paraxial system, but with transformed propagation parameters. Explicitly, one obtains " !# α2 αρ w0 exp [i8(z)] exp i k L − BD J0 U (ρ, z 0 ) = w(z) 2k A + iB/z R 1 ik × exp − − ρ 2 + z 2 α 2 /k 2 , (6.12) w 2 (z) 2R(z) where 8(z 0 ) = − arctan (B/z R ), the beam width is h i1/2 w(z 0 ) = w0 A2 + (B/z R )2 ,
(6.13)
Propagation Through Lens Systems
65
and the radius of wavefront curvature is R(z 0 ) =
A2 + (B/z R )2 . AC + B D/z R2
(6.14)
Clearly, these expressions reduce to Equations (2.37)–(2.40) if free-space propagation is considered, i.e. if we set A = D = 1, C = 0, and B = z.
6.3. Polychromatic Fields in Paraxial Systems The propagation of polychromatic propagation-invariant fields (including stationary fields and coherent pulses) in optical systems is best handled by first computing the output of the system for each frequency component separately and then combining the results. For stationary fields, the frequency components are added incoherently with the appropriate spectral weighting, while for coherent pulses the superposition is coherent, as explained in Section 5.1. Optical systems are typically more or less chromatic, which has to be taken into account when dealing with wideband fields. Let us assume, for simplicity, that the system consists of sections of free space (refractive index equal to unity) separated by thin lenses. The frequency-dependent contributions to the system matrix with elements A(ω), B(ω), C(ω), and D(ω) then arise from the lenses only. The matrix of a thin lens is of the form
A(ω) B(ω) 1 0 = , C(ω) D(ω) −1/ f (ω) 1
(6.15)
where f (ω) is the (frequency-dependent) focal length of the thin lens. For a refractive (singlet) lens f (ω) =
n(ω) ¯ −1 f (ω) ¯ n(ω) − 1
(6.16)
ω f (ω), ¯ ω¯
(6.17)
and for a diffractive lens f (ω) =
where ω¯ is some reference frequency, such as the centre frequency of the spectrum G(ω), and n(ω) is the frequency-dependent refractive index of the lens material. Equations (6.4)–(6.7) show that, to obtain a propagation-invariant output field with a polychromatic propagation-invariant input field, one
66
Propagation-Invariant Optical Fields
must design a system with C(ω) = 0 over the entire frequency range of G(ω). Moreover, to preserve the original dispersion relation, A(ω) must be independent of ω. If the former condition is satisfied only, e.g. at the central frequency ω, ¯ other frequency components will change scale upon propagation. If C(ω) = 0 but the condition A(ω) = constant is not satisfied, the new dispersion relation (in paraxial form) is, from Equation (6.9), β(ω) ≈
ω c α 2 (ω) − . c 2ω A2 (ω)
(6.18)
Thus, in principle, controlled transformations of the dispersion relation are possible with the aid of chromatic effects in afocal lens systems, but no such designs have yet been published.
7. GENERATION METHODS Since all propagation-invariant fields possess infinite energy, it is clear that only their finite approximations can be generated, in the same sense as finite approximations of plane waves are produced. The generation of approximately propagation-invariant fields of the types discussed in Sections 2.7 and 2.8 requires, however, somewhat more complicated schemes than the generation of plane waves. Basically, two approaches are possible: one can either modulate a plane wave (or, e.g. a Gaussian beam) directly, or one can start from the angular spectrum using a setup such as that in Figure 4. Unfortunately the most obvious schemes suffer from low light efficiency, which excludes their use in most of the proposed applications of propagation-invariant fields. Therefore highefficiency methods are often preferable, even if they have limitations such as a non-uniform longitudinal intensity profile over the propagationinvariant range. An early review of generation methods of Bessel fields was presented by Lapointe (1992).
7.1. Spatial Wavefront Modulation Perhaps the most obvious technique to generate a Bessel field, proposed by Durnin and Eberly (1989), is to modulate a plane wave by a filter with complex-amplitude transmittance t (ρ, ψ) = |J0 (αρ)| exp [iφ(ρ)] ,
(7.1)
where φ(ρ) = arg [J0 (αρ)] is a binary function with values zero or π , depending on the sign of J0 (αρ). Such a filter could be constructed, for
Generation Methods TL
DE
67
PL
z F'
F
F
L'
FIGURE 11 Generation of Bessel fields by spatial-frequency modulation: TL is a toroidal lens of focal length F 0 , the diffractive element DE is a complex spatial filter placed in the focal plane of TL, and PL is a positive lens of focal length F
example, by the double-layer photographic method of Chu, Fienup, and Goodman (1973). The problem of low light efficiency associated with this approach arises from the asymptotic 1/r envelope of the Bessel field, which requires considerable amplitude suppression. Basically the same limitation applies to the holographic production scheme proposed by Cox and Dibble (1991). On the other hand, to produce the correct transverse profile over a significant longitudinal range, it is essential to reproduce the phase information in Equation (7.1); neglecting some (or all) of the amplitude information is far less critical. This aspect also arises in connection with the diffractive generation techniques to be discussed in Section 7.3.
7.2. Spatial-frequency Selection and Modulation The method originally proposed by Durnin, Miceli Jr., and Eberly (1987) for generation of J0 Bessel fields employs a ring aperture in the front focal ˜ and Smith (1989) suggested plane of a positive lens, as in Figure 4. Canas a solution to the problem of low light efficiency by forming the ring with a toroidal lens instead of a narrow aperture, as illustrated in Figure 11. The lens TL can be realized as a refractive or a diffractive element, e.g. by diamond turning or microlithography. Assuming that F > F 0 in Figure 11, and that the aperture of PL is large enough not to truncate the field, simple ray optics give the estimate L = F + L 0 = F + F 2 /F 0
(7.2)
for the propagation-invariant range behind PL. The role of DE in Figure 11 is to modulate the angular spectrum so as to produce the azimuthal complex-amplitude transmittance A(φ) in Equation (2.17); with no DE, the J0 field mode is generated. This setup of Figure 11 was used by Lautanen, Kettunen, Laakkonen, and Turunen (2000) to demonstrate a pair of parallel-propagating Bessel fields, one with a J02 and the other with J12 intensity distribution.
Propagation-Invariant Optical Fields
68
R n
z
L
FIGURE 12 Transformation of an incident plane wave into a conical wave by an axicon with apex angle
The techniques of Figures 4 and 11 also permit the generation of electromagnetic propagation-invariant fields if the field of the ring is polarized appropriately. For example, radially and azimuthally polarized fields can be produced using space-variant diffractive polarization components. Partially coherent fields can be realized as well: the general solution (4.4) contains, for example, a class of fields with incoherent angular spectra, which can be generated by replacing DE with a rotating diffuser, as demonstrated by Turunen, Vasara, and Friberg (1991). Such diffusers were used by Bouchal and Courtial (2004) to generate arrays of propagation-invariant fields with variable coherence properties.
7.3. Axicons and Diffractive Elements The axicon, a refractive element with a rotationally symmetric conical surface, was introduced by McLeod (1954, 1960). Figure 12 shows that each incident ray is refracted towards the optical axis at an angle θ = arcsin (n sin ) − ,
(7.3)
where n is the refractive index of the axicon, thus generating a conical wave over the range L = R tan θ . An approximation of the fundamental J0 Bessel field is thus produced, as pointed out by Leith, Collins, Khoo, and Wynn (1980), Roy and Tremblay (1980), and Indebetouw (1989). Fabrication of conical surfaces (see Ferguson, Gutheinz, Mayo III, German, Lowrey, and Konopnicki (1982) and Bickel, H¨ausler, and Maul (1985)) of sufficiently high optical quality to reveal the Bessel pattern was difficult before the emergence of diamond-turning technology; some of the first demonstrations of high-quality axicon-generated fundamental Bessel fields were provided by Scott and McArdle (1992), while higherorder Bessel fields were produced by Arlt and Dholakia (2000) using illumination with Laguerre–Gauss modes. An interesting alternative to diamond turning is the polishing technique of Grosjean, Saleh,
Generation Methods
69
Suarez, Ibrahim, Piquerey, Charraut, and Sandoz (2007) for generating microaxicons on fibre tips. A diffractive axicon is a radially periodic structure formed from the refractive axicon by the zone construction principle of Miyamoto (1961). To generate a conical wave with cone angle θ , a radial grating with period d = λ sin θ is required. The local grating profile determines the diffraction efficiency into the desired conical wave, which is the first (generalized) diffraction order of the radial grating. In general, several spurious convergent and divergent conical waves with different efficiencies are generated as described by Dammann (1969) and Goodman and Silvestri (1970). The calculation of the efficiencies of various diffraction orders can be performed by treating the element locally as a linear grating and applying standard grating theory (see, e.g. Petit (1980) and Turunen (1997, chap. 2)). Then the field behind the diffractive axicon can be determined, for example, as described by Vahimaa, Kettunen, Kuittinen, Turunen, and Friberg (1997). Bessel fields were indeed demonstrated with diffractive optics (Turunen & Wyrowski, 1997, chap. 1) soon after the work of Durnin ˜ and Smith (1989). (1987) by Turunen, Vasara, and Friberg (1988) and Canas ˜ and Smith (1989) used an on-axis setup based on a radial grating Canas with triangular profile, while Turunen, Vasara, and Friberg (1988) used a binary element in an off-axis setup to separate the spurious orders from the desired one. Such a setup also facilitated first demonstrations of higher-order Bessel field modes by Vasara, Turunen, and Friberg (1989b). Other early examples of diffractive generation of higher-order Bessel beams and their superpositions have been provided by Khonina, Kotlyar, Soifer, Shinkaryev, and Uspleniev (1992), Paterson and Smith (1996a), and Kotlyar, Khonina, and Soifer (1995). A characteristic property of fields generated by uniformly illuminated refractive or diffractive axicons is a linear (average) increase of the axial intensity upon propagation. This can be explained by considering Figure 12: light from a ring of (narrow) width 1ρ at a distance ρ from the optical axis crosses the z-axis at a distance z = ρ/ tan θ . Since the area of the ring is 2πρ dρ and the light crosses the axis over a range 1z = 1ρ/ tan θ , a linear increase of intensity up to the distance L = R/ tan θ takes place with uniform illumination. In general, if the incident radial intensity distribution is Iradial (ρ), the resulting axial intensity distribution Iaxial (z) ≈ z tan θ Iradial (z tan θ ). As illustrated in Figure 3, an abrupt truncation of the incident wave field leads to rapid fluctuations, which in the case of axicons perturb the uniformly intensifying longitudinal wave profile. A considerable amount of research has been devoted to the elimination of these defects. In particular, Sochacki, Bar´a, Jaroszewicz, and
70
Propagation-Invariant Optical Fields
Kołodziejczyk (1992), Sochacki, Kołodziejczyk, Jaroszewicz, and Bar´a ´ (1992), Staronski, Sochacki, Jaroszewicz, and Kołodziejczyk (1992), and ´ Sochacki, Jaroszewicz, Staronski, and Kołodziejczyk (1993) have studied the generation of uniform-intensity axial line sections over a range z 1 < z < z 2 using annular-aperture axicons with nonlinear radial phase-delay profiles. Either refractive non-conical surfaces or diffractive elements with radially modulated local period d(ρ) can be used to realize such phase transformations. In both cases the map-transform method introduced by Kurtz, Hoadley, and dePalma (1973) and Bryngdahl (1974) is available to the design of the local phase function. Obviously, any (intentionally introduced) perturbation of the output wave from the ideal conical shape leads to longitudinal variations of the transverse scale of the Bessel field, i.e. α becomes more or less dependent on z. The rapid longitudinal intensity fluctuations can be damped by apodization of the incident field as discussed in Section 2.7 or, equivalently, by the aperture as described by Jaroszewicz, Sochacki, ´ Kołodziejczyk, and Staronski (1993). For refractive axicons a gradedtransmission aperture is needed, but in the case of diffractive axicons the apodization can be achieved simply by modulating the local diffraction efficiency as a function of ρ. The best apodization scheme depends on the type of the local grating profile: Popov, Friberg, Honkanen, Lautanen, Turunen, and Schnabel (1998) considered the schemes available for binary surface-relief profiles. Another method to reduce the longitudinal intensity fluctuations is the use of partially coherent illumination, proposed by Popov and Friberg (1995). This also modifies the transverse intensity profile of the generated field, typically smoothing out the radial intensity distribution to some degree, as shown by Thaning and Friberg (2002). A combination of a collimation lens with the axicon facilitates the conversion of a field radiated by an axial point source into an conical wave. This was first demonstrated by Soroko (1989), while MacDonald, Chrostowski, Boothroyd, and Syrett (1993) employed a combination of a binary reflection-type diffractive lens and an axicon to transform a diodelaser beam into an approximate Bessel field.
7.4. Tandem Optical Systems The use of two optical elements in succession, i.e. a tandem system, offers significant new degrees of freedom for beam shaping, particularly in diffractive optics (see Bartelt (1984, 1985), and Turunen and Wyrowski (1997, chap. 1)). In the case for generation of conical waves, the first element may be used to tailor the field distribution incident on the second element, which performs the final transformation into a conical wave. In
Generation Methods
71
this way the compromises involved with the use of only a single element (such as uneven longitudinal intensity profile) can largely be avoided. Bickel, H¨ausler, and Maul (1985) introduced a telescopic combination of a negative and a positive refractive axicon to project a line focus over a desired longitudinal interval. Davidson, Friesem, and Hasman (1992) presented a tandem diffractive system, in which the first element produces a 1/ρ intensity distribution at the plane of the second element, which corrects the phase distortion and adds the linear axicon phase. Honkanen ¨ and Turunen (1998) and P¨aa¨ kkonen, Simonen, Honkanen, and Turunen (2002) designed and demonstrated a tandem system in which the first element produces an apodized annular illumination wave, allowing the second element to realize a line section with a nearly uniform longitudinal intensity profile. We note that there has been some controversy in the literature relating to tandem systems for generation of Bessel fields. Thewes, Karim, and Awwal (1991) and Iftekharuddin, Awwal, and Karim (1993) proposed refractive tandem systems for Gaussian-to-Bessel transformation, but Lin, Seka, Eberly, Huang, and Brown (1992) criticized this method as not being capable of producing a Bessel field. Their point is that the system does not generate the π phase shifts that occur at the zeroes of the Bessel profile. Arif, Hossain, Awwal, and Islam (1998) proposed to overcome this problem by generating only the central lobe of the Bessel field, but the propagation characteristics of such a field are closer to those of a Gaussian beam than those of a Bessel field. Finally, we note that it is possible to generate almost arbitrary electromagnetic propagation-invariant fields by combining the conical-wave generation capability of axicons with the polarization-modulation capabilities of so-called polarization gratings introduced by Gori (1999) and investigated further by Tervo and Turunen (2000, 2001). Unfortunately such elements are difficult to fabricate for visible wavelengths.
7.5. Optical Resonators Thus far we have considered passive optical elements, which transform a fixed optical beam into an approximately propagation-invariant field. One may proceed by asking if active sources of propagation-invariant fields exist. The most natural approach is to investigate whether laser resonators with Bessel modes can be realized. Friberg and Turunen (1994) showed that propagation-invariant fields are the exact eigenmodes of an infinite-aperture (passive) Fabry–Perot resonator in the sense of the resonator theory of Wolf and Agarwal (1984). They also established that the general expression for the modes of such a resonator is a superposition of self-imaging optical fields. Therefore finite-aperture approximations of self-imaging fields can be
72
Propagation-Invariant Optical Fields
generated by illuminating a Fabry–Perot resonator with a laser beam, and that finite-aperture approximations of Bessel field modes may be generated by selecting one of the Montgomery’s rings, as described by Indebetouw (1989), and examined in more detail by Cox and Dibble (1992). Unfortunately the Gaussian-to-Bessel conversion efficiency in a Fabry–Perot resonator is low. Several laser-resonator configurations for the generation of Bessel modes were described by Durnin and Eberly (1989). The fact that the Fourier transform of such a mode is confined on a ring suggests configurations, in which one of the resonator mirrors possesses a thin annular aperture; a Fourier-transform relationship between the field at this mirror and the field transmitted by the opposite mirror is arranged with either intracavity or external-cavity lenses. This scheme was demonstrated by Uehara and Kikuchi (1989), who obtained a good approximation of a Bessel–Gauss field, but only over a rather limited propagation range. Lin, Seka, Eberly, Huang, and Brown (1992) showed that an unstable laser resonator produces an annular output field, which can be Fourier-transformed with a positive lens to produce an approximately propagation-invariant field over a considerable distance. Jabczynski (1990) introduced a scheme, in which the active medium takes the form of an annular cylinder placed in the middle of a symmetric annular-aperture confocal resonator. However, the propagation-invariant range is only a fraction of the resonator length and the scheme does not appear easily realizable because of difficulties in optical pumping of the annular-shaped active medium. Aspheric-mirror resonators (B´elanger & Par´e, 1991) are capable of producing laser beams with prescribed output intensity distributions (Leger, 1997, chap. 7; Leger, Chen, & Wang, 1994). Such resonators are also suitable for the production of approximations of Bessel fields. ¨ P¨aa¨ kkonen and Turunen (1998) derived analytical expressions for the mirror shapes of a resonator with a fundamental Bessel–Gauss mode: the output mirror, located at the waist of the Bessel–Gauss field, is planar and the second (ideally perfectly reflecting) aspheric mirror conjugates the phase of the field incident on it, thus producing the desired field mode. ¨ Such a resonator was demonstrated by Hakola, Buchter, Kajava, Elfstrom, ¨ Simonen, P¨aa¨ kkonen, and Turunen (2004) and analyzed further by Ling and Li (2006). Several configurations based on axicons as resonator mirrors and intracavity elements have been presented by Rogel-Salazar, New, and Ch´avez-Cerda (2001), Muys and Vandamme (2002), Tsangaris, New, and Rogel-Salazar (2003), Guti´errez-Vega, Rodr´ıgues-Masegosa, and Ch´avez-Cerda (2003), Hern´andez-Aranda, Ch´avez-Cerda, and Guti´errezVega (2003), Wu, Chen, and Guo (2007), and Hern´andez-Aranda and
Generation Methods
73
Guti´errez-Vega (2007). One of the configurations was demonstrated experimentally by Khilo, Katranji, and Ryzhevich (2001). Finally, Erdogan, King, Wicks, Hall, Anderson, and Rooks (1992) have demonstrated approximations of azimuthally polarized Bessel–Gauss fields, produced in optically pumped vertical-cavity laser resonators with circular-grating output mirrors.
7.6. Reconfigurable Devices Sometimes it is of interest to alter the functional form or the transverse scale of the propagation-invariant field electronically in real time. Davis, Guertin, and Cottrell (1993) were the first to demonstrate this by writing diffractive axicons in spatial light modulators (SLMs). Magneto-optic SLMs were used, in which binary phase elements can be written in real time. Davis, Carcole, and Cottrell (1996a) demonstrated the generation of higher-order Bessel fields by this method, while Davis, Carcole, and Cottrell (1996b) produced linear combinations of Jm and J−m fields, which appear much like circular spot arrays. Chattrapiban, Rogers, Cofeld, Hill III, and Roy (2003) generated higher-order Bessel modes using nematic liquid-crystal SLMs with nearly continuous phase-modulation capability, apparently without being aware of previous work on the topic. Fatemi and Bashkansky (2006) did essentially the same with ferroelectric SLMs, which are faster but capable of generating only binary phase profiles and thus having lower efficiency. Spatial light modulators also allow the introduction and modulation of a carrier frequency in the diffractive element, thus enabling real-time pointing and scanning with a propagation-invariant field. Because of the relatively large pixel size of currently available SLMs, real-time control of propagation-invariant fields with acceptable signal-to-noise ratio is possible only for small values of the cone angle θ. Afocal optical systems can be used to scale down the field and also increase the angular scanning range. The available propagation-invariant range (in comparison with the transverse scale) depends on the number of pixels, which is orders of magnitude smaller in SLMs than in passive diffractive devices generated by lithographic methods.
7.7. Generation of Polychromatic and Pulsed Fields All methods described in the previous subsections are, in principle, applicable also for the generation of polychromatic and pulsed propagation-invariant fields, but some are more practical than others. For example, many laser resonators considered above only have gain over a narrow frequency range, and are therefore ruled out in generation of broadband fields. We note that the generation of such fields has not yet been thoroughly investigated and therefore no comprehensive analysis
74
Propagation-Invariant Optical Fields
of all techniques can be given here. Nevertheless, the key question is the realization of the desired dispersion relation, which we proceed to discuss. Let us consider first the original setup of Durnin, Miceli Jr., and Eberly (1987), involving a narrow ring aperture in the front focal plane of a perfectly achromatic lens. If the aperture is illuminated by a polychromatic field, the system generates a field with a fixed cone angle, i.e. the dispersion relations (5.15) and (5.16). Thus X waves are generated with pulsed illumination. The realization of any other dispersion relation in this setup requires a radially frequency-selective aperture in place of the simple ring aperture, which does not seem easy to produce; even if successful, this method would suffer from low light efficiency. Axicon-type elements provide more versatility in the generation of polychromatic fields of different forms, since refractive, diffractive, and hybrid axicons with different chromatic properties can be employed. In the case of a refractive axicon with apex angle , the cone angle of the field is [c.f., Equation (7.3)] θ (ω) = arcsin [n(ω) sin ] − ≈ [n(ω) − 1] ,
(7.4)
where the latter form is applicable in the paraxial approximation. For typical optical materials in the visible region, n(ω) is a smoothly increasing function of frequency. A coloured transverse profile then results, with the longest wavelengths (lowest frequencies) in the spectrum having the largest transverse scale factor α(ω), as observed in the experiments of ´ Fischer, Brown, Morris, Lopez-Mariscal, Wright, Sibbett, and Dholakia ´ (2005) and Fischer, Little, Smith, Lopez-Mariscal, Brown, Sibbett, and Dholakia (2006). Thus, as the field propagates, the shortest wavelengths in the spectrum reach the end of their propagation-invariant range first, and thereafter the field becomes increasingly redshifted. For diffractive axicons, the cone angle θ (ω) is determined by the grating equation, so that in the mth diffraction order at zero angle of incidence 2π c 2π c θ (ω) = arcsin m ≈m , dω dω
(7.5)
where d is the radial grating period, and the latter form is again paraxial. Thus an achromatic field with α(ω) = 2π/d = constant is produced. It should be noted, however, that the propagation-invariant range now depends critically on ω. Thus, contrary to refractive axicons, the longest wavelengths reach the end of the propagation-invariant range first and, as soon as this happens, the field begins to get blueshifted. The grism, illustrated in Figure 13a, is an element made of a refractive and a diffractive surface. It provides the greatest degree of flexibility in
Generation Methods
(a)
75
(b)
i
i
f
FIGURE 13 (a) Dispersion by a grism. (b) Use of a hybrid axicon (circular grism) in conjunction with Durnin’s setup to generate localized waves
control of the dispersion relation. With angle of incidence θi , the deflection angle θ (ω) is given by a sequential use of Snell’s law and the grating equation: sin(θi + ) 2π c θ (ω) = arcsin n(ω) sin arcsin − +m n(ω) dω 2π c ≈ θi − [n(ω) − 1] + m . dω
(7.6)
We now have the structural parameters and d, as well as the dispersion curve n(ω) provided by the choice of the glass material, free for fitting in the desired dispersion curve. Reivelt and Saari (2000) and Reivelt and Saari (2002a) proposed the system depicted in Figure 13b, based on the use of a circular grism (a hybrid refractive-diffractive axicon) for generation of pulsed localized waves. They showed numerically that, by proper choice of the free parameters, good approximations of focus wave modes can be obtained. With different choices, a range of subluminal and superluminal fields can in fact be produced. Unfortunately, no experimental demonstration has been given at the time of writing. Valtna, Reivelt, and Saari (2007) have recently put forward exact methods for generating dispersion relations of superluminal fields using different combinations of conical diffraction gratings and mirrors. The propagation-invariant ranges of the superluminal fields are only of the order of the diameter of the system aperture, but can be expanded using afocal, achromatic lens systems as discussed in Section 6. Simulations and experimental results on production of quasi-stationary and pulsed polychromatic fields have been provided by several groups. ˜ Sonajalg and Saari (1996) discussed the use of refractive axicons to compensate for the dispersion of X waves in air, and the use of a diffractive element with a combined function of a lens and an axicon to suppress
76
Propagation-Invariant Optical Fields
the temporal spread of X waves in more strongly dispersive media. ˜ The latter scheme was demonstrated experimentally by Sonajalg, R¨atsep, and Saari (1997). Saari and Reivelt (1997) demonstrated X waves using Durnin’s original setup with an added interfering plane wave to reveal the X shape. Femtosecond pulses were generated using microaxicon arrays by Grunwald, Griebner, Tschirschwitz, Nibbering, Elsaesser, Kebbel, ¨ ´ Hartman, and Juptner (2000). Fischer, Brown, Morris, Lopez-Mariscal, Wright, Sibbett, and Dholakia (2005) used axicons with Ti:Sapphire lasers, superluminescent diodes, supercontinuum sources, and halogen bulbs to demonstrate pulsed and continuous-wave fields with either almost full or limited spatial and temporal coherence. Grunwald, Bock, Kebbel, Huferath, Neumann, Steinmeyer, Stibenz, N´eron, and Piche (2008) produced ultrashort (and ultra-broadband) Bessel–Gauss pulses using refractive axicons.
8. APPLICATIONS The unique properties of propagation-invariant fields immediately suggest a wide range of applications, some of which were already recognized by Durnin and Eberly (1989). Unfortunately, closer scrutiny has revealed that initial enthusiasm in some of these applications was premature, in the sense that more traditional methods provide at least comparable performance. One example is power transport: Durnin, Miceli Jr., and Eberly (1988) compared the fundamental Bessel mode and a Gaussian beam, both truncated by an aperture of the same size. It turned out that in optimized conditions an approximately equal amount of power can be transported by both methods through an aperture at a fixed distance. Another apparently interesting field of applications is laser machining (Rioux, Trenblay, & B´elanger, 1978), it is tempting to think that the central peak of the J0 field could be used to drill deep, narrow holes in materials. However, since the Bessel field is an interference pattern, the central peak alone is not propagation-invariant and thus begins to spread as soon as the hole gains any appreciable depth. As a result, the drilling characteristics are comparable to the use of a Gaussian beam of the same diameter as the central lobe of the Bessel field. Fortunately, in many applications the advantages of propagationinvariant fields are more obvious than in those relying on power transport. We proceed to briefly discuss some such application areas. No exhaustive coverage is attempted, partly because an excellent review of applications has recently been published by McGloin and Dholakia (2005), and partly because of the continuing progress in this area and the emergence of new applications.
Applications
77
8.1. Precision Instrumentation and Metrology The most obvious applications of Bessel fields, suggested by their high directionality and rapid transverse intensity variation, may be found in optical alignment, pointing, and tracking. Either the fundamental J0 field mode or the J1 mode may be used, of which the latter provides somewhat higher beam-centre definition as discussed earlier. Many of these applications are essentially the same as those of axicons, already presented by McLeod (1954, 1960) and Ferguson, Gutheinz, Mayo III, German, Lowrey, and Konopnicki (1982), before the realization that axicons of sufficiently high quality produce approximations of Bessel field modes. Some such applications were pointed out by Herman and Wiggins (1991) explicitly in connection with Bessel fields. The use of conical waves produced by axicons in triangulation was considered by Bickel, H¨ausler, and Maul (1985). More recently Davis, Carcole, and Cottrell (1996a) proposed the use of two intersecting Bessel fields in triangulation-based range funding. Propagation-invariant fields have been proposed for, e.g. the alignment of integrated optical elements (MacDonald, Chrostowski, Boothroyd, & Syrett, 1993). In alignment applications Bessel fields feature the interesting property that the central spot is self-regenerated shortly behind an obstacle placed on the axis, as noted by DeBeer, Hartmann, and Friedberg (1987), Herman and Wiggins (1991), and Bouchal, Wagner, and Chlup (1998). This is a direct consequence of the conical-wave nature of the field: a spherical obstacle casts a conical shadow but the Bessel beam re-emerges after the shadow, allowing the alignment of several obstacles in a row. Bouchal (2002) demonstrated that the self-reconstruction property is maintained also if phase objects are placed in the propagation path. MacDonald, Boothroyd, Okamoto, Chrostowski, and Syrett (1996) proposed to employ this property, together with the long focal line available with Bessel fields, in realization of free-space optical interconnections to maintain synchronization in multiboard computer systems. This is just one example of applications facilitated by the fact that the pointing accuracy of a Bessel field is retained even if a major part of the interference pattern is obstructed by a three-dimensional distribution of obstructions. Another obvious application of propagation-invariant fields is related to the long focal depths of their finite-aperture approximations. Often, especially in exposure-type applications, the problem is that the sidelobes of the Bessel field reduce contrast. To avoid this problem, high-contrast materials (resists) can be used, which are exposed only if a certain intensity level is exceeded. Nevertheless, it is worthwhile to seek an optimum trade-off between the required propagation-invariant range and the sidelobe level by considering, e.g. Bessel–Gauss fields and other apodized propagation-invariant fields. Some techniques for this purpose
78
Propagation-Invariant Optical Fields
have been proposed by Dong, Yang, Gu, and Ersoy (1996) and by Provost and de Bougrenet de la Tocnaye (1997). Propagation-invariant fields have also found applications in various scanning systems. H¨ausler and Heckel (1988) demonstrated that a propagation-invariant field in combination with a fast scanning mirror can generate a sharp horizontal line with a large depth of focus, which appears to be ideal for surface profilometry using the concept of light sectioning. Use of axicons in a laser-scanning fluorescence microscope is discussed by Arimoto, Saloma, Tanaka, and Kawata (1992). They consider the aberrations arising at non-normal illumination and methods of correcting them. The question of aberrations in propagation-invariant fields is indeed important, but relatively little attention has been devoted to it thus far: some numerical examples are given for diffractive axicons by Vasara, Turunen, and Friberg (1989a) and aberrations in the angular spectrum have been evaluated by Bouchal (1993).
8.2. Manipulation of Atoms and Particles Guiding, collimation, focusing, trapping and cooling of neutral atoms is possible through the dipole force exerted on the atom by near-resonant light. Detuning above resonance forces atoms away from bright regions, while detuning below resonance attracts atoms towards regions of higher intensity. Making use of these effects, Balykin, Letokhov, Minogin, Rozhdetsvensky, and Sidorov (1985) applied reflaxicon fields of the type described by Equations (2.19) and (2.20) to cool atoms, and Florjanczyk and Tremblay (1989) proposed collimation and guiding of atoms within the central lobe of a fundamental J0 Bessel field, while Arlt, Dholakia, Soneson, and Wright (2001) analyze the use of J0 fields in trapping of atoms and J1 fields in guiding them over long distances. Laser acceleration of charged particles (electrons) with the aid of electromagnetic Bessel fields can be based on the inverse free electron ˇ laser (Cai, Bhattacharjee & Marshall, 1988) or inverse Cerenkov radiation focused with an axicon (Fontana & Pantell, 1983). The latter method has been analyzed by Romea and Kimura (1990). Kimura, Kim, Romea, Steinhauer, Pogorelsky, Kusche, Fernow, Wang, and Liu (1995) have demonstrated acceleration rates up to 31 MeV/m with this approach. Scully and Zubairy (1991) consider linearly polarized light (in paraxial approximation) and point out that the z component of the electric field of the fundamental Bessel field is zero on the optical axis. This can, in fact, be seen immediately from the exact (non-paraxial) result, Equation (3.17). To overcome the problem they propose the use of the J1 type linearly polarized field, for which e1z (ρ, ψ) = i
α J0 (αρ) − J2 (αρ) exp (i2ψ) , 2β
(8.1)
Applications
79
and on axis e1z (0, ψ) = iα/2β. These exact results should be compared to eq. (26) of Scully and Zubairy (1991). The rotational nonsymmetry of the field in Equation (8.1) may be a disadvantage. However, it is possible to construct a symmetric circularly polarized field with α e1z (ρ, ψ) = i J0 (αρ). β
(8.2)
Such fields, generated by axicons, have been employed in the accelerator experiments by Fontana and Pantell (1983), and their properties in practical systems have been investigated by Romea and Kimura (1990) and Tidwell, Ford, and Kimura (1992). Particles of macroscopic size (e.g. 1–50 µm) may be manipulated by means of radiation pressure. He, Heckenberg, and RubinsteinDunlop (1995) succeeded in trapping reflective and absorptive particles (aluminium and high Tc superconducting powder, respectively) in the central dark spot of a higher-order Bessel field. Interestingly, in this application it is of advantage to employ Bessel fields of order higher than m = 1 to enlarge the diameter of the central dark region of the doughnut distribution. Because of their helical wave fronts, higher-order Bessel beams exhibit orbital angular momentum, which can be transferred to trapped particles, as illustrated by Arlt, Garces-Chavez, Sibbett, and Dholakia (2001), Garc´ez-Chavez, McGloin, Melville, Sibbett, and Dholakia (2002), Garc´es-Chaves, Volke-Sepulveda, Ch´avez-Cerda, Sibbett, and Dholakia (2002), McGloin, Garcez-Chavez, and Dholakia (2003), Khonina, Kotlyar, Skidanov, Soifer, Jefimovs, Simonen, and Turunen ˇ zm´ar, Siler ˇ (2004), Cisˇ and Zem´anek (2006), and others. The transfer of orbital angular momentum results in a rotation of the particle, which is trapped in the dark region if its refractive index is low and in the bright regions if the index is high. Particles can also be guided in the twodimensional trap formed by the propagation-invariant field, and the selfreconstruction effect can be used to trap several particles simultaneously, even if they are separated longitudinally by distances very large compared to the transverse scale of the field. Finally, Garc´es-Chaves, Roskey, Summers, Melville, McGloin, Wright, and Dholakia (2004) demonstrated levitation of a chain of particles in the main lobe of a vertical J0 beam, based on balancing the radiation pressure and the self-healing effect.
8.3. Nonlinear Optics The work on the use of Bessel fields in nonlinear optics was initiated by Wulle and Herminghaus (1983), who demonstrated phase-matched second-harmonic generation in a KDP crystal with a Q-swithed Nd:YAG laser. The key observation here is that, for a Bessel field, the effective
80
Propagation-Invariant Optical Fields
interaction wavelength λeff can be controlled by varying (with a zoom lens) the cone angle θ and thereby the propagation constant β: λeff = 2π/β = λ/ cos θ . This work raised high hopes of the possibility of enhancing the efficiency of nonlinear optical processes by using Bessel fields instead of traditional Gaussian beams, and indeed highharmonic generation is possible with ultrashort-pulsed Bessel fields (Nisoli, Priori, Sansone, Stagira, Cerullo, deSilvestri, Altucci, Bruzzese, de Lisio, Villoresi, Poletto, Pascolini, & Tondello, 2002). The work on harmonic generation is nicely summarized, with appropriate references, by McGloin and Dholakia (2005). Unfortunately the conclusion is that little or no improvement can be achieved. The reason is basically the same as in power transport applications: what can be gained by the long propagation distance of the non-spreading central spot is lost because of the presence of the ring pattern that carries most of the power. Very recently, exciting developments have taken place in the applications of Bessel beams in nonlinear optics, which do not rely on power transport but on the shape of the Bessel-field interference pattern. In a medium with cubic nonlinearity, the Bessel-field pattern generates a refractive-index profile capable of confining spatial optical solitons; the propagation-invariant range of this pattern can easily extend over the whole length of the crystal. The peak value of the refractive-index modulation depends on the strength of the Bessel field, which can be used to control the confinement and coupling of the solitons in the radially quasi-periodic lattice. Kartashov, Vysloukh, and Torner (2004) were the first to realize that the use of ‘Bessel lattices’ with cylindrical symmetry, instead of square or honeycomb lattices, has a profound effect in the soliton behaviour. The existence of lowest-order solitons in the centre of the Bessel-beam-induced refractive-index lattice, higher-order solitons confined to the side lobes of the lattice, as well as spiraling solitons in lattices produced by interfering Bessel fields, have been predicted. In follow-up work, a rich variety of solitons and their interactions was discovered: see Kartashov, Vysloukh, and Torner (2005a,b), Mihailache, Mazilu, Lederer, Malomed, Kartashov, Crasovan, and Torner (2005), Wang, Chen, and Kevrekidis (2006), Dong and Wang (2007), and Dong, Wang, Zhou, Yang, Lv, and Chen (2008). Experimental observations in photorefractive crystals were reported by Fischer, Neshev, Lopez-Aguayo, Desyatnikov, Sukhorukov, Krolikowski, and Kivshar (2006).
9. CONCLUSIONS In this review the basic theory of propagation-invariant fields has been covered, considering monochromatic scalar and electromagnetic fields as well as spatially partially coherent, polychromatic and pulsed fields.
References
81
Different finite-energy approximations of these fields and experimental methods for their generation has also been discussed, and a brief overview of selected applications was given. Even though new theoretical papers on the phenomenon of propagation invariance continue to emerge, the theory appears rather complete, at least in the scalar domain. The situation is different in electromagnetic optics: no general expression has been found for the class of propagation-invariant coherent fields, and the subject of partially coherent and partially polarized propagation-invariant fields has barely been touched. We also stress that the theory of propagation-invariant pulses is still under active study, and in particular the partially coherent case is poorly understood. Recent years have produced substantial developments in versatile, light efficient experimental techniques to generate Bessel fields. Many of these techniques being based on diffractive or hybrid optics, and active elements such as spatial light modulators are increasingly used for their production. Diamond-turning techniques are expected to lead soon to the demonstration of systems with chromatic control, such as that presented in Figure 13, which will allow optical generation of localized (pulsed) waves at different speeds. It also appears feasible to realize integrated opto-electro-mechanical modules capable of launching Bessel fields, even with tunable transverse shape and scale as well as temporal profile. It is curious to note that the flood of proposed applications of propagation-invariant fields, which followed Durnin’s paper in 1987, appears to have dried up somewhat since the early 1990’s. Nevertheless, we believe that technical applications of propagation-invariant fields will gain new momentum as the theory and production methods are matured.
ACKNOWLEDGMENTS ¨ We express our gratitude to Pertti P¨aa¨ kkonen and Jani Tervo for their generous help in the preparation of the manuscript. The work was partly supported by the Academy of Finland (projects 111701, 129155, 209806, and 215341), the Swedish Foundation for Strategic Research (SSF), the Network of Excellence in Micro-Optics (NEMO, www.micro-optics.org), as well as the Ministry of Education, Finland, through the Research and Development Project on Nanophotonics.
REFERENCES Aagedal, H., Wyrowski, F., & Schmid, M. (1997). In J. Turunen, & F. Wyrowski (Eds.), Diffractive optics for industrial and commercial applications. Berlin: Akademie-Verlag. Alexeev, I., Kim, K. Y., & Milchberg, H. M. (2002). Physical Review Letters, 88, 073901. Arif, M., Hossain, M. M., Awwal, A. A. S., & Islam, M. N. (1998). Applied Optics, 37, 649.
82
Propagation-Invariant Optical Fields
Arimoto, R., Saloma, C., Tanaka, T., & Kawata, S. (1992). Applied Optics, 31, 6653. Arlt, J., & Dholakia, K. (2000). Optical Communications, 177, 297. Arlt, J., Dholakia, K., Soneson, J., & Wright, E. M. (2001). Physical Review A, 63, 063602. Arlt, J., Garces-Chaves, V., Sibbett, W., & Dholakia, K. (2000). Optical Communications, 177, 297. Arsenault, H. A., & Boivin, A. (1967). Journal of Applied Physics, 38, 3988. Bagini, V., Frezza, F., Santarsiero, M., Schettini, G., & Schirripa Spagnolo, G. (1996). Journal of Modern Optics, 43, 1155. Bajer, J., & Hor´ak, R. (1996). Physical Review E, 54, 3052. Balykin, V. I., Letokhov, V. S., Minogin, V. G., Rozhdetsvensky, Yu. V., & Sidorov, A. I. (1985). Journal of the Optical Society of America B, 2, 1776. Bandres, M. A., Guti´errez-Vega, J. C., & Ch´avez-Cerda, S. (2004). Optics Letters, 29, 44. Bartelt, H. (1984). Applied Optics, 23, 1499. Bartelt, H. (1985). Applied Optics, 24, 3811. B´elanger, P. A. (1984). Journal of the Optical Society of America A, 1, 723. B´elanger, P. A., & Par´e, C. (1991). Optics Letters, 16, 1057–1059. B´elanger, P. A., & Rioux, M. (1978). Applied Optics, 17, 1080. Belskii, A. M. (1992). Optics and Spectroscopy, 73, 568. Berry, M. V., & Balazs, N. L. (1979). American Journal of Physics, 62, 264. Bertolotti, M., Sereda, L., & Ferrari, A. (1997). Journal of the European Optical Society A: Pure Applied Optics, 6, 153. Bickel, G., H¨ausler, G., & Maul, M. (1985). Optical Engineering, 24, 975–977. Borghi, R. (1999). IEEE Journal of Quantum Electronics, 35, 849. Borghi, R., Santarsiero, M., & Gori, F. (1997). Journal of the Optical Society of America A, 14, 23. Borghi, R., Santarsiero, M., & Porras, M. A. (2001). Journal of the Optical Society of America A, 18, 1618. Born, M., & Wolf, E. (1999). Principles of optics (7th (expanded) ed.). Cambridge, UK: Cambridge University Press. Bouchal, Z. (1993). Journal of Modern Optics, 40, 1325. Bouchal, Z. (2002). Optics Letters, 27, 1376. Bouchal, Z. (2003). Czechoslovak Journal of Physics, 53, 537. Bouchal, Z., Bajer, J., & Bertolotti, M. (1998). Journal of the Optical Society of America A, 15, 2172. Bouchal, Z., & Courtial, J. (2004). Journal of Optics A: Pure Applied Optics, 6, S184. Bouchal, Z., Hor´ak, R., & Wagner, J. (1996). Journal of Modern Optics, 43, 1905. Bouchal, Z., & Oliv´ık, M. (1995). Journal of Modern Optics, 42, 1555. Bouchal, Z., & Peˇrina, J. (2002). Journal of Modern Optics, 49, 1673. Bouchal, Z., Wagner, J., & Chlup, M. (1998). Optical Communications, 151, 207. Brittingham, J. N. (1983). Journal of Applied Physics, 54, 1179. Bryngdahl, O. (1974). Journal of the Optical Society of America, 64, 1092. Bryngdahl, O., & Wyrowski, F. (1990). In E. Wolf (Ed.), Progr. opt.: 28 (p. 1). Amsterdam: North-Holland. Bunkin, F. V., Korobkin, V. V., Kurinyi, Yu. A., Polonskii, L. Ya., & Pyatnitskii, L. N. (1983). Soviet Journal of Quantum Electronics, 13, 254. Butkus, R., Gadonas, R., Januˇsonis, J., Piskarskas, A., Regelskis, K., Smilgeviˇcius, V., & Stabinis, A. (2002). Optical Communications, 206, 201. Cai, S. Y., Bhattacharjee, A., & Marshall, T. C. (1988). Nuclear Instruments and Methods in Physics Research, A272, 481. ˜ Canas, R. G., & Smith, R. W. (1989). IEE Conference Publications, 311, 160. Campbell, J. A., & Soloway, S. (1990). Journal of Acoustical Society of America, 88, 2467. Chattrapiban, N., Rogers, E. A., Cofeld, D., Hill III, W. T., & Roy, R. (2003). Optics Letters, 28, 2183. Ch´avez-Cerda, S., McDonald, G. S., & New, G. H. S. (1996). Optical Communications, 123, 225.
References
83
Ch´avez-Cerda, S., Padgett, M. J., Allison, I., New, G. H. C., Guti´errez-Vega, J. C., O’Neil, A. T., Macvicar, I., & Courtial, J. (2002). Journal of Optics B: Quantum and Semiclassical Optics, 4, S52. Chu, D. C., Fienup, J. R., & Goodman, J. W. (1973). Applied Optics, 12, 1386. ˇ zm´ar, T., Siler, ˇ Cisˇ M., & Zem´anek, P. (2006). Applied Physics B, 84, 197. Collins, S. A. (1970). Journal of the Optical Society of America, 60, 1168. Colombeau, B., Froehly, C., & Vampoille, M. (1979). Optical Communications, 28, 35. Conti, C. (2003). Physical Review E, 68, 016606. Cottrell, D., Craven, J. M., & Davis, J. A. (2007). Optics Letters, 32, 298. Courtial, J., Whyte, G., Bouchal, Z., & Wagner, J. (2006). Optics Express, 14, 2108. Cox, A. J., & D’Anna, J. (1992). Optics Letters, 17, 232. Cox, A. J., & Dibble, D. C. (1991). Applied Optics, 30, 1330. Cox, A. J., & Dibble, D. C. (1992). Journal of the Optical Society of America A, 9, 282. Dammann, H. (1969). Optik, 53, 409. ´ Dartora, C. A., Nobrega, K. Z., Dartora, A., Viana, G. A., & Filho, H. T. S. (2006). Optical Communications, 265, 481. ´ Dartora, C. A., Zamboni-Rached, M., Nobrega, K. Z., Recami, E., & Hern´andez-Figueroa, H. E. (2003). Optical Communications, 222, 75. Davidson, N., Friesem, A. A., & Hasman, E. (1992). Optical Communications, 88, 326. Davis, J. A., Guertin, J., & Cottrell, D. M. (1993). Applied Optics, 32, 6368. Davis, J. A., Carcole, E., & Cottrell, D. M. (1996a). Applied Optics, 35, 593. Davis, J. A., Carcole, E., & Cottrell, D. M. (1996b). Applied Optics, 35, 599. DeBeer, D., Hartmann, S. R., & Friedberg, R. (1987). Physical Review Letters, 59, 2611. DeNicola, S. (1991). Optical Communications, 80, 299. DeNicola, S. (1996). JEOS A: Pure Applied Optics, 5, 827. DiTrapani, P., Valiulis, G., Piskarskas, A., Jedrkiewicz, O., Trull, J., Conti, C., & Trillo, S. (2003). Physical Review Letters, 91, 093904. Dong, L., & Wang, H. (2007). Optics Express, 15, 1706. Dong, L., Wang, H., Zhou, W., Yang, X., Lv, X., & Chen, H. (2008). Optics Express, 16, 5649. Dong, B.-Z., Yang, G.-Z., Gu, B.-Y., & Ersoy, O. K. (1996). Journal of the Optical Society of America A, 13, 97. Donnelly, R., & Ziolkowski, R. (1993). Proceedings of the Royal Society of London, Series A, 440, 541. Durnin, J. (1987). Journal of the Optical Society of America A, 4, 651. Durnin, J., & Eberly, J.H. (1989). US Patent 4,887,885. Durnin, J., Miceli Jr., J. J., & Eberly, J. H. (1987). Physical Review Letters, 58, 1499. Durnin, J. E., Miceli Jr., J. J., & Eberly, J. H. (1988). Optics Letters, 13, 79. Durnin, J. E., Miceli Jr., J. J., & Eberly, J. H. (1991). Physical Review Letters, 66, 838. Erdogan, T., King, O., Wicks, G. W., Hall, D. G., Anderson, E., & Rooks, M. J. (1992). Applied Physics Letters, 60, 1921. Fagerholm, J., Friberg, A. T., Huttunen, J., Morgan, D. P., & Salomaa, M. M. (1996). Physical Review E, 54, 4347. Fatemi, F. K., & Bashkansky, M. (2006). Optics Letters, 31, 864. Fedotowski, A., & Lehovec, K. (1974). Applied Optics, 13, 2919. Ferguson, T. R., Gutheinz, L. M., Mayo III, J. W., German, J. D., Lowrey, W. H., & Konopnicki, M. J. (1982). Opt. Eng., 21, 959. ´ Fischer, P., Brown, C. T. A., Morris, J. E., Lopez-Mariscal, C., Wright, E. M., Sibbett, W., & Dholakia, K. (2005). Optics Express, 13, 1176. ´ Fischer, P., Little, H., Smith, R. L., Lopez-Mariscal, C., Brown, C. T. A., Sibbett, W., & Dholakia, K. (2006). Journal of Optics A: Pure Applied Optics, 8, 477. Fischer, R., Neshev, D. N., Lopez-Aguayo, S., Desyatnikov, A. S., Sukhorukov, A. A., Krolikowski, W., & Kivshar, Y. S. (2006). Optics Express, 14, 2825. Florjanczyk, M., & Tremblay, R. (1989). Optical Communications, 73, 448. Fontana, R., & Pantell, R. H. (1983). Journal of Applied Physics, 54, 4285.
84
Propagation-Invariant Optical Fields
Friberg, A. T., Fagerholm, J., & Salomaa, M. M. (1997). Optical Communications, 136, 207. Friberg, A. T., & Turunen, J. (1994). Journal of the Optical Society of America A, 11, 227. Friberg, A. T., Vasara, A., & Turunen, J. (1991). Physical Review A, 43, 7079. Garc´ez-Chavez, V., McGloin, D., Melville, H., Sibbett, W., & Dholakia, K. (2002). Nature, 419, 145. Garc´es-Chavez, V., Roskey, D., Summers, M. D., Melville, H., McGloin, D., Wright, E. M., & Dholakia, K. (2004). Applied Physics Letters, 85, 4001. Garc´es-Chavez, V., Volke-Sepulveda, K., Ch´avez-Cerda, S., Sibbett, W., & Dholakia, K. (2002). Physical Review A, 66, 063402. Gbur, G., & Visser, T. (2003). Optics Letters, 28, 1627. Goodman, J. W., & Silvestri, A. M. (1970). IBM Journal of Research and Development, 14, 478. Gori, F. (1980). Optics Acta, 27, 1025. Gori, F. (1999). Optics Letters, 24, 584. Gori, F. (1998). Optics Letters, 23, 241. Gori, F., Guattari, G., & Padovani, C. (1987a). Optical Communications, 64, 311. Gori, F., Guattari, G., & Padovani, C. (1987b). Optical Communications, 64, 491. Gori, F., & Palma, C. (1978). Optical Communications, 27, 185. Gori, F., & Santarsiero, M. (1998). Optics Letters, 23, 241. Gori, F., & Santarsiero, M. (2007). Optics Letters, 32, 3531. Gori, F., Santarsiero, M., & Borghi, R. (2008). Optics Letters, 33, 1857. Gori, F., Santarsiero, M., Borghi, R., & Li, C.-F. (2008). Journal of the Optical Society of America A, 25, 2826. Gori, F., Santarsiero, M., Borghi, R., & Ram´ırez-S´anchez, V. (2008). Journal of the Optical Society of America A, 25, 1016. Gori, F., Santarsiero, M., Vicalvi, S., Borghi, R., & Guattari, G. (1998). JEOS A: Pure Applied Optics, 7, 941. Greene, P. L., & Hall, D. G. (1996). Journal of the Optical Society of America A, 13, 962. Greene, P. L., & Hall, D. G. (1998). Journal of the Optical Society of America A, 15, 3020. Grosjean, T., Saleh, S. S., Suarez, M. A., Ibrahim, I. A., Piquerey, V., Charraut, D., & Sandoz, P. (2007). Applied Optics, 46, 8061. Grunwald, R., Bock, M., Kebbel, V., Huferath, S., Neumann, U., Steinmeyer, G., Stibenz, G., N´eron, J.-L., & Piche, M. (2008). Optics Express, 16, 1077. Grunwald, R., Griebner, U., Tschirschwitz, F., Nibbering, E. T. J., Elsaesser, T., Kebbel, V., ¨ Hartman, H.-J., & Juptner, W. (2000). Optics Letters, 25, 981. Gu´erineau, N., Harchaoui, B., Primot, J., & Heggarty, K. (2001). Optics Letters, 26, 411. Gu´erineau, N., & Primot, J. (1999). Journal of the Optical Society of America A, 16, 293. Guti´errez-Vega, J. C., Iturbe-Castillo, M. D., & Ch´avez-Cerda, S. (2000). Optics Letters, 25, 1493. Guti´errez-Vega, J. C., Iturbe-Castillo, M. D., Ramirez, G. A., Tepich´ın, E., Rodr´ıguez-Dagnino, R. M., Ch´avez-Cerda, S., & New, G. H. C. (2001). Optical Communications, 195, 35. Guti´errez-Vega, J. C., Rodr´ıgues-Masegosa, R., & Ch´avez-Cerda, S. (2003). Journal of the Optical Society of America A, 20, 2113. Hafizi, B., & Sprangle, P. (1991). Journal of the Optical Society of America A, 8, 705. ¨ H., Simonen, J., P¨aa¨ kkonen, ¨ Hakola, A., Buchter, S. C., Kajava, T., Elfstrom, P., & Turunen, J. (2004). Optical Communications, 238, 335. Hall, D. G. (1996). Optics Letters, 21, 9. H¨ausler, G., & Heckel, W. (1988). Applied Optics, 27, 5165. Havelock, T. H. (1904). Proceedings of the London Mathematical Society, 2, 122. He, H., Heckenberg, N. R., & Rubinstein-Dunlop, H. (1995). Journal of Modern Optics, 42, 217. Herman, R. M., & Wiggins, T. A. (1991). Journal of the Optical Society of America A, 8, 932. Herman, R. M., & Wiggins, T. A. (1992). Applied Optics, 31, 5913. Hern´andez-Aranda, R. I., Ch´avez-Cerda, S., & Guti´errez-Vega, J. C. (2005). Journal of the Optical Society of America A, 22, 1909.
References
85
Hern´andez-Aranda, R. I., & Guti´errez-Vega, J. C. (2007). Journal of the Optical Society of America A, 24, 1209. Hern´andez-Figueroa, H. E., Zamboni-Rached, M., & Recami, E. (Eds.) (2008). Localized waves. New York: Wiley. Heyman, E., & Felsen, L. B. (1994). IEEE Transactions on Antennas and Propagation, 42, 1668. Heyman, E., Steinberg, B. Z., & Felsen, L. B. (1987). Journal of the Optical Society of America A, 4, 2081. Hillion, P. T. M. (1992). IEEE Transactions on Antennas and Propagation, 40, 1031. Honkanen, M., & Turunen, J. (1998). Optical Communications, 154, 368. Hor´ak, R., Bouchal, Z., & Bajer, J. (1997). Optical Communications, 133, 315. Hsu, D. K., Margetan, F. J., & Thompson, D. O. (1989). Applied Physics Letters, 55, 2066. Hu, W., & Guo, H. (2002). Journal of the Optical Society of America A, 19, 49. Iftekharuddin, K. M., Awwal, A. A. S., & Karim, M. A. (1993). Applied Optics, 32, 2252. Indebetouw, G. (1989). Journal of the Optical Society of America A, 6, 150. Indebetouw, G. (1992). Journal of the Optical Society of America A, 9, 549. Jabczynski, J. K. (1990). Optical Communications, 77, 292. Jaroszewicz, Z. (1997). In M. Pluta (Ed.), R & D Treatises: Vol. 5. Axicons: Design and propagation properties. Warsaw, Poland, SPIE Polish Chapter. Jaroszewicz, Z., Burvall, A., & Friberg, A. T. (2005). Optics and Photonic News, 16(July), 34. ´ Jaroszewicz, Z., Sochacki, J., Kołodziejczyk, A., & Staronski, L. R. (1993). Optics Letters, 18, 1893. Jiang, Z.-P. (1996). Optical Communications, 125, 207. Jiang, Z., Lu, Q., & Liu, Z. (1995). Applied Optics, 34, 7183. Jordan, R. H., & Hall, D. G. (1994). Optics Letters, 19, 427. Kartashov, Y. V., Vysloukh, V. A., & Torner, L. (2004). Physical Review Letters, 93, 093904. Kartashov, Y. V., Vysloukh, V. A., & Torner, L. (2005a). Physical Review Letters, 94, 043902. Kartashov, Y. V., Vysloukh, V. A., & Torner, L. (2005b). Optics Letters, 30, 637. Kettunen, V., & Turunen, J. (1998). Optics Letters, 23, 1247. Khilo, A. N., Katranji, E. G., & Ryzhevich, A. A. (2001). Journal of the Optical Society of America A, 18, 1986. Khonina, S. N., Kotlyar, V. V., Skidanov, R. V., Soifer, V. A., Jefimovs, K., & Turunen, J. (2004). Journal of Modern Optics, 51, 2167. Khonina, S. N., Kotlyar, V. V., Soifer, V. A., Lautanen, J., Honkanen, M., & Turunen, J. (1999). Optik, 110, 137. Khonina, S. N., Kotlyar, V. V., Soifer, V. A., Shinkaryev, M. V., & Uspleniev, G. V. (1992). Optical Communications, 91, 158. Kimura, W. D., Kim, G. H., Romea, R. D., Steinhauer, L. C., Pogorelsky, I. V., Kusche, K. P., Fernow, R. C., Wang, X., & Liu, Y. (1995). Physical Review Letters, 74, 546. Kiselev, A. P. (2007). Optics and Spectroscopy, 102, 603. Kogelnik, H., & Li, T. (1966). Proceedings of IEEE, 54, 1312. Kotlyar, V. V., Khonina, S. N., & Soifer, V. A. (1995). Journal of Modern Optics, 42, 1231. Kotlyar, V. V., Soifer, V. A., & Khonina, S. N. (1997). Journal of Modern Optics, 44, 1409. Kowarz, M. W., & Agarwal, G. S. (1995). Journal of the Optical Society of America A, 12, 1324. Kurtz, C. N., Hoadley, H. O., & dePalma, J. J. (1973). Journal of the Optical Society of America, 63, 1080. Lapointe, M. R. (1992). Optics and Laser Technology, 24, 315. Lautanen, J., Kettunen, V., Laakkonen, P., & Turunen, J. (2000). Journal of the Optical Society of America A, 17, 2208. Leger, J. R. (1997). In J. Turunen, & F. Wyrowski (Eds.), Diffractive optics for industrial and commercial applications. Berlin: Akademie Verlag. Leger, J. R., Chen, D., & Wang, Z. (1994). Optics Letters, 19, 108. Leith, E. N., Collins, G., Khoo, I., & Wynn, T. (1980). Journal of the Optical Society of America, 70, 141. Lin, Y., Seka, W., Eberly, J. H., Huang, H., & Brown, D. L. (1992). Applied Optics, 31, 2708.
86
Propagation-Invariant Optical Fields
Ling, D., & Li, J. (2006). Journal of the Optical Society of America B, 23, 1574. Liu, Z., & Fan, D. (1998). Journal of Modern Optics, 45, 17. Longhi, S. (2004). Optics Letters, 29, 147. ´ Lopez-Mariscal, C., Bandres, M. A., Ch´avez-Cerda, S., & Guti´errez-Vega, J. C. (2005). Optics Express, 13, 2364. Lu, J., & Greenleaf, J. F. (1990). IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, 37, 438. Lu, J., & Greenleaf, J. F. (1992a). IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, 39, 19. Lu, J., & Greenleaf, J. F. (1992b). IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, 39, 441. ¨ B., & Huang, W. (1996). Journal of Modern Optics, 43, 509. Lu, ¨ B., Huang, W., Zhang, B., Kong, F., & Zhai, Q. (1996). Optical Communications, 131, 223. Lu, MacDonald, R. P., Boothroyd, S. A., Okamoto, T., Chrostowski, J., & Syrett, B. A. (1996). Optical Communications, 122, 169. MacDonald, R. P., Chrostowski, J., Boothroyd, S. A., & Syrett, B. A. (1993). Applied Optics, 32, 6470. Mackinnon, L. (1978). Foundations of Physics, 8, 157. Mandel, L., & Wolf, E. (1995). Coherence and quantum optics. Cambridge, UK: Cambridge University Press. McCutchen, C. W. (1964). Journal of the Optical Society of America, 54, 240. McGloin, D., & Dholakia, K. (2005). Contemporary Physics, 46, 15. McGloin, D., Garcez-Chavez, V., & Dholakia, K. (2003). Optics Letters, 28, 657. McLeod, J. H. (1954). Journal of the Optical Society of America, 44, 592. McLeod, J. H. (1960). Journal of the Optical Society of America, 50, 166. Mihailache, D., Mazilu, D., Lederer, F., Malomed, B. A., Kartashov, Y. V., Crasovan, L.-C., & Torner, L. (2005). Physical Review Letters, 95, 023902. Mishra, S. R. (1991). Optical Communications, 85, 159. Miyamoto, K. (1961). Journal of the Optical Society of America, 51, 17. Montgomery, W. D. (1967). Journal of the Optical Society of America, 57, 772. Mugnai, D., Ranfagni, A., & Ruggeri, R. (2000). Physical Review Letters, 84, 4830. Muys, P., & Vandamme, E. (2002). Applied Optics, 41, 6375. Nisoli, M., Priori, E., Sansone, G., Stagira, S., Cerullo, G., deSilvestri, S., Altucci, C., Bruzzese, R., de Lisio, C., Villoresi, P., Poletto, L., Pascolini, M., & Tondello, G. (2002). Physical Review Letters, 88, 033902. Ohtsuka, Y. (1981). Optical Communications, 39, 283. Ostrovsky, A. S., Nartin´ez-Niconoff, G., & Martinez-San-Juan, J. C. (2001). Opt. Commun., 195, 27. Ostrovsky, A. S., Nartin´ez-Niconoff, G., & Martinez-San-Juan, J. C. (2002). Opt. Commun., 207, 131. Overfelt, P. L. (1991). Physical Review A, 44, 3941. Overfelt, P. L., & Kenney, C. S. (1991). Journal of the Optical Society of America A, 8, 732. ¨ P¨aa¨ kkonen, P., Lautanen, J., Honkanen, M., Kuittinen, M., Turunen, J., Khonina, S. N., Kotlyar, V. V., Soifer, V. A., & Friberg, A. T. (1998). Journal of Modern Optics, 46, 2355. ¨ P¨aa¨ kkonen, P., Simonen, J., Honkanen, M., & Turunen, J. (2002). Journal of Modern Optics, 49, 1943. ¨ P¨aa¨ kkonen, P., & Turunen, J. (1998). Optical Communications, 156, 359. Palma, C. (1997). Applied Optics, 36, 1116. Palma, C., Borghi, R., & Cincotti, G. (1996). Optical Communications, 125, 113. Paterson, C., & Smith, R. (1996a). Optical Communications, 124, 121. Paterson, C., & Smith, R. (1996b). Optical Communications, 124, 131. Patorski, K. (1989). In E. Wolf (Ed.), Progr. opt.: Vol. 27 (p. 1). Amsterdam: North-Holland. Petit, R. (Ed.) (1980). Electromagnetic theory of gratings. Berlin: Springer. Polonskiy, L. (2005). Optics and Photonic News, 16(July/August), 7.
References
87
Ponomarenko, S., Huang, W., & Cada, M. (2007). Optics Letters, 32, 2508. Popov, S., & Friberg, A. T. (1995). Opt. Eng., 34, 2567. Popov, S., Friberg, A. T., Honkanen, M., Lautanen, J., Turunen, J., & Schnabel, B. (1998). Optical Communications, 154, 359. Porras, M. A. (2001). Optics Letters, 26, 1364. Provost, J. N., & de Bougrenet de la Tocnaye, J. L. (1997). Journal of the Optical Society of America A, 14, 2748. Recami, E. (1998). Physica A, 252, 586. ´ Recami, E., Zamboni-Rached, M., Nobrega, K. Z., Dartora, C. A., & Hernand´ez-Figueroa, H. E. (2003). IEEE Journal on Selected Topics in Quantum Electronics, 9, 59. Reivelt, K., & Saari, P. (2000). Journal of the Optical Society of America A, 17, 1785. Reivelt, K., & Saari, P. (2002a). Physical Review E, 65, 046622. Reivelt, K., & Saari, P. (2002b). Physical Review E, 66, 056611. Reivelt, K., & Saari, P. (2004). Optics Letters, 29, 1176. Richards, B., & Wolf, E. (1959). Proceedings of the Royal Society of London, Series A, 253, 358. Rioux, M., Trenblay, R., & B´elanger, P. A. (1978). Applied Optics, 17, 1532. Rogel-Salazar, J., New, G. H. C., & Ch´avez-Cerda, S. (2001). Optical Communications, 190, 117. Romea, R. D., & Kimura, W. D. (1990). Physical Review D, 42, 1807. Roy, G., & Tremblay, R. (1980). Optical Communications, 34, 1. Rubinowicz, A. (1965). In E. Wolf (Ed.), Progr. opt.: Vol. 4 (p. 199). Amsterdam: North-Holland. Ruschin, S. (1994). Journal of the Optical Society of America A, 11, 3224. Ruschin, S., & Leizer, A. (1998). Journal of the Optical Society of America A, 15, 1139. Saari, P. (2008). Optics Express, 16, 10303. Saari, P., & Reivelt, K. (1997). Physical Review Letters, 21, 4135. Saari, P., & Reivelt, K. (2004). Physical Review E, 69, 036612. Salo, J., Fagerholm, J., Friberg, A. T., & Salomaa, M. M. (2000). Physical Review E, 62, 4261. Salo, J., & Salomaa, M. M. (2001a). Journal of Optics A: Pure Applied Optics, 2, 366. Salo, J., & Salomaa, M. M. (2001b). Acoustic Research Letters Online, 2, 31. Santarsiero, M. (1996). Optical Communications, 132, 1. Schechner, Y. Y., Piestun, R., & Shamir, J. (1996). Physical Review E, 54, R50. Scott, G., & McArdle, N. (1992). Opt. Eng., 31, 2640. Scully, M. O., & Zubairy, M. S. (1991). Physical Review A, 44, 2656. Seshadri, S. R. (2008). Journal of the Optical Society of America A, 25, 1. Sezginer, A. (1985). Journal of Applied Physics, 67, 678. Shaarawi, A. M. (1997). Journal of the Optical Society of America A, 14, 1804. Shaarawi, A. M., Ziolkowski, R. W., & Besieris, I. M. (1995). Journal of Mathematical Physics, 36, 5565. Shchegrov, A. V., & Wolf, E. (2000). Optics Letters, 25, 141. Sheppard, C. J. R. (2001). Journal of the Optical Society of America A, 18, 2594. Sheppard, C. J. R. (2002). Journal of the Optical Society of America A, 19, 2218. Sheppard, C. J. R., & Saari, P. (2008). Optics Express, 16, 150. Siviloglou, G. A., Broky, J., Dogariu, A., & Christodoulides, D. N. (2007). Physical Review Letters, 99, 312901. Sochacki, J., Bar´a, S., Jaroszewicz, Z., & Kołodziejczyk, A. (1992). Optics Letters, 17, 7. ´ Sochacki, J., Jaroszewicz, Z., Staronski, L. R., & Kołodziejczyk, A. (1993). Journal of the Optical Society of America A, 10, 1765. Sochacki, J., Kołodziejczyk, A., Jaroszewicz, Z., & Bar´a, S. (1992). Applied Optics, 31, 5326. ˜ Sonajalg, H., & Saari, P. (1996). Optics Letters, 21, 1162. ˜ Sonajalg, H., & Saari, P. (1997). Optics Letters, 22, 310. ˜ Sonajalg, H., R¨atsep, M., & Saari, P. (1997). Optics Letters, 22, 310. Optics Letters 22, 745 (Errata). Soroko, L. M. (1989). In E. Wolf (Ed.), Prog. opt.: Vol. 27 (p. 109). Amsterdam: Elsevier. Soroko, L. M. (1996). Meso-optics: Foundations and applications. Singapore: World Scientific. Sprangle, P., & Hafizi, B. (1991). Physical Review Letters, 66, 837.
88
Propagation-Invariant Optical Fields
´ Staronski, J. R., Sochacki, J., Jaroszewicz, Z., & Kołodziejczyk, A. (1992). Journal of the Optical Society of America A, 9, 2091. Stratton, J. A. (1941). Electromagnetic theory. New York: McGraw-Hill. ˜ Szwaykowski, P., & Ojeda-Castaneda, J. (1991). Optical Communications, 83, 1. Talbot, H. F. (1836). Philosophical Magazine, 9, 401. Tervo, J. (2003). Journal of the Optical Society of America A, 20, 1974. Tervo, J., & Turunen, J. (2000). Optics Letters, 25, 785. Tervo, J., & Turunen, J. (2001). Optical Communications, 192, 13. Tervo, J., Vahimaa, P., & Turunen, J. (2002). Journal of Modern Optics, 49, 1537. Thaning, A., & Friberg, A. T. (2002). Journal of Modern Optics, 49, 1933. Thewes, K., Karim, M. A., & Awwal, A. A. S. (1991). Optics and Laser Technology, 23, 105. Tidwell, S. C., Ford, D. H., & Kimura, W. D. (1992). Opt. Eng., 31, 1527. Tsangaris, C. L., New, G. H. C., & Rogel-Salazar, J. (2003). Optical Communications, 223, 233. Turunen, J. (1997). In H. P. Herzig (Ed.), Micro-optics: Elements, systems and applications. London: Taylor & Francis. Turunen, J. (2002). Journal of Modern Optics, 49, 1795. Turunen, J. (2008). Optics Express, 16, 20283. Turunen, J., & Friberg, A. T. (1993a). JEOS A: Pure Applied Optics, 2, 51. Turunen, J., & Friberg, A. T. (1993b). JEOS A: Pure Applied Optics, 2, 539. Turunen, J., Tervonen, E., & Friberg, A. T. (1990). Journal of Applied Physics, 67, 49. Turunen, J., & Vahimaa, P. (2008). Optics Express, 16, 6433. Turunen, J., Vasara, A., & Friberg, A. T. (1988). Applied Optics, 27, 3859. Turunen, J., Vasara, A., & Friberg, A. T. (1991). Journal of the Optical Society of America A, 8, 282. Turunen, J., & Wyrowski, F. (1997). In J. Turunen, & F. Wyrowski (Eds.), Diffractive optics for industrial and commercial applications. Berlin: Akademie Verlag. Uehara, K., & Kikuchi, H. (1989). Applied Physics B, 48, 125. Vahimaa, P., Kettunen, V., Kuittinen, M., Turunen, J., & Friberg, A. T. (1997). Journal of the Optical Society of America A, 14, 808. Vahimaa, P., & Turunen, J. (2006). Optics Express, 14, 5007. Valtna, H., Reivelt, K., & Saari, P. (2007). Optical Communications, 278, 1. van Nie, A. G. (1964). Philips Research Reports, 19, 378. Vasara, A., Turunen, J., & Friberg, A. T. (1989a). IEE Conference Proceedings Series, 311, 85. Vasara, A., Turunen, J., & Friberg, A. T. (1989b). Journal of the Optical Society of America A, 6, 1748. Vicari, L. (1989). Optical Communications, 70, 263. Volke-Sepulveda, K., Garc´es-Ch´avez, V., Ch´avez-Cerda, S., Artl, J., & Dholakia, K. (2002). J. Opt. B: Quantum Semiclass. Opt., 4, S82. Wang, X., Chen, Z., & Kevrekidis, P. G. (2006). Physical Review Letters, 96, 083904. Whittaker, E. T. (1902). Mathematische Annalen, 57, 333. Wolf, E. (1982). Journal of the Optical Society of America, 72, 343. Wolf, E. (2003). Phys. Lett., A 312, 263. Wolf, K. B. (1988). Physical Review Letters, 60, 757. Wolf, E., & Agarwal, G. S. (1984). Journal of the Optical Society of America A, 1, 541. Wu, F., Chen, Y., & Guo, D. (2007). Applied Optics, 22, 4943. Wulle, T., & Herminghaus, S. (1993). Physical Review Letters, 70, 1401. Zahid, M., & Zubairy, M. S. (1989). Optical Communications, 70, 361. Zamboni-Rached, M. (2004). Optics Express, 12, 4001. Zamboni-Rached, M. (2006a). Journal of the Optical Society of America A, 23, 2166. Zamboni-Rached, M. (2006b). Optics Express, 14, 1804. ´ Zamboni-Rached, M., Nobrega, K. Z., Hern´andez-Figueroa, H. E., & Recami, E. (2003). Optical Communications, 226, 15. Zamboni-Rached, M., & Recami, E. (2008). Physical Review A, 77, 033824. Zamboni-Rached, M., Recami, E., & Hernand´ez-Figueroa, H. E. (2005). Journal of the Optical Society of America A, 22, 2465. Ziolkowski, R. W., Besieris, I. M., & Shaarawi, A. M. (1993). Journal of the Optical Society of America A, 10, 75.
CHAPTER
2 Theoretical Tools for Quantum Optics in Structured Media Dmitri Mogilevtsev and Sergei Kilin Institute of Physics, Belarus National Academy of Sciences, F.Skarina Ave. 68, Minsk 220072, Belarus
Contents
1. 2.
3.
4.
Introduction Exact and Approximate Solutions for System Plus Reservoir 2.1. Spontaneous Emission into a Bosonic Reservoir: The Problem 2.2. The Density-of-States and Photonic Band-Gaps 2.3. Spontaneous Dynamics and the Markovian Approximation 2.4. The Cluster Expansion and Collective Operators 2.5. The Pseudo-Modes Method 2.6. Dephasing into a Bosonic Reservoir Master Equations: Deterministic Approaches 3.1. Projection Operator Method and the Cumulant Expansion 3.2. Validity of Perturbative Approximations: Spontaneous Emission 3.3. A System-Reservoir Border 3.4. Separating the Reservoir 3.5. Solving Master equations I Master Equations: Stochastic Approaches 4.1. Quantum Monte-Carlo Procedure 4.2. Non-Lindblad Quantum Monte-Carlo Procedure 4.3. Pseudomodes and Quantum Monte-Carlo Procedure 4.4. Quantum Diffusion Procedure 4.5. Non-Markovian Quantum Diffusion Procedure
90 92 93 95 97 99 103 105 109 111 114 117 120 123 127 129 133 136 138 141
c 2009 Elsevier B.V. Progress in Optics, Volume 54 ISSN 0079-6638, DOI 10.1016/S0079-6638(10)05407-7 All rights reserved.
89
90
Theoretical Tools for Quantum Optics in Structured Media
5. Concluding Remarks Acknowledgements References
144 145 145
1. INTRODUCTION Interaction of a quantum system with a reservoir represents a fundamental problem of quantum optics and quantum physics in general. One encounters the problem when investigating such phenomena as spontaneous decay and decoherence (Breuer & Petruccione, 2002), emergence of classicality (Paz & Zurek, 2001) and quantum measurements (Braginsky & Khalili, 1992). Nowadays, an ability to find an adequate solution for this problem is of utmost importance for a broad scope of intensively developing fields of research, from quantum information and computing to nanotechnology and chemical physics. Attempts to develop approaches for investigating system-reservoir interaction were undertaken already at the earliest stage of quantum physics development, and some fundamentally important results were obtained (for example, theory of the natural line width (Weisskopf & Wigner, 1930), decay conditions (Krylov & Fock, 1947) and non-exponential asymptotic decay behaviour (Khalfin, 1958)). An extensive investigation of this field, a search for novel and efficient ways to build solutions continues unremittingly up to the present day. The reason is that the problem allows for exact solution in only few specific cases, such as spontaneous emission in the vacuum of the bosonic reservoir or interaction of a harmonic oscillator with a reservoir of harmonic oscillators. When the total number of excitations in the system and the reservoir becomes large, in most cases one cannot obtain an exact solution for neither density matrix elements, nor system operators, nor quantum mechanical averages. Thereof, one has to resort to some approximation. Usually, one develops an approximation by making an assumption about a ‘strength’ of interaction between the system and the reservoir and between the parts of the reservoir. For example, if one assumes that the system-reservoir interaction leads to no more than a few excitations in the reservoir, one might truncate a system of equations for the probability amplitudes and density matrix elements (Lambropoulos, Nikolopoulos, Nielsen, & Bay, 2000) or use a truncated cluster expansion for the operator averages (Kira & Koch, 2006). Up to now, one of the most efficient and widely used approximations is the Weisskopf-Wigner, or Born–Markov approximation (Weisskopf & Wigner, 1930). It can be applied when the coupling between the system and the reservoir is weak, and the reservoir correlation time is small on a
Introduction
91
typical time-scale of the system dynamics. When this approximation holds (actually, it is a very common situation in practice), it is possible to derive a number of effective approaches applicable to a wide range of particular realizations of the system-reservoir interaction problem. For example, one can derive a time-local master equation for the reduced density matrix, develop a formalism of Heisenberg–Langevin equations for systems’ operators, or Fokker–Planck equations for a quasiprobability distribution corresponding to the density matrix averaged over the reservoir (Peˇrina, 1984; Scully & Whitney, 1972). When the system-reservoir interaction cannot be assumed to be small, and/or when the reservoir correlations decay slowly and long-time correlations between the system and the reservoir are being established, one cannot use the Born–Markov approximation. If it is not possible to keep within the frames of approximation allowing one to describe completely the dynamics of reservoir parts, one has to develop a strategy for reducing the complexity of the problem by averaging over the reservoir in such a way as to capture the non-Markovian features of the system dynamics. It is a highly non-trivial task and is intensively pursued now. Recent interest in it was greatly stimulated by experimental progress in creating structured reservoirs such as photonic crystals or microcavities, and the ability to single out and manipulate individual quantum emitters, such as atoms, molecules and quantum dots. For them non-Markovian effects become essential. For example, the dynamics of emitters in photonic crystals might demonstrate very unusual properties like complete ‘freezing’ of upper-levels population after some initial period of decay (Bykov, 1975; John & Wang, 1990). An excitation and coherent control of quantum dots might be strongly affected by nonMarkovian effects of interaction with its surrounding, in particular, the rate of populating oscillations damping becomes dependent on the amplitude of the driving field (Mogilevtsev, Nizovtsev, Kilin, Brandi, Cavalcanti, & Oliveira, 2008; Zrenner, Beham, Stufler, Findels, Bichler, & Abstreiter, 2002). In the present work we review modern theoretical methods suitable for tackling a problem of system-reservoir interaction when the reservoir is exhibiting non-Markovian properties. Here we are trying to represent these methods from a practical point of view, i.e. to present a list of recipes for one facing a particular problem and trying to find a way of obtaining a valid and consistent solution of it. To this end, we demonstrate the machinery of a particular method, and the arguments in its favour, on examples of a few chosen specific models. Namely, we are using here a range of models consisting of a two-level system (TLS) interacting with a reservoir. For these models, the Hamiltonian
92
Theoretical Tools for Quantum Optics in Structured Media
of the system-reservoir interaction, V(t), is represented in the following general form E V(t) = h¯ σE · R(t),
(1)
E where the operator R(t) describes the reservoir (which we shall specify in every particular case); the elements of the vector σE are the operators σ + , σ − and σ z , corresponding to the TLS. The TLS is described by the vectors |+i and |−i corresponding to the upper and lower states; the TLS rising and lowering operators are σ ± = |±ih∓| and σ z = σ + σ − − 1/2. The model described by Equation (1), generally, does not allow for an exact solution. On the other hand, it is of sufficient simplicity for an illustrative demonstration of a whole range of approaches for tackling non-Markovian reservoirs.
2. EXACT AND APPROXIMATE SOLUTIONS FOR SYSTEM PLUS RESERVOIR Exact solutions for the total system plus reservoir dynamics are very valuable for developing theoretical tools. They are starting points for approximations. Also, they allow one to check the validity of developed approaches and to distinguish specific features of dynamics, which might help in specifying an approximation that one can and should apply to build an approach beyond the range of validity of an exact solution. In practice, one can obtain an exact solution for the system plus reservoir dynamics if one can construct a closed system of equations for: (a) probability amplitudes or elements of the density matrix; (b) for system operators or their averages. Generally, one can build a solution of the (a)-type when a space of possible system plus reservoir states is of a finite dimension (for example, when one considers the TLS interacting with the reservoir represented by a finite number of TLS), or when, during the interaction time, the state of the system plus reservoir belongs to some sufficiently small subspace of the total space of possible states. A good indicator of such a possibility is the presence of an integral of motion, of some operator, I, commuting with the Hamiltonian of the problem, [I, H] = 0. If an initial state is an eigenstate of I, then the solution will also be an eigenstate of I corresponding to the same eigenvalue. An important example of I is the operator of the total excitation number in the system plus reservoir, which becomes an integral of motion when one assumes validity of the rotatingwave approximation (RWA), retaining in the interaction Hamiltonian only terms which correspond to simultaneous creation of excitation in one part of the system plus reservoir and annihilation of excitation in another part
Exact and Approximate Solutions for System Plus Reservoir
93
of it (a typical example of the problem solved this way is a spontaneous emission of the TLS into the field reservoir (Louisell, 1977); it is considered below). An example of the case when it is possible to find a solution of the (b)-type is the presence of a set of operators, A j , such that [A j , H] =
X
I jk Ak ,
k
where all the operators I jk commute with the Hamiltonian. A solution for the Jaynes–Cummings model (the TLS plus a resonant quantized mode of the field) can be found this way (Ackerhalt & Rza¸zewski, 1975; Jaynes & ˙ Cummings, 1963) Sometimes, a possibility to find solutions for operators makes it convenient to find an exact solution for a time-evolution operator. For example, a spin-boson model describing dephasing of the TLS due to interaction with the bosonic reservoir can be solved in such a manner (Palma, Suominen, & Ekert, 1996; Unruh, 1995).
2.1. Spontaneous Emission into a Bosonic Reservoir: The Problem As an example of a problem allowing us to obtain a closed set of equations for the probability amplitudes, we consider an interaction of the TLS with the reservoir of electromagnetic field modes. This model describes a fundamental problem of spontaneous emission. Also, it is methodologically important, both for developing ways to describe field reservoirs, and for explaining a difference between what we will call in the subsequent discussion ‘Markovian’ and ‘non-Markovian’ reservoirs. Let us assume the dipole approximation and take the interaction Hamiltonian as a scalar product E · E(E E r , t) V = −d E = d(σ E + + σ − ), where dE is the vector of the TLS dipole moment operator, d E r , t) of the dipole moment of the TLS, and the electric field operator, E(E taken at the TLS position, rE. Let us also assume that in the Coulomb gauge the electric field operator can be defined through the system of eigenmodes (identified by a general index j), uE j (E r ), which are defined as the eigensolutions of Maxwell’s equations as E r) = i E(E
X j
r
h¯ w j ∗ Ď uE j (E r )a j + uE j (E r )a j . 2
94
Theoretical Tools for Quantum Optics in Structured Media
In the rotating-wave (RW) approximation, when one neglects rapidly Ď oscillating terms proportional to σ + a j and σ − a j (which holds in the limit of a weak emitter-field interaction and high frequency TLS transition frequency, ω0 ; see, for example, Louisell (1977)) the Hamiltonian describing the problem is H = h¯ ω0 I + h¯
X
Ď
1 j a j a j + h¯
X
j
Ď g ∗j σ + a j + g j a j σ − ,
(2)
j
where 1 j = w j − ω0 is the detuning of the jth mode of the reservoir with the frequency w j from the TLS transition frequency; r gj = i
wj dE · uE j (E r) 2h¯ Ď
are interaction constants; a j and a j are photon annihilation and creation operators in jth reservoir mode. They satisfy the following Ď commutational relations: [a j , ak ] = δ jk . The operator I = σ +σ − +
X Ď ajaj j
commutes with the Hamiltonian (2). Existence of such an integral of motion defines the structure of the solution and the subspace to which it belongs in the course of dynamics. If the initial wave-vector, |9(0)i, is the eigenvector of I: I|9(0)i = M|9(0)i, then the time-dependent solution will also satisfy I|9(t > 0)i = M|9(t > 0)i, i.e. the solution will belong to the subspace corresponding to the number of excitations M. If the TLS is initially on the upper level, and the reservoir is the vacuum state, i.e. |9(0)i = |+i|vaci R , one has |9(t)i = B(t)|+i|vaci R +
X
A j (t)|−i|1 j i R ,
(3)
j
where the vector |1 j i R corresponds to the state of the reservoir with a single excitation of the jth mode and vacuum in the other modes. In the frame rotating with the TLS transition frequency, the wave-vector of
Exact and Approximate Solutions for System Plus Reservoir
95
the form (3) leads to the following set of equations for the probability amplitudes X d g ∗j A j (t), B(t) = −i dt j d A j (t) = −i1 j A j (t) − ig j B(t). dt
(4)
From this system one obtains a single integro-differential equation for the probability amplitude, B(t), to find the TLS on the upper level d B(t) = − dt
t
Z
dτ K (t − τ )B(τ ),
(5)
0
where K (t) =
X
|g j |2 exp{−i1 j t}
(6)
j
is the reservoir correlation function. Equation (5) is formally equivalent to the system (4). However, from both the methodological and practical points of view it is quite different. To solve the system (4) one needs to know a set of reservoir mode frequencies and interaction constants. In reality, the set of electromagnetic modes is infinite and not discrete. Thus, to solve the system (4) in practice, one needs to approximate a reservoir by a set of a finite number of modes (Busch, Vats, John, & Sanders, 2000; Nikolopoulos & Lambropoulos, 2000). It is a non-trivial procedure, and can lead to a number of errors, especially for long times (Mogilevtsev, Moreira, Cavalcanti, & Kilin, 2006). Whereas Equation (5) shows that for the description of the TLS spontaneous emission into the vacuum of a field reservoir one needs to know only one scalar function K (t). One needs to notice that the properties of K (t) define a kind of approximation which one can apply for a description of emitter-reservoir interactions; this issue will be addressed throughout this chapter.
2.2. The Density-of-States and Photonic Band-Gaps For calculations it is often more convenient to characterize the reservoir not by the time-dependent reservoir correlation function (6), but by it’s Fourier-transform, ρd (w), introduced as K (t) =
Z
+∞
−∞
dwρd (w) exp{−i(w − ω0 )t}.
(7)
96
Theoretical Tools for Quantum Optics in Structured Media
For non-absorbing dielectric media ρd (w) has a simple physical meaning. It is the projected localized density of reservoir states (PLDOS), which defines with how many modes a dipole can interact in the unit frequency interval around its transition and unit volume around the point of its location (Busch, Vats, John, & Sanders, 2000; Sprik, Tiggelen, & Lagendijk, 1996). Notice that to give the quantity ρd (w) a meaning for the density-ofstates one must require that ρd (w) ≡ 0 for w ≤ 0. Generally, calculation of ρd (w) is a challenging computational task. For example, to estimate the PLDOS for the infinite periodic dielectric media one needs to calculate a system of eigensolutions of Maxwell’s equations for a large number of points in the first Brillouin zone, chosen in such a way as to represent, with sufficient precision, an integration over the Brillouin zone (Vats, John, & Busch, 2002; Wang, Gu, Wang, & Xu, 2003a; Wang, Wang, Gu, & Yang, 2003b). Alternatively, one can define the PLDOS operationally, using the fact that for a finite-sized lossless dielectric structure the following relation holds (D’Aguanno, Mattiucci, Centini, Scalora, & Bloemer, 2004; Xu, Lee, & Yariv, 2000): ρd (w) = ρdvac (w)
W (w) , W vac (w)
(8)
where W (w) is the radiation total power emitted in the volume of the structure; W vac (w) the power emitted in the same volume in the unstructured vacuum, and E2 ρdvac (w) = |d|
w3 , 12h¯ ε0 π 2 c3
(9)
is the PLDOS for the vacuum; ε0 stands for the permittivity of the vacuum. The PLDOS defined in this manner does not require finding eigensolutions of Maxwell’s equations; so, one may use, for example, the finite difference time domain method to find it (Rao & Hughes, 2007). As one can see from Equation (9), for the vacuum the PLDOS is smooth; it is zero only at w = 0. Now it is possible to design structures which can (ideally) achieve zero valued PLDOS for w 6= 0. They are so called ‘photonic crystals’ – periodic sets of scatterers – arranged in such a manner that in certain frequency intervals propagation of electromagnetic waves through them is impossible (John, 1987; Yablonovitch, 1987). As we shall see, the spontaneous dynamics of emitters in photonic crystals can be drastically different from that in unstructured media.
Exact and Approximate Solutions for System Plus Reservoir
97
2.3. Spontaneous Dynamics and the Markovian Approximation Generally, a solution of Equation (5) can be represented as (see, for example (Kofman, Kurizki, & Sherman, 1994)): ∞
Z
(10)
dw F(w) exp{−iwt},
B(t) = 0
where ρd (w) F(w) = π
"
w − ω0 − P
Z
∞
0
dxρd (x) x −w
2
#−1 + π 2 ρd2 (w)
,
(11)
and P denotes the principal value of an integral. Equation (10) gives clear hints as to what one should expect in spontaneous dynamics. In particular, a complete decay B(t)|t→+∞ → 0 must occur if the function F(w) is continuous (Krylov & Fock, 1947). If the function F(w) is singular, a complete decay will not take place. This will happen if the following conditions are simultaneously fulfilled (Mogilevtsev & Kilin, 2002): ρd (w) = 0 for w 6= 0, Z ∞ ρd (x) w − ω0 − P dx = 0. x −w 0 In that case a ‘partial’ decay (‘freezing of decay’) might occur, when after some initial period of decay a population of the TLS upper level comes to a stationary value. In particular, it happens when the TLS transition frequency is situated close to the border separating regions with zero and non-zero PLDOS (so-called ‘band-edge’), and is accompanied by creation of the field-emitter bound state (FEBS) (Fain, 1988; John & Wang, 1990, 1991). In Figure 1 an illustration of the upper-level population is given for cases when the TLS transition frequency is far from the band-edge and when it is close to it. One can notice that in the case where ω0 is far from band-edges or other sharp structural features of the PLDOS, the dynamics of the TLS upper-level population seems similar to the exponential decay law. Indeed, for a smooth PLDOS the frequency shift, θ (w) = P
Z
∞
dx 0
ρd (x) x −w
98
Theoretical Tools for Quantum Optics in Structured Media 1
ρd (w)
0.9 0.8
Population
0.7
ω0
0.6 0.5 0.4
ρd (w)
0.3 0.2
ω0
0.1 0
0
5
10
15
Time
FIGURE 1 Examples of the TLS upper-state population dynamics for the TLS transition frequency, ω0 , close to the band-edge (upper thick solid curve and upper inset, depicting the density-of-states and the respective position of ω0 ), and for the TLs transition frequency far from the band-edges (the lower thick solid curve and the lower inset). Thin solid line depicts the absolute value of the normalized reservoir correlation function, |K (t)|, for the case when ω0 is close to the band-edge. Time is given in units of inverse height of the rectangular density-of-states used for the example
in Equation (11) only weakly depends on w (Louisell, 1977; Weisskopf & Wigner, 1930). Thus, neglecting this dependence one arrives at the exponential decay law B(t) ≈ exp{−iθ (ω0 )t − πρd (ω0 )t}. However, the nature of an approximation leading to the exponential decay ¨ law should be carefully considered. First of all, from the Schrodinger equation (or from Equation (10)) it follows that for short times the dynamics cannot be exponential. Indeed, for πρd (ω0 )t 1 the exponential decay law for the TLS lower level population yields the following result: 1 − |P(t)|2 ∝ t. ¨ Whereas the Schrodinger equations dictates 1 − |P(t)|2 ∝ t 2 .
Exact and Approximate Solutions for System Plus Reservoir
99
This feature precisely underlies the so-called ‘quantum Zeno effect’, when frequent measurements, checking whether the TLS decayed or not, ‘freeze’ the TLS on the upper level (Khalfin, 1968; Sudarshan & Misra, 1977). Also, for long times πρd (ω0 )t 1 the decay dynamics deviate strongly from the exponential law and acquire a dependence like |P(t)|2 ∝ t −x , where x > 0 (Khalfin, 1958). Actually, the emitter might undergo a spontaneous decay of a form quite different from the exponential one for all evolution times (Jittoh, Matsumoto, Sato, Sato, & Takeda, 2005). The key to understanding the kind of evolution that one might expect for the TLS is given by the temporal properties of the reservoir correlation function, K (t). When ω0 is far from regions of sharp change of the PLDOS, K (t) decreases much faster than the TLS population decays. However, when one has ω0 close to the band-edge, the decrease time of K (t) is comparable with the time-scale of the population dynamics (see Figure 1). The time of K (t) decay is termed as the ‘reservoir correlation time’, because, as can be seen below, it is the decay time of inter-reservoir correlations arising in the course of the emitter-reservoir interaction. As seen from Equation (5), an approximation leading to the exponential decay law is equivalent to assuming a reservoir correlation time negligibly small on the time-scale of the TLS dynamics, i.e. K (t − τ ) ∝ δ(t − τ ). This approximation reduces the time-nonlocal Equation (5) to the time-local one, assuming ‘no memory effects’ of the reservoir, and for that reason an approximation of negligible reservoir correlation time is commonly called the ‘Markovian approximation’. The main goal of this work is to describe methods suitable for tackling problems arising when this approximation is invalid.
2.4. The Cluster Expansion and Collective Operators From the Hamiltonian (2), in the frame rotating with the TLS transition frequency, ω0 , one can obtain the following set of Heisenberg equations for the TLS and the reservoir operators: d a j = −i1 j a j − ig j σ − , dt X d − σ = −i(2σ + σ − − 1) g ∗j a j , dt j X d + − Ď σ σ a j = −i1 j σ + σ − a j − i gk∗ (σ + a j ak − ak σ − a j ), dt k
(12)
It can be seen from the system (12) that any equation for a product of the TLS and the reservoir operators contains products of higher order. So, generally, the system of equations (12) for such products is
100
Theoretical Tools for Quantum Optics in Structured Media
infinite. For the case of the reservoir consisting of only a single mode (Jaynes–Cummings system) one still can obtain an exact solution for operators (Ackerhalt & Rza¸zewski, 1975). However, for an arbitrary ˙ number of reservoir modes an exact solution of such a kind is not possible. Yet, the system of equations might give one a clue as how to truncate it. Indeed, one can notice that an equation for a product of M excitation creating operators and M + 1 excitation destroying operators (such as, for Ď example, (a j ) M (ak ) M σ − ) will contain products of M +1 excitation creating operators and M +2 excitation destroying operators (such as, for example, Ď (a j ) M+1 (ak ) M+1 σ − ). So, if the initial state of the system contains no more than M + 1 excitations, terms with M + 2 excitation annihilation operators will give zeros when acting on the TLS plus reservoir state. Thus, one can obtain an exact truncated set of equations for averages of operator products. It is to be noted that the example given above is a particular case of the so-called ‘cluster expansion’ method for treating a system-reservoir interaction, widely applied in quantum chemistry, many-body problems and problem of an emitter interaction with radiative and phonon reservoirs in semiconductor physics (Cyzek, 1966; Harris, Monkhorst, & Freeman, 1992; Kira & Koch, 2006). This method is generally suitable when equations for correlation functions of a certain order (i.e. quantum mechanical averages of operator products) contain correlation functions of higher order, and it is possible to assume that these correlations are not negligible only up to some order. For our TLS plus reservoir problem, an applicability of the cluster-expansion approximation means also that ¨ (going back to the Schrodinger equation) the density matrix of a system can be truncated and limited to some subspace corresponding to no more than a certain finite number of excitations. However, here one has to note that the number of equations to solve in cluster-like approximations grows quickly with the increase of the dimension of these subspaces. Indeed, for example, for the TLS interacting with N field modes, and for an initial state with M excitations one has X
|9(t)i = P
n j =M−1
+
X P
n j =M
B(n 1 . . . n N , t)
Y
|n j i|+i
j
A(n 1 . . . n N , t)
Y
|n j i|−i.
j
Thus, one has to solve about N M equations for probability amplitudes. The same holds for the number of equations for correlation functions in the cluster expansion. For a large number of systems in the reservoir the problem becomes intractable even for few excitations in the reservoir.
Exact and Approximate Solutions for System Plus Reservoir
101
In practice, one usually limits oneself to a correlation function up to the third order (Kira & Koch, 2006). One also might add here, that for the problem of the TLS spontaneous emission into the field reservoir, exact solutions were obtained with up to three excitations in the reservoir (Mogilevtsev & Kilin, 2002; Nikolopoulos, Bay, & Lambropoulos, 1999). The number of equations to be solved can be reduced if it is possible to assume that some part of the reservoir is only weakly excited (for example, with frequencies far from the TLS transition frequency), or neglect some correlation functions (i.e. corresponding to correlation of spectrally distant parts of the reservoir). Such a possibility gives rise to a number of approximate methods which are considered further in this chapter. For the spontaneous emission problem, one can give a clear illustration of the ‘cluster expansion’ approach for solving equations for operator averages, with the help of collective operators introduced in such a way as to represent the Hamiltonian (2) in the diagonal form (Loudon, 1973):
H=
N X
Ď
λjCjCj,
C j = U0 j σ − +
j=0
N X
Uk j bk ,
(13)
k=1
where the elements, Uk j , of the unitary matrix U are
Uk j
gk U0 j = , λ j − wk
" k > 0;
U0 j = 1 +
N X k=1
|gk |2 (λ j − wk )2
#−1/2 ,
(14)
and eigenvalues, λ j , are given by the following equation
λ j = ω0 +
N X k=1
|gk |2 . λ j − wk
(15)
The collective operators, C j , satisfy the equation of motion in the form N X d Ď C j = −iλ j C j + 2iU0 j λk U0k U0l U0m Cl Ck Cm , dt k,l,m=0
(16)
which directly points to the representation of the solution in terms of a
102
Theoretical Tools for Quantum Optics in Structured Media
cluster expansion over collective operator products (Mogilevtsev & Kilin, 2002): C j (t) = C j (0)e−iλ j t +
(2)
Ď
f klm (t)Cl (0)Ck (0)Cm (0)
k,l,m=0 Ď Ď (3) f klmnp (t)Cl (0)Cn (0)Ck (0)Cm (0)C p (0) + · · ·
N X
+
N X
(17)
k,l,m,n, p=0 (M)
where f kl... (t) are time-dependent scalar functions. It is interesting to note that in the single-excitation subspace (i.e. for the spontaneous emission problem considered above) the formalism of collective operators also gives a clear illustration of the mechanism underlying ‘freezing’ of emitter population decay. Indeed, if the TLS is initially in the upper level, and the reservoir is in the vacuum state, from Equation (17) one has Ď
Ď
hCn (t)Ck (t)i = hCn (0)Ck (0)i exp{i(λn − λk )t}. Thus, the upper-level population of the TLS can be expressed as 2 N X 2 2 |B(t)| = U00 exp{−iλ0 t} + U0 j exp{−iλ j t} . j=1 2
(18)
The feature underlying ‘freezing’ is an appearance of an eigenvalue, i.e. λ0 , inside the gap, and a sharply pronounced dominance of the coefficient, U00 , corresponding to it, |U00 |2 |U0 j |2 for j > 0 (which is illustrated in Figure 2). So, from Equation (18) it follows that the asymptotic ‘frozen’ upper-level population is given by U00 : |B(t → +∞)|2 = |U00 |4 . Also, the collective operators approach allows one to illustrate the physics underlying ‘freezing’: only one from the single-excitation eigenstates of the Hamiltonian (2), |ψ j i = U0 j |+i|vaci R +
N X
(19)
Uk j |−i|1k i R
k=1
shows a significant degree of the TLS-field entanglement. Indeed, one may represent the field parts of each collective operator as bosonic collective operators: C j = U0 j σ − +
q
1 − U02j B j ,
Bj =
X k=1
bk q
Uk j 1 − U02j
.
(20)
Exact and Approximate Solutions for System Plus Reservoir
103
–3
2
x 10
1.8
2
U 00 =0.7784
1.6 1.4
ρd(w)
2
U0w
1.2 1
ω0
0.8 0.6 0.4 0.2 0
λ0 2
4
6 w– ω 0
8
10
FIGURE 2 An example of the coefficients U02j distribution for the density-of-states shown in the inset and for the TLS transition frequency, ω0 , close to the band-edge. Values of the detunings, w − ω0 , are given in units of the height of the rectangular density-of-states taken for the example.
Introducing, with the help of B j , the single-photon Fock states, |1 j i = Ď B j |vaci R , and representing, with the help of them, states (19), one sees that q the concurrence of the state |ψ j i is 2U0 j 1 − U02j . Thus, the concurrence of the state |ψ0 i can far exceed the concurrencies of other states |ψ j i, j > 0. The state |ψ0 i corresponds to the ideal ‘field-emitter bound state’ (FEBS), which describes a field localized in the spatial vicinity of the emitter and entangled with it (Kofman, Kurizki, & Sherman, 1994; Mogilevtsev, Kilin, & Cavalcanti, 2004a). It will be shown below that identification of this state allows one to separate the reservoir into parts ‘strongly’ and ‘weakly’ interacting with the TLS, and thus to develop an effective approximation. In particular, it allows us to build a cluster approximation for the problem of the TLS-reservoir interaction in the presence of additional reservoirs which produce Markovian-like dynamics of the TLS in the absence of the first reservoir (Mogilevtsev, Moreira, Cavalcanti, & Kilin, 2006).
2.5. The Pseudo-Modes Method The previous subsection demonstrates that to make a solution of the problem easier, sometimes it is advantageous to consider, instead of the original set of reservoir modes, a set of modal superpositions (like those in the expression for the collective operators (16)). For particular
104
Theoretical Tools for Quantum Optics in Structured Media
densities-of-states one can reach very significant simplification of the problem in this way. The essence of the method discussed here is in the following observation: if the PLDOS, ρd (w), has only simple poles on the lower part of the complex plane of w-values, and this set can be numbered (say, poles at wl − iγl with corresponding residues rl /2π ), then, assuming ρd (w) negligibly small for w wl , ∀l, and extending the integration in Equation (7) (neglecting any asymptotical deviation in dynamics occurring because of this, see Section 2.3) one can represent the reservoir correlation function as K (t) = −i
X
rl exp{−(γl + i(wl − ω0 ))t}.
(21)
l
Then, introducing collective amplitudes Z t p A¯ l (t) = −i −irl dτ exp{−(γl + i(wl − ω0 ))(t − τ )}B(τ ),
(22)
0
one can re-write the system Equation (4) as a system of equations for these ˘ collective amplitudes (Garraway, 1997a; Imamoglu, 1994): X d B(t) = −i g¯l A¯ l (t), dt l d ¯ Al (t) = (−i1l − γl ) A¯ l (t) − ig¯l B(t), dt
(23)
√ where g¯l = −irl . Obviously, if the number of poles is not too large, then the system (23) is much simpler to solve than the system (4). Also, one does not need to discretize the density-of-states: the reservoir structure is already accounted for by the representation (22). Since Equation (22) bears formal similarity to the expression for the single-photon modal amplitude, A j (t) = −ig j
Z
t
dτ exp{−i(w j − ω0 )(t − τ )}B(τ ),
0
the amplitude A¯ l (t) was termed an amplitude of the so called ‘pseudo˘ mode’ (Garraway, 1997a,b; Imamoglu, 1994). This pseudo-mode might correspond to the physical entity. For example, for a leaky cavity and the TLS transition frequency close to the cavity resonance, the PLDOS can be
Exact and Approximate Solutions for System Plus Reservoir
105
described by a Lorentzian: ρd (w) ∝
γ , (w − w p )2 + (γ /2)2
where w p is the centre frequency of the PLDOS, and γ describes the width of the PLDOS. This PLDOS has just one simple pole in the lower half-plane, so one will have only one collective amplitude in the system (23). One can consider this amplitude as a probability amplitude having a photon in a collective field mode, and reformulate the problem as a problem of interaction of the TLS with just one quantized cavity mode (‘quasi-mode’), which is coupled to the external reservoir (Dalton, Barnett, ˘ 1994). By establishing a & Garraway, 2001; Garraway, 1997a,b; Imamoglu, correspondence between pseudo-modes and quasi-modes there exists a possibility of deriving master equations, valid even for cases of multiple excitation of the reservoir (Dalton, Barnett, & Garraway, 2001). But, one should notice that, in general, pseudo-modes do not correspond to realistic damped quantized field modes, because the interaction constants in Equation (23), g¯l , can be complex (for example, for the case where one has the PLDOS consisting of Lorentzians with positive and negative weights having the gap at a single point (Garraway, 1997b)). In Section 4 we shall consider the derivation of a master equation based on the method of pseudo-modes.
2.6. Dephasing into a Bosonic Reservoir In conclusion of this section we consider a simple exactly solvable model, which is relevant for a wide range of actual problems, and demonstrates the rather subtle interplay between the Markovian and non-Markovian features of the TLS dynamics. It is a model of the TLS dispersively coupled to the bosonic reservoir. It is described by the following Hamiltonian: H=
X X 1 1 Ď Ď h¯ ω0 σ z + h¯ w j a j a j + h¯ σ z g ∗j a j + g j a j . 2 2 j j
(24)
The model described by the Hamiltonian (24) was investigated as a simple example of a decoherence influence on a quantum computation (DiVincenco, 1995; Palma, Suominen, & Ekert, 1996; Unruh, 1995). Also, it has a direct correspondence with the experimental reality describing the dephasing of localized semiconductor structures due to interaction with the reservoir of phonons (such as quantum dots and shallow donor defects (Brandes, 2005; Hayashi, Fujisawa, Cheong, Jeong, & Hirayama, 2003; Kira & Koch, 2006; Zrenner, Beham, Stufler, Findels, Bichler, & Abstreiter, 2002).
106
Theoretical Tools for Quantum Optics in Structured Media
The Hamiltonian (24) allows one to obtain exact solutions for operators describing the whole system. Obviously, populations of the TLS levels stay constant, σ z (t) = σ z (0), and one can simply obtain the following solution for operators: a j (t) = a j (0) exp{−iw j t} − α j (t)σ z (0), (25) ( ) X Ď σ + (t) = σ + (0) exp {iω0 t + iδω(t)} exp i (α j (t)a j (0) + α ∗j (t)a j (0)) , j
where the time-dependent frequency shift is δω(t) =
t
Z
dt¯ 0
t
Z
dτ I m(K (t¯ − τ ))θ (t¯ − τ ),
0
the functions are α j (t) =
gj (1 − exp{−iw j t}), 2w j
and K (t) is the reservoir correlation function (6); θ (x) is the step-function. One can see that conditions for Markovianity of the dephasing reservoir and exponentiality of the TLS polarization decay are more stringent in the current case than in the case of the spontaneous emission into the field reservoir. First of all, properties of the reservoir density-of-states near w = 0 become important. Now one cannot assume the density-of-state to be smooth enough to allow for the Markovian approximation, because dephasing can be strongly affected by features of the density-of-states, which were significant only for very long times, in the case of spontaneous emission. In addition, even a weak deviation from Markovianity can strongly influence the TLS dynamics, in the case, for example, when one drives the TLS with a sufficiently strong classical field (Apanasevich, Kilin, & Nizovtsev, 1988) (see Section 3). The effect of non-Markovianity and strong TLS-reservoir correlation, arising as a result of it, can be illustrated if one considers the unitary timeevolution operator in the interaction picture, ( ) Z t X i ← − Ď ∗ U(t) = T exp − σ z dτ (gk a j exp{iw j τ } + gk a j exp{−iw j τ }) , 2 0 j ← − where T denotes the operator of the time-ordering. The time-dependent operator in the exponential of this equation has a c-number commutator
Exact and Approximate Solutions for System Plus Reservoir
107
with itself in different time-moments, so that one can represent the operator U(t) as: U(t) = exp {−iδω(t)}
o n Ď exp σ z (α j (t)a j + α ∗j (t)a j ) .
Y
(26)
j
If the reservoir is initially in the vacuum state, and the TLS is initially uncorrelated with it, for example, |9(0)i = (cos ϕ|+i + sin ϕ|−i)
Y
|0 j i,
j
then, in the process of interaction, the reservoir modes are in the superpositions of coherent states with opposite phases, |α j (t)i and | − α j (t)i, corresponding to the TLS level entangled with it: |9(t)i exp {iδω(t)} = cos ϕ|+i
Y
|α j (t)i + sin ϕ|−i
j
Y
| − α j (t)i.
(27)
j
In the case where |α j (t)| ≥ 1, the state (27) exhibits a significant emitterfield entanglement, and by projecting this state on some specific emitter ¨ state (for example, sin ϕ|+i + cos ϕ|−i), one obtains Schrodinger-cat states of the reservoir modes. However, the density-of-states must be sufficiently narrow to achieve this. One might add here that the Hamiltonian (24) can be viewed as a particular example of a dispersive emitter-field interaction scheme of a kind developed recently in cavity quantum electrodynamics for generating large self-Kerr and cross-Kerr nonlinearities (Brandao, ˘ Hartmann, & Plenio, 2008; Fleischhauer, Imamoglu, & Marangos, 2005). Such schemes can indeed be used for generating non-classical quantum ¨ states and Schrodinger-cat states in particular (Glancy & de Vasconcelos, 2008; Tyc & Korolkova, 2008). It is worth mentioning that, for certain states of the reservoir, the Markovian approximation is applicable even for sharply structured density-of-states. Let us take, for example, the thermal initial state described by the following density matrix ( ρ r es (0) = N
−1
exp h¯ β
X j
Ď wjajaj
)
" ,
(
N = Tr exp h¯ β
X
Ď wjajaj
)# ,
j
where β = 1/k B T , T is the temperature and k B is the Boltzmann constant.
108
Theoretical Tools for Quantum Optics in Structured Media
Then, by averaging the solution (25) over the reservoir, we get σ + (t) = σ + (0) exp {iω0 t + iδω(t) − γ (t)} , where γ (t) =
X |g j |2 j
w2j
coth
h¯ w j 2k B T
(1 − cos(w j t)).
(28)
Let us consider the dephasing rate, γ (t), for Ohmic density-of-states, ρ(w) = Dw exp{−w/ }, commonly used in spin-boson models, where D is the dimensionless constant describing the TLS reservoir interaction strength, and is the cut-off frequency describing the effective width of the reservoir spectrum. Then, assuming the effective width far exceeds the inverse thermal correlation time, t B = h¯ β/π, one obtains from Equation (28) the following equation exp{−w/ } h¯ w γ (t) = D dw coth (1 − cos(wt)) w 2k B T 0 o sinh(t/t B ) D n ln 1 + 2 t 2 + D ln ≈ . 2 t/t B Z
+∞
(29)
From Equation (29) one can see that for such small temperatures or times that t t B , the TLS polarization dynamics is far from simple exponential decay. Indeed, for very small times, t −1 , the decay rate is proportional to the square of the time, γ (t) ∼ 2 t 2 (which is a correct quantum-mechanical prediction). If the times are not so small, but still far less than the thermal time, −1 t t B , one has a power correlation D law decay, γ (t) ≈ ln (t) . And only if the time far exceeds the thermal correlation time, t t B , does the Markovianity emerge: γ (t) ≈ Dt/t B . Obviously, the higher the temperature is, the quicker the Markovian regime will appear. It is for that reason that the ‘high-temperature limit’ in spin-boson models is commonly being equalled to an assumption of Markovianity. In that regime the influence of the spectral width of the reservoir is negligible, and for arbitrarily narrow densities-of-states one is able to reach the Markovian regime (and obliterate system-reservoir correlations) just by increasing the temperature.
Master Equations: Deterministic Approaches
109
3. MASTER EQUATIONS: DETERMINISTIC APPROACHES As it was discussed in the previous section, one is seldom able to find exact solutions for both the system and the reservoir variables. However, a problem of the system-reservoir interaction in a wide variety of practical cases can be solved with the help of averaging over the reservoir or a part of it. Thus one can reduce an initial problem to one of treating only a few variables describing a system, with the help of the ‘reduced’ density matrix averaged over reservoir variables. It is to be noted here that in a few cases one is able to perform an averaging over the reservoir exactly. As naturally expected, it can be done when an exact solution for the systemreservoir variables can be found (for example, such an averaging can be done for the spontaneous emission of the TLS into the vacuum of the field reservoir (Anastopoulos & Hu, 2000), or for the dephasing example considered in Section 2.6 (Doll, Zueco, Wubs, Kohler, & Hanggi, 2008)). Sometimes, it is possible to derive a master equation for the reduced density matrix in a closed form, in cases where the exact solution cannot be found by other methods (for example, when the reservoir is the bosonic one in a Gaussian initial state, and the system-reservoir coupling part of the Hamiltonian is V ∝ SR(t), where S is the system’s operator, and R(t) describes the reservoir (Ishizaki & Tanimura, 2008)). However, in the majority of practical cases one can perform an averaging over the reservoir by only using certain approximations. These approximations imply general assumptions about the respective strength of the system-reservoir interaction (it is usually supposed to be weak), the state of the reservoir, and about the reservoir correlation time. We will next illustrate the whole set of implications in obtaining the master equation for the reduced density matrix, using a traditional textbook example (see (Louisell, 1977; Scully & Zubairy, 1997)). Let the problem be described by the Hamiltonian H = H0 + λV,
(30)
where H0 describes the system and reservoir in the absence of interaction, V is the interaction Hamiltonian, and λ is the parameter describing a respective ‘strength’ of the interaction. In the interaction picture one has the following equation for the complete system+reservoir density matrix: d i ρ c (t) = − λ[V(t), ρ c (t)] = λL(t)ρ c (t), dt h¯
(31)
where we have introduced the Liouvillian superoperator L(t). Formally solving this equation and substituting the result in the right hand side of
110
Theoretical Tools for Quantum Optics in Structured Media
Equation (31) yields d ρ (t) = λL(t)ρ c (0) + λ2 dt c
t
Z 0
dτ L(t)L(τ )ρ c (τ ).
(32)
Now let us start the sequence of approximations. Firstly, let us suppose that the initial states of the system (described by the density matrix ρ) and the reservoir (described by the density matrix ρ r es ) are independent, ρ c (0) = ρ(0) ⊗ ρ r es (0). Secondly, let us assume that the system-reservoir interaction is so weak that the state of the reservoir is only weakly affected, ¯ ⊗ ρ r es (0) + O(λ). ρ c (t) ≈ ρ(t) We have assumed that changes induced by the interaction (systemreservoir entanglement and, in general, a deviation of the reservoir state from the initial one) are of the order of λ. This assumption constitutes the so-called ‘Born approximation’. Averaging over the reservoir (which is denoted by the trace operation Trr {. . .}), and neglecting terms ∼ O(λ3 ), we come to the closed equation for the reduced density matrix d ¯ ρ(t) ≈ λTrr {L(t)ρ r es (0)}ρ(0) dt Z t ¯ ) ⊗ ρ r es (0)}. + λ2 dτ Trr {L(t)L(τ )ρ(τ
(33)
0
This equation can already be (and quite commonly is) used in practice for estimating the reduced density matrix. Further simplification can be made if one makes an assumption about the correlation time of the reservoir. If this time is much shorter than the typical times of the system dynamics, ¯ ) ≈ ρ(t) ¯ one commonly assumes ρ(τ in the integrand, such that Equation (33) reduces to a time-local form (i.e. ‘Redfield equation’ (Redfild, 1955)). Assuming that one considers evolution for times greatly exceeding the reservoir correlation time, one might replace the upper limit of integration by infinity (which completes the set of assumptions commonly known as ‘the Born–Markovian approximation’). The procedure described above can be considered as a practical guideline for derivation of master equations. However, in each particular case care should be taken in making approximations. Actually, ‘weakness’ of the system-reservoir interaction does not necessarily imply absence
Master Equations: Deterministic Approaches
111
of long-time system-reservoir correlations (as was shown in Sections 2.3 and 2.6). Besides, Equation (33) can violate the condition of positive semi-definiteness of the reduced density matrix (Barnett & Stenholm, 2001). And even the possibility of assuming the reservoir correlation time to be negligible might not preclude the possibility of accounting for the reservoir structure (for example, see Florescu and John (2004); Mogilevtsev, Nizovtsev, Kilin, Brandi, Cavalcanti, and Oliveira (2008)). In this section we describe a number of more strict and controllable ways of deriving master equations, analyzing the validity of approximations and the applicability of the derived equations for dealing with the reservoir structure.
3.1. Projection Operator Method and the Cumulant Expansion As a particular realization of the idea of looking for a reduced density matrix, while the reservoir state during the system-reservoir interaction remains close to a certain known state, let us consider here such a flexible and universal tool as the projection operator technique (Nakajima, 1958; Zwanzig, 1960). To realize it, one introduces the projection operator, P, acting on the system plus reservoir density matrix in the following way: Pρ c = Trr {ρ c } ⊗ ρ r .
(34)
This operator satisfies the condition P 2 = P, and the reduced density matrix is ρ¯ = Trr {Pρ c }. Introducing another projection operator Q = 1−P, from Equation (31) one obtains the system of equations d Pρ c (t) = λPL(t)Pρ c (t) + λPL(t)Qρ c (t), dt d Qρ c (t) = λQL(t)Pρ c (t) + λQL(t)Qρ c (t). dt
(35) (36)
Solving Equation (36) for Qρ c (t) and substituting the result in Equation (35) one gets d Pρ c (t) = λPL(t)Pρ c (t) + dt
Z
t
dτ K(t, τ )Pρ c (τ ) + F(t, 0)ρ c (0), (37)
t0
where K(t, τ ) = λF(t, τ )L(τ ),
Z t ← − F(t, τ ) = λPL(t) T exp λ dxQL(x) Q. τ
112
Theoretical Tools for Quantum Optics in Structured Media
By averaging Equation (37) over the reservoir, it is possible to obtain a closed master equation for the reduced density matrix. Generally, due to the impossibility of performing an averaging over the reservoir explicitly (apart from the few cases mentioned earlier), such an equation is still quite far from being really useful for practical calculations. However, it allows one to develop approximations in a more controlled manner, in particular, to put a stricter meaning into empirical expansion in powers of the parameter λ used in the beginning of this section. One can expand F(t) as F(t, τ ) = λPL(t)Q + λ PL(t) 2
Z τ
t
dxL(x)Q + · · ·
(38)
When the reservoir state, described by the density matrix ρ r , is initially uncorrelated with the system, one has Qρ c (0) = 0, and the nonhomogeneous term in Equation (37) disappears. Also, in practice, it often holds that the projections of odd number of Liouvillians are zero, PL(t1 )L(t2 ) . . . L(t2k+1 )P = 0 (which holds, for example, for the bosonic reservoir initially in the thermal state and linear coupling, as in the Hamiltonian (2)). Then, in the expansion for K(t, τ )P, terms corresponding to the odd powers of λ disappear (together with the first term in the right-hand part of Equation (37)). Using the projection operator technique one is also able to derive a time-convolutionless form of the master equation (37). For that, one needs to express the matrix ρ c (τ ) through its value at a later time moment, t, with the help of the backward propagator, G(t, τ ), as ρ c (τ ) = G(τ, t)ρ c (t),
Z t − → G(τ, t) = T exp −λ dxL(x) , τ
− → where T denotes the operator of anti-chronological ordering. With the help of the backward propagator, it is possible to represent the solution for Qρ c (t) from the second equation of the system (36) in time-local form, and with it to obtain a time-convolutionless form of the master equation (Chaturvedi & Shibata, 1979; Shibata & Takahashi, 1977): d ¯ ¯ Pρ c (t) = K(t)Pρ c (t) + F(t)ρ c (0), dt where ¯ K(t) = λPL(t) [1 − J (t)]−1 P,
(39)
Master Equations: Deterministic Approaches
113
Z t − −1 ← ¯ dxQL(x) Q, F(t) = λPL(t) [1 − J (t)] T exp λ
(40)
0
and J (t) = λ
Rt 0
n R o ← − t dτ T exp λ τ dxQL(x) QL(τ )G(τ, t).
Both the time-nonlocal and time-convolutionless master equations can also be derived using the ordered cumulant expansion (Carruthers & Nieto, 1968; Kubo, 1962; Van Kampen, 1974, 2007). For a factorized initial state of the system and reservoir, describing the dynamics of the reduced density matrix with the help of the forward propagator averaged over the reservoir (Green’s matrix), G(t, 0), ¯ ρ(t) = G(t, 0)ρ(0),
G(t, 0) = Trr
Z t ← − T exp λ dxL(x) ρ r (0) , 0
one builds the cumulant expansion representing
ln G(t, 0) =
∞ X λn n=1
n!
Mn (t, 0),
(41)
dtn mn (t1 , . . . tn ).
(42)
where Mn (t, 0) =
t
Z
dt1 . . .
0
Z
tn−1
0
Operators mn (t1 , . . . tn ) are cumulants. They can be defined in different ways giving the possibility of obtaining both time-nonlocal and time-convolutionless forms of master equations (see, for example, (Apanasevich, Kilin, & Nizovtsev, 1988; Kilin & Nizovtsev, 1986)). Socalled partially ordered cumulants, defined as m1 (t1 ) = Trr L(t1 )ρ r (0) ,
m2 (t1 , t2 ) = Trr L(t1 )L(t2 )ρ r (0) , m3 (t1 , t2 , t3 ) = Trr L(t1 )L(t2 )L(t3 )ρ r (0) , m4 (t1 , t2 , t3 , t4 ) = Trr L(t1 )L(t2 )L(t3 )L(t4 )ρ r (0) − Trr L(t1 )L(t2 )ρ r (0) Trr L(t3 )L(t4 )ρ r (0) − Trr L(t1 )L(t3 )ρ r (0) Trr L(t2 )L(t4 )ρ r (0) − Trr L(t1 )L(t4 )ρ r (0) Trr L(t2 )L(t3 )ρ r (0) , ...
114
Theoretical Tools for Quantum Optics in Structured Media
allow one to obtain a time-convolutionless master equation equivalent to Equation (39).1 Provided PL(t)P = 0, the completely ordered cumulants mn (t1 , . . . tn ) = Trr L(t1 )QL(t2 )Q . . . QL(tn )ρ r (0) lead to the Green’s matrix satisfying Equation (37), and, as a consequence, to the time-nonlocal master equation coinciding with the one obtained via the projection operator method (Yoon & Deutch, 1975). Here one should notice that the expansion (41) need not always exist. For example, one cannot build it for some cases of strong system-reservoir coupling, when the reduced density matrix at time t = 0 is not uniquely defined through its value at t > 0 (Breuer & Petruccione, 2002). To conclude, from considerations given above it follows that, for derivation of an approximate master equation, one needs to hazard two guesses: firstly, one needs to guess correctly which state the reservoir is close to during the considered period of the system-reservoir interaction. Here one has to notice that, if one manages to guess such a state, then, even if it is a correlated reservoir-system state, there is a possibility to account for it by introducing correlated projection operators (Breuer, Gemmer, & Michel, 2006). For interaction with structured reservoirs such a state guessing is highly non-trivial. For example (as demonstrated in the previous Section), for bosonic reservoirs with band-gaps, one might end up with a strongly correlated system-reservoir state starting with a completely uncorrelated one. Secondly, one needs to guess how strong the interaction is, i.e. how many terms should be retained in the power series expansion in λ. Such a truncation itself turns out to be rather problematic. It might happen that the second-order approximation results in the equation coinciding with the exact one (Ishizaki & Tanimura, 2008). From other side, higher-order terms of the expansion can lead to timedivergent terms in the resulting master equation. Even well-behaved highorder terms do not guarantee that one is able to account for stronger and more ‘non-Markovian’ system-reservoir interactions with the help of them. Generally, the validity of the expansion in λ essentially depends on the reservoir correlation time. In addition, when truncating expansions, one is not guaranteed that the resulting equation will lead to a nonnegatively defined density matrix (Barnett & Stenholm, 2001).
3.2. Validity of Perturbative Approximations: Spontaneous Emission In this subsection we illustrate the conditions for the validity of approximate master equations and their applicability for treating 1 Of course, this equivalence holds only for exact equations. Approximations of the same order of λ can be different.
Master Equations: Deterministic Approaches
115
problems involving non-Markovian reservoirs, and compare the results obtained via different methods. In particular, we will focus on timenonlocal and time-convolutionless forms of the master equations in order to treat a TLS spontaneous emission into the vacuum of the field reservoir. Notice, that in the limit of zero reservoir correlation time, both forms lead to the same master equation. For our example one has a master equation of the following form d ¯ ρ(t)] ¯ ¯ ¯ ρ(t) = −i[H, + γ D(σ − )ρ(t), dt
(43)
¯ = δωσ + σ − , δω is the frequency shift, and the superoperator where H (‘dissipator’) of the operator x, is Ď Ď ¯ ¯ ¯ − ρ(t)x ¯ D(x)ρ(t) = 2xρ(t)x − xĎ xρ(t) x.
The shift, δω, and the damping rate, γ , are, respectively, the imaginary and real parts of the integrated reservoir correlation function, γ + iδω =
Z
+∞
dt K (t).
0
The master equation (43) is of so-called ‘Lindblad’ form, which guarantees positive semi-definiteness of the reduced density matrix (Lindblad, 1975). Under the Born approximation one has different master equations for time-nonlocal and time-convolutionless methods considered in the previous Subsection. For the time-nonlocal method from Equation (38), one has Z t d ¯ ρ(t)] ¯ ¯ ¯ )σ + ρ(t) = −i[H, + dτ 2Re{K (t − τ )}σ − ρ(τ dt 0 ¯ ) − K ∗ (t − τ )ρ(τ ¯ )σ + σ − , − K (t − τ )σ + σ − ρ(τ
(44)
and for the time-convolutionless approach, one has Equation (44) with ¯ ) replaced with ρ(t). ¯ ρ(τ These master equations look very different. In particular, it might seem that the time-nonlocal form accounts for some ‘memory effects’ of the reservoir, making the density matrix dependent on its value at earlier times, whereas the time-convolutionless equation is apparently unable to do that. In fact, within the validity range of the Born approximation one should expect similar results from both equations (at least, for times much exceeding the reservoir correlation time) (Paz & Zurek, 2001). One can illustrate a difference in the results produced by
116
Theoretical Tools for Quantum Optics in Structured Media 1
γ T = 0.1
0.9 0.8
Population
0.7 0.6 0.5
γ T=1
0.4 0.3 0.2 0.1
0
1
2
3 γt
4
5
6
FIGURE 3 Example of the TLS population dynamics as given for the exact solution (47) (solid lines); time-nonlocal solution (dotted lines) and time-convolutionless solution (dashed lines).
both equations in the case of a strongly non-Markovian reservoir with the help of a simple ‘kick-back’ reservoir correlation function K (t) = γ (δ(t) − δ(t − T )) ,
(45)
where T corresponds to the reservoir correlation time. It mimics features of the reservoir correlation function for reservoirs with band-gaps in the density-of-states (Mogilevtsev, Kilin, & Onischenko, 2004b). As follows from Equation (5), this correlation function leads to the following equation for the probability amplitude of the TLS upper-level occupation: d B(t) = −γ B(t) + γ B(t − T )2(t − T ), dt
(46)
with the exact solution B(t) =
∞ X n=0
exp{−γ (t − nT )}
γ n (t − nT )n 2(t − nT ). n!
(47)
The solution (47) reproduces the typical features of the TLS population dynamics leading to the ‘freezing’ of the TLS population (Figure 3).
Master Equations: Deterministic Approaches
117
The time-nonlocal Equation (44) leads to the equation for the TLS upper ¯ level population, ρ++ (t) = h+|ρ(t)|+i, similar to Equation (46), and its solution by Equation (47), where one has replaced B(t) → ρ++ (t) and γ → 2γ . For the time-convolutionless version of Equation (44) one has d ρ++ (t) = −2γ (1 − 2(t − T ))ρ++ (t) dt
(48)
with the solution ρ++ (t) = exp{−2γ t} for t ≤ T and ρ++ (t) = exp{−2γ T } for t > T . At first glance, the time-nonlocal form of the master equation produces similar results as the exact equation (Figure 3). However, let us consider the values of the ‘frozen’ population given by the exact, timenonlocal and time-convolutionless solutions. The exact equation yields ρ++ (∞) = [1 + γ T ]−2 , the time-nonlocal master equations leads to ρ++ (∞) = [1 + 2γ T ]−1 , and the time-convolutionless one gives ρ++ (∞) = exp{−2γ T }. All three solutions are close in the limit γ T 1. When this condition does not hold, the time-nonlocal master equation is liable to yield no lesser deviations from the exact equation than the timeconvolutionless one. So, the validity of the derived master equations depends not only on the interaction strength, but also on the reservoir correlation time. The product of these two quantities plays the role of a small parameter validating the expansions considered in the previous subsection. If one deals with reservoirs having a rather large correlation time (for example, band-gaps ones), one can hardly expect that including a few more higher-order terms in the expansion (38) will lead to significant improvements of the result. Besides, it can happen that including higherorder terms in master equations of different forms (i.e. time-nonlocal and time-convolutionless ones) will lead to a significant difference in the results obtained with the help of these equations. Generally, properties of the reservoir and its initial state define which kind of higher-order approximate master equation will give results closer to the true solution (Apanasevich, Kilin, & Nizovtsev, 1988; Kilin & Nizovtsev, 1986).
3.3. A System-Reservoir Border When deriving the master equation, it is essential to separate the system from the reservoir in the correct manner. First of all, the system-reservoir border should be drawn through everything included in the Hamiltonian. Thus, all entities not included in the reservoir should be considered as a compound system interacting with it. One cannot derive a master equation for a certain system and reservoir, and then assume that it is valid for another system, provided that the reservoir is the same. If one has derived the master equation, for example, for the TLS interacting with
Theoretical Tools for Quantum Optics in Structured Media
118
the reservoir, it does not mean that the derived equation is valid for the TLS interacting with the same reservoir plus some additional system. Also, a ‘phenomenological’ inclusion of an additional term (such as a classical driving) into the derived master equation, might lead to breaking of non-negative definiteness of the solution (Barnett & Stenholm, 2001; Ishizaki & Tanimura, 2008). One can include any number of additional Hamiltonian terms in the Lindblad equation (for example, into Equation (43)) without breaking the non-negative definiteness of the density matrix. However, such a procedure will give a result close to the correct one only if the interaction of the system with additional systems is sufficiently weak for the reservoir not to be influenced much with it. To illustrate this let us consider an example of the TLS interaction with the dephasing reservoir described by the operator Rd (t) in presence of a quasiclassical resonant driving. In the basis rotating with the TLS transition frequency the Hamiltonian for this problem is H(t) = h¯ σ + σ − Rd (t) + h¯ g(σ + + σ − ),
(49)
where g is the driving field-TLS interaction constant. If one assumes that the reservoir is Markovian, and performs formally the derivation procedure, leaving aside the term corresponding to the driving field-TLS interaction, one arrives at the following master equation d ¯ ¯ ¯ ρ(t) = −ig[σ + + σ −, ρ(t)] + dD(σ + σ − )ρ(t), dt
(50)
where Z
+∞
d=
dthRd (0)Rd (t)i.
0
We have assumed, for simplicity’s sake, that hRd (t)i = 0. However, if one includes the term corresponding to the driving fieldTLS interaction in the system, by ‘dressing’ the Hamiltonian (49) with it V(t) = h¯ S+ (t)S− (t)Rd (t),
S− (t) = eig(σ
+ +σ − )t
σ − e−ig(σ
+ +σ − )t
,
and uses the Born approximation in the time-convolutioness projection operator technique, one arrives at the following master equation for the
Master Equations: Deterministic Approaches
119
‘undressed’ reduced density matrix (Apanasevich, Kilin, & Nizovtsev, 1988): d ¯ ¯ ρ(t) = −ig[σ + + σ − , ρ(t)] dt Z t ¯ − σ + σ − ρ(t)P(τ ¯ − dτ hR(t)R(τ )i σ + σ − P(τ − t)ρ(t) − t) 0 Z t + − ¯ ¯ − dτ hR(t)R(τ )i ρ(t)P(τ − t)σ + σ − − P(τ − t)ρ(t)σ σ , (51) t0
where it is assumed that hR(τ )R(t)i = hR(t)R(τ )i P(t) =
1 1 [1 + cos(2gt)]σ + σ − − (σ + + σ − ) sin(2gt). 2 2
Obviously, an applicability of the Markovian approximation to Equation (51) is defined by whether or not it is possible to assume hR(t)R(τ )iP(τ − t) ∼ δ(τ − t), i.e. the Markovian approximation that led to the Lindblad equation (50) means that the spectral density of the reservoir correlation function, ρd (w) (defined by Equation (7)), is smooth in the vicinity of ω0 . Whereas an applicability of the Markovian approximation to Equation (51) means that ρd (w) is smooth in the vicinity of three frequencies: ω0 , and ω0 ± 2g. In other words, driving probes the structure of the reservoir: the more intense the driving is, the broader the spectral region of the reservoir that becomes involved in the system-reservoir interaction. Of course, the procedure of the master equation derivation, neglecting an adequate system-reservoir border, is unable to capture this feature. To clarify this in more detail, to which observable consequences will lead to a difference between Equation (50) and the Markovian version of Equation (51), let us consider equations for the density matrix elements following from them. Both master equations lead to formally similar systems of equations (Mogilevtsev, Nizovtsev, Kilin, Brandi, Cavalcanti, & Oliveira, 2008): d ρ++ (t) = ig[ρ−+ (t) − ρ+− (t)], dt d ρ+− (t) = κρ+− (t) + ig[ρ ¯ −− (t) − ρ++ (t)]. dt
(52)
120
Theoretical Tools for Quantum Optics in Structured Media
But for the system obtained from Equation (50) one has κ = πρd (ω0 ), g¯ = g, whereas for the system obtained from Equation (51) one has κ=
π π ρd (ω0 ) + (ρd (ω0 + 2g) + ρd (ω0 − 2g)), 2 4 π g¯ = g − (ρd (ω0 + 2g) − ρd (ω0 − 2g)). 2
(53)
As follows from Equation (53), if ρd (w) changes significantly in the interval [ω0 − 2g, ω0 + 2g], the dephasing rate, κ, acquires a dependence on the driving amplitude. This phenomenon leads to observable consequences, such as driving-dependent decay of the Rabi oscillations of the TLS population (Zrenner, Beham, Stufler, Findels, Bichler, & Abstreiter, 2002). It is useful to mention here that an approximation of local Markovianity, when one approximates terms like K (t − τ ) exp{−iw j (t − τ )} in the master equation with delta-functions, is nowadays used even for bandgap reservoirs with the emitter’s transition frequency close to the bandedge (Florescu & John, 2004). It can be justified when the detunings w j are large enough, such that ω0 + w j are far away from the band-edge (of course, one should bear in mind that such an approximation of the local Markovianity is unable to account fully for the effects of the emitterreservoir correlations).
3.4. Separating the Reservoir As was seen from the previous subsection, defining a border between the system and the reservoir is very important for deriving master equations. Here we demonstrate that when defining this border, in some cases one is also able to account for even long-time system-reservoir correlations, such as the emitter-field bound state considered in Section 2.4. The idea underlying this procedure is simple: if the reservoir exhibits a nonMarkovian feature, maybe, it is possible to separate the part ‘responsible’ for this feature, from the remaining, Markovian part? This separation looks obvious when the reservoir consists of several parts of different nature (for example, in a measurement simulation scheme, when the non-Markovian reservoir consists of the single TLS plus the Markovian reservoir coupled to it (Salo, Barnett, & Stenholm, 2006)). Sometimes, such separation can be also achieved in the reservoir of objects of a similar nature, if certain structural features point to the way of performing it. For example, in the quasi-modes method (considered in the Section 2.5) one can consider certain bosonic reservoirs as a countable number of collective modes coupled to separate Markovian reservoirs (the master equation derivation for this case will be considered in the next Section).
Master Equations: Deterministic Approaches
121
Generally, the presence of sharp reservoir features like peaks, band-gaps etc. serves as an indicator that one should look for some superpositions of the reservoir modes with frequencies in the vicinity of these features, which can exhibit long-time correlations with the system. Also, an existing exact solution can provide a clue for the reservoir separation. As example of a reservoir separation, let us consider the derivation of the master equation for the spontaneous emission into a band-gap reservoir under the condition when the ‘freezing’ of population decay is possible. A clue for the procedure is provided by the properties of collective emitter-field operators introduced in Section 2.4 by Equation (13). As shown there, for cases when the ‘freezing’ of decay is possible, only one of collective operators C j , namely, C0 , describes a strong emitterfield coupling, and the single-excitation eigenvector, |ψ0 i, satisfying Ď C0 C0 |ψ0 i = |ψ0 i, corresponds to the ideal ‘frozen’ state, i.e. the ‘frozen’ state with maximally achievable stationary upper-level population of the 2 ) (Mogilevtsev, Kilin, & Cavalcanti, 2004a). As Equation (20) TLS (it is U00 shows, the field parts of each collective operator may be represented as bosonic collective operators B j . As illustrated in Figure 1, the collective operator B0 is composed mostly from the operators b j corresponding to the reservoir modes with frequencies close to the band-edge. To separate the reservoir part described by the operator B0 from the rest of the reservoir, let us introduce the collective operator C in such manner that the ideal ‘frozen’ state, |ψ0 i, is the zero-eigenvalue eigenstate of the following operator C|ψ0 i = σ − − B0 U00 p |ψ0 i = 0,
(54)
q 2 . Every collective operator can be represented with where p = 1/ 1 − U00 the help of this operator and the collective bosonic operators B j as q C j = U0 j C + B j 1 − U02j + B0 U00 U0 j p = U0 j C + B´ j ,
(55)
Ď
where B´ j commute with B0 for j > 0; so the Hamiltonian (13) can be recast into the following form X Ď Ď ´ j + U0 j (CĎ B´ j + B´ Ďj C) . H = λ0 σ + σ − + B0 B0 − CĎ C + λ j B´ j B j=1
This formula points directly to the way of obtaining a master equation for the problem. Namely, one needs to average over the reservoir of all
Theoretical Tools for Quantum Optics in Structured Media
122
modes B´ j , which can be achieved in the standard manner discussed above provided the Born–Markov approximation is valid. However, for bandgap reservoirs, it is not always the case. The correlation function of the reservoir of modes B´ j can exhibit the same ‘long-tail’ behaviour and rather large correlation time as the correlation function of the reservoir of modes b j , (which can be guessed from the fact that far from the band-edge one has λ j U0 j ≈ g j ) (Mogilevtsev, Kilin, & Cavalcanti, 2004a; Mogilevtsev, Moreira, Cavalcanti, & Kilin, 2006). The situation can be remedied by a closer look at the structure of the reservoir of B´ j modes. They are not independent, and evolve without an interaction with the TLS, being coupled with each other through a collective mode A: √ d ´ B j = −iλ j B´ j − iAU0 j ν A p. dt where A = ν −1 A
j=1 λ j U0 j B j ;
ν 2A =
4 p. − λ20 U00 One can separate A out from the reservoir of B´ j modes by introducing new reservoir operators independent of A:
P
´
P
2 2 j=1 λ j U0 j
Ď ´ ¯ B¯ j = ν −1 j j (B j − ν j A A) H⇒ [B j , A ] = 0,
where 2 ν A j = ν −1 A U j0 λ j + λ0 U00 p ,
ν 2jk = δ jk − ν A j ν Ak − U0 j U0k p.
Using operators B¯ j and A, the Hamiltonian (13) can be represented in the following way Ď H = λ0 σ + σ − + B0 B0 − CĎ C + ω A AĎ A + G(CĎ A + AĎ C) X X Ď ¯ Ďj B¯ j + + λ¯ j B g¯ j (AĎ B¯ j + B¯ j A), j=1
(56)
j=1
P where ω A = l=1 λl ν 2Al , g¯ j = λ j ν A j ν j j , λ¯ j = λ j ν j j , G = ν A . For a wide variety of the densities-of-states with band-gaps (in particular, for the most commonly used for modelling isotropic, step-like and anisotropic densities-of-states (John & Wang, 1991)), the reservoir of B¯ j modes can be treated as the Markovian one (Mogilevtsev, Kilin, & Cavalcanti, 2004a; Mogilevtsev, Moreira, Cavalcanti, & Kilin, 2006). Additionally, for extended densities-of-states (which is usually the case for realistic densities-of-state, and for the model ones mentioned above),
Master Equations: Deterministic Approaches
123
the dynamics of the collective mode A is very fast on a time-scale of the TLS dynamics. Thus, the mode A can be adiabatically eliminated, and one ¯ finally arrives at the master equation for the reduced density matrix, ρ(t), which is the result of the reservoir averaging over all modes, apart from one corresponding to the operator B0 : d ¯ Ď C, ρ(t)] ¯ ¯ ρ(t) = i[(λ0 + δ)C dt
Ď Ď ¯ ¯ − ρ(t)C ¯ + γ¯ 2Cρ(t)C − CĎ Cρ(t) C ,
(57)
where γ¯ = ν 2A Re(X ), δ¯ = ν 2A I m(X ), and X = 1/[γ + i(ω A + δ)]. The parameters γ and δ (respectively, the relaxation rate and the frequency shift of the collective mode A) are defined through the density-of-states, %(w), for the original reservoir described by the set of modes b j , in the following way γ ≈
π 2 ω A %(ω A + ω0 ), ν 2A
δ≈
1 P ν 2A
Z dw
(w − ω0 )2 %(w) . w − ω0 − ω A
Obviously, the master Equation (57) is able to capture the effect of decay ‘freezing’. For any single-photon initial state of the TLS and the mode B0 , asymptotically the density matrix will be ¯ ρ(∞) = p|ψ0 ihψ0 | + (1 − p)|vacihvac| + q|ψ0 ihvac| + q ∗ |vacihψ0 |, where |vaci = |−i|0i0 , and |0i0 denotes the vacuum state of the B0 mode. Coefficients p, q are defined by the initial conditions; for example, p = hψ0 |ρ(0)|ψ ¯ 0 i. For the initially completely excited TLS and the mode B0 in the vacuum state, one has the ‘frozen’ upper-level TLS population equal 4 , which is precisely the result predicted by the exact solution (18). to U00 One has to notice here that the master equation, very similar to Equation (57), was derived for the single-point gap using the method of the pseudomodes (Bay, Lambropoulos, & Mølmer, 1998).
3.5. Solving Master equations I If one has been able to derive a master equation for a chosen small subsystem, one has already greatly simplified the problem of describing the dynamics of this subsystem. However, even a solution of the system of equations for the elements of the density matrix in a chosen basis may be a challenging task. Indeed, for an example of Equation (57): for a subsystem consisting of the TLS and a single field mode with an initial state with up
Theoretical Tools for Quantum Optics in Structured Media
124
to N photons one has to solve a system of about 4(N + 1)2 differential (or integro-differential) equations. For large N this task becomes quite difficult. First of all, to avoid the problem of solving a large system of differential equations for a subsystem including harmonic oscillators, one can use quasiprobabilities to describe their states. In this way one has to deal with a system of a few partial differential equations for quasiprobability distributions. The s-ordered quasiprobability distribution is generally defined as (Cahill & Glauber, 1969a,b): Ws (α, α ∗ ) =
1 π2
Z
d 2 β exp{αβ ∗ − α ∗ β + s|β|2 /2}Tr{D(β)ρ} ( ) Ď 2 s+1 a a Ď = Tr ρD(α) D (α) , π(1 − s) s−1
(58)
where D(α) = exp{αaĎ −α ∗ a} is the coherent shift operator; the parameter s belongs to the interval [−1, 1]. The s-ordered quasiprobability distribution (58) is completely equivalent to the modal density matrix. Products of modal operators with the density matrix after averaging (58) are transformed in the following way s+1 ∂ (a) ρ → α + 2 ∂α ∗ n
n
Ws ,
n s−1 ∂ n Ď ∗ a ρ→ α + Ws . 2 ∂α
Thus, for example, a Lindblad dissipator for the modal photon loss into the vacuum of the Markovian reservoir Dρ = 2aρaĎ − aĎ aρ − ρaĎ a, turns into the following differential term Dρ(t) →
∂ ∂2 ∂ ∗ α + ∗ α + (1 − s) ∗ ∂α ∂α ∂α ∂α
! Ws (α, α ∗ , t).
(59)
Writing equations for quasiprobability distributions, one should rememeber that, for different s, the functions (58) have drastically different properties. For s = 1, one obtains a highly singular quasiprobability distribution (the so called Glauber-Sudarshan, or P-function). The P-function corresponding to the coherent state is a delta-function, P-functions of Fock states are expressed as derivatives of delta-functions. By contrast, for s = −1, the quasiprobability function (Husimi, or Q-function) is nonsingular and non-negative, being proportional to an average of the density matrix over the coherent state of the oscillator, W−1 (α, α ∗ ) ∼ hα|ρ|αi. Despite its singular character, due to the simplicity of the equations it obeys,
Master Equations: Deterministic Approaches
125
the P-function is often used for modelling purposes (as an example we would like to mention here an investigation of bi- and multi-stability in damped driven Jaynes–Cummings and Tavis-Cummings models (Kilin & Krinitskaya, 1991, 1993), where analytic estimates of the quasiprobability functions were derived). For a more complete review of the quasiprobability distribution uses for this purpose we refer to Refs. (Gardiner & Zoller, 1999; Hillery, O’Connell, Scully, & Wigner, 1984; Peˇrina, 1984)). In some important cases exact solutions of master equations can be obtained by using algebraic methods (see also a discussion in the beginning of the Section 2). If the master equation can be represented in the form X d ρ(t) = L j ρ(t), dt j where a finite set of the superoperators L j forms a closed algebra (i.e. it satisfies the conditions [L j , Lk ] =
X
C jkl Ll ,
(60)
l
and (super)operators C jkl commute with all L j ), then it is possible to make a factorization ( ) X Y exp Ljt = exp F j (t)L j , j
j
where the superoperators F j (t) are defined from the relations (60) (with the help of a generalization of the Campbell-Baker-Hausdorff formula, see, for example (Louisell, 1977; Puri, 2001)). As an application of the technique described above, let us consider again the problem of an interaction of a single field mode with the Markovian dissipative reservoir. To deal with such problems it is useful to resort to ‘thermofield’ notations allowing us to deal with superoperators in a manner quite similar to operators (Chaturvedi & Srinivasan, 1991). For our example, introduction of the ‘thermofield’ notationPmeans that instead of an operator (i.e. a density matrix), say, ρ = k,l ρkl |kihl|, acting in the Hilbert space H, where |ki is the Fock state with k photons, P ˜ we consider a state vector |ρi = k,l ρkl |ki|li in the doubled space N ∗ ˜ is the Fock state with l photons in H∗ . So, when in H H , where |li a master equation modal operators a and aĎ act on the density matrix from the right, one introduces operators a˜ and a˜ Ď in such a manner that
126
Theoretical Tools for Quantum Optics in Structured Media
˜ |kihl|aĎ ⇐⇒ a˜ |ki|li. ˜ Obviously, operators a and aĎ |kihl|a ⇐⇒ a˜ Ď |ki|li, Ď commute with a˜ and a˜ . In the thermofield notation, the master equation with the the dissipator D(a) can be represented in the ‘Hamiltonian’ form as γa d γa γa d ρ = D(a)ρ H⇒ |ρi = H(a)|ρi = (2a˜a − aĎ a − a˜ Ď a˜ )|ρi, (61) dt 2 dt 2 2 where γa /2 is the modal decay rate. Now it is not hard to notice that superoperators A+ ≡ aĎ a˜ Ď ,
A3 ≡ (aĎ a + a˜ Ď a˜ + 1)/2
A− ≡ a˜a,
(62)
generate an su(1, 1) algebra with the Casimir invariant A0 ≡ aĎ a − a˜ Ď a˜ . Applying the Campbell–Baker–Hausdorff formula to Equation (61), one arrives at the solution |ρ(t)i = exp
nγ o a t exp[−γa tA3 ] exp[(1 − e−γa t )A− ]|ρ(0)i. 2
(63)
If the closeness relation (60) does not hold, sometimes it is convenient to explore a Laplace-transformed solution by evaluating a resolvent, namely, the superoperator (z − λL)−1 , for the master equation d ρ(t) = Lρ(t). dt For example, for L = D(X), where the operator X satisfies commutation relations [X, XĎ X] = X and [X, XĎ ] = p(XĎ X); p(x) is a polynomial of x, one can find the resolvent exactly (Klimov & Romero, 2003). Another algebraic approach, which can be of some use for obtaining exact solutions, is the method of damping basis (Briegel & Englert, 1993). For the master equation with the Liouvillian L one can find a solution by expanding the density matrix in terms of the eigenoperators of L. Right eigenoperators, R j are introduced as LR j = r j R j , where r j are right eigenvalues. Representing the initial density matrix as ρ(0) =
X
SjRj,
j
where S j - scalar coefficients, and assuming non-degenerate and non-zero
Master Equations: Stochastic Approaches
127
eigenvalues r j , one can represent the solution as ρ(t) =
X
(64)
S j R j exp{r j t}.
j
One can determine the coefficients S j in Equation (64) by finding the set ´ jL = rjR ´ j , because for them, one should have of the left eigenoperators, R ´ l R j } = δl j . Tr{R For the example of the TLS damping into the Markovian reservoir, described by the equation d ρ(t) = γ D(σ − )ρ(t), dt one can introduce a damping basis of right eigenoperators in the following way: D(σ − )(1 − σ z ) = 0,
D(σ − )σ z = −2γ σ z ,
D(σ − )σ ± = −γ σ ± .
The left eigenoperators are 2(σ z − 1), σ ∓ and the unit operator. With the help of this damping basis, the solution of the master equation is represented as ρ(t) = 1 − σ z + S1 σ z exp{−2γ t} + (S2 σ − + S2∗ σ + ) exp{−γ t}. In the general case finding damping bases is rather difficult. However, it can be quite useful for systems with few degrees of freedom. For example, it allows us to obtain solutions for such non-trivial problems as thermally driven Jaynes–Cummings systems and single-atom lasers (Briegel & Englert, 1993; Ginzel, Briegel, Martini, Englert, & Schenzle, 1993).
4. MASTER EQUATIONS: STOCHASTIC APPROACHES A stochastic approach to solving problems of the system-reservoir interaction is based on the possibility of simulating the influence of certain types of reservoirs on the system by a classical noise, and to ¨ derive the stochastic Schrodinger equation for a wave function dependent on this classical noise. Advantages of such an approach are twofold: ¨ firstly, stochastic Schrodinger equations can be used for derivation of ¨ exact or approximate master equations; secondly, stochastic Schrodinger equations can provide a convenient way to solve master equations.
128
Theoretical Tools for Quantum Optics in Structured Media
In this section we give examples of both of these uses of the stochastic approach. We consider the ‘quantum Monte-Carlo’ approach, which allows for certain classes of master equations (first of all, for Lindblad ones) to model a system dynamics with a set of ‘quantum trajectories’, which change deterministically according to the effective Hamiltonian, but in random time-moments undergo instantaneous changes (‘quantum jumps’) (Carmichael, 1993; Mølmer, Castin, & Dalibard, 1993; Plenio & Knight, 1998). One obtains any averages from system variables by averaging over each of these functions and subsequent classical averaging over results of these ‘trajectory averages’. Thus, to find a solution for an individual trajectory one needs to solve only N equations instead of N 2 −1 for the density matrix elements. This approach can be extended to some non-Lindblad master equations (Breuer, Kappler, & Petruccione, 1999), and used for derivation of an exact master equation from the equations of motion obtained via the pseudo-modes method (Bay, Lambropoulos, ˘ & Mølmer, 1998; Garraway, 1997a,b; Imamoglu, 1994). This approach gives one the possibility of finding a solution for such a problem as the resonance fluorescence of a three-level emitter into a band-gap reservoir, when one of transition frequencies in near the band-edge, and the other is far from the band-edge, being situated within the flat continuum (Bay, ¨ Lambropoulos, & Mølmer, 1997). The stochastic Schrodinger equation can also be derived in such a way that random changes of the wave function are continuous (the so called ‘quantum diffusion’) (Percival, 2005). Quantum diffusion approach for interaction of the emitter with bosonic reservoirs makes it possible to derive both exact and approximate master equations even for essentially non-Markovian problems (Strunz, Diosi, & Gisin, 1999; Strunz & Yu, 2004; Yu, Diosi, Gisin, & Strunz, 1999). It should be noted that stochastic approaches for solving a problem of system-reservoir interaction have stemmed from such physical problems as resonance fluorescence of single emitters (Dehmelt, 1975, 1987; Plenio & Knight, 1998) and continuous quantum measurements (Davies, 1976; Kilin, 1990; Srinivas & Davies, 1981). For example, in the latter case a measurement of a certain number of ‘clicks’ with a photodetector (which occurs stochastically) conditions the emitter to some particular quantum state. This state can be considered as a ‘quantum trajectory’. Generally, in the case of Markovian reservoirs one can always interpret the corresponding stochastic wavefunction as one describing an evolution conditioned on results of certain measurements performed on the system (Carmichael, 1993). However, for an arbitrary non-Markovian reservoir such an interpretation is problematic.
Master Equations: Stochastic Approaches
129
4.1. Quantum Monte-Carlo Procedure Here we consider a quantum Monte-Carlo procedure for solving Lindblad master equations. As we shall see below, this procedure allows for an effective solution of a non-Markovian master equation written in the Lindblad form (such, for example, as the one given by Equation (57)) and can be generalized for certain types of time-local non-Lindblad master equations. Now let us consider the following master equation i d ρ(t) = − [H, ρ(t)] + 2Lρ(t)LĎ − LĎ Lρ(t) − ρ(t)LĎ L. dt h¯
(65)
We want to investigate the evolution of the system from the initial time t = 0 to t = T , and for that purpose we discretize this time interval with N + 1 sufficiently small intervals, [tn , tn+1 ], tn+1 − tn = δtn , n = 0, 1, . . . N . For simplicity’s sake, for the moment, let us also assume that the initial state of the system is pure, ρ(0) = |9(0)ih9(0)|. Then in the simplest version of the quantum Monte-Carlo method (given, for example, by Mølmer, Castin, and Dalibard (1993)), one models the system’s dynamics with a bunch of M quantum trajectories, |9m (t)i. A classical average over all these stochastic trajectories (denoted as E{· · ·}), in the limit of an infinite number of them, should give the density matrix of the system, E{|9m (tn )ih9m (tn )|} =
lim
M→+∞
M 1 X |9m (tn )ih9m (tn )| = ρ(tn ). M m=1
These trajectories are generated in the following way: it is assumed that in each time interval [tn , tn+1 ] the stochastic wave function, |9m (t)i, changes either deterministically according to the equation |9m (tn+1 )i = |9m (tn )i −
iδtn H|9m (tn )i − LĎ L|9m (tn )i, h¯
(66)
or undergoes a ‘quantum jump’ to the state |9m (tn+1 )i =
L|9m (tn )i h9m (tn )|LĎ L|9m (tn )i
.
(67)
130
Theoretical Tools for Quantum Optics in Structured Media
This jump occurs with the probability δp(tn ) = 2δtn h9m (tn )|LĎ L|9m (tn )i. Thus, it is important for practical simulations to choose small intervals δtn to have δp(tn ) much less than unity. In practice, for each interval one generates a random number, r , uniformly distributed from 0 to 1, and compares it with the calculated δp(tn ). If δp(tn ) > r , the jump occurs; otherwise the system changes deterministically according to Equation (66). It is easy to check that in the limit of infinitely small δtn , an evolution described by Equations (66) and (67) is completely equivalent to one described by the master Equation (65). Indeed, in the time moment tn+1 one can represent a projector ρ (m) (tn+1 ) = |9m (tn+1 )ih9m (tn+1 )| as a statistical mixture of evolution results given by both Equations (66) and (67): ρ (m) (tn+1 ) = 2δtn Lρ (m) (tn )LĎ i i + 1 − δtn H − δtn LĎ L ρ (m) (tn ) 1 + δtn H − δtn LĎ L . h¯ h¯ Neglecting terms proportional to [δtn ]2 and averaging over all trajectories, in the limit δn → 0, one arrives at Equation (65). For a finite number of trajectories and finite δn one estimates the average of an arbitrary operator (say, A), as hA(tn )i ≈ hA(M) (tn )i =
M 1 X h9m (tn )|A|9m (tn )i. M m=1
(68)
For practical applications of the scheme described by Equations (66) and (67) one needs to estimate, firstly, how dense a time discretization should be to avoid errors arising from such a procedure and, secondly, to estimate the number of trajectories required for a sufficiently accurate evaluation of a quantity of interest. The first question can be answered by considering Equation (68). Obviously, one should require a statistical error, δhA(tn )i, satisfying δhA(tn )i hA(tn )i.
Master Equations: Stochastic Approaches
131
1
〈 σ +(t )σ –(t )〉
0.8
0.6
0.4
0.2
0
0
0.5
1
1.5 γt
2
2.5
3
FIGURE 4 An example of the quantum trajectories simulation of the dynamics of the upper level of a two-level emitter interacting with the vacuum of the Markovian √ reservoir (described by Equation (65) with H ≡ 0 and L = σ − γ /2). The thick solid line corresponds to the exact solution of Equation (65); dotted, dashed, dash-dotted and thin solid lines correspond M = 1, 10, 50, 500 trajectories, respectively. The time interval was discretized with N = 100 points.
In our case all trajectories are equally probable, which gives √ δhA(tn )i ≈ 1hA(tn )i/ M, where 1hA(tn )i is the variance of the operator A in the state described by the density matrix ρ(tn ) (Mølmer, Castin, & Dalibard, 1993). Thus, for the required number of trajectories one has the condition √
M 1hA(tn )i/hA(tn )i.
(69)
So, the number of required trajectories depends strongly on the state of the system. For example, if the emitter is a harmonic oscillator with the state close to the coherent one, for estimation of an average number of photons the condition (69) gives M 1, i.e. in that case one needs the same number of trajectories for all evolution times when the state is close to the coherent one. From the other side, for the upper-level population, P(t), of the TLS spontaneously decaying into the vacuum, one has M P(t)−1 − 1, i.e. the number of required trajectories increases significantly with extension of the time-interval of interest. In Figure 4 one can see an illustration of approximating the spontaneous upper-state dynamics with
132
Theoretical Tools for Quantum Optics in Structured Media
different numbers of trajectories. Time-intervals tn must be chosen such that 2δtn hLĎ (tn )L(tn )i 1. One can make a procedure of discretization adaptable by defining an interval between jumps (and preventing the possibility of two or more jumps in a pre-defined interval) in the following manner: firstly, one generates a random number, r , uniformly distributed between 0 and 1, and then finds a time-moment of the next jump, tn+1 , from the equation r = |h9m (tn+1 )|9m (tn )i|2 , where the function |9m (tn+1 )i is given by a normalized solution of Equation (66) (Plenio & Knight, 1998). The simplest version of the quantum Monte-Carlo method described above, can be straightforwardly generalized for the cases of mixed initial states of the system and a simultaneous coupling to a number of reservoirs. If the initial state is mixed, for an application of the method it should be represented in a diagonal form, ρ(0) =
X
λ j |φ j ihφ j |.
j
Then, the initial stochastic wave functions should be randomly chosen from the set {|φ j i} according to the distribution of λ j . If the system is coupled to a number of reservoirs with corresponding Lindblad P Ď operators L j , in Equation (66) one needs to replace LĎ L with j L j L j . For determining a kind of a jump (choosing L j in Equation (67)) one needs to generate a set of probabilities Ď
δp j (tn ) = 2δtn h9m (tn )|L j L j |9m (tn )i, and then to choose randomly, according to the normalized distribution of p j (tn ). Here one should also notice that the quantum Monte-Carlo method can be easily generalized for a calculation of correlation functions (Mølmer, Castin, & Dalibard, 1993; Plenio & Knight, 1998). Considerations given above can be also straightforwardly generalized for the case of a time-dependent Hamiltonian, H(t), and Lindblad operators, L j (t). However, one must require that the probability of jumps, δp j (t), always
Master Equations: Stochastic Approaches
133
remains positive. If this condition is not satisfied, one can still generalize the method in such a way as to circumvent this difficulty. This will be considered in the next subsection. It should also be mentioned that the time-dependence of the Lindblad operators might be an obstacle for choosing δt j sufficiently small; these intervals should not exceed typical times of significant variation of the Lindblad operators. This problem can be circumvented by an appropriate scaling of quantum trajectories (which can also be useful for optimizing the number of discretization points and trajectories for simulation) (Piilo, Maniscalco, Messina, & Petruccione, 2005). The practical procedure described by Equations (66) and (67) can be ¨ formalized by writing it as a stochastic Schrodinger equation
idt d|9(t)i = − H − dtLĎ L + δp(t) + h¯
Ldt − 1 dP(t) |9(t)i, δp(t)
where dP(t) is the stochastic Poissonian increment satisfying the following equations (Gardiner & Zoller, 1999): [dP(t)]2 = dP(t),
E{dP(t)} = δp(t).
¨ A procedure for deriving a stochastic Schrodinger equation from the master equation is called the ‘unravelling’ of the master equation. This procedure is not unique (Carmichael, 1993; Gardiner & Zoller, 1999). Here one has to mention that the quantum Monte-Carlo approach given above can be used also for problems involving non-Markovian reservoirs. For example, it might be applied to find a solution for the case when one has an emitter coupled to the band-gap reservoir and additionally subjected to dissipation into the Markovian reservoir. If one can write a wave-function for the emitter plus band-gap reservoir in a closed form, then one can treat it as the stochastic one, and consider the Markovian reservoir as a source of non-unitarity and jumps. For example, in this manner the solution was found for the three-level 3system emitter with one transition coupled to the band-gap reservoir, and another to the Markovian reservoir (plus an additional classical driving in that transition) (Bay, Lambropoulos, & Mølmer, 1997).
4.2. Non-Lindblad Quantum Monte-Carlo Procedure The stochastic procedure described above can be generalized for nonLindblad time-local master equations with the help of an ‘embedding’ scheme, i.e. assuming that the system in question is coupled to an additional quantum object in such a way that the dynamics of the system
134
Theoretical Tools for Quantum Optics in Structured Media
plus additional object can be described by the Lindblad equation, and the density matrix of the system can be obtained by averaging the density matrix of the extended system by the state of the additional quantum object. As an example let us consider the following master equation: X d ρ(t) = A(t)ρ(t) + ρ(t)BĎ (t) + C j (t)ρ(t)D j (t), dt j
(70)
where time-dependent operators A(t), B(t), C j (t), D j (t) are such that the density matrix ρ(t) remains non-negative for arbitrary times. An example of the physical system described by such an equation is one describing non-Markoivan TLS decay (which can be derived by the timeconvolutionless projection operator technique described in Section 3.1): d ρ(t) = γ (t) 2σ − ρ(t)σ + − ρ(t)σ + σ − − σ + σ − ρ(t) , dt where time-dependent relaxation rate, γ (t), may be negative. Then one can take A(t) = B(t) = −γ (t)σ + σ − , and C(t) = 2σ − , D(t) = γ (t)σ + , transforming the master equation to the form (70). A desired extension can be achieved by assuming that an additional TLS is coupled to the system in question, and that the quantum trajectory for an extended system can be represented as (Breuer, Kappler, & Petruccione, 1999): |9(t)i = |ψ(t)i|+ia + |φ(t)i|−ia ,
(71)
where |ψ(t)i and |φ(t)i are stochastic wave-functions of the system; |±ia are upper and lower states of the additional TLS. The vector (71) is chosen in such a way that the density matrix of the system in question can be found as ρ(t) = h+|a ς (t)|−ia = h+|a E {|9(t)ih|9(t)} |−ia = E {|ψ(t)ihφ(t)|} , (72) where ς (t) is the density matrix of the extended system. By introducing operators X j (t) = σ a+ σ a− C j (t) + (1 − σ a+ σ a− )D j (t), and taking into account a representation (72), one can write the master equation for the density matrix, ς(t), of the extended system as the
Master Equations: Stochastic Approaches
135
Lindblad one d ς (t) = −i[σ a+ σ a− A(t) + (1 − σ a+ σ a− )B(t), ς (t)] dt X Ď Ď Ď + 2X j (t)ς (t)X j (t) − ς (t)X j (t)X j (t)(t) − X j (t)X j (t)ς (t) . j
Notice that the jump probabilities for operators X j (t) are always positive now: Ď
2δth9(t)|X j (t)X j (t)|9(t)i = δp j (t) ≥ 0. The embedding method can be generalized to more complicated master equations than Equation (70). For example, using an additional threelevel system, one can obtain a Lindblad equation from the following one (Breuer, 2004a): X d i Ď Ď ρ(t) = − [H(t), ρ(t)] + 2 C j (t)ρ(t)D j (t) + D j (t)ρ(t)C j (t) dt h¯ j X Ď Ď − C j (t)D j (t) + D j (t)C j (t) ρ(t) j
− ρ(t)
X Ď Ď C j (t)D j (t) + D j (t)C j (t) . j
It is interesting to notice that one can deal with negative probabilities, which might occur for non-Markovian time-local master equations without the embedding, by assuming that negative probabilities mean that a jump has occurred in the opposite time-direction (i.e. that a trajectory for a given time moment has instantly changed to another one from the whole ensemble corresponding to a previous timemoment) (Piilo, Maniscalco, Harkonen, & Suominen, 2008). This method clearly illustrates a non-Markovianity as the ‘memory’ of the reservoir. Despite the fact that for a numerical realization one needs to calculate simultaneously a number of trajectories to which an ‘time-reversed’ jump might occur, the method might still appear more efficient then calculation in doubled or tripled Hilbert space. It is also worth noting that a nondiagonal representation, similar to the one used in Equation (72), can be used for exact simulation of a complete system plus reservoir dynamics with the help of quantum trajectories (Breuer, 2004b; Lacroix, 2005). To that end one represents a trajectory as a product of parts corresponding to
136
Theoretical Tools for Quantum Optics in Structured Media
the system and the reservoir, |9(t)i = |ψ(t)isystem |χ (t)ireservoir , and for the density matrix one has ´ ρ(t) = E{|9(t)ih9(t)|}, where the classical averaging is assumed to be done on a diadic composed from different stochastic wave-vectors. It is curious that such Monte-Carlo formulation of exact dynamics allows one to represent each trajectory as the evolution of uncorrelated system and reservoir states. In this picture an entanglement between the system and the reservoir arises as a result of averaging over the set of trajectories.
4.3. Pseudomodes and Quantum Monte-Carlo Procedure Quantum Monte-Carlo methodology can be used to derive a master equation. A pseudo-modes method considered in Section 2.5 offers an especially simple and straightforward way of doing it. A system of Equation (23) for the amplitudes of pseudomodes, A¯ j (t), and the probability amplitude to have the TLS in the upper state, B(t), can be obtained from the following effective Hamiltonian (Garraway, 1997a; ˘ Imamoglu, 1994): Heff = h¯
X X Ď Ď (1 j − iγ j )A j A j + h¯ g¯ j (A j σ − + σ + A j ), j
(73)
j
Ď
provided the interaction constants g¯ j are real; here A j , A j are the creation and annihilation operators of pseudomodes. One can easily interpret the Hamiltonian (73) as describing a deterministic evolution between jumps (as was done in Section 4.1). So, one naturally comes up with the following master equation for the density matrix, ρ(t), of the TLS plus the set of pseudomodes " # X X d Ď Ď − + ρ(t) = −i (1 j A j A j + g¯ j (A j σ + σ A j )), ρ(t) + γ j D(A j )ρ(t). dt j j However, when the interaction constants are imaginary, one needs to apply an orthogonal transformation to the pseudomode amplitudes, A j (t), in such a way that the system of equations for the transformed amplitudes corresponded to real pseudomodes-TLS interaction constants
Master Equations: Stochastic Approaches
137
(Garraway, 1997a,b). Here we show an example of such a procedure for a single-point density-of-states, %d (w), composed of two Lorentzians (Garraway, 1997a): W1 γ1 W 2 γ2 %d (w) = D − , (w − ω0 )2 + (γ1 /2)2 (w − ω0 )2 + (γ2 /2)2
(74)
where weights W1,2 are positive, and a positive constant D describes the strength of the TLS-field interaction. For the single-point gap one has W1 /γ1 = W2 /γ2 . The two Lorentzians in Equation (74) corresponds to two pseudomodes (with amplitudes a1,2 (t),√respectively). However, one of the interaction constants (namely, g¯ 2 = i DW2 ) is imaginary. One can circumvent this problem by introducing new pseudomode amplitudes by the following transformation C1 (t) = f (A1 (t) cos{µ} + iA2 (t) sin{µ}), C2 (t) = f (A2 (t) cos{µ} − iA1 (t) sin{µ}),
(75)
where f = (cos{2µ})−1/2 , and the angle µ is defined as exp{4iµ} = −W2 /W1 . Equations for the new amplitudes are d d B(t) = −iDC2 (t), C1 (t) = −γ¯1 C1 (t) − ivC2 (t), dt dt √ d C2 (t) = −γ¯2 C2 (t) − ivC1 (t) − i D B(t), dt
(76)
where new relaxation rates for the pseudomodes are γ¯1 = γ2 W1 − γ1 W2 , γ¯2 = γ1 W1 − γ2 W2 ; the interaction √ constant of pseudomodal coupling induced by the rotation (75) is v = W1 W2 (γ1 − γ2 )/2. For the single-point gap, one has γ¯1 = 0, which leads to the following master equation h i √ d Ď ρ(t) = −i C2 (vC1 + Dσ − ) + h.c., ρ(t) + γ¯2 D(C2 )ρ(t), dt Ď
where C1,2 , C1,2 are creation and annihilation operators of the new pseudomodes. If γ¯2 is so large that the pseudomode C2 remains practically unpopulated (which corresponds to the assumption that the ‘positive’ Lorentzian is much wider than the ‘negative’ one), it is possible to adiabatically eliminate the pseudomode C2 , arriving at the master
138
Theoretical Tools for Quantum Optics in Structured Media
equation very similar in form to the collective operator Equation (57): v2 d Ď Ď ¯ ¯ ¯ − ρ(t)C ¯ ρ(t) = 2Cρ(t)C − CĎ Cρ(t) C , dt γ¯2
(77)
where the collective operator √ C = C1 +
D − σ , v
and ρ(t) ¯ denote the density matrix, ρ(t), averaged over the states of the pseudomode C2 . It is to be noted that the master equation of the form (77) was derived by the pseudomode method for the Fano-like density-ofstates in the work (Bay, Lambropoulos, & Mølmer, 1998).
4.4. Quantum Diffusion Procedure Unravelling of the master equation can be done, assuming that the state of the systems changes continuously under the stochastic action of the reservoir. It can be done using the so called ‘quantum diffusion’ method (Percival, 2005). This method can be applied to any problem described by Lindblad master equations. To perform the unravelling, one takes in any time-moment, t, an infinitesimal change in the systems stochastic wave function, d|9(t)i, can be represented as d|9(t)i = dt|ϕ(t)i + |φ(t)idξ(t),
(78)
where wave-functions |ϕ(t)i and |φ(t)i are orthogonal to each other and change independently of the stochastic process, which has the increment dξ(t). Now let us demonstrate that one can build an unraveling of the Lindblad master Equation (65) using continuously changing quantum trajectories (78). We assume that the process ξ(t) is a complex Wiener process satisfying the following conditions (Gardiner & Zoller, 1999): E{ξ(t)} = E{ξ(t)∗ } = 0,
E{ξ(t)ξ(τ )∗ } = 2δ(t − τ )
with the increment satisfying E{dξ(t)} = E{dξ(t)∗ } = 0, 2
E{[Re(dξ(t))][I m(dξ(t))]} = 0,
E{[Re(dξ(t))] } = E{[I m(dξ(t))]2 } = dt.
(79)
Master Equations: Stochastic Approaches
139
It is to be noted, that the properties (79) imply an ‘unconventional’ rule of calculating increments from noise-dependent quantities, say, A and B. Namely, it is d[AB] = [dA]B + A[dB] + [dA][dB]. They are the increments in the so-called ‘Ito form’ (Gardiner & Zoller, 1999). As before, we require that the statistical average of the projector |9(t)ih9(t)| should give the density matrix of the system, ρ(t). Then one has d ρ(t) = E {|9(t)ihϕ(t)| + |ϕ(t)ih9(t)|} + 2|φ(t)ihφ(t)|. dt
(80)
Now, if one requires that the master Equation (65) be satisfied, not only by the density matrix, ρ(t), but also by every projector |9(t)ih9(t)|, from Equations (65) and (80), one obtains the following equation of the quantum state diffusion (Percival, 2005): i Ď d|9(t)i = dt − H − LĎ L + hLiψ (2L − hLi9 ) |9(t)i h¯ + (L − hLi9 ) |9(t)idξ(t),
(81)
where an averaging over an individual trajectory is denoted by hLi9 = h9(t)|L|9(t)i. ¨ The stochastic Schrodinger equation obtained in Ito form (81) is equivalent to the Lindblad Equation (65). For practical applications of Equation (81) one should take account of the fact that the Ito form of the stochastic process increment imposes a specific form of time-discretization necessary for handling the problem numerically. An integral over the stochastic variable is then defined by Z
t
t0
f (τ )dξ(τ )
= lim I
N →∞
N X
f (tn )(ξ(tn+1 ) − ξ(tn )),
(82)
n=0
where tn = t0 + (n − 1)(t − t0 )/N , where f (t) is taken to be the ‘nonanticipation’ function, i.e. the one that does not depend on ξ(τ ) for τ > t. If one assumes that the stochastic integration is to be done using the following discretization scheme (so called ‘Stratonovich form’ of the
Theoretical Tools for Quantum Optics in Structured Media
140
stochastic integration) Z
t
t0
f (τ )dξ(τ )
= lim S
N →∞
N X 1 n=0
2
( f (tn+1 ) − f (tn ))(ξ(tn+1 ) − ξ(tn )), (83)
then stochastic increments are to be defined differently and in more ‘conventional’ form: dξ(t)dξ ∗ (t) = 0; thus one will have the usual differentiation rule d[AB] = [dA]B + A[dB]. Both forms (82) and (83) have their practical uses. Obviously, the ¨ Stratonovich form is more suitable for deriving stochastic Schrodinger equations directly from the Hamiltonian describing the system-reservoir interaction (see, for example, (Gardiner, Parkins, & Zoller, 1992); also, an example of such a derivation is given in the next Subsection). By contrast, the Ito form (as can be seen by comparing Equations (82) and (83)), is better suited for a numerical simulation. The Equation (81) can be transformed from the Ito form to the Stratonovich form, and vice versa(Gardiner, Parkins, & Zoller, 1992; Percival, 2005). Also, for practical applications one needs to know how dense the discretization should be and how many trajectories one needs to take to achieve the desired accuracy in a given interval of time. The answer to the latter question is the same as in Section 4.1; the analysis given there is independent of the particular kind of a stochastic process used for building the unravelling. To answer the former question, one should require that δtn be so small that the norm of a change in the trajectory caused by the diffusion be much less than the norm itself, i.e. hφ(tn )|φ(tn )i = δtn hLĎ (tn )L(tn )i − hLĎ (tn )ihL(tn )i 1. In the case of too rapid diffusion one can avoid using an exceedingly large number of discretization points by implementing the non-normalized version of Equation (81),
i Ď d|9(t)i = dt − H − L L + 2hLi9 L |9(t)i + L|9(t)idξ(t), h¯ where all the terms causing changes along the vector |9(t)i were dropped (Breuer, Kappler, & Petruccione, 1998). The quantum diffusion can be generalized for the case of coupling to different reservoirs in a way similar
Master Equations: Stochastic Approaches
141
to that done for quantum Monte-Carlo: each of the Lindblad operators, L j are put in correspondence with an independent Wiener process, ξ j (t). Also, in a similar manner, generalization to the case of mixed initial states can be made. To conclude, one can consider quantum diffusion as a limiting case of the quantum Monte-Carlo, when one takes the limit of an infinite number of jumps infinitesimally changing the system’s state.
4.5. Non-Markovian Quantum Diffusion Procedure The method of quantum diffusion can be generalized for an arbitrary non-Markovian reservoir composed of harmonic oscillators. Also, the non-Markovian quantum diffusion equation obtained can be used for derivation of both exact and approximate non-Markovian master equations. To illustrate the procedure of deriving the stochastic ¨ ¨ Schrodinger equation, let us start from the usual Schrodinger equation in the interaction picture (Diosi & Strunz, 1997; Strunz, Diosi, & Gisin, 1999): d 1 H + LĎ R(t) + RĎ (t)L |9(t)itot , |9(t)itot = −i h¯ dt
(84)
where the reservoir operator, R(t) =
X
a j g j exp{−i(w j − ω0 )(t − t0 )},
j
Ď
with bosonic creation and annihilation operators, a j , a j . The reservoir is initially supposed to be in the vacuum state, so one has hR(t)i = 0, and hR(t)RĎ (τ )i = K (t − τ ), where the reservoir correlation function is introduced by Equation (6). Now let us use the Bargmann representation for the parts of the total wave-function describing the reservoir Z |9(t)itot =
|9({α j }; t)i
Y d2 α j j
π
exp{−|α j |2 }|α j ),
(85)
where |9({α j }; t)i =
Y
α j |9(t)itot ,
j
Ď
and the unnormalized coherent states are |α j ) = exp{α j a j }|0 j i. Using the representation (85), one obtains from Equation (84) the
Theoretical Tools for Quantum Optics in Structured Media
142
following equation: i d |9({α j }; t)i = − H|9({α j }; t)i dt h¯ X − iL g j exp{i(w j − ω0 )(t − t0 )}α ∗j |9({α j }; t)i j
− iLĎ
X
g j exp{−i(w j − ω0 )(t − t0 )}
j
∂ |9({α j }; t)i. (86) ∂α ∗j
This equation can be conveniently rewritten by introducing the variable r (t) = i
X
α j g j exp{−i(w j − ω0 )(t − t0 )}.
j
Using the functional derivative over this variable, instead of Equation (86) one has the following equation (Diosi & Strunz, 1997; Strunz, Diosi, & Gisin, 1999): d i |9({α j }; t)i = − H + Lr ∗ (t) |9({α j }; t)i h¯ dt Z t δ − LĎ dτ K (t − τ ) ∗ |9({α j }; t)i. δr (τ ) t0
(87)
Formally, Equation (87) is no more than an equivalent form of the ¨ Schrodinger Equation (84). However, the form (87) can be considered ¨ as the stochastic Schrodinger equation for the particular trajectory |9({α j }; t)i, describing a realization of the reservoir plus system state. To that end one considers the function r (t) as a stochastic process (the so called ‘coloured noise’) satisfying (Gardiner & Zoller, 1999): E{r (t)} = 0,
E{r (t)r ∗ (τ )} = K (t − τ ).
(88)
Indeed, averaging over all the trajectories one will have ρ(t) =
Z |9({α j }; t)ih9({α j }; t)|
Y d2 α j j
π
exp{−|α j |2 } = E{|9(t)ih9(t)|},
where we have denoted a stochastic wave function as |9(t)i to emphasize the difference between the stochastic and deterministic approaches. An application of Equation (87) for practical calculations can be greatly
Master Equations: Stochastic Approaches
143
simplified by representing the derivative over the stochastic process, r (t), as the result of the action of some linear operator on the trajectory δ δr ∗ (τ )
|9(t)i = O(r, t, τ )|9(t)i.
(89)
Using such a representation, one can derive an exact master equation from Equation (87) (Yu, Diosi, Gisin, & Strunz, 1999): d i ¯ t)}] − [LĎ , E{O ¯ Ď (r, t)P9 }], (90) ρ(t) = − [H, ρ(t)] + [L, E{P9 O(r, dt h¯ where ¯ t) = O(r,
t
Z
dτ K (t − τ )O(r, t, τ ),
0
P9 = |9(t)ih9(t)|.
The possibility of using both Equations (87) and (90) depends on the possibility of finding a sufficiently simple exact form of the operator O(r, t, τ ) or an appropriate approximation of it. Apart from a few trivial cases (such as, for example, the problem of dephasing considered in Section 2.6), finding an exact form of O(r, t, τ ) is rather difficult. However, one can build a number of approximations for it by expanding O(r, t, τ ) with respect to the noise (i.e. the function r (t)), or time difference, taking into account that for t = τ one has simply O(r, t, t) = L (Yu, Diosi, Gisin, & Strunz, 1999). For example, an expansion of O(r, t, τ ) up to terms of the first order of the difference t − τ gives ¯ t) = f 0 (t)L − i f 1 (t) [H, L] − f 2 (t)[LĎ , L]L, O(r, h¯
(91)
where Z t dτ K (t − τ ), f 1 (t) = dτ (t − τ )K (t − τ ), 0 0 Z tZ τ f 2 (t) = dτ dx(t − τ )K (t − τ )K (τ − x).
f 0 (t) =
Z
t
0
0
¯ t) is independent of noise, so now The approximation (91) means that O(r, one obtains from Equation (90) a much simpler equation i d ¯ t)] + [LĎ , O ¯ Ď (r, t)ρ(t)]. ρ(t) = − [H, ρ(t)] + [L, ρ(t)O(r, dt h¯
(92)
144
Theoretical Tools for Quantum Optics in Structured Media
Obviously, by construction, the time-local Equation (92) can be adequate only for reservoirs with small correlation times (for that reason it was termed as ‘post-Markovian’ (Yu, Diosi, Gisin, & Strunz, 1999)). However, even very simple approximations for O(r, t, τ ) do not necessarily lead to such a restriction. For example, using a simple ‘nonlocal’ approximation O(r, t, τ )|9(t)i ≈ σ − |9(τ )i for our standard problem of the TLS damping into the bosonic reservoir, one gets the familiar time-nonlocal master Equation (44), obtained in the Born approximation. To conclude, the non-Markovian quantum diffusion method seems a versatile and efficient tool potentially able to tailor approximations to particular features of the reservoir structure.
5. CONCLUDING REMARKS The history of attempts to deal with non-Markovian reservoirs goes back for more than a half a century. In every field of quantum physics such problems continue to emerge, and intensive research efforts are being applied to develop methods of adequate consideration of the systemreservoir interaction, which allow us to take into account the effect of reservoir memory and a back-action of the reservoir on the system. One often needs to deal with non-Markovianity when considering emitters in photonic crystals and microcavities, when treating problems of coherent control and single-emitter spectroscopy. One needs to account for nonMarkovianity when describing an interaction of quantum dots, defects and vacancies in solid-state matrices with the electromagnetic field. However, one should admit that until now no universal recipe has been found for the solution of non-Markovian system-reservoir problems. Moreover, no universally applicable easy-to-use and well controllable approximation exists for this purpose. Up to now, one cannot safely resort to any set of rules to build equations of motion with ‘phenomenological’ terms responsible for the reservoir memory effects (even supposedly weak ones) in the same manner that one can construct a system of Bloch equations for emitter density matrix elements. One is not even guaranteed safe from obtaining non-physical, negatively defined density matrices after performing standard approximations (for example, the Born one) in a well-developed method for deriving master equations, such as the projector-operator method. On the other hand, for some particular problems remarkable progress has been achieved. For example, an exactly solvable model of a harmonic oscillator interacting with a reservoir of harmonic oscillators has been extensively and deeply investigated. It became the testing ground for asserting the validity of
References
145
various approximations, and for methods of deriving master equations. Also, for the problem of TLS interaction with the reservoir, a number of effective approaches have been suggested (and even an exact master equation has been built for the interaction of the TLS with the harmonic oscillators reservoir initially in the Gaussian state). Exact master equations can be built for reservoirs with densities-of-states of some specific shape, (for example, superpositions of Lorenzians), or with some specific form of the system-reservoir interaction (for example, the dispersive one, such as the problem of the TLS dephasing into a reservoir of harmonic oscillators). As things are at present, for every particular problem one should aim at developing an individual approach tailored to account for the particular features of the reservoirs present in the problem. For building approximations it is crucially important to derive correctly a border between the system of interest and the reservoir, where interaction with the system can be taken as sufficiently weak so as not to disturb significantly the state of this reservoir. Then, features of the structure of the defined reservoir (the correlation time, presence of gaps or other regions of sharp change in the density-of-states, etc.) may define a choice of the approximation necessary for capturing effects which might stem from these features. Exactly solvable cases for reservoirs with such structural features are able to provide clues as to what these effects might be.
ACKNOWLEDGEMENTS The authors are grateful to J. Peˇrina for his suggestion to write this review, and to V. Shatokhin and D. Horoshko for careful reading of the manuscript and for making numerous corrections. Grateful acknowledgement is given for partial financial support by EU under EQUIND project of 6FP IST-034368 and BRFFI.
REFERENCES Ackerhalt, J. R., & Rza¸zewski, (1975). Physics Review A, 12, 2549. ˙ Anastopoulos, C., & Hu, B. L. (2000). Physics Review A, 62, 033821. Apanasevich, P. A., Kilin, S. Ya., & Nizovtsev, A. P. (1988). Journal of Applied Spectroscopy, 47, 1213. Barnett, S. M., & Stenholm, S. (2001). Physics Review A, 64, 033808. Bay, S., Lambropoulos, P., & Mølmer, K. (1997). Physics Review Letters, 79, 2654. Bay, S., Lambropoulos, P., & Mølmer, K. (1998). Physics Review A, 57, 3065. Braginsky, V. B., & Khalili, F. Y. (1992). Quantum Measurement. New York: Cambridge University Press. Brandao, F. G. S. L., Hartmann, M. J., & Plenio, M. B. (2008). New Journal of Physics, 10, 043010. Brandes, T. (2005). Physics Reports, 408, 315. Breuer, H. P., Kappler, B., & Petruccione, F. (1998). Journal of Physics A, 31, L147. Breuer, H. P., Kappler, B., & Petruccione, F. (1999). Physics Review A, 59, 1633.
146
Theoretical Tools for Quantum Optics in Structured Media
Breuer, H. P., & Petruccione, F. (2002). The theory of open quantum systems. Oxford: Oxford University Press. Breuer, H. P. (2004). Physics Review A, 70, 012106. Breuer, H. P. (2004). Physics Review A, 69, 022115. Breuer, H. P., Gemmer, J., & Michel, M. (2006). Physics Review E, 73, 016139. Briegel, H.-J., & Englert, B.-G. (1993). Physics Review A, 47, 3311. Busch, K., Vats, N., John, S., & Sanders, B. C. (2000). Physics Review E, 62, 4251. Bykov, P. (1975). Soviet Journal of Quantum Electronics, 4, 861. Carruthers, P., & Nieto, M. M. (1968). Reviews of Modern Physics, 40, 411. Cahill, K. E., & Glauber, R. J. (1969). Physics Review, 177, 1857. Cahill, K. E., & Glauber, R. J. (1969). Physics Review, 177, 1882. Carmichael, H. J. (1993). An open system approach to quantum optics. In Lecture notes in physics. Berlin: Springer. Chaturvedi, S., & Shibata, N. H. F. (1979). Zeitschrift fur ¨ Physik B, 35, 297. Chaturvedi, S., & Srinivasan, V. (1991). Physics Review A, 43, 4054. Cyzek, J. (1966). Journal of Chemical Physisc, 45, 4256. D’aguanno, G., Mattiucci, N., Centini, M., Scalora, M., & Bloemer, M. J. (2004). Physics Review E, 69, 057601. Dalton, B. J., Barnett, S. M., & Garraway, B. M. (2001). Physics Review A, 64, 053813. Davies, E. B. (1976). Quantum theory of open systems. New York: Academic. Dehmelt, H. G. (1975). Bulletin of the American Physical Society, 20, 60. Dehmelt, H. G. (1987). Nature (London), 325, 581. Diosi, L., & Strunz, W. T. (1997). Physics Letters A, 235, 569. Divincenzo, D. P. (1995). Physics Review A, 51, 1015. Doll, R., Zueco, D., Wubs, M., Kohler, S., & Hanggi, P. (2008). Chemical Physics, 347, 243. Fain, B. (1988). Physics Review A, 37, 546. ˘ Fleischhauer, M., Imamoglu, A., & Marangos, J. P. (2005). Reviews of Modern Physics, 77, 633. Florescu, M., & John, S. (2004). Physics Review A, 69, 053810. Gardiner, C. W., Parkins, A. S., & Zoller, P. (1992). Physics Review A, 46, 4363. Gardiner, C. W., & Zoller, P. (1999). Quantum noise: A handbook of Markovian and nonMarkovian quantum stochastic methods with applications to quantum optics. Berlin: Springer. Garraway, B. M. (1997a). Physics Review A, 55, 2290. Garraway, B. M. (1997b). Physics Review A, 55, 4636. Ginzel, C., Briegel, H.-J., Martini, U., Englert, B.-G., & Schenzle, A. (1993). Physics Review A, 48, 732. Glancy, S., & de Vasconcelos, H. M. (2008). Journal of the Optical Society of America B, 25, 712. Harris, F. E., Monkhorst, H. J., & Freeman, D. L. (1992). Algebraic and diagrammatic methods in many-Fermion theory. New York: Oxford Press. Hayashi, T., Fujisawa, T., Cheong, H. D., Jeong, Y. H., & Hirayama, Y. (2003). Physics Review Letters, 91, 226804. Hillery, M., O’Connell, R. F., Scully, M. O., & Wigner, E. P. (1984). Physics Reports, 106, 121. ˘ Imamoglu, A. (1994). Physics Review A, 50, 3650. Ishizaki, A., & Tanimura, Y. (2008). Chemical Physics, 347, 185. Jaynes, E. T., & Cummings, W. T. (1963). Proceedings of IEEE, 51, 89. Jittoh, T., Matsumoto, S., Sato, J., Sato, Y., & Takeda, K. (2005). Physics Review A, 71, 012109. John, S. (1987). Physics Review Letters, 58, 2486. John, S., & Wang, J. (1990). Physics Review Letters, 64, 2418. John, S., & Wang, J. (1991). Physics Review B., 43, 12772. Khalfin, L. A. (1958). JETP, 6, 1053. Khalfin, L. A. (1968). JETP Letters, 8, 65. Kilin, S. Ya., & Nizovtsev, A. P. (1988). Journal of Physics B, 19, 3457. Kilin, S. Ya. (1990). Quantum optics: Fields abd their detection. Minsk: Nauka i Technica (in Russian). Kilin, S. Ya., & Krinitskaya, T. B. (1991). Journal of the Optical Society of America B, 8, 2289.
References
147
Kilin, S. Ya., & Krinitskaya, T. B. (1993). Physics Review A, 48, 3870. Kira, M., & Koch, S. W. (2006). Progress in Quantum Electronics, 30, 155. Klimov, A. B., & Romero, J. L. (2003). Journal of Optics B, 5, 316. Kofman, A. G., Kurizki, G., & Sherman, B. (1994). Journal of Modern Optics, 41, 353. Krylov, N. S., & Fock, V. A. (1947). JETP, 17, 93. Kubo, R. (1962). Journal of the Physical Society of Japan, 17, 1100. Lacroix, D. (2005). Physics Review A, 72, 013805. Lambropoulos, P., Nikolopoulos, G. M., Nielsen, T. R., & Bay, S. (2000). Reports on Progress in Physics, 63, 455. Lindblad, G. (1975). Commun. Math. Phys., 40, 147. Loudon, R. (1973). The quantum theory of light. Oxford: Clarendon. Louisell, W. H. (1977). Radiation and noise in quantum electronics. New York: Krieger. Mogilevtsev, D., & Kilin, S. (2002). Optics and Spectroscopy, 93, 405. Mogilevtsev, D., Kilin, S., & Cavalcanti, S. B. (2004). PNFA, 2, 161. Mogilevtsev, D., Kilin, S., & Onischenko, N. (2004). Laser Physics, 14, 692. Mogilevtsev, D., Moreira, F., Cavalcanti, S. B., & Kilin, S. (2006). Laser Physics Letters, 3(7), 327–344. Mogilevtsev, D., Nizovtsev, A., Kilin, S., Brandi, S., Cavalcanti, S. B., & Oliveira, L. (2008). Physics Review Letters, 100, 017401. Mølmer, K., Castin, Y., & Dalibard, J. (1993). Journal of the Optical Society of America, 10, 524. Nakajima, S. (1958). Progress of Theoretical Physics, 20, 948. Nikolopoulos, G. M., Bay, S., & Lambropoulos, P. (1999). Physics Review A, 60, 5079. Nikolopoulos, G. M., & Lambropoulos, P. (2000). Physics Review A, 61, 053812. Palma, G. M., Suominen, K.-A., & Ekert, A. K. (1996). Proceedings of the Royal Society of London, Series A, 452, 567. Paz, J. P., & Zurek, W. H. (2001). Environment-induced decoherence and the transition from quantum to classical. In R. Kaiser, C. Westbrook, & F. David (Eds.), Course 8 of Les Houches lectures session LXXII: Coherent atomic matter waves. Berlin: Springer. Percival, I. (2005). Quantum state diffusion. Cambridge: Cambridge University Press. Peˇrina, J. (1984). Quantum statistics of liner and nonlinear optical phenomena. Dordrecht: D. Reidel Publishing Company. Piilo, J., Maniscalco, S., Messina, A., & Petruccione, F. (2005). Physics Review E, 71, 056701. Piilo, J., Maniscalco, S., Harkonen, K., & Suominen, K.-A. (2008). Physics Review Letters, 100, 180402. Plenio, M. B., & Knight, P. L. (1998). Reviews of Modern Physics, 70, 101. Puri, R. R. (2001). Mathematical methods of quantum optics. Berlin: Springer. Manga rao, V. S. C., & Hughes, S. (2007). Physics Review Letters, 99, 193901. Redfild, A. G. (1955). Physics Review, 98, 1787. Salo, J., Barnett, S. M., & Stenholm, S. (2006). Optical Communications, 259, 772. Scully, M. O., & Whitney, K. G. (1972). Tools of theoretical quantum optics. In E. Wolf (Ed.), Progress in optics: Vol. X (pp. 89–135), Amsterdam: North-Holland. Scully, M. O., & Zubairy, M. S. (1997). Quantum optics. Cambridge: Cambridge University Press. Shibata, N. H. F., & Takahashi, Y. (1977). Journal of Statistical Physics, 17, 171. Sprik, R., Van Tiggelen, B. A., & Lagendijk, A. (1996). Europhysics Letters, 35, 265. Srinivas, M. D., & Davies, E. B. (1981). Optics Acta, 28, 981. Strunz, W. T., Diosi, L., & Gisin, N. (1999). Physics Review Letters, 82, 1801. Strunz, W. T. (2004). Physics Review A, 69, 052115. Sudarshan, E. C. G., & Misra, B. (1977). Journal of Mathematical Physics, 18, 756. Tyc, T., & Korolkova, N. (2008). New Journal of Physics, 10, 023041. Unruh, W. G. (1995). Physics Review A, 51, 992. Van Kampen, N. G. (1974). Physica, 74, 215; Van Kampen, N. G. (1974). Physica, 74, 239. Van Kampen, N. G. (2007). Stochastic processes in physics and chemistry (3rd ed.). Amsterdam: North-Holland Personal Library, Elsevier.
148
Theoretical Tools for Quantum Optics in Structured Media
Vats, N., John, S., & Busch, K. (2002). Physics Review A, 65, 043808. Wang, X.-H., Gu, B.-Y., Wang, R., & Xu., H.-Q. (2003). Physics Review Letters, 91, 113904. Wang, R., Wang, X.-H., Gu, B.-Y., & Yang, G.-Z. (2003). Physics Review B, 67, 155114. Weisskopf, V. F., & Wigner, E. (1930). Z. Physik, 63, 54. Xu, Y., Lee, R. K., & Yariv, A. (2000). Physics Review A, 61, 033807. Yablonovitch, E. (1987). Physics Review Letters, 58, 2059. Yoon, B., & Deutch, J. M. (1975). J. Chem. Phys., 62, 4687. Yu, T., Diosi, L., Gisin, N., & Strunz, W. T. (1999). Physics Review A, 60, 91. Zrenner, A., Beham, E., Stufler, S., Findels, F., Bichler, M., & Abstreiter, G. (2002). Nature, 418, 612. Zwanzig, R. (1960). J. Chem. Phys., 33, 1338.
CHAPTER
3 Polarization and Coherence Optics: Historical Perspective, Status, and Future Directions Christian Brosseau Universit´e Europ´eenne de Bretagne, Universit´e de Brest, Lab-STICC and D´epartement de Physique, CS 93837, 6 avenue Le Gorgeu, 29238 Brest Cedex 3, France
Contents
1. Introduction 2. From Bartholinus (1669) to Stokes (1852) 3. From Stokes (1852) to Poincare´ (1892) 4. From Poincare´ (1892) to Wolf (1954) 5. From Wolf (1954) to Wolf (2007) 6. Summary and Prospects Acknowledgments References
149 151 163 172 188 197 199 199
1. INTRODUCTION Despite an abundant literature, the polarization of light remains a somewhat mysterious property: it is an aspect of the visual world that is detected by insects and by many vertebrates other than mammals but is hidden from us.1 It is difficult to single out who first discovered polarized light. Since the first report of a polarization effect dating back from 1669, the polarization and coherence properties of light are a topic of enduring
E-mail address:
[email protected]. 1 Much of the light around us is polarized, but our eyes are not sensitive to this property and can detect
it only with the aid of special filters such as crystals or sheet polarizers. c 2009 Elsevier B.V. Progress in Optics, Volume 54 ISSN 0079-6638, DOI 10.1016/S0079-6638(10)05408-9 All rights reserved.
149
150
A History of Polarization and Coherence Optics
interest (Born & Wolf, 1980; Shurcliff, 1962; Strong, 1958). Today such studies are an area of intensive research that is increasingly leading to technological applications and scientific discoveries, e.g. light polarization reveals exoplanet characteristics (Berdyugina, Berdyugin, Fluri, & Piirola, 2008). Before getting to the presentation of the chief historical events over a period of more than three centuries that will be the focus of our review, a note is in order. Most students, and many of their mentors, pay little attention to the history of the problems they are trying to solve, assuming that it is of more interest to historians than to working scientists. The historian makes sense of things mainly by comparison, which comes in many forms. Nevertheless, it is hard to initiate a comparison of individuals who are unequal in talent and impact. When historians query the past—and it is usually people who are their object—they bring a sense of significance, a critical interest, to the dense texture they study. They mould what they find according to their temperament, which can be transformed, if the enterprise is carried out faithfully. On the other hand, the scientist may connect observations about the past in a serial fashion. In reality, nothing is more helpful to the novice scientist than a deep understanding of how his or her subject has evolved (French, 1990). Although this earlier literature is extremely interesting, the author does not advocate that all new workers in polarization theory set out to acquaint themselves with the minute details of these papers. For polarization and coherence optics, the chain of thought extends from the late 1660s to nowadays. In light of the above, the purpose of this review is to summarize some of the most important milestones, of the past three and a half centuries, along the road towards increased understanding of polarization optics. As we shall see, the nascent germ of many novel ideas in polarization and coherence optics can be traced back directly or indirectly to Stokes, and Poincar´e and other physicists from times long gone by. The story is divided into four main steps: from Bartholinus (1669) to Stokes (1852): polarization is a geometric property of light; from Stokes (1852) to Poincar´e (1892): polarization is an electromagnetic property of light; from Poincar´e (1892) to Wolf (1954): polarization is a statistical property of light; and from Wolf (1954) to Wolf (2007): polarization and coherence form a unified corpus for the study of light. The goal of this essay is not to report particularly noteworthy original research results but rather to give an extended and largely nontechnical introduction to basic ideas behind the concepts of polarization and coherence of light. It is designed to present a general perspective that would be understandable to anyone with an
From Bartholinus (1669) to Stokes (1852)
151
elementary knowledge of classical optics. Here we present some historical remarks, catalog relevant central concepts and ideas in the development of understanding polarization optics with which we consistently deal in their classical form. The word “classic” is used here to mean “before quantization”. It is important to keep in mind that, while the development of quantum physics has revealed a world quite different from the one depicted by classical physics, quantum coherence and polarization effects are outside the scope of the present review. We are forced to recognize that the progression in the process of gaining knowledge presented here is an order imposed in retrospect by the author. We will see that the list of contributors includes some of the greatest names in physics. In most of these cases the motivation of the research work was fundamental knowledge. Dates refer to the fundamental papers or reports. Ample references to most original literature are given at the end of this review for those who desire source material and those who want to place in context and appreciate the considerable achievements of people who have contributed solidly in their own fields. Little detail is provided on personal lives, the emphasis being placed on the technical developments. It is an entirely subjective account since it is a difficult task to appropriately weight individual contributions in retrospect. This review is organized in five parts, corresponding to the four steps mentioned above, and a summary where future prospects are also discussed.
2. FROM BARTHOLINUS (1669) TO STOKES (1852) Let us begin our excursion by reviewing a few facts about the scientific context of the seventeenth century. Optics in the seventeenth century was crucial for the intellectual development of Europe. At that time, purely geometrical (Euclidian) considerations sufficed to explain most major optical observations, i.e. the rectilinear propagation of light through transparent media (Fermat’s principle). Much less had been formulated concerning the physical principles underlying the very nature of light. Light was considered as the instantaneous propagation of an action (of mechanical origin) through a kind of subtle ethereal matter (luminiferous aether) which fills all space (Ronchi, 1957; Sabra, 1967). It is in this context that the story of Polarization Optics starts. It is rather difficult to tell who was the first in Polarization Optics. Although it is a part of the established folklore that the Vikings (circa 700) might have used polarization of skylight to find their way across the Atlantic using the famed sunstone mentioned in the medieval sagas, (LaFay, 1970; Wehner, 1976) this belief should be treated with extreme caution (Roslund & Beckman,
A History of Polarization and Coherence Optics
152
(a)
(b)
FIGURE 1 (a) Erasmus Bartholinus. (b) The front page of the 60-page Erasmus Bartholinus memoir
1994).2 The discovery of polarization is usually attributed to Bartholinus (1670). Erasmus Bartholinus (Figure 1(a)), a Danish mathematician at the University of Copenhagen, is credited with the first, scientifically presented, observation of a polarization effect. In 1669 he made the pivotal discovery of polarization by double refraction using a crystal of Iceland spar (i.e. calcite, a rhombohedral crystalline form of calcium carbonate CaCO3 which was discovered in Iceland and then known as Iceland spar) and published his observations in a 60-page memoir (Figure 1(b)) entitled “Experimenta crystalli Islandici disdiaclastici quibus mira et insolita refractio delegitur” (Bartholinus, 1670). In point of fact he observed the splitting of an incident beam into two beams of equal intensity after passing through a crystal. One of these rays, 2 The sunstone theory was proposed in 1966 by the Danish archaeologist Thorkild Ramskou. Yet the only hint that Vikings looked towards the sky through rock crystals to give them direction comes from a mention in a Viking legend, known as a saga. No one had tested the polarimetric navigation hypothesis until recently. In 2005, Horvath and a group of the Biooptics Laboratory at the Eotvos University in Hungary (Hegedus, Akesson, Wehner, & Horvath, 2007) sailed the Arctic Ocean aboard the Swedish icebreaker Oden and found that sunstones could indeed light the way in foggy and cloudy conditions. In previous studies Horvath and colleagues showed that Vikings might have required some kind of device, other than just the naked eye, to accurately guess the position of the sun on cloudy days. However, unlike with Viking sundials, archaeologists have yet to find any remnants of sunstones. For now, physics cannot rule one way or another until further research reveals more about the Vikings’ culture and technology— anything else is speculation. But the political aspect of speculation is intriguing in itself: it is easy to romanticize the past, to project modern knowledge onto ancient peoples, without realizing that we may be making too many assumptions. On the other hand, we have often underestimated the abilities of past peoples, and been surprised at discovering that some of our modern inventions have been invented more than once.
From Bartholinus (1669) to Stokes (1852)
153
FIGURE 2 Christian Huyghens
i.e. the ordinary ray, obeys the usual law of refraction, the other does not, and hence is called the extraordinary ray. At this point, it is worth noting that the challenge of accounting for the optical properties of crystals gave the starting impetus to the study of polarization optics. In 1672 Christian Huyghens (Figure 2), a Dutch physicist, interpreted the double refraction phenomenon from his conception of a spherical light wave (i.e. envelope construction) and observed that each of the two beams arising from the double refraction phenomenon can be extinguished by passing through a second calcite crystal which is rotated about direction of the beam (Huyghens, 1690, 1950; Ziggelaar, 1980). His investigations showed also, that the two beams have different polarization directions, and that the condition for the separation of light not to occur was that light traversed the crystal parallel to the crystallographic axis. He subsequently completed, in 1690, his famous book on light, “Trait´e de la Lumi`ere”, in which he developed the geometrical theory required to interpret all optical phenomena known to this time as: reflection, refraction and double refraction (Huyghens, 1690, 1950). His reasoning was based upon the principle named after him, according to which, light behaves like a spherical wave emitted from a source; each point of the aether on which light falls becomes the source of spherical wavelets (secondary waves) that propagate in all directions and that combine to form the wavefront (the envelopes of the spherical waves) perpendicular to the direction of propagation. That was a major conceptual advance. Note that the ¨ finiteness of the velocity of light demonstrated by Ole Romer in 1676 was
A History of Polarization and Coherence Optics
154
(a)
(b)
FIGURE 3 (a) Sir Isaac Newton. (b) The front page of Newton’s book on optics
one of the two basic ingredients of Huyghens’ analysis, the other being the wave nature of light (Born & Wolf, 1980; Ronchi, 1957; Sabra, 1967). In 1704 Sir Isaac Newton (Figure 3(a)) published his celebrated treatise (Figure 3(b)) on optics (“Opticks”) which was the standard treatise on this subject for many decades (Newton, 1952; Shapiro, 1994). Newton was one of the most famous “emissionists”, i.e. a supporter of the corpuscular or particle theory of light (Buchwald, 1989; Skolnik, 1970; Whittaker, 1960). In this view a beam of light comprises many rays, identified with geometrical lines, which are in turn composed of streams of particles. The differences in colours come from the differences in particle size and mass. In this scheme refraction and reflection are understood in terms of Newtonian forces acting between these particles and the interface. Since these ideas were embedded in his theory of gravitation (a theory that was verified in 1758 by the return of the Halley comet, predicted in 1705), it was essentially this view that prevailed through the eighteenth century because of the great prestige lent it by Newton. The seed of a scientific breakthrough may slumber for a time without generating much interest. It is not until the start of the nineteenth century that we see the next progress in the field. This period marked the turning of the tide in favor of the view that light vibrations are transverse to the direction of propagation, the evidence for which had been advanced vigorously by Thomas Young (Figure 4) (James & Wolf, 1991; Young, 1802, 1804). The double-slit optical interference experiment of Young, performed in 1801, was a key discovery to the understanding of the wave theory of light, according to which he established that beams of light can constructively and destructively interfere. Young’s
From Bartholinus (1669) to Stokes (1852)
155
FIGURE 4 Thomas Young
experiments were crucial because destructive interference was not a possibility according to the Newtonian corpuscular theory. Another big advance in the understanding of the nature of light came when Thomas Young showed in 1803 that polarization phenomena arise from the transverse nature of light. The latter recognition gave impetus to the wave theory of light. Bear in mind that this theory predates the electromagnetic theory of light by more than six decades. The next great progress came with Malus (Figure 5). In 1808 EtienneLouis Malus, a military French engineer discovered the polarization of natural light by reflection while experimenting with a crystal of Iceland spar and light reflected by the windows of the Palais du Luxembourg in Paris. By extensive experimentation, he showed by purely geometric reasoning how to express the intensity of light emerging from a polarizing crystal when the light it receives is linearly polarized along a direction making a specific angle with its axis, in a paper entitled “Sur une propri´et´e de la lumi`ere r´efl´echie” (Bohren, 1988; Malus, 1808). It has been customary to give to physical effects names that honour the scientists who first draw attention to them. This discovery is no exception and is known as Malus’s law. In 1812 Sir David Brewster (Figure 6), a Scottish clergyman-turnedphysicist at Saint Andrews College, enunciated the tangent law named after him in a couple of celebrated papers: “On the laws of polarization and double refraction in regularly crystallized bodies” and “On the laws which regulate the polarization of light by reflection from transparent
156
A History of Polarization and Coherence Optics
FIGURE 5 Etienne-Louis Malus
FIGURE 6 Sir David Brewster
bodies” (Barr, 1988; Brewster, 1815, 1816, 1987; Lakhtakia, 1989). What Brewster did was to show how the polarizing angle (the angle of incidence at which unpolarized light becomes completely polarized upon reflection by an optically smooth, planar interface between two dissimilar transparent media) depends on composition (by way of refractive index). Like Malus, he investigated polarization by reflection of natural light by planar interfaces between two media. One of his early achievements was his discovery that at a particular angle of incidence, the reflected light from glass passing through a calcite crystal could be
From Bartholinus (1669) to Stokes (1852)
157
FIGURE 7 Augustin Jean Fresnel
extinguished. This angle is called the Brewster angle. Moreover complete polarization occurs when the angle of incidence equals Brewster’s angle. Photoelasticity (strain birefringence) was also described by Brewster in 1816: the possibility that isotropic transparent optical media could be made optically anisotropic by the application of mechanical stress is a topic of profound significance on which Brewster has had a seminal influence. The theory was worked out subsequently by Neumann (1840). Brewster also invented and patented the kaleidoscope (Brewster, 1987).3 A major advance in the understanding of light polarization was made by Augustin Jean Fresnel (Figure 7). In 1823 he derived, on the basis of the elastic theory of aether (the concept of an ethereal medium, filling space, was formulated by Descartes two centuries earlier), his famous reflection and transmission formulas for a plane wave that is incident upon a static and plane interface between two dielectric isotropic media, in an article entitled “M´emoire sur la double r´efraction que les rayons lumineux e´ prouvent en traversant les aiguilles de cristal roche suivant des directions parall`eles a` l’axe” (Fresnel, 1825, 1866). It is important to note that Fresnel’s equations require arbitrary sign conventions which may led to confusion: a fact already mentioned by Fresnel as the “petite difficult´e”. Fresnel gave also a heuristic explanation of the phenomenon of optical activity and recognized the phenomenon of optical rotation arising from circular birefringence, i.e. such material has an index of refraction for right-circularly polarized light which is different from the index for left-circularly polarized light. His pioneering work concerning the diffraction caused by small apertures and screens is also fundamental. 3 The kaleidoscope was the television of the 1850s and no respectable home would be without a kaleidoscope in the middle of the library.
158
A History of Polarization and Coherence Optics
FIGURE 8 Dominique Franc¸ois Arago
To Fresnel must be assigned the credit for discovering the modern concept of polarization of light and for stimulating the efforts which put the wavetheory of light on firm foundations. In the years 1812–1815 came an important milestone by the French physicist Dominique Franc¸ois Arago (Figure 8) at the Paris Observatory. Arago performed many fundamental investigations on optical activity in quartz, i.e. the continuous rotation of the direction of vibration of light by propagation along the optic axis of a crystal. Arago invented the pile-of-plates, sometimes credited, incorrectly, to Stokes, whose paper on the pile-of-plates-polarizer followed Arago’s by more than 20 years (Arago, 1854). On the theoretical side, his principal contribution was the discovery of the interference laws published in a joint paper with Fresnel, “Sur l’action que les rayons lumineux polaris´es exercent les uns sur les autres”, which played a key role in the demonstration of the transverse nature of light waves propagating in free space (Arago & Fresnel, 1819). Arago and Fresnel’s contribution was to enunciate four interference laws governing the interference of polarized light. These laws were determined experimentally and are summarized by the following statements (in modern version): (1) Two waves, linearly polarized in the same direction, can interfere. (2) Two waves, linearly polarized with orthogonal polarizations, cannot interfere. (3) Two waves, linearly polarized with orthogonal polarizations, if derived from the orthogonal components of unpolarized light and then brought into the same plane of oscillation, cannot interfere. (4) Two waves, linearly polarized with orthogonal polarizations, if derived from the same linearly polarized wave and then brought into the same plane, can interfere (Barakat, 1993; Collett, 1971; Mujat, Dogariu, & Wolf, 2004).
From Bartholinus (1669) to Stokes (1852)
159
FIGURE 9 Jean-Baptiste Biot
In 1815 there soon followed the discovery, by Jean Baptiste Biot (Figure 9), of the optical activity of some materials described in the article “Sur un mode particulier de polarisation qui s’observe dans la tourmaline”; he performed experiments of rotatory polarization to distinguish between left- and right-handed rotation of the direction of vibration of light (Ronchi, 1957; Sabra, 1967). Shortly after, in 1822, Sir John Herschel, astronomer at Cambridge University, presented evidence that the rotatory polarization of quartz originates from two different crystallographic structures: a phenomenon called eniantomorphism (Herschel, 1841). In point of fact the first decades of the nineteenth century were a very exciting time in optics. Many of the basic designs of innovative optical instruments were improved and made many discoveries possible. For example, the development of the optical microscope opened new realms to human investigation. This encouraged opticians to make numerous measurements on a wide variety of systems and at the same time made it possible to observe many subtle effects. Many landmark ideas come out of these studies. The first birefringent polarizer was realized in 1828 by the Scottish physicist William Nicol, at the University of Edinburgh. Nicol is known as being the inventor of the polarizing prism that now carries his name, described in an article “On a method of so far increasing the divergency of the two rays in calcareous-spar that only one image may be seen at a time” (Nicol, 1829). The first polarizing microscope was built by H. F. Talbot in 1834 and used by Brewster for studying minerals. Several ingenious optical designs have been carried out in an attempt to improve the quality of measurements. Significant
160
A History of Polarization and Coherence Optics
FIGURE 10
Louis Pasteur
contributions were made including the polariscope in 1830 by A. Seebeck, ¨ the “pince a` tourmalines” by J. Muller, the compensators by Babinet and ¨ Senarmont, the Norrenberg reflecting polariscope and the Glan-Foucault polarizers and the Wollaston and Rochon prisms (Ronchi, 1957; Sabra, 1967). In 1846, Wilhelm Haidinger, an Austrian mineralogist, discovered the remarkable brush phenomenon that now bears his name. This optical curiosity is observable and apparent to almost anyone: it is manifested as specific patterns that can be seen when one looks at polarized skylight (Haidinger, 1846, 1847; Minnaert, 1954; Misson, 1993; Shurcliff, 1955; Sloan, 1955). The following year, Haidinger was the first to report an ¨ observation of circular dichroism in an article “Uber den pleochroismus des amethysts”. The first to realize that molecular symmetry is responsible for the phenomenon of optical rotation was Louis Pasteur (Figure 10) in his doctoral research done in 1848 (Charney, 1979; Pasteur, 1907). Examining different salts of tartaric acid he noticed the formation of two types of crystal, each one being a mirror image of the other (eniantomorphs). Separating the two types of crystals, then dissolving these separately in water, he found that one solution caused the direction of polarization of linearly polarized light to rotate clockwise, the other counter-clockwise. The remarkable discovery of the ability of left-handed and right-handed molecules to rotate light differently is at the basis of what is termed now chiral asymmetry that plays a fundamental role in the chemistry of life. It is likewise instructive to note that William Bird Herapath, a physician and surgeon in Bristol, England, observed, for the first time, linear dichroism using crystals of quinine sulphate periodide (now known as
From Bartholinus (1669) to Stokes (1852)
FIGURE 11
161
Sir George Gabriel Stokes
herapathite).4 The high point of Herapath‘s scientific career was the discovery in 1852, published in the Philosophical Magazine under the title “On the optical properties of a newly discovered salt of quinine which crystalline substance possesses the power of polarizing a ray of light, like tourmaline and at certain angles of rotation of depolarizing it like solenite” (Herapath, 1852, 1855). A major advance to the field came by Sir George Gabriel Stokes (Figure 11). Stokes introduced four measurable quantities that now bear the name of Stokes parameters for describing the properties of polarized light. He published the seminal paper, “On the composition and resolution of streams of polarized light from different sources”, in 1852 (Stokes, 1852, 1901). The Stokes parameters describe light of any state of polarization, from unpolarized to partially polarized to completely polarized. Despite the impressive insight and relatively simple concepts underlying Stokes’s original paper, this remarkable study received little attention at that time but would become fully appreciated many decades later. Although his basic motivation for the introduction of these four parameters was to describe mathematically unpolarized light, his parametrization is applicable to any state of polarization, i.e. to partially polarized, as well as, unpolarized and completely polarized waves. In a sense the modern description of polarization optics can be said to have started 4 Herapath, whose pupil, a Mr. Phelps, had found that when he dropped iodine into the urine of a dog that has been fed quinine, little scintillating green crystals formed in the reaction liquid. Phelps went to his teacher, and Herapath then looked at the crystals under a microscope and noticed that in some places they were light where they overlapped and in some places they were dark. He was shrewd enough to recognize that here was a remarkable phenomenon, a new polarizing material.
162
A History of Polarization and Coherence Optics
FIGURE 12
John Tyndall
from the time of the publication of papers by Stokes. This was an entirely new and unexplored point of view because his description of light was formulated in terms of intensities rather than field vectors, thus dealing with measurable quantities at optical frequencies. The first parameter represents the total intensity of the field and the three others describe the state of polarization. He also formulated a mathematical statement for unpolarized light: the intensity of unpolarized light is unaffected by any rotation of the axes of the reference coordinate system and any phase change introduced in one of the wave components. Stokes was eventually the best interpreter of the interference laws that Fresnel and Arago discovered. To polarization, Stokes brought also, in 1849, the principle of reciprocity (reversibility) involving reflection and transmission of light beams upon an interface between two dielectric media (Stokes, 1849). In 1860 Giorgo Govi, an Italian astronomer, published his finding that the light scattered by small particles can be polarized (in his experiment, he filled a closed room with smoke and directed a beam of sunlight onto it) and rediscovered several observations first made by Arago (Govi, 1860a,b). His conclusion, that the scattering of light by small particles is a phenomenon which is independent of the reflection of light (a common idea shared at the time) opened the science of the scattering of radiation by matter. The early work in environmental monitoring science was led by the Scottish physicist John Tyndall (Figure 12). His brilliant experiments evidencing that the scattering of light by particles depends significantly on the size of the scatterers are particularly noteworthy. He was the first to find that a cloud of very fine particles polarizes perfectly the beam at 90◦ scattering angle. Tyndall’s 1869 study of photochemical smogs, published under the title “On the blue colour
From Stokes (1852) to Poincar´e (1892)
163
of the sky, the polarization of sky light and on the polarization of light by cloudy matter in general”, would be quite decisive to the subsequent analysis of Rayleigh (Tyndall, 1869a,b). More specifically, he was the first to observe the bluish colour of light after scattering of natural light. Moreover he invented many instruments and used them to conduct nephelometric research into the colour of water and sky. Let me interrupt the story here to say something about the subject before the entry of the seminal electromagnetic theory of Maxwell. Two features are of special pertinence to the following discussion. First, the transverse wave theory of light could provide an understanding of the major optical phenomena discovered at the time of Maxwell’s treatise: propagation, polarization, diffraction and interference. The majority of scientists became convinced of the superiority of the wave theory in the late 1820s and early 1830s, even though this theory said nothing about the source of the optical field (Buchwald, 1989; Skolnik, 1970; Whittaker, 1960). Second, in spite of many difficulties, the mechanical theory of the elastic aether persisted.
3. FROM STOKES (1852) TO POINCARE´ (1892) A new chapter in this story was the entry of Maxwell’s theory of electrodynamics. The electromagnetic nature of light was established and the connections between radiation of many different kinds was discovered. The story involved many of the intellectual giants of the nineteenth century. This period began with the Danish physicist Oersted in 1820. By experimenting with the effect of a current on a suspended magnetic needle he made the fundamental discovery that magnets interact with electric currents (Ronchi, 1957; Sabra, 1967; Scott, 1963). The second third of the nineteenth century brought a profound change in physics. A chief contributor to the fundamental aspects of early electromagnetic theory was the British physicist Michael Faraday (Figure 13). Faraday was one of the greatest experimental scientists of all time. In 1831, he discovered electromagnetic induction, subsequently explained para- and dia-magnetism and interpreted them through his field theory. When Faraday postulated the physical laws of electromagnetism he had in mind a mechanical picture, deduced from geometric reasoning, using the two concepts of lines of force traversing all space and actions-at-a-distance exerted between the particles in a medium (Faraday, 1964). In 1845 he discovered the phenomenon, which now bears his name, of magnetically induced optical rotation of the direction of vibration of light upon passing through a medium in a direction parallel to the magnetization. His eponymous law has become the earliest indication of the relationship between electromagnetism and light (Faraday, 1964).
164
A History of Polarization and Coherence Optics
FIGURE 13
Michael Faraday
FIGURE 14
James Clerk Maxwell
According to Einstein, Faraday (along with Maxwell) was responsible for the greatest change in the axiomatic basis of physics since Newton. In 1864, James Clerk Maxwell (Figure 14) completed a six-page memoir entitled “A dynamical theory of the electromagnetic field”, in which he developed the mathematical theory required for the description of how electromagnetic waves propagate (Harman, 1990; Maxwell, 1864, 1962; Siegel, 1991).
From Stokes (1852) to Poincar´e (1892)
165
The formulation of Maxwell’s theory certainly ranks as one of the prime events of nineteenth century physics: his eponymous equations have become almost as immutable as Newton’s law of gravity. These equations summarize the fundamental relations between electricity and magnetism and became the cornerstone upon which generations of scientists have based their theoretical studies. Maxwell put Faraday’s concepts into the elegant mathematical form of four differential equations and one of his major innovations was to introduce the notion of a displacement current. Indeed, were it not for the displacement current, it would not be possible to deduce from Maxwell’s equations that electromagnetic waves have the property of light. In the preface to his monumental Treatise on Electricity and Magnetism, he observed that “As I proceeded with the study of Faraday, I perceived that his method of conceiving the phenomena was also a mathematical one, though not exhibited in the conventional form of symbols”. By Maxwell’s time, scientists distributed the results of their research by means of correspondence. The reader may wish to consult Harman for an interesting discussion of letters Maxwell wrote to Faraday where he described his incorporation of his mechanical concepts to the theory of the electromagnetic field (Harman, 1993; Rosenfeld, 1957). According to Maxwell’s mechanical model, the electric field represents a physical stress in the aether and the magnetic field represents the rate of change in the stress field. Thus electromagnetic radiation is a result of electric oscillations. The Maxwell’s equations allow for the possibility of transverse wave solutions, and the corresponding phase velocity, expressed in terms of electric and magnetic quantities, is remarkably close to the free-space speed of light. Moreover, the transversality of the vibrations is consistent with the observation of the Arago-Fresnel interference experiment. Electromagnetism is often viewed as a concept-driven revolution in the sense proclaimed by Thomas Kuhn, i.e. a revolutionary new way of looking at nature (HoymingenHuene, 1993; Kuhn, 1970). The determination of Rudolph Kohlrausch and Wilhelm Weber of the velocity of the electromagnetic waves was found in accordance with the velocity of light carried out by Henry Fizeau: The principle then emerged that light has the properties of electromagnetic waves (Kohlrausch & Weber, 1856; Ronchi, 1957). These fundamental researches constituted the seed corn that has led to a scientific harvest of inestimable value today. With the basic physics introduced by Faraday and Maxwell, the theory of electrodynamics followed the development of the necessary mathematics with contributions from Poisson, Green, Thomson (Lord Kelvin) and others (Mach, 1926; Ronchi, 1957). Although now largely forgotten, significant progress in the study of polarized light was reported by Emile Verdet in 1869 (Verdet, 1865, 1869). Verdet may be
166
A History of Polarization and Coherence Optics
singled out for mention as having been the first to develop a rigorous analysis, surprisingly modern in tone, of unpolarized light. Another accomplishment of Verdet in this area concerns the Faraday effect. Verdet’s constant of proportionality links the angle of rotation and the product of the magnetic field strength with the distance the light travels through the medium. Also worthy of note are the investigations of unpolarized light by J. Stefan published in 1864 (Stefan, 1864a,b). On the experimental side, a large number of important experiments were made, permitting the observation and the study of natural optical rotation, and natural and magnetic linear and circular dichroism. In 1875, John Kerr, a Scottish physicist, discovered the quadratic electro-optic effect that bears now his name. Indeed the full title of Kerr’s article was: “A new relation between electricity and light: dielectrified media birefringent” (Kerr, 1875; Yariv & Yeh, 1984). By experimenting with a block of glass between crossed polarizers and applying an electric field (∼105 V m−1 ) normal to the optic axis, Kerr found that the glass becomes birefringent (proportional to the square of the field) and part of the light is transmitted. In 1893 the German physicist Friedrich Karl Pockels discovered a new (linear) electro-optic effect: in the Pockels effect, the field is parallel to the direction of light. When a dc electric field (∼105 V m−1 ) is applied to a noncentrosymmetrical crystal, e.g. Lithium niobate LiNbO3 , the refractive index changes linearly with the applied voltage, and incident linearly polarized light becomes elliptically polarized (Kerr, 1875; Yariv & Yeh, 1984). A variety of other magnetooptic phenomena were discovered over the years: the Kerr magnetooptic effect is the analogue of the Faraday effect in the case of light reflected off a magnetic or magnetized material; the Cotton–Mouton effect is double refraction of light passing perpendicular to the magnetic field applied to the material, i.e. the magnetic equivalent of the Kerr effect (Kerr, 1875; Yariv & Yeh, 1984). Materials that show large electro-optic and magneto-optic effects have become important in the transmission, storage, and retrieval of digital data. A major experiment was performed by O. Wiener in 1892 that demonstrated that the electric field vector of an electromagnetic wave exerts a greater force on electrons than the magnetic field does (provided that one considers nonrelativistic speeds). Thus the electric field is responsible for changes of the polarization state upon interaction of light with matter. This discovery is at the basis of the convention to choose the direction of the electric field as the polarization direction of the lightwave (Wiener, 1890). A great experimental support to Maxwell’s theory of electromagnetism came with Hertz’s experimental work. In the years 1887 and 1888 Heinrich Rudolf Hertz (Figure 15), a German physicist at the Technical University in Karlsruhe, produced and detected electric waves in air, thus
From Stokes (1852) to Poincar´e (1892)
FIGURE 15
167
Heinrich Rudolf Hertz
demonstrating the application of the concepts of the electromagnetism theory to the microwave and radio regions of the spectrum. Among Hertz’s achievements was the experimental confirmation of the transverse nature of electromagnetic waves through experiments on reflection, refraction, interference and polarization, in accordance with the predictions of Maxwell’s theory (Hertz, 1888, 1896; Hertz, Hertz & Susskind, 1977; Mulligan, 1987). Hertz’s experimental work lead to the technology of radio and microwave communications, detection, and wire-grid polarizers to test the properties of radiowaves. In addition to being a remarkable experimentalist he was a first-rate theoretical physicist: His fundamental work helped lay the foundation for quantum theory and relativity. On the one hand, Hertz’s observation of the photoelectric effect in 1887 played an important role in the development of quantum physics and his works on cathode rays with Lenard was crucial for Wilhelm Roentgen’s discovery of X- rays in 1895. On the other hand, Hertz’s investigations into the electrodynamics of moving bodies advanced ideas that eventually led to Einstein’s special theory of relativity. In the late 1890s appreciation of the significance of the electromagnetic theory was increasing. Within this context, in 1896, Zeeman identified a fundamental rule concerning the broadening of the D-lines of sodium atoms in a magnetic field. Then, on the basis of the experiments by Zeeman, Hendrik Antoon Lorentz quickly developed a theory and was able to predict the polarization behaviour of the spectral lines. These predictions were experimentally verified by Zeeman. Thus, all optical phenomena can be accounted for by the interaction with electromagnetic fields. This success had a vast and immediate impact on the acceptance of the electromagnetic theory of light
168
A History of Polarization and Coherence Optics
(a) FIGURE 16
(b)
´ (b) Henri Poincare´ with Marie Curie (a) Henri Poincare.
within the optics community. Indeed, Maxwell’s successful theoretical explanation of the observed phenomena of light propagation did not lead to a universal acceptance of the electromagnetic hypothesis, which was by no means evident to many leading scientists in the end of the nineteenth century (Buchwald, 1989; Whittaker, 1960). That work won Lorentz and Zeeman the 1902 Nobel prize in physics. Several other works are noteworthy. The reciprocity theorem, due to Hermann von Helmholtz in 1881, is fundamental in scattering processes. It expresses the relationship existing when the incident and scattered beams are interchanged. Around 1889 Drude rederived Fresnel’s formulae from Maxwell’s equations and this constitutes another important step in the acceptance of Maxwell’s theory (Buchwald, 1989; Cantor, 1975; Ronchi, 1957). A major breakthrough was achieved in 1890 with the publication of a 96 page-memoir “Upon the reflection and refraction of light by a nonmagnetic transparent homogeneous sphere of arbitrary size and refractive index”, by the mathematical physicist Ludwig V. Lorenz of Copenhagen which predated the classic work of Mie (1908) about particle light scattering and absorption (Kragh, 1991; Logan, 1965; Lorenz, 1890). History justly positions the French mathematician and professor of mathematical physics at the Sorbonne, Paris, Henri Poincar´e (Figure 16(a); (b)) as a dominant figure in mathematical physics. His ideas, discoveries and techniques pervade all areas of theoretical physics. Poincar´e published in 1892 a mathematical treatment of polarized light where he introduced the Poincar´e sphere and the complex plane representations, to specify the state of polarization (a convenient summary of much of his
From Stokes (1852) to Poincar´e (1892)
169
work is contained in his magnum opus on optics Th´eorie Math´ematique de la Lumi`ere which was the standard treatise on optics for many years) (Poincar´e, 1892). It was Poincar´e, more than any other, who truly saw the physical implications of geometry in polarization optics. Using a stereographic projection, he mapped each point on the plane into a sphere whose points are in one-to-one correspondence with all the possible states of polarization of a light beam. One of the conveniences of the Poincar´e sphere is that it provides an intuitively geometric view of the transformation of polarized light when it interacts with optical devices in terms of rotations of states. Oddly enough the importance of the Poincar´e sphere in polarization optics was not realized until the second-half of the twentieth century. Poincar´e’s large body of work on optics is a prime example of his elaboration of ideas in new fields. It is remarkable how many important ideas he put forward, establishing the basic concepts of modern chaos and dynamical systems theory. He was a prolific author (Poincar´e’s long and diverse bibliography contains more than 500 articles) and wrote several fine books on the theory of differential equations and dynamics which were widely used texts. His accomplishments as a theoretical physicist make best sense when they are placed in this broader perspective of his scientific interests, giving any modern scientist ample cause for inspiration. At the turn of the twentieth century, John William Strutt, the third Baron Rayleigh (Figure 17), more familiar as Lord Rayleigh, Cavendish professor of experimental physics at the Cavendish Laboratory in Cambridge and then professor of natural philosophy5 at the Royal Institution of Great Britain in London, published many fascinating articles in optics. Although he received the Nobel prize for physics in 1904 for his discovery of argon, Rayleigh had a long and extraordinarily productive career. Lord Rayleigh contributed to almost every branch of classical physics: optics, acoustics, electromagnetism, thermodynamics, statistical mechanics, and others. His collected papers contain more than 400 papers. An important characteristic of Lord Rayleigh’s way of doing physics was his close attention to experiment (Howard, 1964; Strutt (Lord Rayleigh), 1871a,b,c, 1881, 1900, 1920; Young, 1982). One of his major contribution came in 5 There was a time when science and philosophy were essentially the same discipline. The world produced natural philosophers – not physicists, or chemists. As knowledge and understanding of nature expanded, scientists and philosophers drifted apart and the former fragmented into different disciplines and subdisciplines. In that not too distant past, it was not uncommon for “the man at the bench” to write his own account of the history of his field. Today, however, most working scientists display little interest in philosophical discussions about the scientific method or the role of hypotheses in discovery, nor do they appear particularly knowledgeable about past events. We now have a bevy of historians of science chronicling and interpreting the various facts of discovery, investigating sociological attitudes or looking at the role of religious belief in the formulation of hypotheses. There appears to be little dialogue between these historians and philosophers who write about science and those who actually do the work of discovery.
170
A History of Polarization and Coherence Optics
FIGURE 17
Lord Rayleigh
1871 in his famous paper entitled “On the scattering of light from small particles”, where he derived the polarization at 90◦ law (polarization of the light scattered by the atmosphere when viewed at a right angle to the direction of incidence), the inverse fourth power law for the intensity of light scattered by particles, not necessarily spherical, whose size is much smaller than the wavelength of the light, and explained that the degree of polarization of the scattered light depends on the angle of scattering from the elastic-solid theory of the luminiferous aether (Howard, 1964; Strutt (Lord Rayleigh), 1871a,b,c, 1881, 1900, 1920; Young, 1982). Single scattering from air molecules is known to result in the polarization of scattered light in a cloudless blue sky. In 1881 he re-derived these results (for a dielectric cylinder and a small sphere) from the Maxwell’s theory of electromagnetism, which in fact did not change his major conclusions. For a more detailed exposition of Rayleigh’s legacy in physics the reader should consult Refs. (Howard, 1964; Strutt (Lord Rayleigh), 1920; Uslenghi, 1978; Young, 1982). In his book on the Theory of Sound, he established a principle of reciprocity in scattering which was subsequently extended to optics by Perrin. Several major advances, both theoretical and experimental, quickly followed Rayleigh’s work. Love investigated the problem of scattering of electromagnetic waves by a sphere. Then Gustav Mie in 1908 (“Beitr¨age ¨ ¨ zur Optik truber Medien, speziell kolloidaler Metallosungen”) and, independently, P. Debye (“Der Lichtdruck auf Kugeln von beliebigem material”) in 1909 found an analytical solution to the scattering problem of a plane, time-harmonic electromagnetic wave of arbitrary polarization and frequency by an homogeneous sphere of arbitrary size, provided that the sphere material is isotropic and is characterized by a permittivity which can be complex and frequency dependent, usually referred to as Mie theory (Debye, 1909; Lilienfeld, 1991; Logan, 1962; Mie, 1908).
From Stokes (1852) to Poincar´e (1892)
171
A list of classic highlights in polarization optics would not be complete without tribute to the following research contributions: The effect of magnetically induced birefringence in colloidal suspensions of ferromagnetic particles was first investigated by Ettore Majorana in 1902 (Majorana, 1932). Umov reported in 1905 on polarization aspects in the reflection of light by rough surfaces. Since their initial discovery in 1888 by Reinitzer, liquid crystalline materials have been investigated intensely for their scientific and technological potentials. Perhaps the earliest instance of the important polarization aspects in liquid crystals was presented by Mauguin as earlier as 1911 in his article entitled “Sur les cristaux liquides de Lehmann” (Mauguin, 1911). He is given due credit as the first to present a Poincar´e sphere construction (which was new at the time) to study the deformation of twisted nematic liquid crystals. Schuster and Schwarzschild initiated the first radiative transfer theory in 1905 for explaining the appearance of absorption and emission lines in stellar spectra (Schuster, 1905). George Hale reported early interesting observations on the polarization of sunlight and detected circular polarization in the wings of a Fraunhofer line from a sunspot in 1908 (Rozenberg, 1960a,b). Significant improvements in optical instrumentation were made in the first decades of the twentieth century, e.g. the polarizing microscope. Circular dichroism in solutions was extensively investigated by Aim´e Cotton (Cotton, 1895) and Georges Bruhat (Bruhat, 1929a,b). As an aside it is worthy to note the finding of the circular polarization properties of certain families of beetles by A. Michelson in 1911. He observed colourful effects resulting from the property of converting unpolarized incident light from the sun into ¨ left circularly polarized reflected light (Kattawar, 1994; Konnen, 1985; Michelson, 1911). The end of the nineteenth century was a period of excitement about investigations of different kinds of radiation. In 1895 Wilhelm Conrad ¨ Rontgen (Figure 18), a professor of physics at the Julius Maximilian ¨ University of Wurzburg, Germany, published his famous article entitled “Eine neue Art von Strahlen” in which he presented the discovery of high-energy electromagnetic radiations: the X-rays (X for unknown). Note that the existence of “invisible” high-frequency electromagnetic radiations had been predicted by von Helmholtz a few years earlier (Glasser, 1934; ¨ Nitske, 1971; Rontgen, 1895). That work won Rontgen the 1901 (and ¨ first) Nobel prize in physics. From Rontgen’s discovery to the present time, X-ray crystallography has proved essential in relating the atomic structure of condensed matter to their functions and physical properties. The transverse nature of X- rays was first described and demonstrated in ¨ a paper “Polarization in secondary Rontgen radiation” published in 1906 by the Scottish physicist Charles G. Barkla (Barkla, 1906). This advance
172
A History of Polarization and Coherence Optics
¨ FIGURE 18 Wilhelm Conrad Rontgen
was recognized (eleven years later) when Barkla was the recipient of the 1917 Nobel prize for physics. Before continuing with this narrative, it is perhaps appropriate to recall a few important facts. By the end of the nineteenth century and the beginning of the twentieth century, we are in a time of great change: Maxwell constructed a theory of electromagnetic radiation which was successful in explaining all of the observed phenomena of propagation and scattering of light but the detailed understanding of the processes of emission and absorption by matter required one more step in the theoretical description of the interaction between field and matter. An essential question that arose was this: What is the origin of light? The question can be clarified only by an appropriate quantum treatment.
4. FROM POINCARE´ (1892) TO WOLF (1954) The development of quantum theory, in the first third of the twentieth century brought a profound change in physics and had a major impact on our notions of the nature of light. The theory of electromagnetic radiation in thermal equilibrium was a source of mystery in the context of the physics prevailing at the turn of the century. In addition to providing clues about the physical principles underlying all optical phenomena, the depth of the new ideas (the dual nature of light as both corpuscule and wave) provided by the quantum theory in 1924–1925 soon followed by quantum field theory in the late 1920s, sparked a revolutionary way of looking at nature, a paradigm change. In the years up to about 1905 the subject of polarization optics seemed to have reached a sort of maturity: the impressive success of Maxwell’s
From Poincar´e (1892) to Wolf (1954)
173
FIGURE 19 Max Planck
electrodynamics made it clear to optical physicists that the laws behind the phenomena had been comprehended. At the same time researchers developed a variety of clever methods for the analysis of the polarization of light. A large fraction of the efforts in the field were being devoted to the study of novel techniques and instruments, rather than to the elucidation of the theoretical foundations concerning the nature of light. In spite of these efforts it appears that the need to introduce a mathematical formalism for describing radiation fluctuations, and particularly the correlation between the electric field at two space points, was fundamental to an understanding of the nature of light. This subject was taken up by several workers. Max Planck (Figure 19), one of the towering figures who built the edifice of quantum mechanics, gave a precise expression of the entropy of a polarized “pencil of radiation”, (Ore, 1955; Planck, 1959; Rosen, 1954) although this concept was first conceived by Wien in 1894. The statistical description of optical fields played a key role in the development of quantum physics and had a central role in the efforts to understand the nature of light. The origin of the subject lies in von Laues’s (who was Planck’s student) pioneering work in 1907 (von Laue, 1906, 1907a,b). Prior to this time there had been no clear recognition of the fact that the variation of the electric field vector of light is a random process both in time and in space. In a paper devoted mainly to the thermodynamics of radiation, Max von Laue introduced a measure of the coherence between two pencils of radiation. Although von Laue’s ideas diffused rapidly, they apparently did not receive the amount of attention
174
A History of Polarization and Coherence Optics
FIGURE 20
Max Berek
they deserved since the next step in coherence theory was done two decades later. By the early 1920s much had changed. Many scientists had become seriously concerned with the analysis of radiation fluctuations. This shift of understanding and attention may eventually be regarded as one of the twentieth century’s pivotal developments in polarization and coherence theories. Developments in statistical optics were then proceeding at an impressive rate. In 1926, Max (Figure 20) introduced another measure of the coherence of light, which was called the degree of consonance (Berek, 1926a,b,c, 1927; Wartmann, Wolff-Fischer, & Berek, 1926). Another highly influential result was the introduction of the density matrix formalism by von Neumann (1955, 1961), which has much to do with the coherency matrix formalism pioneered by Wiener. van Cittert calculated the second-order correlation at points on a screen illuminated by an incoherent quasi-monochromatic source (van Cittert, 1934, 1939). A great step was taken in 1938 when Frederik Zernike (Figure 21), a Dutch physicist, was able to show in a remarkable paper that the degree of coherence could be measured by studying the visibility of interference fringes. van Cittert (1934, 1939) and Zernike (1938), independently, formulated a central theorem concerning the propagation of coherence for stationary sources. The essence of the theorem due to van Cittert and Zernike is that the two-point coherence function in the far field of a quasi-monochromatic spatially incoherent light source is proportional
From Poincar´e (1892) to Wolf (1954)
FIGURE 21
Frederik Zernike
FIGURE 22
Norbert Wiener
175
to the Fourier transform of the source intensity distribution. This theorem has applications for long-baseline interferometry in astronomy, in which one uses a measurement of the coherence function to obtain the source intensity distribution. Zernike’s great contribution, for which he won the Nobel prize in 1953, was the invention of the phase contrast microscope. The basic idea is to introduce a phase shift in the light scattered by a small object and to make it interfere with the light of a coherent background. In this way the phase difference is thus converted into an amplitude difference. After the first third of the twentieth century several factors conspired to shift the centre of gravity of polarization and coherence theories toward the US. An important example came with Wiener (Figure 22).
176
A History of Polarization and Coherence Optics
In 1930 Norbert Wiener, at the MIT Mathematics Department, developed a rigorous mathematical basis for the theory of the coherence of light. Although his analysis is rather complicated, he was able to derive a measure of coherence in his historic paper “Coherency matrix and quantum theory” (Wiener (1929), Wiener (1930), Wiener (1933), Wiener (1953), Bulletin of the American Mathematical Society (1966) (issue dedicated to the memory of Wiener), Khintchine (1934)). Note that Wiener researches on generalized harmonic analysis were motivated partly by investigations in the theory of Brownian motion and partly to give a rigorous interpretation of white light. He was dissatisfied with the conventional description of unpolarized light because this ignores the statistical nature of light waves. What is seldom appreciated is that Wiener belongs to that small group of theoretical physicists who shaped modern coherence theory. Autocorrelation and cross-correlation functions between field variables at two space–time points were introduced for describing correlations of random processes in electromagnetic fields. Wiener, and a few years later, Khintchine formulated a fundamental theorem of the theory of random processes which is central to the analysis of optical coherence: the Fourier transform of the autocorrelation function of a stationary random process is the spectral density (power spectrum) of the process. Although Wiener gave real grounds for the concept of coherence, his ideas did not gain general recognition. To quote Levinson “most of Wiener’s important work was inspired by physics or engineering and in this sense he was very much an applied mathematician. He formulated his theories in the framework of rigorous mathematics and as a consequence his impact on engineering was very much delayed”. The close relation to the mathematical description of polarization states with spinors was noted by Jordan (1927), thus indicating the possibility of using the wealth of mathematical techniques developed in quantum mechanics for the treatment of problems involving partially polarized electromagnetic radiation, although Ugo Fano (Figure 23) is recognized as being the first to have popularized the connection between the quantum treatment of polarization and the classical description of polarization of light in his paper “Remarks on the classical and quantum-mechanical treatment of partial polarization” (Fano, 1949, 1954, 1957). The credit for pioneering modern sheet polarizers belongs to Land (perhaps best known as an inventor for conceiving and developing commercial applications for instant photography as early as 1947) (Land (1951), Land and West (1946), Optics and Photonic News (1994) (issue
From Poincar´e (1892) to Wolf (1954)
FIGURE 23
Ugo Fano
FIGURE 24
Edwin H. Land
177
dedicated to the memory of Land), McElheny (1998)).6 In 1927 Edwin H. Land (Figure 24), then a freshman (19-year-old) physics major at Harvard College, invented the first synthetic polarizers by orienting crystalline needles of herapathite in a sheet of plastic through electro-, magneto-static as well as mechanical forces (J-sheet polarizer). Note that the material, i.e. sulfate of iodo-quinine, that was the starting point for Land’s invention of a plastic sheet polarizer was discovered by William Herapath in the mid 6 As a scientist and inventor, Land was fascinated by light and its interactions with matter. As entrepreneur and businessman, Land founded and inspired the Polaroid Corporation, renowned for its innovative products and enlightened social conscience. As advisers to Presidents, he played a pivotal role in the development of high-altitude aircraft and satellite reconnaissance. In later years Land’s passion was to understand how we see colour.
178
A History of Polarization and Coherence Optics
FIGURE 25
Paul Soleillet
1800s (Herapath, 1852, 1855). Land’s strongest motivation in developing the sheet polarizer was his desire to equip automobiles with glareless headlights (Billings & Land, 1948; McElheny, 1998; Robson, 1984). The list of applications of these polarizers is extensive and we do not attempt to review it here.7 In the years 1938–1939 Land invented H and K (with Howard Rogers) synthetic sheet polarizers which do not contain dichroic crystals but form molecular analogs of the wire-grid polarizer. These molecular polarizers made of polyvinyl alcohol stained with iodine for H-Sheet and polyvinylene for K-Sheet possess a good stability to light and heat, can be large in size and can act over a wide angular range. The developments of these optical devices required several advances in material research: polymer science had played a major role in this story early on. Land took out more than five hundred US patents (second only to Edison) relating to different areas of research, including polarization, photography and human colour vision. The development of polarization applications of light focused attention on the importance of investigating methods for describing the changes in polarization state that a beam of light undergoes in interacting with optical elements. Several remarkable contributions were carried out successively by Paul Soleillet (Figure 25), Robert Clark Jones, Francis
7 World War II brought new uses and challenges for sheet polarizers. In the North Atlantic, the search for German submarines became a matter of life and death. Aboard ships and aircraft, lookouts scanned the sea for periscopes and shadows moving under the waves. But observation was difficult, complicated by a moving craft, a bright sun, and reflections dancing on the water. Sheet polarizers were fitted into glasses, and the reflections disappeared – polarized sunglasses have been with us ever since (Robson, 1984).
From Poincar´e (1892) to Wolf (1954)
179
Perrin and Hans Mueller. Individually, each had important significance. Collectively, they heralded what we know as “Jones and Mueller polarization matrices”. Soleillet’s deep insight into this subject area is already evident in a remarkable (little-known) paper published in 1929, “Sur les param`etres caract´erisant la polarisation partielle de la lumi`ere dans les ph´enom`enes de fluorescence” (Soleillet, 1929). Less well known is that he was the first to suggest how one might characterize an optical device by the linear relation between the output Stokes parameters and the input Stokes parameters. Although the 4 × 4 polarization matrix relating the output Stokes parameters to the input Stokes parameters is now traditionally termed the Mueller formalism, Paul Soleillet (Figure 25) was evidently the first to consider it. His paper could have become the cornerstone upon which optical scientists would base their theoretical studies. But apparently Soleillet’s contribution did not attract the attention it deserved and the matrix formalism was later named for Hans Mueller, who rediscovered the results which can be found in Soleillet’s original paper. The power and fecundity of what is now termed the Jones calculus in polarization optics was demonstrated by Robert Clark Jones.8 In 1941 to 1942, he had a remarkable year at Land’s laboratory where, as an undergraduate, he proved his first results. The exceptional pertinence of Clark Jones can be appreciated in the reading of the set of 8 publications noted for their clarity and insight, “A new calculus for the treatment of optical systems”, on the problems related to the effect of various non-image forming optical instruments on an incident polarized electromagnetic wave using the formalism of matrix calculus (Clark Jones, 1941a,b,c, 1942, 1947a,b, 1948, 1956). The state of polarization is represented by a 2 × 1 vector and each optical element is represented by a 2 × 2 matrix. His formalism was initially devised to study the effect of material anisotropy on the state of polarization of light and led to the identification of eight types of canonical behaviours associated with birefringence and dichroism. This matrix method, however has limitations: the wave is normally incident, completely polarized, and the theory is valid under the paraxial approximation. Moreover, this method does not consider reflection from the surface of an anisotropic medium, or the effect of multiple reflections within an anisotropic plate. The Jones matrix method was later extended by Wolf to deal with partial polarization (Wolf, 1954a,b, 1955, 1957, 1959). Further significant extensions and elaborations of Jones’s work were subsequently published 8 There are many things about Robert Clark Jones that are out of the ordinary. For example, in his publications he uses the name R. Clark Jones, whereas he was known to his friends as Bob. Many scientists refer to him as Clark-Jones, hyphen and all. Clark Jones was an inveterate rider of railroad locomotives.
180
A History of Polarization and Coherence Optics
FIGURE 26
Francis Perrin
by Berreman and others and by Vernon and Huggins, to treat the problem of reflection by and transmission through a slab of inhomogeneous anisotropic material assuming polarized light obliquely incident upon the slab (Berreman, 1972; Ling & Teitler, 1984; Teitler & Henvis, 1970; Vernon & Huggins, 1970). One notable contribution to the study of scattering of polarized light by optical media was the work of Francis Perrin (Perrin, 1942). In his 1942 landmark paper “Polarization of light scattered by isotropic opalescent media”, Perrin (Figure 26) extended the previous work of R. Krishnan (Krishnan, 1938a,b,c, 1939a,b,c) and noted that the symmetry properties of the optical medium permit the reduction of the number of independent parameters of polarization matrices. He recognized the importance of the concept of symmetry in elastic scattering by isotropic suspensions of particles and he rigorously exploited the constraints required by symmetry. Later Perrin and Abragam reported several symmetry relations for the special cases of scattering by spherical particles and for molecular scattering (Perrin & Abragam, 1951). The matrix method to represent the effect of nonimage forming optical instruments and scattering media on incident Stokes parameters is commonly termed the Mueller formalism after Hans Mueller, then a professor of physics at MIT. Apparently he had never seen Soleillet’s original paper. Mueller used such matrices in lectures on optics at the MIT during 1945 to 1948 but did not publish his results in the archival literature except in a short note: “The foundations of optics” Mueller (1943, 1948). A MIT report is cited by several authors but I was not able to locate any other published works. Progress worth noting was also made by N. G. Parke (Parke III, 1949) who investigated the connection between the Jones
From Poincar´e (1892) to Wolf (1954)
181
and Mueller matrix formalisms in his Ph.D. work done at MIT under the guidance of N. Wiener. On the experimental side, progress was rapid. As early as 1928, the French astronomer Bernard Ferdinand Lyot, gave the first impetus to the use of polarization in remote sensing of planetary surfaces (Billings, 1951; Egan, 1985; Loeber, 1982; Lyot, 1928, 1944; Richter, 1979). In 1928 Lyot’s paper on what is now called the Lyot depolarizer was published. This depolarizer consists of two retardation plates, with retardation in the ratio 1:2 and with their fast axes oriented at 45◦ to one another. He also invented several optical devices such as the double-refraction filter (1944). In the years 1934 to 1935 important papers by Langsdorf Jr., Wood, Birge and DuBridge deserve special mention because they drew attention to the fact that the description of unpolarized light departs from the totally polarized case (Birge, 1935; Du Bridge, 1935; Hariharan & Singh, 1961; Langsdorf Jr. & Du Bridge, 1934; Wood, 1934). In 1947 Goos and H¨anchen discovered experimentally the effect that now bears their names, related to the lateral spatial displacement of beams on reflection at interfaces between two dielectric media (Goos & H¨anchen, 1947, 1949). As earlier as 1943, Harold H. Hopkins demonstrated the possibility of controling the spatial coherence of light for practical applications, e.g. microscopy (Hopkins, 1943, 1951, 1953, 1957). After years of effort, ellipsometric measurements occured that changed the study of surface and thin-film phenomena. Ellipsometry is a technique for in-situ nondestructive characterization of surface (interfacial) phenomena using the change in the state of polarization of a light-wave probe. The terminology ellipsometry is fairly recent and was coined by Alexandre Rothen in 1945 (Hall, 1969; Rothen, 1945, 1964; Rzhanov, Svitashev & Semenenko, 1979). The earliest applications of ellipsometry in biology were investigated by Rothen. In 1949, the Austrian biologist Karl von Frisch (later Nobel Prize winner) discovered the polarization vision of honey bees, and showed that these insects are able to navigate by means of the sky polarization when the sun is hidden by clouds, as long as there are some clear, blue sky regions. The eyes of these invertebrates have birefringent properties and have the ability to sense the direction of polarization of skylight (Bernard & Wehner, 1977; Cameron & Pugh, 1991; Hawryshyn, 1992; Rossel & Wehner, 1969; von Frisch, 1950; Wehner, 1976). Because vision dominates human perception, it is quite surprising that the human visual system is almost completely insensitive to polarization. However, William Shurcliff (Figure 27) called attention to the fact that the human eye has the ability to distinguish circularly polarized from unpolarized light, in relation to the phenomenon of Haidinger’s brush. Shurcliff is also noted for an important monograph “Polarized light” (Shurcliff, 1971).
182
A History of Polarization and Coherence Optics
FIGURE 27
William Shurcliff
On the theory side Luneberg developed, in 1944, an analytical formulation of diffraction theory taking into account polarization phenomena (Luneberg, 1944). A list of classic highlights would not be complete without tribute to Henry Hurwitz. In 1945, an important paper by Hurwitz, “The statistical properties of unpolarized light”, was published and attracted notice. Specifically, Hurwitz was able to compute the average values of many polarization parameters, e.g. the ellipticity angle (Hurwitz, 1945). In 1946, the prominent physicist Denis Gabor (Figure 28) invented the concept of an analytic signal for making a unique complex representation of a real signal and used it to treat problems arising in communication theory. This representation is a generalization of one that is used for monochromatic radiation bringing to the field of optical coherence theory a thorough theoretical foundation (Gabor, 1946). Gabor’s name is also on the lips of optical scientists for the invention of the technique of holography (the wavefront reconstruction process) in 1948; he received a Nobel prize in Physics for this breakthrough in 1971. The required physical basis of the theory of electromagnetic energy propagation in random media was suggested by Subrahmanyan Chandrasekhar (Figure 29), a professor in astronomy and astrophysics at the University of Chicago in 1946, in his book “Radiative Transfer”. In this monograph, he used the Stokes parameters method in his solution of the problem of radiative transfer (the Boltzmann equation approach) for the scattering of partially polarized light (Chandrasekhar, 1946, 1950; Chandrasekhar & Elbert, 1954). This work launched the theory of vector radiative transfer to model light propagation in anisotropically scattering media. Thanks to his profound scholarship, he clearly detected the trends in the development of modern physics and was awarded the Nobel prize in 1983 for his pioneering achievements in astrophysics.
From Poincar´e (1892) to Wolf (1954)
FIGURE 28
Denis Gabor
FIGURE 29
Subrahmanyan Chandrasekhar
183
Another big advance in the study of radiation scattering by matter came when van de Hulst (Figure 30) explained the polarization characteristics of the glory. van de Hulst did much for the field by writing an important monograph, “Light scattering by small particles”, emphasizing the symmetry constraints in the analysis of polarization matrices (van de Hulst, 1957). Other theoretical progress needs to be mentioned. To our knowledge, the first appearance in the optics literature, of a geometric–algebra relying on quaternions (introduced by William Rowan Hamilton in 1843) to compute rotations and retardations of pure
184
A History of Polarization and Coherence Optics
FIGURE 30
Hendrik Christoffel van de Hulst
states of polarization, was in a paper by Richartz and Hsu¨ in 1949, for ¨ predicting the change of polarization by optical devices (Richartz & Hsu, 1949). However their formalism cannot describe the action of a polarizer or even a partial polarizer (Pellat-Finet, 1990, 1992). In the years 1951 to 1952, the propagation of polarized light in optically active crystals was analyzed in a series of important articles of Ramachandran and Ramaseshan (Ramachandran & Chandrasekharan, 1951; Ramaseshan, 1951; Ramachandran & Ramaseshan, 1952, 1961). The application of the density matrix formalism in the description of phenomena connected with electron and photon polarization was discussed by Tolhoek (1956). The subject was taken up by several other contributors: Falkoff and McDonald (1951), Westfold (1959), and, independently, Jerrard (1954) developed a formalism based on the complex number representation (i.e. the ratio between the complex amplitudes of the components of the electric vector in an orthogonal coordinate system normal to the direction of wave propagation) to represent the state of polarization of polarized radiation transmitted through birefringent media. Here we pause to make a brief comment. For many reasons the history of science in the former Soviet Union is not fully known in the West. For examples, Rozenberg (1946, 1960a,b) and Stamov (1953) were pioneers in developing the study of polarized light scattering from the atmosphere. These works were overlooked in countries other than Russia for several years but their importance has since been appreciated. During World War II, there had been a substantial increase of interest and technological development of radar into the microwave range for aircraft detection. The bridges between pure science and military technologies that have been created in and for war continued in
From Poincar´e (1892) to Wolf (1954)
185
various forms during the postwar years. The wartime research laid the groundwork for the development of polarization radar. In the early 1950s, experimental efforts to make coherent polarimetric phase information the basis for technologically important devices a reality were undertaken. An afflux of investigators with electrical engineering backgrounds led to work on polarization radar where the game is to guess the shape of an unknown scattering object from the analysis of the scattering data. One of the earlier pioneers in this area was Sinclair at the Ohio State University (Sinclair, 1950). The state of polarization of an electromagnetic wave scattered by a target differs, in general, from that of the incident wave. To characterize the scattering properties of the target, Sinclair introduced, in 1950, the 2 × 2 coherent backscattering matrix formalism (the 2 × 2 Sinclair matrix is analogous to the 2 × 2 Jones matrix for the case of transmission). These efforts were pursued by Edward Morton Kennaugh who introduced in 1952 his radar target characteristic operator theory based on the optimal polarization null concept (Kennaugh, 1949, 1952). He demonstrated that there exist polarization states for which the radar receives minimum or maximum power. Other works which have pushed this area of development very hard and need be cited are those of Deschamps (1951); Deschamps and Mast (1973) and Huynen (1970). More specifically, Jean Richard Huynen made a seminal contribution to polarimetric radar theory techniques by introducing the fork concept basis to study the polarimetric radar target optimisation problem [98]. In subsequent years, various methods for studying the optimal reception problem, namely that of finding incident polarization states to optimize for a given scattering or Mueller matrix system and the intensity at the receiver, were developed and large amounts of data were accumulated (Boerner & Yan, 1991; Graves, 1956). These methods have also found application in such fields as geological surveys and medical and remote sensing diagnostics. During the years following the war, theoretical progress had impressive success in interpreting the statistical properties of light. A great step on coherence theory was carried out by Emil Wolf (Figure 31) in the years 1954 to 1955 in his landmark papers “Optics in terms of observable quantities” and “Coherence properties of partially polarized electromagnetic radiation”, giving a treatment of partial coherence which has a great deal in common with the theory of partial polarization (Wolf, 1954a,b, 1955, 1957, 1959, 2001). They must be considered along with the studies of Wiener as the classic papers in the field. The foundations of the modern polarization theory were laid. The starting point for most calculations having a bearing on optical coherence theory is the 2 × 2 Wolf’s coherency matrix. His coherency matrix differs from that of Wiener in that Wolf has chosen to take as his fundamental quantities temporal
186
A History of Polarization and Coherence Optics
FIGURE 31
Emil Wolf
covariance functions, unlike Wiener who used the integrated power spectra. Mathematically these two approaches are functionally related through a Fourier-Stieljes transform. In a series of subsequent widely quoted papers and two important books co-authored by Max Born (Born & Wolf, 1980) (“Principles of Optics”) and Leonard Mandel (Mandel & Wolf, 1965, 1995) (“Optical Coherence and Quantum Optics”), noted for their clarity and insight, he provided a solid basis, which is transparent in terms of the physical interpretation of the theoretical understanding of the concept of coherence. Wolf introduced also more general 3-dimensional tensors for dealing with non-plane waves (Roman & Wolf, 1960a,b). Under his tutelage a large school of scientists in the field of optical physics emerged, and they continue to do important work to the present day. Among the numerous contributions of Leonard Mandel in the subject, we mention an analysis of coherence when the spectral width of the radiation is not narrow, and the introduction of the concept of cross-spectral purity (Mandel, 1961). Within this context, researchers in increasing numbers began to study optical coherence. These works have resulted in the explanation of a wealth of phenomena and the development of myriad new results; we refer the reader to Blanc-Lapierre and Dumontet (BlancLapierre & Dumontet, 1954, 1955), Gamo (1956, 1957a,b) and Parrent Jr. (1959a,b); Parrent Jr. and Roman (1960). Bourret was the first to calculate the coherence properties of blackbody radiation (perfect thermal emitter) in equilibrium in an enclosure, which permit the characterization of the spatial and temporal fluctuations of a (thermal) electromagnetic field (Bourret, 1960). Subsequently Richard Barakat (Figure 32) added, in his 1963 paper “Theory of the coherency matrix for light of arbitrary spectral bandwidth”, the concept of a spectral
From Poincar´e (1892) to Wolf (1954)
FIGURE 32
187
Dick Barakat
coherency matrix from a somewhat different point of view than the one introduced by Wiener (Barakat, 1963). Note that this seminal paper, which may be considered as a classic publication since it was chosen as one of the “Benchmark Papers in Optics, Polarized Light” (Swindell, 1975), is probably Barakat’s most cited paper on the subject. Among his other finest results in the field are the derivations of the probability-density functions of the Stokes parameters (Barakat, 1993), the von Neumann entropy of N interacting pencils of radiation in terms of scalar invariants (Barakat & Brosseau, 1993), the generalized Verdet–Stokes–Barakat conditions for light to be natural (unpolarized) (Barakat, 1996), and the necessary and sufficient conditions for a Mueller matrix to be derivable from a Jones matrix in the presence of measurement and modelling error (Anderson & Barakat, 1994). His accomplishments as an optical physicist make best sense when they are placed in the broader perspective of his scientific interest (Brosseau, 2000). More than 30 years after his 1963 paper, a question central to the experimenters – the equivalence between the Mueller matrix and the Jones matrix descriptions of a linear optical system, has resurfaced in 1994. Barakat’s 1994 paper with Don Anderson (Anderson & Barakat, 1994) on this issue is a classic. The main interest of this work is that it establishes the mathematical requirement for the necessary and sufficient conditions (the trace formula), which are capable of experimental verification, for an optical system characterized by a Mueller matrix to be characterized also by a Jones matrix, i.e. the socalled Mueller–Jones equivalence issue (Brosseau, 1998). By themselves, the technicalities of this derivation provide a a powerful procedure for analyses of practical problems, i.e. calibration of polarimetric instruments, estimation of experimental errors, and testing computational procedures.
188
A History of Polarization and Coherence Optics
Observe that an important current of research in areas as diverse as imaging, biological optics, and oceanic and atmospheric optics, deals with establishing the information content of the various matrix elements. I was quite fortunate to work with Dick Barakat when I went to the Division of Applied Physics at Harvard as a postdoctoral fellow. At a personal level Dick was a man of unbounded energy and endurance: he worked hard and demanded the same from others (Brosseau, 2000). My intellectual journey with him has been exciting and rewarding (Brosseau & Barakat, 1990; Brosseau, Rockower, & Barakat, 1991; Brosseau & Barakat, 1991, 1992). With Dick I worked out the SU(N) expansion of the density matrix to calculate the von Neumann entropy of polarization, which is a measure of the degree to which an optical field is polarized (Barakat & Brosseau, 1993). This concept of entropy was much later used by Brosseau and Bicout (Brosseau & Bicout, 1994) to introduce depolarization lengths which characterize elastic multiple scattering of polarized light by a dense random collection of uncorrelated and noninteracting spherical dielectric particles. Monte Carlo simulations indicated that the amount of depolarization generated from multiple scattering depends on such factors as the size and shape distribution and index of refraction (Bicout, Brosseau, Martinez, & Schmitt, 1994). In a medium containing particles small compared to the wavelength (Rayleigh regime), the characteristic length of depolarization for incident linearly polarized light is found to exceed that for incident circularly polarized light, while the opposite is true in a medium composed of particles large compared to the wavelength (Mie regime). In this context, depolarization by multiple scattering is connected to entropy production. It should be emphasized that the group of 2 × 2 special unitary transformations SU(2) can be found in the optics literature (Pancharatnam, 1956a; Ramaseshan & Nityananda, 1994). As a result of these and other considerations, the general expectation in the optics community is that the Jones matrix calculus associated with the coherency matrix formalism, on the one hand, and the Mueller matrix calculus associated with the Stokes formalism, on the other hand, provide two efficient formulations of the theory of polarization in algebraic terms. Thus in the early 1960s, the conceptual basis of the coherency (density) matrix was rather thoroughly formulated.
5. FROM WOLF (1954) TO WOLF (2007) Further important theoretical contributions to the subject were made by the great Indian physicist S. Pancharatnam and can be singled out. For example, in 1954, Pancharatnam introduced the concept of spectral
From Wolf (1954) to Wolf (2007)
189
functions to deal for the description of the polarization properties of a polychromatic light beam (Pancharatnam, 1956a,b, 1963a,b, 1975). As early as 1956, in a classic paper, he was able to predict a remarkable effect, now usually called the geometric phase which concerns the phase change of a light beam whose polarization state is made to trace out a cycle on the Poincar´e sphere. He further observed this phase in a study devoted to the interference of two coherent light beams. It was quickly realized by Pancharatnam that the origin of this phase is topological. It is remarkable that when Pancharatnam discovered this important effect, he was only 22 years of age. It is also interesting to note that Pancharatnam belonged to the distinguished dynasty that includes the spectroscopist Raman, the astrophysicist Chandrasekhar, and the crystallographer Ramaseshan. In addition to being important in its own right this discovery has stimulated many subsequent developments including what is termed in the contemporary literature as Berry phase (Berry, 1987a,b). Here it is also worth mentioning the important contribution of Saxon on reciprocity and symmetry theorems which complements the earlier works of Rayleigh, Krishnan and Perrin (Saxon, 1955). Advances in instrumentation came on many different fronts. B. Lyot ¨ (Lyot, 1933, 1944) and Ohman (1938a,b, 1958) designed birefringent filters for solar research, i.e. the distribution of hydrogen may be measured by photographing the solar corona in the light of the Hα line. Solc (1953, 1960, 1965) and, independently, Evans (Evans, 1958, 1963) discovered polarization interference filters composed of a stack of identical birefringent plates (each oriented at a prescribed azimuth angle) placed between crossed polarizers. This discovery has found use in important applications where filters of extremely narrow bandwidth with wide angular fields or tuning capability are required, including electro-optic tunable filters and electrically tunable lasers. Extensive polarization measurements in the area of astrophysics were pioneered by Hiltner and Dollfus and led to many important results concerning the remote sensing of planetary surfaces and atmospheric constituents (Dollfus, 1956; Hiltner, 1947, 1949, 1951; Lenoble, 1957). In the 1950s considerable improvement in instrumentation focused attention on the importance of investigating higher order than two correlation effects in optical fields. In the years 1952 to 1957, Hanbury-Brown (HanburyBrown, 1991, 1974) and Hanbury-Brown with Twiss (Hanbury-Brown & Twiss, 1956a,b,c) have shown that, when a light wave illuminates two photoelectric detectors, the outputs from the detectors are in general correlated; this correlation being proportional to the square of the degree of coherence of the light vibrations at the detectors. They built the first optical interferometer to measure the apparent size of visible stars and performed successful experiments demonstrating that correlations
190
A History of Polarization and Coherence Optics
between intensities (i.e. fourth-order effect in optical fields) can also be measured (Forrester, 1961; Sanders, 1959). These correlation phenomena were used to determine the velocity of light and the profiles of spectral lines. The important point made by Hanbury–Brown and Twiss is that the degree of coherence at two points in a radiation field can be inferred from correlation measurements of the fluctuating signals appearing at radio antennae placed at the two points. Certain periods of history bear the name of the materials that have shaped the development of civilization, e.g. the stone age. An appropriate characterization of optics in the second half of the twentieth century would be the laser age since the laser is hailed as one of the most significant inventions that have truly revolutionized the field. The experimentum crucis came in 1954 when Charles Townes and his coworkers realized the first maser (the acronym for microwave amplification by stimulated emission of radiation). Townes and his co-workers designed a microwave oscillator based on stimulated emission by molecules (NH3) in an excited state (Gordon, Zeiger, & Townes, 1954, 1955). Two years later, Nicolaas Bloembergen proposed the first tunable three-level solidstate maser (Bertolotti, 1983; Hecht, 1992). Subsequently in 1960, Arthur Schawlow and Charles Townes extended the concept of the maser to higher frequencies of the electromagnetic spectrum, “Infrared and Optical Masers”, based on a plane and parallel Fabry–Perot interferometer as the optical resonator to provide a strong coupling between light and the amplifying laser materials (Schawlow & Townes, 1958). This laid the groundwork for development of the laser. In recognition of their pioneering researches, Charles Townes shared the 1964 Nobel prize for physics with Nicolai Basov and Aleksandr Prokhorov for their fundamental works in quantum electronics that led eventually to the construction of oscillators and amplifiers based on the maser–laser principle. Note that Bloembergen and Schawlow will receive the Nobel prize for physics in 1981 for their contributions to the development of laser spectroscopy. Then, Theodore H. Maiman realized the first laser using the fluorescence of a ruby crystal pumped by a xenon discharge lamp (Maiman, 1960a,b). Shortly thereafter followed the operation of the first gas (Helium–Neon) laser by Ali Javan (Javan, 1965; Javan, Bennett, & Herriot, 1961) and the following year Robert Hall and his co-workers described the first semiconductor laser (Bertolotti, 1983). By contrast with a usual thermal light source, e.g. mercury arc lamp, a laser source produces coherent light, amplified by stimulated emission in an active medium inside an optical resonator, whose spectrum is exceedingly narrow and which is highly directional. The historical development of lasers is well documented in several books, e.g. (Bertolotti, 1983; Hecht, 1992). In physics, the laser has galvanized the science of the
From Wolf (1954) to Wolf (2007)
191
interaction of radiation with matter. Its imprint stamps a huge range of experimental techniques and has posed interesting new problems in the search to understand the nature of light, e.g. the fluctuation properties of such radiation cannot be described by a Gaussian random process. In technology it has resulted in expanded techniques and key contributions to communication systems, e.g. guiding dielectric structures. In the more technical arena, a number of interesting devices were fabricated in the 1960s. George R. Bird and Maxfield Parrish Jr. designed and demonstrated wire-grid infrared polarizers (Bird and Shurcliff, 1959; Bird and Parrish Jr., 1960). A wire-grid polarizer consists of a grid of parallel metal wires which reflects one polarization (e.g. the electric field parallel to the grid lines) of incident radiation field while transmitting the other (e.g. the electric field perpendicular to the grid lines), provided the period of the grid is smaller than the wavelength of the wave. Measurements of the Mueller matrix were pioneered by Pritchard and Elliott (1960) and by Rozenberg (Gorchakov & Rozenberg, 1967; Rozenberg, 1968) for the characterization of the optical properties of the atmosphere. Inou´e, Hyde and Kubota carried out the theory of the diffraction image in the polarizing microscope, based on the fact that refraction at the optical surfaces of a lens system modifies the polarization of the wave transmitted through it (Inou´e & Hyde, 1956; Kubota & Inou´e, 1959; Kubota & Saito, 1960; McCrone, McCrone, & Delly, 1978). In the early 1960s many developments culminated in the establishment of the quantum theory of light which gave the basic understanding of laser light. Quantization is necessary to show the bosonic nature of photons. In 1963 Roy J. Glauber of Harvard University introduced quantum analogs of the correlation functions of the classical theory as expectation values of normal ordered products of the creation and annihilation operators of the electromagnetic field (Blandin & Cohen-Tannoudji, 1965; Glauber, 1963a,b, 1965). The quantum theory of coherence phenomena was mainly originated by Glauber. He was awarded the 2005 Nobel Prize in Physics “for his contribution to the quantum theory of optical coherence” (Glauber, 2007). He gave the quantum state of the laser field its “official” name: the coherent state. E.C.G. Sudarshan (Sudarshan, 1963) investigated the tradeoff between the classical and quantum-mechanical descriptions of coherence and found that quantum correlation functions take the same form as the classical versions if a particular positive definite phase–space function (termed the P-distribution by Glauber) is used for the description of the statistical properties of the field. Mandel and Wolf did much to develop the foundations of the quantum statistical properties of radiation (Mandel & Wolf, 1965, 1995). Jauch and Rohrlich presented a quantummechanical approach to polarized light and introduced Stokes operators which contain all the relevant information about the polarization of the
192
A History of Polarization and Coherence Optics
quantized field (Jauch & Rohrlich, 1955). Building on the base laid by quantum mechanics, photoelectron counting attracted new converts to the field and physicists formulated the theoretical foundations for studying the interactions of the electromagnetic field with electrons (Barakat & Blake, 1980). The theoretical formulation of nonlinear optics was done by Nicolaas Bloembergen at Harvard University in 1956 and became a fundamental piece of work (Bloembergen, 1965). Following Bloembergen’s work, researchers began to understand that nonlinear optical materials could have properties quite different from those of conventional linear materials. In 1961 the key discovery of generation of the second harmonics was made by Franken and his colleagues. They detected UV light (0.347 mm) at twice the frequency of a ruby laser beam (0.694 mm) when this beam traversed a quartz crystal (Franken, Hill, Peters, & Weinreich, 1961). This key discovery was one of the earliest nonlinear optical phenomena to be observed experimentally. Maker, Terhune, and Savage were the first to demonstrate that a strong elliptically polarized pulse propagating within a nonlinear isotropic medium induces a refractive-index change which results in a continuous precession of the orientation angle of the polarization ellipse, i.e. the so-called self-induced ellipse rotation, while leaving its shape and handedness unchanged (Askar’yan, 1962; Maker, Terhune, & Savage, 1964; Mayer & Gires, 1964). The discovery of this optically induced birefringence was the spark that ignited the explosion of energy devoted to the nonlinear optical properties of materials. For example, optical parametric down-converters serve as tunable infrared coherent oscillators and microwave electro-optic modulators contribute to the development of optical fibre communication systems. But a discussion of these topics is beyond the scope of this review. From the theoretical standpoint, Pauli algebra to describe the polarization of a plane wave was used by Hestenes (1966, 1990) to show that the electromagnetic theory takes a simple form in the geometric (Clifford) algebra of real three-dimensional space (Jancewicz, 1988). Although not well known, Whitney (1971) made also an important contribution to this problem by providing applications of the Pauli spin matrices as the base on which to expand the coherency (density) matrix and the polarization Jones matrices. A book by O’Neill (1963) and Takenaka (1973) brought important results on the algebraic description of polarized states and their interactions with optical devices. We do not try to review the vast literature that has accumulated since the mid-1990s in its entirety. That would be analogous to charting vast unmapped territory which has not been fully explored. Instead, we suggest “travel diaries” of the participants of the exploratory expedition. Recent important publications, e.g. (Wolf, 2003, 2007), facilitate our
From Wolf (1954) to Wolf (2007)
193
FIGURE 33 Some attendees at SPIE AM 100 Conference, A Tribute to Emil Wolf, Engineering Legacy of Physical Optics, (San Diego, USA, 2003). From left to right: F. Wyrowski, G. Savant, S. Peet, J. Foley, A. Friberg, A. Dogariu, C. L. Nikias, N. Savant, T. Jannson, J. Kunc, K. M. Johnson, K. Oughstun, M. Wolf, J. Bilbro, D. Fisher, E. Wolf, R. Borghi, C. Brosseau, T. Visser, P. Smid, A. van der Meulen-Visser, M. Ciftan, G. S. Agarwal, M. Bennahmias, Z.- H. Gu
task, since they present the current developments in this field from a complementary point of view. Wolf (Figure 33) introduced a unified theory of partial coherence and polarization of the electromagnetic field where complete coherence is possible for totally unpolarized fields. Until the publication of this paper, the theory of polarization of light and the theory of coherence had always been treated independently of each other. According to Wolf’s theory, the state of coherence and polarization of a stochastic light beam changes on propagation as a result of the correlations in the emission of radiation within the source. The study of the polarization properties of a general (nonplane) electromagnetic field has received rather little attention from a theoretical perspective, with the notable exception of work by Fainman and Shamir (1984). Only in the last few years have experimental reports appeared which have stimulated many theoretical studies. This effort is largely driven by the developing of many areas of physics ranging from near-field microscopy to medical imaging and biological physics where such fields are encountered. Recent developments in nanooptics have sparked a number of experimental and theoretical studies (see Carozzi, Karlsson, and Bergman (2000), Set¨al¨a, Shevchenko, Kaivola, and Friberg (2002), Lindberg, Set¨al¨a, Kaivola, and Friberg (2004), Set¨al¨a, Lindfors, Kaivola, Tervo, and Friberg (2004), Lindfors, Set¨al¨a, Kaivola, and Friberg (2005), Vahimaa and Tervo (2004), Dennis (2004),
194
A History of Polarization and Coherence Optics
Nieto-Vesperinas and Garcia (1996), Kreibig and Vollmer (1995), Takahara and Kobayashi (2004), Barnes, Dereux, and Ebbesen (2003), and references therein). These examples represent just a few of the many examples that have recently prompted the optics community to investigate the case where the wave front of the electromagnetic wave is not planar. Thus an overview of our present understanding of the theoretical aspects of polarization of an arbitrary electromagnetic wavefield is particularly timely. Recently proposed ideas (Allen & Eberly, 1975; Baylis, Bonenfant, Derbyshire, & Huschilt, 1993; Bhandari, 1988; Brosseau, 1998; Brosseau & Dogariu, 2006; Meier & Zakharchenya, 1984; Schlichter, 1990; Urbantke, 1991; Zapasskii & Kozlov, 1995) tried to describe a polarization state as a pseudospin. An interesting connection has been established with the geometric phase. The geometric phase, as encountered when guiding light in an optical fibre, is a special case of the Berry phase (Arvind, Mallesh, & Mukunda, 1997; Berry, 1984, 1987a,b; Ramaseshan & Nityananda, 1994; Zwanziger, Koenig, & Pines, 1990) which results when any parameter describing a property of the system under concern, here the wave vector characterizing the propagation of the light field, undergoes a cyclic evolution in the state space, e.g. a geodesic triangle on the Poincar´e sphere. The geometric phase solely depends on the amount of the solid angle enclosed by the evolution path. It is worth noting that Pancharatnam anticipated the geometric phase in a classic paper on phase shifts in polarized light (Pancharatnam, 1956a; Ramaseshan & Nityananda, 1994). Pancharatnam’s 1956 and Berry’s 1984 papers were followed by a great number of papers in polarization optics, e.g. Bhandari (1988), Aravind (1992), Wagh and Rakhecha (1995), Zwanziger, Koenig, and Pines (1990), Ramaseshan and Nityananda (1994) and Arvind, Mallesh, and Mukunda (1997). In (Pancharatnam, 1956a), Pancharatnam studied the problem of determining the phase change undergone by polarized light after it has passed through a sequence of polarizers such that its final polarization is the same as its initial polarization. Certain topological properties of vector fields have been also studied using a space curve formalism (Balakrishnan, Bishop, & Dadoloff, 1990). In recent decades, the description of wave singularities led to new applications of polarization and coherence theory. As Nye (1983, 1999, chap. 12 and 13); Nye and Hajnal (1987) has pointed out, with light whose state of polarization varies with position, different generic singularities can occur, associated with the vector nature of light. For example, since for circular polarization the electric field vector never passes through zero, there is no obvious time origin for phase reference. These polarization singularities are lines, on which the electric (or magnetic) polarization ellipse is purely circular (C lines) or purely linear (L lines). These singular
From Wolf (1954) to Wolf (2007)
195
lines of polarization optics were detected experimentally by Hajnal (1990), using the modulated scatterer technique. The local structure of these singularities is very rich, and provides information about the polarization structure of the rest of the field. Berry and Dennis (2001) and Berry (2000) have shown that the patterns of ellipse-fields are different for purely paraxial and fully three-dimensional fields. In many situations the electromagnetic field does not meet the paraxial requirements of a beam. Friberg and colleagues (Lindberg, Set¨al¨a, Kaivola, & Friberg, 2004; Lindfors, Set¨al¨a, Kaivola, & Friberg, 2005; Set¨al¨a, Kaivola, & Friberg, 2002; Set¨al¨a, Lindfors, Kaivola, Tervo, & Friberg, 2004) tried to calculate a degree of polarization for an arbitrary electromagnetic field, based on the 3 × 3 spectral coherency matrix. But in the process, they had to face up to a curious feature, i.e. the value of the threedimensional degree of polarization for an unpolarized transverse wave is not equal to zero but to 12 , whose explanation is based on the fact that being a two-dimensional field the electric field vector at every point in space lies in a plane and consequently this wavefield is not random in the direction orthogonal to that plane. Dennis (2001, 2002) proposed a geometric interpretation of the 3 × 3 coherency (density) matrix for nonparaxial polarization by emphasizing its invariance with respect to spatial rotations. Under these transformations, the coherency (density) matrix decomposes into a real symmetric positive definite matrix, interpreted as the moment of inertia of the ensemble of electric field vectors, and a real axial vector, corresponding to the mean angular momentum of the ensemble. Related lines of questioning have been followed for a number of years, the goal there being mostly to characterize and control the statistical properties of nonplane electromagnetic fields. It should be also noted that Agarwal (2005) presented an analysis of some difficulties associated with the 3 × 3 density matrix for partially polarized transverse electromagentic fields. Ellis, Dogariu, Ponomarenko, and Wolf (2005) and Ellis and Dogariu (2005a) calculated a degree of polarization in a completely different way by considering the eigenvalues of the spectral density matrix. Defining the polarization states of a statistically stationary narrowband electromagnetic field as pseudospin states with SU(N) symmetry, Brosseau and Dogariu (2006) showed that the geometrical description of the convex set of states faithfully reflects the possibilities of relevant polarization descriptors. Within this perspective, the convexity property of the density matrix permits one to “see” geometrically how the pure states (completely polarized) and the mixed states (partially polarized) of different ranks are positioned relative to each other in the convex set of states. The SU(N) symmetry of the density matrix is similar to that of a pseudospin system, i.e. an N-level system in statistical physics. The geometric–algebra of the complex Hilbert space
196
A History of Polarization and Coherence Optics
unifies the Stokes parameters, the density (coherency) matrix, the entropy, the Jones and Mueller matrices, etc, in a single formalism and constitutes a descriptor of the polarization states of an arbitrary wavefield. The SU(3) symmetry gives a consistent description of the richness of phenomena in nonparaxial light situations, where in general there is no well defined propagation direction. The Poincar´e sphere is a particular case, i.e. a plane wave, of illustrating the general fact that pure states form the extremal points of the convex set. At the time when intense studies of this problem have begun, another exciting experimental direction has spun off. This is the physics of nanooptics where the properties of the optical field reveal the complex light–matter interaction. Based on a seminal idea of Soleillet (1929), Morales and Navarro (2003) also introduced recently a description of a polarized (plane) wave and (ideal) polarizers based on the Minkowski space–time formalism of special relativity, leading to a “kinematic” interpretation of the degree of polarization of the wave. Hence the Minkowskian geometry can be relevant to dealing with the properties of a partially polarized wave and of the changes of its degree of polarization. The complete determination of the polarimetric characteristics of a three-dimensional field requires measurements of the field correlations in three mutually perpendicular directions. This could be realized by using, for instance, three orthogonal dipolelike probes which are overlapped spatially and which are detected simultaneously. In the optical domain however, such an approach seems very difficult to implement. In a recent experiment, Ellis and Dogariu (2005b) demonstrated that the full threedimensional polarimetry of a general state can be obtained by placing in the point of interest a probe which couples all three components of the field and then re-emits the radiation. This probe acts as a secondary source for the radiation which is being sensed by a conventional detector, placed away from the point where measurements are made. By changing the coupling characteristics of such a probe, sufficient information is generated for determining all the elements of the cross-spectral density matrix. In quantum optics, the scalar Stokes parameters have to be replaced by Hermitian Stokes operators but the non-commutability of these operators precludes the simultaneous measurement of the quantities represented by them. The interested reader may consult (Case, 1988; Jauch & Rohrlich, 1955; Luis, 2002, 2005; Robson, 1974; Viana Ramos, 2005) for interesting discussions of the differences between classical and quantum descriptions of the density matrix relevant to polarized light. For example, Luis (2002) reported that a full characterization of polarization by means of higherorder than two correlations is required in quantum optics because no quantum state has a definite polarization ellipse for the same reasons that quantum particles do not follow definite classical trajectories. Among
Summary and Prospects
197
other applications, quantum effects associated with the polarization of an optical wavefield are of great interest for fundamental research in teleportation, quantum information processing (Burkalov & Klyshko, 1999; Burkalov, Chekhova, Karabutova, Klyshko, & Kulik, 1999; Karasev & Masalov, 1993), and non-classical states, e.g. squeezed states of light with suppressed fluctuations of one of the Stokes parameters (Alodzhants & Arakelyan, 1998, 2004) and states of light with “hidden” polarizations (Klyshko, 1997).
6. SUMMARY AND PROSPECTS In this brief historical survey, we have spanned the more than threecentury period from Bartholinus’s memoir of 1669 to Wolf’s book on the unified theory of coherence and polarization of light published in 2007. A large number of physicists have been involved in these efforts and their work has resulted in a vast edifice of scientific knowledge and technique that has led to many technological revolutions. Since an enormous volume of archival literature, both theoretical and experimental, that was concerned in the properties not only of polarized light but of other frequencies of the electromagnetic spectrum, has appeared. I have taken some trouble to sketch these highlights because the fundamental science involved in understanding polarization of light has great intellectual and practical value. The history has been enriched because of the many diverse fields, including optical communications, optical signal processing, spectroscopy and imaging, in which this body of knowledge has found engineering applications. A further enumeration of the current trends would be a formidable undertaking. Most of these developments have not been fully explored on either experimental or theoretical fronts and are much too recent to be written up as ’history’. But with all of the important accomplishments, are there any things left for the future? Most definitively yes. The last word on polarization optics has not yet been heard. There is very good reason to believe that with our increasingly powerful methods we are going to find surprises not only in optical physics but in technology as well. If we have gone so far in our understanding of the nature of polarized light in this century, it is in a large part by standing on the shoulders of intellectual giants such as Poincar´e, Faraday and Maxwell. If Fresnel, Arago and their colleagues had been told in 1800 to 1820 that they were laying the foundations for technological processes such as optical fibre communications, they would have been somewhat startled. I am confident that the future of polarization optics will be as exciting and fruitful as the past has been. What next? The nearly five decades that have passed since Wolf and others invented the coherency (density) matrix formalism have
198
A History of Polarization and Coherence Optics
borne witness to a proliferation of innovations and applications. The field of coherence and polarization optics is at an exciting stage of development. Even though some of the basic concepts regarding optical coherence theory have been around for a while, much remains to be understood and discovered. It is only very recently that measurement techniques have matured to the point that experiments can be successfully conducted. There do of course remain a number of points to be investigated, regarding the fundamental issue of precisely how a geometric–algebra picture lives with other work in polarization optics. A better understanding of how a 3-dimensional approach can be related to near-field optics and nanooptics problems. We are at the threshold of an exciting period in nanoscale optics because now we apparently have available an optical technique which allows us to investigate certain material properties on a length scale much smaller than the wavelength of light. One important question in this regard is how this 3-dimensional approach can be related to near-field optics and nanooptics problems (Keller, 2000; Ohtsu and Kobayashi, 2004; Nieto-Vesperinas and Garcia, 1993; van Hulst and Lewis, 1998; Pohl and Courjon, 1993). For example, related to the emerging nanotechnology, a recent experiment has demonstrated nanoscale control of optical properties on a length scale ¨ ¨ much less than a light wavelength (Muller, Sonnichen, von Poschingen, von Plessen, Klar, & Feldmann, 2002). The development of the scanning near-field optical microscope has revolutionized optical imaging and led to applications ranging from biophysics to optical data storage, e.g. (Betzig, Trautman, Harris, Weiner, & Kostelak, 1991). The full potential of most of the more recent polarization optics developments has yet to be realized. We may not need to wait very long for answers. Another exciting arena for polarization states with SU(3) symmetry is nonlinear optics (Kolokolov & Turitsyn, 2004; Menyuk & Wai, 1994; Svirko & Zheludev, 2000). For example, Mayer (1991) has shown that such states can be generated during the quantum parametric decay of a photon of a pump wave into two photons, namely, an idler wave and a signal wave, upon the formation of quantum states of an optical field with SU(3) symmetry in an anisotropically cubic nonlinear medium. Many people, ourselves included, expect that optics will have increasingly important things to say about optical biomedical imaging (Bowen, Treps, Schnabel, & Lam, 2002; Tuchin, 2000; Yodh & Chance, 1995). This consideration relies on that fact that, of the four basic forces in Nature, only the electromagnetic force is effective on a scale comparable to that of biological organisms as exemplified by the human visual and nervous systems. Diffusing waves are now used to view body and structure. This is made possible by a spectral window that exists within tissues in the 700–900 nm region, in which wave transport is
References
199
dominated by scattering rather than absorption. Near-field imaging of tissue characterization in turbid media using diffusing light fields offers new non-invasive opportunities for clinical diagnosis of the human body’s structure and function. By analyzing how light interacts with tissues, these “optical body scanners” would help doctors diagnose lesions such as tumours in the breast, and would determine whether a tumour is malignant or benign. For that purpose, understanding how light propagates through a highly scattering medium is an important step in resolution of the problem. Light is both absorbed and scattered by biological tissues, resulting in diffuse transmittance. Characteristics such as intensity, coherence, and polarization of the incident light change as it is absorbed and scattered inside the tissue. The extent of these changes depends on the nature of light being used, the type of tissue through which it propagates, and the tissue thickness. Thus, the light that emerges from the medium has very different characteristics from the incident light. Recent advances in techniques for studying optical fields with subwavelength features for which the longitudinal component of the electric field vector becomes important now allow a meaningful dialogue between experimentalists and theorists. Surely, this subject can keep any of us busy for some time.
ACKNOWLEDGMENTS This paper is a much belated revision and adaptation of lecture notes I provided for the Frontiers in Optics 2008/Laser Science XXIV Special Symposium entitled Polarized Light: 200 Years since Malus’ Discovery held in October 2008 at Rochester (USA). I thank Professors Taco D. Visser, Department of Physics and Astronomy, Vrije Universiteit, Amsterdam, The Netherlands, and Thomas G. Brown, Institute of Optics, Rochester University, USA, who organized this symposium. It is also a pleasure to acknowledge useful discussions with Professor Emil Wolf. Most of the photographs were obtained from the Internet. Laboratoire des Sciences et Technologies de l’Information, de la Communication et de la Connaissance (Lab-STICC) is UMR CNRS 3192.
REFERENCES Agarwal, G. S. (2005). Journal of Modern Optics, 52, 651. Allen, L., & Eberly, J. H. (1975). Optical resonance and two-level atoms. New York: Wiley. Alodzhants, A. P., & Arakelyan, S. M. (1998). Zhurnal Eksperimental’noi i Teoreticheskoi Fiziki, 113, 1235; JETP 86, 672. Alodzhants, A. P., & Arakelyan, S. M. (2004). Optics and Spectroscopy, 97, 424; Optika i Spektroskopiya, 97, 453. Anderson, D., & Barakat, R. (1994). Journal of the Optical Society of America, 11, 2305.
200
A History of Polarization and Coherence Optics
Arago, F. (1854). Œuvres compl`etes. Paris: Gide. Arago, F., & Fresnel, A. J. (1819). Annales de Chimie Physique, 2, 288. Aravind, P. K. (1992). Optics Communications, 94, 192. Arvind, K., Mallesh, S., & Mukunda, N. (1997). Journal of Physics A, 30, 2417. Askar’yan, G. A. (1962). Soviet Physics - JETP, 15, 1088. Balakrishnan, R., Bishop, A. R., & Dadoloff, R. (1990). Physical Review Letters, 64, 2107. Barakat, R. (1963). Journal of the Optical Society of America, 53, 317. Barakat, R. (1993). Journal of the Optical Society of America A, 10, 180. Barakat, R. (1996). Optics Communications, 123, 443. Barakat, R., & Blake, J. (1980). Physics Reports, 60, 225. Barakat, R., & Brosseau, C. (1993). Journal of the Optical Society of America, 10, 529. Barkla, G. G. (1906). Proceedings of the Royal Society of London Series A, 77, 247. Barnes, W. L., Dereux, A., & Ebbesen, T. W. (2003). Nature (London), 424, 824. Barr, E. S. (1988). Optics News, 16, 8. Bartholinus, E. (1670). Hafniae, reprinted in Swindell, W., 1975, (Ed.), Benchmark papers in Optics, Polarized Light, (Dowden, Huchtinson and Ross, Stroudsburg). Erasmus Bartholinus’s manuscript Experimenta crystalli Islandici disdiaclastici quibus mira et insolita refractio delegitur has disappeared”. But in the Royal Library in Copenhagen we find a manuscript entitled “Nova experimenta Crystalli Islandici disdiaclastici, quibus mira et insolita refractio detegitur”. It is a shorter, preliminary description of Erasmus Bartholinus’ experiments with the Iceland double refractive calcite crystals, perhaps written by a secretary. The differences in arrangement of the material in the manuscript and the printed treatise together with certain omissions and additions throw some light upon the final form in which Erasmus Bartholinus wished to publish his famous treatise on the double refractive Iceland spar. Baylis, W. E., Bonenfant, J., Derbyshire, J., & Huschilt, J. (1993). American Journal of Physics, 61, 534. Berdyugina, S. V., Berdyugin, A. V., Fluri, D. M., & Piirola, V. (2008). Astrophysical Journal, 673, L83. Berek, M. (1926a). Zeitschriff fur ¨ Physik, 36, 675. Berek, M. (1926b). Zeitschriff fur ¨ Physik, 36, 824. Berek, M. (1926c). Zeitschriff fur ¨ Physik, 37, 387. Berek, M. (1927). Zeitschriff fur ¨ Physik, 40, 420. Bernard, G. D., & Wehner, R. (1977). Vision Research, 17, 1019. Berreman, D. W. (1972). Journal of the Optical Society of America, 62, 502. Berry, M. V. (1984). Proceedings of the Royal Society of London Series A, 392, 45. Berry, M. V. (1987a). Proceedings of the Royal Society of London Series A, 414, 31. Berry, M. V. (1987b). Journal of Modern Optics, 34, 1401. Berry, M.V. (2000). Singular optics. (pp. 1–12), Alushta, Crimea: SPIE. Berry, M. V., & Dennis, M. R. (2001). Proceedings of the Royal Society of London Series A, 457, 141. Bertolotti, M. (1983). Masers and lasers: A historical approach. Bristol: Adam Hilger. Betzig, E., Trautman, J., Harris, T., Weiner, J., & Kostelak, R. (1991). Science, 251, 1468. Bhandari, R. (1988). Physical Letters A, 133, 1. Bicout, D., Brosseau, C., Martinez, A., & Schmitt, J. M. (1994). Physical Review E, 49, 1767. Billings, B. H. (1951). Journal of the Optical Society of America, 41, 966. Billings, B. H., & Land, E. H. (1948). Journal of the Optical Society of America, 38, 819. Bird, G. R., & Shurcliff, W. A. (1959). Journal of the Optical Society of America, 49, 235. Bird, G. R., & Parrish Jr., M. (1960). Journal of the Optical Society of America, 50, 886. Birge, R. T. (1935). Journal of the Optical Society of America, 25, 179. Blanc-Lapierre, A., & Dumontet, P. (1954). Comptes Rendus de l’ Academie des Sciences, Paris, 238, 1005. Blanc-Lapierre, A., & Dumontet, P. (1955). Review Optics, 34, 1.
References
201
Blandin, A., & Cohen-Tannoudji, C. (Eds.) (1965). Quantum optics and electronics. New York: Gordon and Breach. Bloembergen, N. (1965). In C. de Witt, A. Blandin, & C. Cohen-Tannoudji (Eds.), Les Houches lectures 1964, Nonlinear optics. New York: Gordon and Breach. Boerner, W. M., & Yan, W.-L. (1991). In B. de Neumann (Ed.), Electromagnetic modelling and measurements for analysis and synthetic problems. Amsterdam: Kluwer. Bohren, C. F. (1988). Weatherwise, 41, 175. for interesting historical details about the discoveries of Malus. Born, M., & Wolf, E. (1980). Principles of optics (6th ed.). Oxford: Pergamon Press, is among the most frequently cited references in the optics research literature. Bourret, R. C. (1960). Nuovo Cimento, 18, 347. Bowen, W. P., Treps, N., Schnabel, R., & Lam, P. K. (2002). Physical Review Letters, 89, 253601. Brosseau, C., & Barakat, R. (1990). Comptes Rendus de l’ Academie des Sciences, Paris B, 310, 181. Brosseau, C., Rockower, E., & Barakat, R. (1991). Optics Communications, 82, 204. Brosseau, C., & Barakat, R. (1991). Optics Communications, 84, 127. Brosseau, C., & Barakat, R. (1992). Optics Communications, 91, 408. Brosseau, C., & Bicout, D. (1994). Physical Review E, 50, 4997. Brosseau, C. (1998). Polarized light : A statistical optics approach. New York: Wiley. Brosseau, C. (2000). Journal of Optics A: Pure and Applied Optics, 2, R9. Brosseau, C., & Dogariu, A. (2006). Symmetry properties and polarization descriptors for an electromagnetic wavefield. In E. Wolf (Ed.), Progress in optics: Vol. 49 (pp. 315–380), Amsterdam: Elsevier. Brewster, D. B. (1815). Philosophical Transactions of Royal Society (London), 105, 125. Brewster, D. B. (1816). Philosophical Transactions of Royal Society (London), 105, 60. Brewster, D. B. (1987). The Kaleidoscop: Its history, theory and construction, reprinted by Van Cort Instruments, Holyoke, MA. Bruhat, G. (1929a). Review Optics, 8, 365. Bruhat, G. (1929b). Review Optics, 8, 413. Buchwald, J. Z. (1989). The rise of the wave theory of light. Chicago: University of Chicago Press. Bulletin of the American Mathematical Society, (1966), 72, 1. Dedicated to the memory of Norbert Wiener in recognition of his pioneering contributions to science. Burkalov, A. V., & Klyshko, D. N. (1999). Pis’ma v Zhurnal Experimental’noi i Teoreticheskoi Fisika, 69, 795; JETP Letters 69, 839. Burkalov, A. V., Chekhova, M. V., Karabutova, O. A., Klyshko, D. N., & Kulik, S. P. (1999). Physical Review A, 60, R4209. Cameron, D. A., & Pugh, E. N. (1991). Nature, 353, 161. Cantor, G. (1975). The reception of the wave theory of light in Britain. In R. Mc Cormach (Ed.), Historical studies in the physical sciences: Vol. 6. Princeton: Princeton University Press. Carozzi, T., Karlsson, R., & Bergman, J. (2000). Physical Review E, 61, 2024. Case, K. M. (1988). Transport Theory and Statistical Physics, 17, 635. Chandrasekhar, S. (1946). Astrophysical Journal, 104, 110. Chandrasekhar, S. (1950). Radiative transfer. Oxford: Clarendon Press. Chandrasekhar, S., & Elbert, D. (1954). Transactions of American Philosophical Society, 44, 643. Charney, E. (1979). The molecular basis of optical activity: Optical rotatory dispersion and circular dichroism. New York: Wiley. Clark Jones, R. (1941a). Journal of the Optical Society of America, 31, 488. Clark Jones, R. (1941b). Journal of the Optical Society of America, 31, 493. Clark Jones, R. (1941c). Journal of the Optical Society of America, 31, 500. Clark Jones, R. (1942). Journal of the Optical Society of America, 32, 486. Clark Jones, R. (1947a). Journal of the Optical Society of America, 37, 107. Clark Jones, R. (1947b). Journal of the Optical Society of America, 37, 110. Clark Jones, R. (1948). Journal of the Optical Society of America, 38, 671. Clark Jones, R. (1956). Journal of the Optical Society of America, 46, 126. Collett, E. (1971). American Journal of Physics, 39, 1483.
202
A History of Polarization and Coherence Optics
Cotton, A. (1895). Comptes Rendus de l’ Academie des Sciences, Paris, 120, 989. Debye, P. (1909). Annals of Physics, 30, 59. Dennis, M. R. (2001). Topological singularities in wave fields. Ph. D. Thesis, Bristol University. Dennis, M. R. (2002). Optics Communications, 213, 201. Dennis, M. R. (2004). Journal of Optics A: Pure and Applied Optics, 6, S26. Deschamps, G. A. (1951). Proceedings of IRE, 39, 540. Deschamps, G. A., & Mast, P. E. (1973). IEEE Transactions, AP-21, 474. Dollfus, A. (1956). Annals of Astrophysics, 19, 83. Du Bridge, L. A. (1935). Journal of the Optical Society of America, 25, 182. Ellis, J., Dogariu, A., Ponomarenko, S., & Wolf, E. (2005). Optics Communications, 248, 333. Ellis, J., & Dogariu, A. (2005a). Optics Communications, 253, 257. Ellis, J., & Dogariu, A. (2005b). Physical Review Letters, 95, 203905. Egan, W. G. (1985). Photometry and polarization in remote sensing. New York: Elsevier. Evans, J. W. (1958). Journal of the Optical Society of America, 48, 142. Evans, J. W. (1963). Applied Optics, 2, 193. Fainman, Y., & Shamir, J. (1984). Applied Optics, 23, 3188. Falkoff, D. L., & McDonald, J. E. (1951). Journal of the Optical Society of America, 410, 861. Fano, U. (1949). Journal of the Optical Society of America, 39, 859. Fano, U. (1954). Physical Review, 93, 121. Fano, U. (1957). Reviews of Modern Physics, 29, 74. Faraday, M. (1964). Experimental researches in electricity: Faraday’s collected works: 3 vols.. New York: Dover. Forrester, A. T. (1961). Journal of the Optical Society of America, 51, 253. Franken, P., Hill, A. E., Peters, C. W., & Weinreich, G. (1961). Physical Review Letters, 7, 118. French, A. P. (1990). The role of history in physics teaching. In J. Roche (Ed.), Physicists look back: Studies in the history of physics. Bristol: Adam Hilger. Fresnel, A. J. (1825). Annales de Chimie Physique, 28, 147. Fresnel, A. J. (1866). Œuvres Compl`etes, (L’Imprimerie Imp´eriale, Paris). Gabor, D. (1946). Journal of the Institution of Electrical Engineers. (London), 93, 429. Gamo, H. (1956). Journal of Applied Physics (Japan), 25, 431. Gamo, H. (1957a). Journal of Applied Physics (Japan), 26, 414. Gamo, H. (1957b). Journal of the Optical Society of America, 47, 976. Glasser, O. (1934). Wilhelm Conrad Rontgen and the early history of X-rays. Baltimore, MD: Thomas. Glauber, R. J. (1963a). Physical Review, 130, 2529. Glauber, R. J. (1963b). Physical Review, 131, 2766. Glauber, R. J. (1965). Optical Coherence and Photon Statistics. Les Houches Lectures 1964, de Witt, C. R. J.. Glauber, R. J. (2007). Quantum theory of optical coherence: Selected papers and lectures. Weinheim: Wiley-VCH. Goos, F., & H¨anchen, H. (1947). Annals of Physics, 6, 333. Goos, F., & H¨anchen, H. (1949). Annals of Physics, 6, 251. Gorchakov, G. I., & Rozenberg, G. V. (1967). Izvetsia Academy of Science USSR, Atmospheric and Oceanic Physics, 3, 394. Gordon, J. P., Zeiger, H. J., & Townes, C. H. (1954). Physical Review, 95, 282. Gordon, J. P., Zeiger, H. J., & Townes, C. H. (1955). Physical Review, 99, 1164. Govi, G. (1860a). Comptes Rendus de l’ Academie des Sciences, Paris, 51, 360. Govi, G. (1860b). Comptes Rendus de l’ Academie des Sciences, Paris, 51, 669. Graves, C. D. (1956). Proceedings of IRE, 44, 248. Haidinger, W. (1846). Annalen der Physik, 67, 435. Haidinger, W. (1847). Annalen der Physik, 70, 531. Hajnal, J. V. (1990). Proceedings of the Royal Society of London A, 430, 413. Hall, A. C. (1969). Surface Science, 16, 1.
References
203
Hanbury-Brown, R., & Twiss, R. Q. (1956a). Nature, 177, 27. Hanbury-Brown, R., & Twiss, R. Q. (1956b). Nature, 178, 1046. Hanbury-Brown, R., & Twiss, R. Q. (1956c). Nature, 177, 1447. Hanbury-Brown, R. (1991). Boffin. London: IOP. Hanbury-Brown, R. (1974). The intensity interferometer. London: Taylor and Francis. Hariharan, P., & Singh, R. G. (1961). Journal of the Optical Society of America, 51, 1148. Harman, P. M. (1990). The scientific letters and papers of James Clerk Maxwell: Vol. 1. New York: Cambridge University Press. Harman, P. M. (1993). European Journal of Physics, 14, 148. Hawryshyn, C. W. (1992). American Scientist, 80, 164. Hecht, J. (1992). Laser pioneers. New York: Academic Press. Hegedus, R., Akesson, S., Wehner, R., & Horvath, G. (2007). Proceedings of the Royal Society of London A, 463, 1081; See also Barta, A., Horvath, G., & Meyer-Rochow, B. (2005). Journal of Optical Society of America, A 22, 1023. Herapath, W. B. (1852). Philosophical Magazine, 3, 161. Herapath, W. B. (1855). Philosophical Magazine, 9, 366. Herschel, J. W. (1841). Encyclopedia metropolitana: Vol. 4. Light (p. 341). Hertz, H. (1888). Wiedemann’s Annalen der Physik und Chemie, 34, 551. Hertz, H. (1896). Miscellaneous papers. London: Macmillan. Hertz, J., Hertz, M., & Susskind, C. (1977). Heinrich hertz: Memoirs, letters, diaries. San Francisco: San Francisco Press. Hestenes, D. (1966). Spacetime algebra. New York: Gordon and Breach. Hestenes, D. (1990). New foundations for classical mechanics. Dordrecht: Reidel. Hiltner, W. A. (1947). Astrophysical Journal, 106, 231. Hiltner, W. A. (1949). Astrophysical Journal, 109, 471. Hiltner, W. A. (1951). Astrophysical Journal, 114, 241. Hopkins, H. H. (1943). Proceeding of the Physical Society London, 55, 116. Hopkins, H. H. (1951). Proceedings of the Royal Society of London A, 208, 263. Hopkins, H. H. (1953). Proceedings of the Royal Society of London A, 217, 408. Hopkins, H. H. (1957). Journal of the Optical Society of America, 47, 508. Hoymingen-Huene, P. (1993). Reconstructing scientific revolutions: Thomas S. Kuhn’s philosophy of science. Chicago: University Chicago Press. Howard, J. N. (1964). Applied Optics, 3, 1091. This paper presents a sketch of the life and principal scientific contributions of Lord Rayleigh, with emphasis on his research in optics. Hurwitz, H. (1945). Journal of the Optical Society of America, 35, 525. Huyghens, C. (1690). Trait´e de la Lumi`ere. Leiden, Van der Aa. Huyghens, C. (1950). Œuvres compl`etes. Nijhoff, La Hague: Soci´et´e Hollandaise des Sciences. Huynen, J. R. (1970). Phenomenological theory of radar targets. Ph.D. Dissertation, Delft University of Technology, Delft, The Netherlands. Inou´e, S., & Hyde, W. L. (1956). Journal of the Optical Society of America, 46, 372. James, D., & Wolf, E. (1991). Physical Letters A, 157, 6. Jancewicz, B. (1988). Multivectors and clifford algebras in electrodynamics. Singapore: World Scientific. Jauch, J. M., & Rohrlich, F. (1955). Theory of photons and electrons. Reading, MA: AddisonWesley. Javan, A. (1965). In C. de Witt, A. Blandin, & C. Cohen-Tannoudji (Eds.), Les Houches lectures 1964, Gaseous optical masers. New York: Gordon and Breach. Javan, A., Bennett, W. R., & Herriot, D. R. (1961). Physical Review Letters, 6, 106. Jerrard, H. G. (1954). Journal of the Optical Society of America, 44, 634. Jordan, P. (1927). Zeitschriff fur ¨ Physik, 44, 292. Karasev, P., & Masalov, A. V. (1993). Optika i Spektroskopiya, 74, 928; Optics and Spectrocopy, 74, 551]. Kattawar, G. W. (1994). Optics and Photonic News, 6, 43.
204
A History of Polarization and Coherence Optics
Keller, O. (2000). Jurnal of Chemical Physics, 112, 7856. Kennaugh, E. M. (1949). Effects of type of polarization on echo characteristics. Antennas Laboratory, The Ohio State University, Columbus, Ohio. Kennaugh, E. M. (1952). Polarization properties of radar reflections. M.Sc. Thesis. The Ohio State University, Columbus, Ohio. Kerr, J. (1875). Philosophical Magazine, 50, 337. Khintchine, A. (1934). Mathematische Annalen, 109, 604. Klyshko, D. N. (1997). Zhurnal Eksperimental’noi i Teoreticheskoi Fiziki, 111, 1955; JETP 84, 1065. Kohlrausch, R., & Weber, W. (1856). Poggendorff’s Annalen der Physik und Chemie, 99, 10. Kolokolov, I. V., & Turitsyn, K. S. (2004). Zhurnal Eksperimental’noi i Teoreticheskoi Fiziki, 125, 395; JETP 98, 348. ¨ Konnen, G. P. (1985). Polarized light in nature. Cambridge: Cambridge University Press. Kragh, H. (1991). Applied Optics, 30, 4688. Kreibig, U., & Vollmer, M. (1995). Optical properties of metal clusters. New York: Springer. Krishnan, R. S. (1938a). Proceeding of Indian Academy of Sciences A, 8, 442. Krishnan, R. S. (1938b). Proceeding of Indian Academy of Sciences A, 7, 91. Krishnan, R. S. (1938c). Proceeding of Indian Academy of Sciences A, 7, 878. Krishnan, R. S. (1939a). Kolloid-Zeitschriff , 84, 2. Krishnan, R. S. (1939b). Kolloid-Zeitschriff , 84, 18. Krishnan, R. S. (1939c). Proceeding of Indian Academy of Sciences A, 10, 395. Kubota, H., & Inou´e, S. (1959). Journal of the Optical Society of America, 49, 191. Kubota, H., & Saito, H. (1960). Journal of the Optical Society of America, 50, 1020. Kuhn, T. S. (1970). The Structure of Scientific Revolution. University of Chicago, Chicago. LaFay, H. (1970). Nature Geography, 137, 528. Lakhtakia, A (1989). Optics News, 15, 14. Land, E. H. (1951). Journal of the Optical Society of America, 41, 957. Land, E. H., & West, C. D. (1946). Colloid Chemistry, 6, 160. for details about the discovery of dichroic polarizers. Langsdorf Jr., A., & Du Bridge, L. A. (1934). Journal of the Optical Society of America, 24, 1. Lenoble, J. (1957). Journal de Physique et Le Radium, 18, 47. Lilienfeld, P. (1991). Applied Optics, 30, 4696. Lindberg, J., Set¨al¨a, T., Kaivola, M., & Friberg, A. T. (2004). Journal of Optics A: Pure and Applied Optics, 6, S59. Lindfors, K., Set¨al¨a, T., Kaivola, M., & Friberg, A. T. (2005). Journal of the Optical Society of America, 22, 561. Ling, P. J., & Teitler, S. (1984). Journal of the Optical Society of America A, 1, 703. Loeber, A. P. (1982). Journal of the Optical Society of America, 72, 650. Logan, N. A. (1962). Journal of the Optical Society of America, 52, 342. Logan, N. A. (1965). Proceedings of IEEE, 53, 773. Lorenz, L. V. (1890). Kongelige Danske Videnskabernes Selskab Forhandlinger, 6, 1. Luneberg, R. K. (1944). Mathematical Theory of Optics. Brown University, Providence. Lyot, B. (1928). Annales de l’Observatoire de Paris, 8, 102. Lyot, B. (1933). Comptes Rendus de l’ Academie des Sciences, Paris, 197, 1593. Lyot, B. (1944). Annals of Astrophysics, 7, 31. Luis, A. (2002). Physical Review A, 66, 013806. Luis, A. (2005). Optics Communications, 253, 10. Mach, E. (1926). The principles of physical optics: A historical and philosophical treatment. London: Methuen. McElheny, V. (1998). Insisting on the impossible: The life of Edwin land. Reading, MA: Perseus Books. Maiman, T. H. (1960a). Nature, 187, 493. Maiman, T. H. (1960b). British Communications and Electronics, 7, 674. Majorana, E. (1932). Comptes Rendus de l’ Academie des Sciences, Paris, 135, 159. Maker, P. D., Terhune, R. W., & Savage, C. M. (1964). Physical Review Letters, 12, 507.
References
205
Malus, E. (1808). M´emoires de la Soci´et´e d’Arcueil, 1, 113. Mayer, G., & Gires, F. (1964). Comptes Rendus de l’ Academie des Sciences, Paris, 258, 2039. Mandel, L. (1961). Journal of the Optical Society of America, 51, 1342. Mandel, L., & Wolf, E. (1965). Reviews of Modern Physics, 37, 231. Mandel, L., & Wolf, E. (1995). Optical coherence and quantum optics. New York: Cambridge University Press. Mauguin, C. (1911). Bulletin de la Societe Francaise de Mineralogie, 34, 71. Maxwell, J. C. (1864). Proceedings of the Royal Society of London Series, 13, 531. Maxwell, J. C. (1962). A treatise on electricity and magnetism. New York: Dover. Mayer, A. A. (1991). Kvantovaya Elektronika (Moscow), 18, 1264. McCrone, W. C., McCrone, L. B., & Delly, J. G. (1978). Polarized light microscopy. Ann Arbor: Ann Arbor Science Publ. Meier, F., & Zakharchenya, B. (Eds.) (1984). Optical orientation. New York: North-Holland. Menyuk, C. R., & Wai, P. K. A. (1994). Journal of the Optical Society of America B, 11, 1288. Michelson, A. A. (1911). Philosophical Magazine, 21, 554. Mie, G. (1908). Annalen der Physik (Leipzig), 25, 376. Minnaert, M. (1954). Light and color in the open air. New York: Dover. Misson, G. P. (1993). Ophthalmic and Physiological Optics, 13, 392. Morales, J. A., & Navarro, E. (2003). Physical Review E, 67, 026605. Mueller, H. (1943). Report No.2 of OSRD, Project OEMsr-576., cited by R. C. Jones as a declassified report in which Mueller explained the principle of his calculus. However the author has been unable to locate this report in the collections of the librairies to which he had access. Mueller, H. (1948). Journal of the Optical Society of America, 38, 661. Mujat, M., Dogariu, A., & Wolf, E. (2004). Journal of the Optical Society of America, 21, 2414. ¨ ¨ Muller, J., Sonnichen, C., von Poschingen, H., von Plessen, G., Klar, T. A., & Feldmann, J. (2002). Applied Physics Letters, 81, 171. Mulligan, J. F. (1987). American Journal of Physics, 55, 711. Neumann, F. E. (1840). Berichte Koniglich Preussischen Akademie der Wissenschten, 71, 255. Newton, I. (1952). Opticks (edition based on the 4th ed.). Dover Publications, Royal Society, London, 1704. Nicol, W. (1829). Edinburgh New Philosophical Journal, 6, 83. Nieto-Vesperinas, M., & Garcia, N. (Eds.) (1993). Optics at the nanometer scale, imaging and storing with photonic near fields. Dordrecht: Kluwer. Nieto-Vesperinas, M., & Garcia, N. (Eds.) (1996). Optics at the nanometer scale. Dordrecht: Kluwer Academic Publishers. Nitske, W. R. (1971). The life of Wilhelm Conrad Rontgen, discoverer of the X ray. Tucson: University of Ariz. Press. Nye, J. F. (1983). Proceedings of the Royal Society of London Series A, 389, 279. Nye, J. F., & Hajnal, J. V. (1987). Proceedings of the Royal Society of London A, 409, 21. Nye, J. F. (1999). Natural focusing and fine structure of light. Bristol: IoPP. O’Neill, E. L. (1963). Introduction to statistical optics. MA: Addison-Wesley, Reading. ¨ Ohman, Y. (1938a). Nature, 41, 157. ¨ Ohman, Y. (1938b). Nature, 41, 291. ¨ Ohman, Y. (1958). Arkiv fuer Astronomi, 2, 165. Ohtsu, M., & Kobayashi, K. (2004). Optical near fields. Berlin: Springer-Verlag. Optics and Photonic News (1994), 5, 9. Special issue dedicated to the memory of Edwin Land in recognition of his pioneering contributions to science and technology. Ore, A. (1955). Physical Review, 98, 887. Pancharatnam, S. (1956a). Proceeding of Indian Academy of Sciences A, 44, 247. Pancharatnam, S. (1956b). Proceeding of Indian Academy of Sciences A, 44, 398. Pancharatnam, S. (1963a). Proceeding of Indian Academy of Sciences, 57, 218. Pancharatnam, S. (1963b). Proceeding of Indian Academy of Sciences, 57, 231. Pancharatnam, S. (1975). Collected works. Oxford: Oxford University Press.
206
A History of Polarization and Coherence Optics
Parrent Jr., G. B. (1959a). Journal of the Optical Society of America, 49, 787. Parrent Jr., G. B. (1959b). Optica Acta, 6, 285. Parrent Jr., G. B., & Roman, P. (1960). Nuovo Cimento, 15, 370. Parke III, N. G. (1949). Journal of Mathematical Physics (MIT), 28, 131. Pasteur, L. (1907). Uber die Asymmetrie bei Naturlich Vorkommenden Organischen Verbrindunen. Leipzig: Teubner. Pellat-Finet, P. (1990). Optik, 84, 169. Pellat-Finet, P. (1992). Optik, 90, 101. Perrin, F. (1942). Jurnal of Chemical Physics, 10, 415. Perrin, F., & Abragam, A. (1951). Journal de Physique et Le Radium, 12, 69. Planck, M. (1959). The theory of heat radiation. New York: Dover. Pohl, D. W., & Courjon, D. (Eds.) (1993). NATO advanced study institute, Series E: applied sciences: vol. 242. Near field optics. Dordrecht: Kluwer. Poincar´e, H. (1892). Th´eorie Math´ematique de la Lumi`ere: Vol. 2. Paris: Georges Carr´e. Pritchard, B. S., & Elliott, W. G. (1960). Journal of the Optical Society of America, 50, 191. Ramachandran, G. N., & Chandrasekharan, V. (1951). Proceeding of Indian Academy of Sciences A, 33, 199. Ramaseshan, R. (1951). Proceeding of Indian Academy of Sciences A, 34, 32. Ramaseshan, S., & Nityananda, R. (1994). Current Science in India, 67, 217. Ramachandran, G. N., & Ramaseshan, S. (1952). Journal of the Optical Society of America, 42, 49. ¨ Ramachandran, G. N., & Ramaseshan, S. (1961). In S. Flugge (Ed.), Handbuch der Physik: Vol. XXV/1. Crystals optics. Berlin: Springer Verlag. ¨ H. Y. (1949). Journal of the Optical Society of America, 39, 136. Richartz, M., & Hsu, Richter, P. H. (1979). Journal of the Optical Society of America, 69, 460. Robson, B. A. (1974). The theory of polarization phenomena. Oxford: Clarendon Press. Robson, D. P. (1984). Chemical Matters, 8, 12. Roman, P., & Wolf, E. (1960a). Nuovo Cimento, 17, 462. Roman, P., & Wolf, E. (1960b). Nuovo Cimento, 17, 477. Ronchi, V. (1957). Histoire de la Lumi`ere. Paris: Colin; French translation Taton, J. for an account of the history of the development of the theory of light. Rontgen, W. (1895). Sitzungber Phys. -Med. Gesellschaft Wurzburg, ¨ 132, english translation in Nature 53, 1896, 274. Rosen, P. (1954). Physical Review, 96, 555, for details about the entropy of a polarized “pencil of radiation”. Rosenfeld, L. (1957). Nuovo Cimento Suppliment, 4, 1630. Roslund, C., & Beckman, C. (1994). Applied Optics, 33, 4754. Rossel, S., & Wehner, R. (1969). Nature, 323, 128. Rothen, A. (1945). Review of Scientific Instruments, 16, 26. Rothen, A. (1964). In E. Passaglia, A. A. Stromberg, & J. Kruger (Eds.), Natl. Bur. Std. Misc. Publ.: Vol. 256. Ellipsometry in the measurement of surfaces and thin films. Washington: US Govt. Printing Office. Rozenberg, G. V. (1946). Doctoral Dissertation, Moscow University. Rozenberg, G. V. (1960a). Soviet Physics-Uspekhi, 3, 346. Rozenberg, G. V. (1960b). Uspekhi Fizicheskii Nauk, 71, 173. Rozenberg, G. V. (1968). Advances in Physical Sciences, USSR, 11, 353. Rzhanov, A. V., Svitashev, K. K., & Semenenko, A. I. (1979). Principles of ellipsometry. Novosibirsk: Nauka. Sabra, A. (1967). Theories of light from descartes to Newton. London: Oldbourne. Sanders, J. H. (1959). Nature, 183, 312. Saxon, D. S. (1955). Physical Review, 100, 1771. Schawlow, A. L., & Townes, C. H. (1958). Physical Review, 112, 1940. Schlichter, C. P. (1990). Principles of magnetic resonance (3rd ed.). New York: Springer. Schuster, A. (1905). Astrophysical Journal, 21, 1.
References
207
Scott, W. T. (1963). American Journal of Physics, 31, 819. This review provides a full bibliography on the evolution of the electromagnetic field concept. Set¨al¨a, T., Shevchenko, A., Kaivola, M., & Friberg, A. T. (2002). Physical Review E, 66, 016615. Set¨al¨a, T., Kaivola, M., & Friberg, A. T. (2002). Physical Review Letters, 88, 123902. Set¨al¨a, T., Lindfors, K., Kaivola, M., Tervo, J., & Friberg, A. T. (2004). Optics Letters, 29, 2587. Shapiro, A. E. (1994). The optical papers of Isaac Newton. Cambridge: Cambridge University Press. Shurcliff, W. A. (1955). Journal of the Optical Society of America, 45, 399. Shurcliff, W. A. (1962). Polarized light: Production and use. Cambridge: Harvard UniversityPress. Shurcliff, W. A. (1971). Polarized light and optical measurement. Braunschweig: Pergamon Press. Siegel, D. M. (1991). Innovation in Maxwell’s electromagnetic theory. Cambridge: Cambridge University Press. Sinclair, G. (1950). Proceedings of IRE, 38, 148. Skolnik, M. E. (1970). Radar handbook. New York: Mc Graw-Hill. Sloan, L. L. (1955). Journal of the Optical Society of America, 45, 402. Solc, I. (1953). Cescoslovensky Casopis pro Fyziku, 3, 366. Solc, I. (1960). Cescoslovensky Casopis pro Fyziku, 10, 16. Solc, I. (1965). Journal of the Optical Society of America, 55, 621. Soleillet, P. (1929). Annalen de Physique, 12, 23. Stamov, D. G. (1953). Doctoral Dissertation, Moscow University, Moscow. Stefan, J. (1864a). Sitzungberichte der Kaiserlichen Akademie der Wissenschaften, Wien, 50, 380. Stefan, J. (1864b). Poggendorff’s Annalen der Physik und Chemie, 124, 623. Stokes, G. G. (1849). Cambridge and Dublin Mathematical Journal, 4, 1849. Stokes, G. G. (1852). Transactions of the Cambridge Philosophical Society, 9, 399. Stokes, G. G. (1901). Mathematical and Physical Papers. London: Cambridge University Press. Strong, J. (1958). Concepts in classical optics. San Francisco: Freeman. Strutt, J. W., (Lord Rayleigh), (1871a). Philosophical Magazine, 41, 107. Strutt, J. W., (Lord Rayleigh) (1871b). Philosophical Magazine, 41, 274. Strutt, J. W., (Lord Rayleigh), (1871c). Philosophical Magazine, 41, 447. Strutt, J. W., (Lord Rayleigh), (1881). Philosophical Magazine, 41, 81. Strutt, J. W., (Lord Rayleigh) (1900). Philosophical Magazine, 49, 157. Strutt, J. W., (Lord Rayleigh), (1920). Collected Scientific Papers. Cambridge University Press, Cambridge. Swindell, W. (1975). Benchmark papers in optics, polarized light. Stroudsburg: Dowden, Huchtinson, Ross. (Collection of reprints of 35 papers on polarized light selected from the literature on this subject of the last 3 centuries). Sudarshan, E. C. G. (1963). Physical Review Letters, 10, 277. Svirko, Y. P., & Zheludev, N. I. (2000). Polarization of light in nonlinear optics. New York: Wiley. Takahara, J., & Kobayashi, T. (2004). Optics and Photonic News, 15, 55. Takenaka, H. (1973). Nouvelle Revue d’Optique, 4, 37. Teitler, S., & Henvis, B. W. (1970). Journal of the Optical Society of America, 60, 830. Tolhoek, H. A. (1956). Reviews of Modern Physics, 28, 277. Tuchin, V. V. (2000). Tissue optics: Light scattering methods and instruments for medical diagnosis: vol. TT38. Bellingham, WA: SPIE Optical Engineering Press. Tyndall, J. (1869a). Proceedings of the Royal Society of London Series, 17, 223. Tyndall, J. (1869b). Philosophical Magazine, 37, 384. Urbantke, H. (1991). American Journal of Physics, 59, 503. Uslenghi, P. L. E. (Ed.) (1978). Electromagnetic scattering. New York: Academic. Vahimaa, P., & Tervo, J. (2004). Journal of Optics A: Pure and Applied Optics, 6, S41. van Cittert, P. H. (1934). Physica, 1, 201. van Cittert, P. H. (1939). Physica, 6, 1129. van de Hulst, H. C. (1957). Light scattering by small particles. New York: Wiley.
208
A History of Polarization and Coherence Optics
van Hulst, N., Lewis, A. (eds.) (1998). Proceedings of the Fourth International Conference on Near Field Optics; [Ultramicroscopy 71, 1], and references therein. Verdet, E. (1865). Annales Scientifiques de l’Ecole Normale Sup´erieure, 2, 291. Verdet, E. (1869). Le¸cons d’ Optique Physique. Paris: Imprimerie Imp´eriale. Vernon, R. J., & Huggins, B. D. (1970). Journal of the Optical Society of America, 70, 1364. Viana Ramos, R. (2005). Journal of Modern Optics, 52, 2093. von Frisch, K. (1950). Bees: Their vision, chemical senses and language. Ithaca: Cornell University Press. von Laue, M. (1906). Annals of Physics, 20, 365. von Laue, M. (1907a). Annals of Physics, 23, 1. von Laue, M. (1907b). Annals of Physics, 23, 795. von Neumann, J. (1955). Mathematical foundations of quantum mechanics. Princeton: Princeton University Press. von Neumann, J. (1961). In A. H. Traub (Ed.), Collected works. New York: Pergamon. Wagh, A. G., & Rakhecha, V. C. (1995). Physical Letters A, 197, 112. Wartmann, R., Wolff-Fischer, E., & Berek, M. (1926). Leitz Mitteillungen fuer Wissenschaft und Technik, 9, 7. Wehner, R. (1976). Scientific American, 235, 106. Westfold, K. C. (1959). Journal of the Optical Society of America, 49, 717. Whitney, C. (1971). Journal of the Optical Society of America, 61, 1207. Whittaker, E. (1960). A history of the theories of aether and electricity. New York: Harper & Row, vols. I and II. Wiener, N. (1930). Acta Mathematica, 55, 117. Wiener, N. (1929). Journal of the Franklin Institute, 207, 525. Wiener, N. (1953). Journal of the Optical Society of America, 43, 225. Wiener, N. (1933). The Fourier integral and certain of its applications. New York: Cambridge University Press. Wiener, O. (1890). Annals of Physics, 40, 203. Wolf, E. (1954a). Proceedings of the Royal Society of London A, 225, 96. Wolf, E. (1954b). Nuovo Cimento, 12, 884. Wolf, E. (1955). Proceedings of the Royal Society of London A, 230, 96. Wolf, E. (1957). Philosophical Magazine, 2, 351. Wolf, E. (1959). Nuovo Cimento, 13, 1165. Wolf, E. (2001). Selected works of Emil Wolf: With commentary. New York: World Scientific. Wolf, E. (2003). Physical Letters A, 312, 263. Wolf, E. (2007). Introduction to the theory of coherence and polarization of light. New York: Cambridge University Press. Wood, R. W. (1934). Journal of the Optical Society of America, 24, 4. Yariv, A., & Yeh, P. (1984). Optical waves in crystals. New York: Wiley. Yodh, A., & Chance, B. (1995). Physics Today, 48, 34. Young, A. (1982). Physics Today, 35, 2. Young, T. (1802). Philosophical Transactions of Royal Society (London), 92, 387. Young, T. (1804). Philosophical Transactions of Royal Society (London), 94, 1. Zapasskii, V. S., & Kozlov, G. C. (1995). Optics and Spectroscopy, 78, 88. Zernike, F. (1938). Physica, 5, 785. Ziggelaar, A. (1980). Annals of Science, 37, 179. Zwanziger, Z. W., Koenig, U., & Pines, A. (1990). Annual Review of Physical Chemistry, 41, 601.
CHAPTER
4 Optical Quantum Computation Tim C. Ralph a and Geoff J. Pryde b a Centre for Quantum Computer Technology, Department of
Physics, University of Queensland, St Lucia 4072, Australia b Centre for Quantum Computer Technology, Centre for Quantum Dynamics, Griffith University, Nathan, 4111, Brisbane, Australia Contents
1.
2.
3.
4. 5.
6.
7.
Introduction 210 1.1. Quantum Computation 210 1.2. Quantum Optics 213 Optical Qubits 215 215 2.1. Dual-Rail Encoding 2.2. Single-Rail Encoding 217 Universal Optical Quantum Gate Sets 218 219 3.1. Nonlinear Gates 223 3.2. Linear Optics Gates 3.3. Coherent State Gates 229 Cluster States 234 Fault Tolerance 236 5.1. Loss Tolerance 237 5.2. Telecorrection 238 5.3. Thresholds and Resource Counts for Optical 239 Schemes Experimental Demonstrations 242 6.1. Linear Optics, Particle Encodings 242 6.2. Circuits and Protocols Using Linear Optics Gates 249 6.3. Linear Optics, Field Encodings 250 6.4. Nonlinear Optical Quantum Computing Experiments 251 Auxiliary Components 252 7.1. Detectors 253 7.2. Sources 255 7.3. Memory 259 7.4. Integrated Optics 261
c 2009 Elsevier B.V. Progress in Optics, Volume 54 ISSN 0079-6638, DOI 10.1016/S0079-6638(10)05409-0 All rights reserved.
209
210
Optical Quantum Computation
8. Conclusion and Summary Acknowledgments References
262 263 263
1. INTRODUCTION At the most fundamental level, physical processes are described by quantum mechanics. Quantum mechanical systems possess unique properties that enable new ways of communicating and processing information (Nielsen & Chuang, 2000). Large scale coherent processing of information via quantum systems is referred to as quantum computation. However, to achieve quantum computation, physical systems with very special properties are required. For example, it is essential that the quantum system evolves coherently and thus must be well isolated from the surrounding environment. Simultaneously, in order that the information stored in the system can be processed and read out, it must also be possible to produce very strong interactions between the system and classical meters and control elements. The invention of the laser in the early 1960s and its subsequent development led to an unprecedented increase in the precision with which light could be produced and controlled, and hence enabled the systematic investigation of the quantum properties of optical fields. It was soon realized that quantum optics offered a unique opportunity, not previously available to experimentalists, to test fundamentals of quantum theory (Walls & Milburn, 1994) and later quantum information science (Bachor & Ralph, 2004). It is natural, then, to consider quantum optics as a physical platform for quantum computation. In this article we review the progress in achieving quantum information processing in optics and the prospects for building a large scale optical quantum computer.
1.1. Quantum Computation The development of computers with greater and greater power during the nineteen sixties and seventies made many more problems in quantum physics tractable to analysis. Nevertheless, it soon became clear that, in general, the simulation of quantum systems is a hard problem – that is, the computing resources required grow exponentially with the size of the quantum system. It was Feynman (1986) who first noted that this bug might be turned into a feature. He pointed out that one quantum system could efficiently simulate another. Hence, a quantum simulator could efficiently solve problems that were intrinsically hard for a classical simulator. This raised the possibility that other computer algorithms may exist that could be more efficiently processed by quantum systems
Introduction
211
than classical systems. Although toy examples of such algorithms were suggested by Deutsch (1985) soon after, it was not till the mid-1990s that Shor (1994) showed that an important problem, the determination of the prime factors of an integer, could be solved in exponentially less time using a quantum computer. Equally importantly, it was shown shortly afterwards that quantum error correction was possible (Shor, 1995; Steane, 1996). This enables coherent correction of the logical errors which will inevitably creep into any calculation on a real physical device. Another influential algorithm, showing speed up for the searching of an unsorted data base, was subsequently developed by Grover (1997). These developments showed that fault tolerant quantum computers (i.e. where errors can be corrected in the presence of imperfect gate operations) were in principle possible and that such machines could solve interesting problems more efficiently than a conventional computer. This led to an explosion of interest in the field of quantum information. Quantum information was originally framed in terms of binary systems. Consider a two-level quantum system. This could be: the spin states of an electron (up or down); two well-isolated energy levels of an atomic system; or one of many other possibilities, including various optical field states. It is clear that such two level systems could be used to carry bits of information. For example, we could assign the value ‘zero’ to one of the states, writing it in Dirac notation (Sakurai, 1985) as |0i, and ‘one’ to the other state, writing |1i. These states are typically taken to be eigenstates of the Pauli Z operator with eigenvalues ±1 (Nielsen & Chuang, 2000). An ordered collection of such objects could then faithfully represent an arbitrary bit string. However, quantum objects offer more possible manipulations than classical carriers of bits. In particular, not only can we have zeros and ones, but we can also√have superpositions of zeros and ones such as the diagonal state |+i = (1/ 2)(|0i + |1i). Indeed bits can just as effectively be encoded in these superposition basis states, √ for example using |+i as a zero and the anti-diagonal state |−i = (1/ 2)(|0i − |1i) as a one. Following our convention, these states are eigenstates of the Pauli X operator. We will refer to the Z basis as the computational basis and the X basis as the diagonal basis. In general we can form any superposition of the form α|0i + β|1i where α and β are arbitrary complex numbers obeying normalization (|α|2 + |β|2 = 1). Because of these extra degrees of freedom we refer to information digitally encoded on quantum systems as quantum bits or qubits (Schumacher, 1995). A feature of qubits is their ability to span all different bit string values simultaneously. For example suppose we start with two qubits in the state |0i|0i where the first ket represents the first qubit and the second ket the second qubit and a tensor product between their two Hilbert spaces is
212
Optical Quantum Computation
implied. If we rotate both of them into their diagonal states we end up with the state 1 (|0i|0i + |0i|1i + |1i|0i + |1i|1i) 2
(1)
which is an equal superposition of all four possible two bit-values. This generalizes to n qubits, where the same operation of rotating every individual qubit leads to an equal superposition of all 2n bit string values. Even greater power comes from the ability to place qubits into certain special superpositions of the bit values – specifically, superpositions of correlated bit values. For example consider the two qubit state 1 √ (|0i|0i + |1i|1i). 2
(2)
Now only two of the four possible combinations are present. A pair of spatially separated quantum systems are said to be entangled if the state that describes the joint system cannot be factored into a product of states describing the individual systems. The state of Equation (2) clearly cannot be factored into contributions from the individual qubits, and thus it is entangled. Such a state has no classical analogue. Entanglement is thought to be a key ingredient in the information processing speed-up offered by quantum computation. If we wish to perform information processing using qubits we need to introduce quantum gates. The action of a quantum gate on a qubit state, |φi, can be represented by the action of a unitary operator, U , on the state, i.e. |φi → U |φi. Some quantum gates have classical counterparts, for example the NOT gate, X , takes |0i to |1i and vice versa. Other gates have no √ classical analogue, such as the √ Hadamard gate, H , for which H |0i = (1/ 2)(|0i + |1i) and H |1i = (1/ 2)(|0i − |1i). An arbitrary rotation by an angle θ about the j-axis of the Bloch sphere, where j = x, y, z, is effected by the unitary Jθ = cos θ I + i sin θ J where J = X, Y, Z , and I is the identity operator. We also require two-qubit gates such as the controlledNOT (CNOT) which performs the NOT operation on one qubit (the target) only if the other qubit (the control) has ‘one’ as its logical value. Eventually, if large arrays of gate operations can be implemented efficiently, and fault tolerantly, on many qubits, one could consider performing quantum computation. In more recent years quantum information research has been extended to systems with Hilbert space dimensions greater than two (Bullock, OLeary, & Brennen, 2005). For example a 3-level quantum system can encode three separate ‘trit’ values, |0i, |1i and |2i, plus any superposition
Introduction
213
of them, α|0i+β|1i+γ |2i. Such a system is called a qutrit and, generalizing, a d-level quantum system is said to form a qudit. There has also been considerable interest in infinite-dimensional Hilbert spaces and the quantum information properties of continuous degrees of freedom such as position and momentum (Braunstein & Pati, 2003; Braunstein & van Loock, 2005). It is usual to take the computational basis states toR be the position eigenstates |xi and hence the momentum states | pi = dx eix p |xi form the diagonal basis where, as expected, each diagonal basis state spans all computational basis values. Quantum computation proposals based on continuous variables have been developed (Lloyd & Braunstein, 1999; Menicucci, Loock, Gu, Weedbrook, Ralph, & Nielsen, 2006). Although these are theoretically universal in the sense that gate sets can be identified to efficiently simulate any process, a question mark exists over fully continuous schemes because of the lack of general error correction protocols. A fruitful way around this problem is to encode qubit states into the continuous spectrum (Gottesman, Kitaev, & Preskill, 2001; Lund, Ralph, & Haselgrove, 2008). Although considerable progress has been made, in many different physical platforms, including ion traps (Haeffner, Roos, & Blatt, 2008), superconductors (Schoelkopf & Girvin, 2008) and solid state systems (Gaebel, Domhan, Popa, Wittmann, Neumann, Jelezko, Rabeau, Stavrias, Greentree, Prawer, Meijer, Twamley, Hemmer, & Wrachtrup, 2006), the realization of quantum computation experimentally still remains a long way off. Indeed it is still quite unclear what physical platforms, if any, are compatible with the task of building a full scale quantum computer. In this article we present the case for optics.
1.2. Quantum Optics Light can be described quantum mechanically in terms of the mode annihilation operator a, ˆ its conjugate, the creation operator aˆ Ď , and the electromagnetic field mode ground state, or vacuum state |0i, defined by a|0i ˆ = 0 (Dirac, 1958). The mode operators obey the commutation relation [a, ˆ aˆ Ď ] = 1. The action of the creation operator on the vacuum state is to create a single photon number state, in√a single spatio-temporal mode, i.e. aˆ Ď |0i = |1i. In general aˆ Ď |ni = n + 1|n + 1i where n = 0, 1, 2, . . .. Similarly the annihilation operator annihilates a single photon in ˆ = √ a particular single spatio-temporal mode and in general a|ni n|n − 1i. The number states form an ortho-normal basis convenient for representing arbitrary states of the field. The spatio-temporal mode operators can be further decomposed into Ď single wave-vector operators, aˆ k , with the property [aˆ k 0 , aˆ k ] = δ(k 0 −k). For example the mode operator representing a plane wave mode propagating
214
Optical Quantum Computation
in the plus x direction can be written Z a(t, ˆ x) =
dk G(k) ei(kx−ωk t) aˆ k
(3)
where the optical frequency is given by ωk = c|k|. G(k) is a normalised spectral mode distribution function centred around some positive wave number, k0 (corresponding to an optical frequency), and is required to be zero for k < 0. Orthogonal spatio-temporal modes are characterized by having mode operators that commute. For example, if we take the mode function of Equation (3) and displace it to a new longitudinal position x 0 and consider its same-time commutator with the original mode we obtain [a(t, ˆ x 0 ), a(t, ˆ x)Ď ] =
Z
0
dk|G(k)|2 eik(x −x) .
(4)
For a suitably large interval |(x − x 0 )| the right hand side of Equation (4) will go to zero and hence we will have orthogonal modes. Effectively we have created a pair of well-separated pulses. We may also create orthogonal modes as a function of transverse displacement, varying transverse or longitudinal mode shapes or differing polarizations. In this article we will simply declare suitably labelled mode operators to be orthogonal and avoid explicit decompositions. The optical observables we will be interested in are the photon number, nˆ = aˆ Ď a, ˆ and the quadrature amplitude, Xˆ θ = eiθ aˆ + e−iθ aˆ Ď . Photon number is proportional to intensity for bright fields and can be measured by photo-detectors. For dim fields individual photons can be resolved with photon counters. The quadrature amplitude of the field can be measured by beating the signal field with a bright, phase reference field at the same optical frequency, a local oscillator (LO), and then measuring it with photo-detection. This is known as homodyne detection. The angle θ is the phase difference between the signal and the LO and is usually taken to be in-phase (θ = 0) or in-quadrature (θ = π/2), giving two conjugate (i.e. non-commuting) observables analogous to position and momentum. As well as the number states, another key state in quantum optics is the coherent state (Glauber, 1962). The coherent states are displaced vacuum states defined by ˆ |αi = D(α)|0i
(5)
Optical Qubits
215
where the displacement operator is ˆ aˆ Ď α ∗ ) ˆ D(α) = ei(aα+ .
(6)
The coherent states are eigenstates of aˆ with eigenvalue α. This leads to average values for their quadrature observables that are the same as for a classical field with the same amplitude. Hence the coherent state is often thought of as the quantum mechanical state which is the closest approximation to a classical optical field. The output of a well stabilized laser is a mixed state which can be approximately decomposed into an ensemble of coherent states with fixed magnitude but random phases (Mølmer, 1997). However, in situations where the phase is unimportant, or when the LO is derived from the same laser as the signal such that the phase is common mode, it is convenient to model the laser output as being in a single coherent state of fixed magnitude and phase.
2. OPTICAL QUBITS We now consider how quantum information can be carried by light. An obvious choice is to consider photons as particles and to encode information onto some bipartite degree of freedom of individual photons such as polarization. Such an encoding always requires 2 distinct optical modes to be present, so we will refer to particle-like encodings as dualrail encodings (Knill, Laflamme, & Milburn, 2001). As the particle state is an energy eigenstate, i.e. a single photon Fock state, dual-rail qubits are stationary states that do not evolve as they propagate. Alternatively, we could consider the field mode as the fundamental object and encode information in different field states, for example two distinct Fock states. In this encoding only one quantum optical mode is used, thus we will refer to field-like encodings as single-rail encodings (Lund & Ralph, 2002). Notice, however, that now we are inevitably considering situations in which our optical modes are in superpositions of energy eigenstates and hence experience a phase evolution as they propagate. As a result, a co-propagating classical mode is implicitly needed as a phase reference for single-rail encodings. In the following, we will describe these encoding techniques in more detail and discuss a number of examples.
2.1. Dual-Rail Encoding Consider two orthogonal optical modes represented by the annihilation operators aˆ and bˆ and the vacuum modes |0ia and |0ib . For brevity we will write |0ia ⊗ |0ib ≡ |00i. We define our logical qubits as |0i = aˆ Ď |00i = |10i
216
Optical Quantum Computation
(a)
θ
a
(b)
η b
φ
λ /2 λ /4 λ /2
PBS
FIGURE 1 Manipulation and detection of dual-rail qubits. (a) Beam splitter and phase-shifter circuit for producing an arbitrary single qubit evolution on a spatial dual-rail qubit. Detection in the computational basis is achieved by measuring which spatial mode holds the photon. (b) Combination of half- and quarter-wave plates oriented at particular angles achieves arbitrary single qubit evolution on a polarization dual-rail qubit. Detection in the computational basis is achieved via a polarizing beamsplitter (PBS) and photon counting
and |1i = bˆ Ď |00i = |01i. That is, single photon occupation of one mode represents a logical zero, whilst single photon occupation of the other mode represents a logical one. This is dual-rail encoding. For example, suppose |0ia and |0ib are spatio-temporal modes with identical profiles, polarization and centre frequency, synchronized in time, but spatially separated in the transverse direction. Arbitrary single qubit operations can be achieved using a beamsplitter and two phase shifters, as illustrated in Figure 1(a). A beamsplitter is a partially reflecting mirror that can coherently combine two optical modes in a set ratio. The interaction in the figure produces the following Heisenberg evolution of the mode operators: p √ aˆ → ηaˆ + eiθ 1 − ηbˆ p p bˆ → eiφ 1 − ηaˆ − eiθ 1 − ηbˆ
(7)
where η is the intensity reflectivity of the beamsplitter. We have assumed the optical elements are lossless, a reasonable first approximation for modern components. We have also assumed perfect mode matching between the two input modes to the beamsplitter, something rather more difficult to arrange in practice. Equation (7) implies the following qubit evolution (Bachor & Ralph, 2004): p √ |10i → η|10i + eiφ 1 − η|01i p p |01i → eiθ 1 − η|10i − eiφ 1 − η|01i
(8)
Optical Qubits
217
which corresponds to an arbitrary single qubit unitary operation. Detection in the computational basis simply corresponds to measuring the photon number in each spatial mode. More commonly, two identical spatio-temporal modes, but with different polarizations, say horizontal and vertical, are used as the dual rails. Then we may write |0i = |10i = |H i and |1i = |01i = |V i. Half- and quarter-wave plates replace the phase shifters and beamsplitters in achieving arbitrary unitaries (Dodd, Ralph, & Milburn, 2003). In particular, the Hadamard gate is implemented by a half-wave plate oriented at 22.5 degrees to the optical axis. Detection in any basis can be achieved via wave plates and polarizing beamsplitters, of which the latter effectively converts polarization encoding into spatial encoding (see Figure 1(b)). The ease of manipulation and phase stabilty of polarization states has made this encoding the most popular in optics (see Section 6.1). Other possibilities are: temporal encodings in which the dual rails are spatio-temporal modes which are identical except for a time displacement (Stucki, Gisin, Guinnard, Ribordy, & Zbinden, 2002) (see example in Section 1.2); and frequency encodings in which the dual-rail modes are identical except for a frequency off-set (Huntington & Ralph, 2004). Qubit initialization for these dual-rail schemes amounts to the ability to produce single-mode, single-photon states in a controlled way. Considerable progress has been made towards achieving this goal, as will be discussed in Section 7.2. Two qubit entangling gates turn out to be quite a challenge for dual-rail schemes, however the ease of single qubit manipulation (especially for polarization and spatial encodings) and measurement make these encodings strong contenders for large scale quantum computation.
2.2. Single-Rail Encoding Single-rail encoding requires only a single quantum mode, that can be prepared in the states |0i = |φi and |1i = |ψi or any superposition of them. The only requirement on these states |φi, |ψi is that they are orthogonal, i.e. that hφ|ψi = 0. In general such qubits will be non-stationary, and therefore a good ‘clock’ (i.e. a LO) is required in order to detect and manipulate them. Perhaps the simplest choice for |φi and |ψi are the vacuum and single photon states, such that |0i = |0i and |1i = |1i. Producing and manipulating superposition states of the form µ|0i + ν|1i is not so easy, however a universal set of non-deterministic operations has been described (Lund & Ralph, 2002). Also, we shall see that some two-qubit dual-rail gates are actually built from more fundamental single-rail gates of this type.
218
Optical Quantum Computation
Another possible choice for |φi and |ψi are two different coherent states, such that |0i = |αi and |1i = |βi. In general such states will not be orthogonal but their overlap is given by |hα|βi|2 = exp[−|α − β|2 ] which is very small for quite modest differences in the amplitudes of the coherent states. This is a continuous-variable-type encoding as we are carving out qubits from a continuous Hilbert space. A popular choice is to take β = −α (Cochrane, Milburn, & Munro, 1998). By choosing α ≥ 2 a small overlap is achieved. The computational states, |αi and | − αi can be distinguished via homodyne detection. A useful feature of this choice is that the equal superposition state |αi + | − αi (|αi − | − αi) contains only even (odd) photon number terms and so these orthogonal diagonal states can be distinguished by photon counting. As with the single photon single-rail scheme, single-qubit unitaries are difficult with this encoding. On the other hand, entanglement production is relatively easy. Splitting a superposition state like |αi + | − αi many times on a beamsplitter leads to multi-mode entanglement. It turns out that this feature, i.e. easy entangling (and disentangling) operations, compensates sufficiently for the greater difficulty of performing single qubit unitaries to make the coherent state single-rail scheme a serious contender for large scale quantum computation (Lund, Ralph, & Haselgrove, 2008). Furthermore there has been considerable recent progress in producing diagonal-basis resource states for this encoding, as will be discussed in Section 7.2. A more exotic single-rail scheme, in which the qubit states are comprised of superpositions of multiple, evenly spaced squeezed states, has also been suggested (Gottesman, Kitaev, & Preskill, 2001). This scheme has the full continuous variable feature that transforming between the computational and diagonal bases is equivalent to transforming between the position and the momentum bases. In addition, this structure provides a natural way for general error correction to be implemented. However it currently appears that the greater technical requirements of this approach outweigh these appealing features (Glancy & Knill, 2006).
3. UNIVERSAL OPTICAL QUANTUM GATE SETS A universal quantum gate set enables any n-qubit unitary transformation to be implemented to arbitrary accuracy, for any n (Nielsen & Chuang, 2000). Many different universal gate sets exist. A sufficient universal gate set is comprised of arbitrary single-qubit unitary operations plus a maximally-entangling two-qubit gate. An example of the latter is the Controlled-NOT (CNOT) gate. The CNOT gate is a two-qubit gate in which one qubit plays the role of a control and the other a target. When the control qubit is in the zero
Universal Optical Quantum Gate Sets
219
state, |0ic , the value of the target qubit |0it or |1it is unchanged. However, when the control is one, |1ic , the value of the target qubit is flipped, zero to one and vice versa. We can see that this gate is maximally entangling by considering its effect on superposition states. These can be calculated by simply making superpositions of the aforementioned transformations. For example if the control is in the diagonal basis we get the following transformations √ 1/ 2(|0ic + |1ic )|0it √ 1/ 2(|0ic + |1ic )|1it √ 1/ 2(|0ic − |1ic )|0it √ 1/ 2(|0ic − |1ic )|1it
√ → 1/ 2(|0ic |0it + |1ic |1it ) √ → 1/ 2(|0ic |1it + |1ic |0it ) √ → 1/ 2(|0ic |0it − |1ic |1it ) √ → 1/ 2(|0ic |1it − |1ic |0it ).
(9)
Notice that separable states are transformed into maximally entangled states. In fact, the two-qubit spanning set of entangled states on the right of Equation (9) is given a special name – the Bell states. A closely related maximally-entangling two-qubit gate is the Controlled-Sign gate (CZ), for which all state components are unchanged except for |1i|1i → −|1i|1i. A CZ gate can be transformed into a CNOT gate by placing Hadamard gates before and after the CZ gate on the target qubit. In the following we will look at how universal gate sets can be realized in various ways for the two most promising optical qubits we have discussed: single photon, dual rail (spatial or polarization) and coherent state single rail.
3.1. Nonlinear Gates We start our discussion with the conceptually simplest, but practically most difficult gate sets – those based on highly nonlinear, in-line optical interactions. 3.1.1. Kerr Nonlinearities for Dual Rail Arbitrary single qubit unitaries come for ‘free’ with dual-rail schemes as they require only simple linear interactions. Thus our job is complete if we can implement a maximally entangling two-qubit gate. So how might such an interaction between two photons be implemented? One solution is to use a χ (3) nonlinear medium to induce a cross-Kerr effect between two photon modes, as first suggested by Milburn (1989). Ideally the cross-Kerr ˆ where aˆ effect will produce the unitary evolution Uˆ K = exp[iχ aˆ Ď aˆ bˆ Ď b], ˆ represents one optical mode and b another. Consider the schematic set-up of Figure 2(a). Two polarization encoded qubits are converted into spatial dual-rail qubits using polarizing beamsplitters. One mode from each of
220
Optical Quantum Computation PBS
PBS
(a) a
Kerr b
(b)
PBS
PBS
PBS
PBS
a K
c
H
H
K
K b PBS
PBS
FIGURE 2 Schematic of the implementation of an optical CZ gate between two optical polarization qubits, a and b, using a strong cross-Kerr nonlinearity: (a) using an all optical cross-Kerr nonlinearity and (b) mediated by an atomic two-level system, c. K are effective atomic/optical cross Kerr interactions, H are Hadamard rotations of the atomic two-level system and PBS are polarizing beam splitters
the qubits is sent through the cross-Kerr material. The operation of this device on an arbitrary two qubit input state is given by the following evolution: |ψi → Uˆ K |ψi Ď
ˆĎ ˆ
= eiχ aˆ 2 aˆ 2 b1 b1 (α|01ia |01ib + β|10ia |10ib + γ |10ia |01ib + δ|01ia |10ib ) = α|01ia |01ib + β|10ia |10ib + γ |10ia |01ib + eiχ δ|01ia |10ib .
(10)
Only when the modes passing through the Kerr material are both occupied is a phase shift induced. If we now choose the strength of the nonlinearity such that χ = π, the effect is to flip the sign of one element of the superposition. Thus we directly apply a CZ gate to our dual-rail qubits, completing our universal gate set. A somewhat more complicated version of the gate uses a coherent state as a quantum bus to mediate the gate (Nemoto & Munro, 2004) allowing a CZ gate to be implemented with χ ≈ π/50. The problem with this idea in practice is that typical nonlinear materials have values of χ that are far too small. One might consider making the interaction region of the material very long in order to boost the nonlinearity, but such a strategy generally leads to very high levels of loss, which negate the desired effect. Even if the loss were negligible,
Universal Optical Quantum Gate Sets
221
undesirable phase noise induced by the nonlinearity in the bulk can inhibit the effect (Shapiro, 2006). Non-linearities close to those required can be realized in cavity quantum electro-dynamic (QED) situations featuring single emitters in cavities of extremely high finesse and small volume. This occurs in the so-called strong coupling regime, in which the dipole coupling between the cavity field and the emitter is significantly greater than the relaxation rates of both the cavity and the dipole. These are difficult conditions to achieve, as will be discussed in Section 6.4. Many schemes have been put forward for utilizing cavity QED to realize optical quantum gates. A conceptually simple scheme with several technical advantages was put forward by Duan and Kimble (2004). This scheme utilizes the ability of a single atom in the strong coupling regime to shift a cavity into or out of resonance with an optical field as a function of its internal state. An optical field reflected off a single-ended cavity will acquire a π phase shift if the cavity is on resonance, but will suffer no phase shift if the cavity is off resonance. Suppose the cavity is on resonance for atomic state |gi but is pushed off resonance for atomic state |ei and an optical pulse in a superposition of vacuum and single photon states is reflected from the cavity. An effective cross-Kerr nonlinearity, with strength χ = π is established between the atom and the singlerail photonic qubit – that is, all state components are left the same except for |1i|gi → −|1i|gi. Figure 2(b) shows schematically how this effective Kerr effect between photon mode and cavity atom can lead to a CZ gate between photons via multiple interactions with the cavity. The major challenges with this scheme are realizing strong coupling conditions in a near ideal single-ended cavity configuration and finding parameter ranges in which spectral distortion of the photon modes is minimized. 3.1.2. Two Photon Absorption A quite different nonlinearity that can implement a CZ gate between dualrail photons is two photon absorption, as first suggested by Franson, Jacobs, and Pittman (2004). They proposed using a pair of optical fibres weakly evanescently coupled and doped with two-photon absorbing atoms to implement the gate. As the photons in the two fibre modes couple, the occurence of two photon state components in either of the modes is suppressed by the presence of the two-photon absorbers via the Zeno effect (Misra & Sudarshan, 1977). After a length of fibre corresponding to a complete swap of the two modes a π phase difference is produced between the |11i term and the others. If the fibre modes are then swapped back by simply crossing them, a CZ gate is achieved. This system can be modelled as a succession of n weak beamsplitters followed by 2-photon absorbers, as shown in Figure 3 (Leung & Ralph,
222
Optical Quantum Computation
PBS
PBS
a
2-photon absorbers
b
2-photon absorbers
PBS
PBS
FIGURE 3 Model of an optical CZ gate between two optical polarization qubits, a and b, using 2-photon absorption and the Zeno effect. Implementation is envisaged using evanescently coupled fibres doped with two photon absorbers. PBS are polarizing beam splitters
2006). As n → ∞ the model tends to the continuous coupling limit envisaged for the physical realization. After the PBSs the central pair of modes in Figure 3 are in some combination of vacuum or one-photon states. After the first beam splitter, the four possible photon number state combinations become: |00i → |00i |01i → eiδ (cos θ |01i ± i sin θ |10i) |10i → eiδ (±i sin θ|01i + cos θ|10i) i |11i → ei2δ cos 2θ |11i ± √ sin 2θ (|02i + |20i) . 2
(11)
Assuming ideal two-photon absorbers, i.e. absorbers that completely block the two-photon state components but do not cause any single photon loss, propagation through the first pair of ideal two-photon absorbers gives the mixed state (1)
ρ (1) = Ps(1) |φi(1) hφ|(1) + P f |vacihvac|
(12)
where |φi(1) is the evolved two-mode input state obtained for the case of a no two-photon absorption event and |vaci is the vacuum state obtained in the case when a two-photon absorption event occurs. The individual
Universal Optical Quantum Gate Sets
223
(1)
components of Ps |φi(1) transform as |00i → |00i |01i → eiδ (cos θ |01i ± i sin θ |10i) |10i → eiδ (±i sin θ |01i + cos θ |10i)
(13)
|11i → ei2δ cos 2θ |11i. Equation (13) describes the transformation of each unit, hence repeating the procedure n times gives, |00i → |00i |01i → einδ (cos nθ|01i ± i sin nθ |10i) |10i → einδ (±i sin nθ|01i + cos nθ |10i)
(14)
|11i → ei2nδ (cos 2θ )n |11i describing the transformations giving the evolved input state after n units, π and going to the |φi(n) . By choosing the conditions nθ = π2 , δ = 2n continuous limit n → ∞, the transformation of Equation (14) tends to (1) an ideal CZ gate and P f → 0. The main experimental challenge is to find a medium that exhibits very strong 2-photon absorption with negligible amounts of linear loss.
3.2. Linear Optics Gates Although in-line nonlinear interactions are an efficient way to implement optical quantum computation in principle, we have already noted that there are many difficulties with this approach. If linear optics could be used, many of these difficulties would be reduced. In fact any quantum unitary can be simulated using linear optics and a single photon (Reck, Zeilinger, Bernstein, & Bertani, 1994). However, this is achieved using a unary encoding in which the number of modes required to simulate an n qubit circuit grows as 2n . As a result this scheme has an exponential resource overhead and cannot in general be used for quantum computation.1 The first researchers to show that a scalable linear optics scheme was possible were Knill, Laflamme, and Milburn (2001). We will refer to their scheme as KLM. The standard dual-rail qubit encoding was used, but arbitrary processing was predicted to be possible without 1 A special case is Grover’s algorithm as discussed in Ralph (2006).
224
Optical Quantum Computation PBS c
PBS v1
vo1
p1
η1
η2
1/ 2
1/ 2
η1 t
co
po1
v2
η2 vo2
p2
PBS
to
po2 PBS
FIGURE 4 Schematic representation of a non-deterministic CZ gate. Polarization encoded qubits are injected at c and t. Ancilla photons are injected at p1 and p2. Successful operation is heralded by the detection of no photons at outputs vo1 and vo2 and the detection of one and only one photon at each of outputs po1 and po2. PBS are polarizing beam splitters
in-line nonlinearity or an exponential overhead. Instead, the KLM toolbox comprises: linear optical elements; single photon sources; photoncounting detectors; and electro-optic feed-forward. The introduction of single-photon ancilla and their subsequent measurement leads to measurement induced nonlinearities being applied to the qubits. Another way of viewing this is that KLM trades in-line nonlinearities for off-line nonlinearities in the form of single photon sources and detectors. KLM led to a surge of research activity in dual-rail linear optical schemes. We refer to such schemes in general as linear optical quantum computing (LOQC) (Kok, Munro, Nemoto, Ralph, Dowling, & Milburn, 2007). 3.2.1. KLM We begin by reviewing the original KLM scheme. The KLM scheme can be broken up into three tiers: non-deterministic entangling gates; nondeterministic teleportation gates; and error encoding against teleportation failure. Non-Deterministic Entangling Gates At the first level, KLM introduced two-qubit gates that could take separable, single photon inputs, and produce entangled outputs. In particular, KLM showed how to make a CZ gate that was nondeterministic, but heralded. That is, the gate does not always work, but an independent signal heralds successful operation. A somewhat simplified version of this gate is shown in Figure 4 (Ralph, White, Munro, & Milburn, 2002). In addition to the single-photon polarization qubits incident at ports c (control) and t (target), the gate also has ancilla inputs comprising
Universal Optical Quantum Gate Sets
225
two vacuum input ports, v1 and v2, and two single photon input√ports, p1 and p2.√ The beamsplitter reflectivities are given by η1 = 5 − 3 2 and η2 = (3 − 2)/7. It can be shown that when no photons are detected at outputs vo1 and vo2, and one and only one photon is detected at each of po1 and po2, then the gate has succeeded and the photon qubits exiting through co and to have had the CZ transformation applied to them. The probability of successful operation is η22 ≈ 0.05. Even at this first level the technical requirements are demanding. Four photons must simultaneously enter the circuit. The detectors at po1 and po2 have to distinguish between zero, one or two photons. Any inefficiency in the production or detection of photons will lead to mistakes and rapidly erase the operation of the gate. High-visibility single photon interference and two photon interference (Hong, Ou, & Mandel, 1987) are required simultaneously: as a result excellent mode-matching and photon indistinguishability are essential. Since KLM, several different suggestions for non-deterministic linear optical CZ (or CNOT) gates have been made (Bao, Chen, Zhang, Yang, Zhang, Yang, & Pan, 2007; Hofmann & Takeuchi, 2002; Knill, 2003; Pittman, Fitch, Jacobs, & Franson, 2003; Ralph, Langford, Bell, & White, 2002). There has also been recent work on numerically identifying optimal probabilities of success for LOQC gates (Uskov, Kaplan, Smith, Huver, & Dowling, 2009). Teleportation Gates We now discuss the second tier of the KLM scheme. Although the gates discussed in the previous section give us access to non-trivial two-qubit operations and small scale circuits, they are ultimately not scalable in their own right. A cascaded sequence of such non-deterministic gates would be useless for quantum computation because the probability of many gates working in sequence decreases exponentially. In order to make a scalable system we must move to teleportation gates. The idea that teleportation can be used for universal quantum computation was first proposed by Gottesman and Chuang (1999). Teleportation uses an entangled Bell pair as a resource to transfer a qubit state from one mode to another (Bennett, Brassard, Crepeau, Jozsa, Peres, & Wootters, 1993). To achieve this, a two-mode measurement in the Bell basis (see Section 3) is made between the input state and one of the particles from the Bell pair. The result of the Bell measurement is used to make bit flip and/or phase flip corrections to the other member of the pair which is transformed into the input state. In gate teleportation, the desired gate transformation is performed on the entangled state. The transformed entangled state is then used to teleport the qubit state, which subsequently acquires the transformation (modulo certain commutation requirements between the transformation and the measurement corrections). The main
226
Optical Quantum Computation
c
“1”
or
“0” “1”
“0”
co
|0101> + |1001> + |0110> – |1010> “1” t
“0”
“0” or
to
“1”
FIGURE 5 Schematic representation of optical CZ gate operation via teleportation. Success is heralded by a single photon being detected at each of the two pairs of detectors in one of the patterns shown. Depending on the particular detection patterns, Pauli corrections may be necessary. If zero (two) photons are detected at one of the detector pairs then the corresponding qubit has been measured to be in the zero (one) logical state and the gate has failed. The probability of success of the gate is 25%
point is that the application of the gate need not be deterministic. Nondeterministic gates can be used in a trial-and-error manner to produce the required entangled state, which is then used to teleport the gate onto the qubit(s). The simplest teleportation gate is shown in Figure 5. The heart of the gate is a teleported single-rail CZ gate (see Section 2.2). The entangled resource is the state 1 (|0101i + |0110i + |1001i − |1010i) 2
(15)
which can be interpreted as two single-rail Bell states which have had a CZ gate applied between. Indeed the circuit of Figure 4 could be used to produce this state from separable single photon inputs. Alternatively one can recognize this state as the dual-rail Bell state |0101i + |1010i with a Hadamard gate applied to the second qubit. Such a state can be generated directly (but still non-deterministically) by parametric down conversion (see Section 7.2.2). This latter interpretation is due to Pittman, Jacobs, and Franson (2001). Unfortunately, Bell measurements, as required for the teleportation, can only be carried out non-deterministically with linear optics ¨ (Lutkenhaus, Calsamiglia, & Suominen, 1999). For the simplest scheme (the beamsplitters in Figure 4) these fail 50% of the time, thus the total probability of success of this gate is 25%. KLM showed how to increase this probability of success using more complicated entangled states in the
Universal Optical Quantum Gate Sets
227
teleporter, however only modest improvements are practical due to the rapidly increasing overheads needed to produce these entangled states. Error Encoding Against Teleportation Failure We have seen that teleportation gates can be implemented which have higher probability of success than the first-tier non-deterministic gates. A key feature of the teleportation gates is that failure results in the measurement of the logical values of the qubits. KLM introduced an error correction code to protect against such computational basis measurements (Z-measurements) of the qubits. A logical qubit can be encoded across two physical qubits as (Knill, Laflamme, & Milburn, 2001) |φi(2) = α(|0i|0i + |1i|1i) + β(|0i|1i + |1i|0i).
(16)
This is a parity encoding – the ‘zero’ state is represented by an equal superposition of all the even parity combinations of the two qubits, whilst the ‘one’ state is represented by all the odd parity combinations. Notice that if a Z-measurement is made on either of the physical qubits of the state in Equation (16) and the result ‘0’ is obtained, then the state collapses to an unencoded qubit, however the superposition is preserved. Similarly if the measurement result is ‘1’, a bit-flipped version of the unencoded qubit is the result, but again the superposition is preserved so the qubit can be recovered. This encoding thus enables recovery from teleportation gate failure and so improves the probability of success of the gate by allowing second attempts. However, notice that a two-qubit (and thus non-deterministic) gate is needed to produce the parity encoding. It is not immediately obvious that producing encoded states non-deterministically, which then can be used to improve the performance of more non-deterministic gates, is a winning strategy. KLM showed however, that provided you start with teleporters with a probability of success greater than 50%, this strategy does improve the gates success. For example, a 2/3 teleporter used with the parity encoding leads to a CZ gate success probability of about 58% (as opposed to 44% without encoding). In order to further improve the probability of success, KLM concatenates the two qubit parity code. For example, using the nomenclature of Equation (16), the next level up logical qubit is given by |φi L4 = α(|0i(2) |0i(2) + |1i(2) |1i(2) ) + β(|0i(2) |1i(2) + |1i(2) |0i(2) ).
(17)
High probabilities of success are obtained after a few levels of concatenation, leading to the claim of a scalable system.
228
Optical Quantum Computation
3.2.2. Parity States The KLM result was a major step forward both in opening the door to small-scale demonstrations of optical quantum circuits, and in pointing the way towards a scalable system. However, in its original form the resources required for scale-up were exorbitant. For example, one can estimate that tens of thousands of Bell pairs are needed to implement a single CZ gate with 95% probability of success using the original KLM approach. We now discuss a modification of KLM that massively reduces this overhead. An alternative way to scale up the parity states, introduced by Hayes, Gilchrist, Myers, and Ralph (2004), is not to concatenate the code as per Equation (17), but instead to increase it incrementally. Hence a logical qubit can be encoded across n qubits by representing logical ‘zero’ by all the even parity combinations of the n qubits and logical ‘one’ by all the odd parity combinations. This code retains the feature that, if the logical qubit is encoded across n physical qubits, then a computational basis measurement on any one of the qubits reduces the state to a logical qubit encoded across (n − 1) physical qubits (with the possible need for a bit-flip). Specifically, this parity encoding is given by √ |0i(n) ≡ (|+i⊗n + |−i⊗n )/ 2 √ |1in ≡ (|+i⊗n − |−i⊗n )/ 2,
(18)
√ where |±i = (|0i ± |1i)/ 2. There are two operations which are easily performed on parity encoded states: an arbitrary X rotation, i.e. X θ = cos(θ/2)I + i sin(θ/2)X , which can be performed by applying that operation to any of the physical qubits; and a Z operation, which can be performed by applying Z to all the physical qubits (since the odd-parity states will acquire an overall phase flip). The teleportation gates are reduced to just partial single-rail and dualrail Bell-state measurements. A dual-rail Bell measurement can be used to add n physical qubits to a parity encoded state using a resource of |0i(n+2) . This is referred to as type-II fusion ( f I I ) (Browne & Rudolph, 2005). The result of f I I is f I I |ψi(m) |0i(n+2) →
|ψi(m+n) (success) |ψi(m−1) |0i(n+1) (failure).
(19)
When successful (with probability 1/2), the length of the parity qubit is extended by n. A phase flip correction may be necessary depending on the outcome of the Bell-measurement. If unsuccessful, a physical qubit
Universal Optical Quantum Gate Sets
229
is removed from the parity encoded state, and the resource state (which may be recycled) is left in the state |0i(n+1) . This encoding procedure is equivalent to a gambling game where we either lose one level of encoding, or gain n, depending on the toss of a coin. Clearly, if n ≥ 2 this is a winning game. The required resource states can be built from Bell pairs using a combination of single-rail Bell measurements (type I fusion) and f I I . The remaining gates required to achieve a universal gate set (a Z 90 = I +iZ and a CNOT gate) can be efficiently performed using these fusion techniques (Gilchrist, Hayes, & Ralph, 2007). The resource overhead for performing gates in this way is of the order 100 Bell pairs per gate.
3.3. Coherent State Gates Linear optical protocols are also possible for coherent state qubits. In fact they are arguably the simplest linear optics schemes known. The off-line resources in this case are cat states, i.e. superpositions of distinct coherent states. The greater difficulty involved in producing these states somewhat off-sets the greater simplicity of these schemes. As with KLM, photon resolving measurements and feedfoward are also required. We will refer to schemes of this type as Coherent State Quantum Computing (CSQC). 3.3.1. Coherent State Qubits with Large Amplitudes We will first consider the scheme of Ralph, Munro, and Milburn (2002) in which it is assumed that very large amplitude cat states are available as resources. The qubits are coherent states with amplitudes of zero (i.e. the ground or vacuum state) and α. These qubits are not exactly orthogonal, but the approximation of orthogonality is good for even moderately large 2 α as hα|0i = e−α /2 . In this discussion it is assumed that α 1 so that the qubits have negligible overlap. As we have seen, two-qubit gates represent a formidable challenge in single-photon dual-rail schemes. Surprisingly, for this coherent state encoding, a non-trivial two-qubit gate can be implemented using only a single beamsplitter. Consider the beamsplitter interaction given by the unitary transformation U B S = exp[iθ (abĎ + a Ď b)], where a and b are the annihilation operators corresponding to two coherent state qubits |γ ia and |βib , with γ and β taking values of α or 0. It is well known that the output state produced by such an interaction is U B S |γ ia |βib = | cos θ γ + i sin θβia | cos θβ + i sin θ γ ib
(20)
where cos2 θ (sin2 θ ) is the reflectivity (transmissivity) of the beamsplitter. Now assume that θ is sufficiently small, so that θ 2 α 2 1 but α is sufficiently large, so that θ α 2 is of order one. Physically this corresponds
230
Optical Quantum Computation
to large coherent states impinging on an almost perfectly reflecting beamsplitter. Under these conditions it can be shown that, to high fidelity, U B S |γ ia |βib ≈ exp[2iθ γβ]|γ ia |βib .
(21)
If we further require that θ α 2 = π/2, then this transformation produces a CZ gate. For universal computation we require, in addition to the above maximally-entangling two-qubit gate, the ability to perform arbitrary single-qubit unitaries. Arbitrary unitaries can be constructed given the ability to do arbitrary rotations around the z-axis (Z φ ), bit-flips, plus the Hadamard gate (Nielsen & Chuang, 2000). A bit flip (or X gate) is equivalent to a displacement of −α followed by a π phase shift of the coherent amplitude. The action of the Z rotation gate is Z φ (µ|0i L + ν|1i L ) = µ|0i L +eiφ ν|1i L . It can be implemented, to a good approximation, by imposing a small phase shift on the qubit: Ď
U ()|αi = eia a |αi 2
≈ eiα |αi = Z φ |αi
(22)
with φ = α 2 and, as before, we require 2 α 2 1. In addition to these gates, a Hadamard gate is required in order to achieve an arbitrary qubit rotation. This can be achieved using an √ ancilla cat state, 1/ 2(|0i + |αi), the CZ gate, a diagonal (or cat) basis measurement and a possible X correction. A cat basis measurement can be implemented by first displacing by −α/2. This transforms our ‘0’, ‘α’ superposition into ‘α/2’, ‘−α/2’ superposition. These new states are parity eigenstates. A photon number measurement determines the sign of the superposition but not the sign of the amplitude as required for a diagonal basis measurement. The complete gate set is summarized in Figure 6. Its simplicity is clear, however this comes at a price. Recall that we require the conditions θ 2 α 2 1 whilst θ α 2 = π/2 in order to realize a CZ gate with high fidelity. This in turn implies α π/2. For high fidelity Ralph, Munro, and Milburn (2002) showed that α > 20 is required. Not only does this present great difficulties in terms of producing the required cat states but also places very high technical requirements on the detectors and linear manipulations. 3.3.2. Coherent State Qubits with Small Amplitudes The requirement for large coherent amplitudes in the Ralph, Munro, and Milburn (2002) scheme can be relaxed by introducing gate teleportation in
Universal Optical Quantum Gate Sets π
–α
X
231
D
θ/ α2
Z(θ )
C– S
δ
H 〈 +|, 〈 – |
|+〉
X
FIGURE 6 Schematic summary of the linear optics coherent state universal gate set of Ralph, Munro, and Milburn (2002). D is a displacement and triangles represent phase shifts. The beamsplitter reflectivity is δ = cos2 θ where θ α 2 = π/2. The gates only operate correctly if α π/2
a similar way to the KLM scheme. Gate teleportation can be carried out directly with a high probability of success in the coherent state scheme, thus we end up with a two-tier scheme that is still significantly simpler than the three-tier KLM scheme. This was first shown by Ralph, Gilchrist, Milburn, Munro, and Glancy (2003) where it was still assumed that α ≥ 2 so that any overlap between the coherent state qubits could be neglected. Lund, Ralph, and Haselgrove (2008) further generalized the scheme, so that the gates would work non-deterministically for coherent amplitudes of any size, and with sufficient probability of success to be scaleable provided α > 1.2. We now describe this scheme. These later papers used the more familiar α, −α encoding such that an arbitrary coherent state qubit is represented by the state Nµ,ν (α)(µ|αi + ν| − αi).
(23)
√ As noted before, transformation between this basis and the 0, α 0 = 2α basis requires only a displacement. We now allow for small values of α by
Optical Quantum Computation
232
(a)
(b)
[α]
(c)
[α ]
[α ]
[α ] [ α ] |0〉
|+〉 [ β ]
[α ]
[α ]
FIGURE 7 Schematic summary of (a) unambiguous qubit measurement, (b) unambiguous Bell measurement and (c) teleportation, for coherent state qubits. In (a), one input is an arbitrary qubit and the other is in the zero logical state (both with absolute values of their amplitudes [α]). The arbitrary qubit is measured to be in |αi if counts register at the top detector and | − αi if counts register at the bottom detector. In (b) both inputs are arbitrary qubits. The four Bell states are determined by which detector fires, and whether the number of counts registered is even or odd. In (c) the top input is an arbitrary qubit but the √ lower input is in the logical plus state with the absolute value of its amplitude β = 2α. The arbitrary qubit is teleported to the output, modulo corrections dependent on which detector fires and whether the counts are odd or even. In all three cases the measurement/operation fails if no counts are registered at the detectors
including the exact normalization 2
Nµ,ν (α) = (|µ|2 + |ν|2 + e−2α (µν ∗ + νµ∗ ))−1/2 .
(24)
As with KLM, measurements play a key role. As this scheme does not restrict the size of |α|, the Z-basis and Bell-state measurements must, in general, distinguish as best as possible between nonorthogonal states. This can be achieved using linear optics and photon counting. The computational or Z-basis measurement is shown in Figure 7(a) and the Bell state measurement is shown in Figure 7(b). For the measurement to be unambiguous and error-free, it must have a failure outcome (Ivanovic, 1987). This occurs in both measurements when no photons are detected. The probability of failure tends to zero as |α| increases. The procedure for teleporting coherent state qubits was first described by van Enk and Hirota (2001) and Jeong, Kim, and Lee (2001) and is shown in Figure 7(c). The teleporter uses unambiguous Bell state measurements which have five outcomes. Four outcomes correspond to successfully identifying the respective Bell states, and the remaining one is the failure mode. When the appropriate Pauli corrections are made, the input qubit is successfully transferred to the output. The fifth outcome corresponds to the measurement failure whose probability again decreases to zero as |α| increases. Upon failure, the output of the teleporter is unrelated to the input and hence the qubit is erased. Gate teleportation, as described
Universal Optical Quantum Gate Sets
233
below, can be used to produce a universal set of gates. The ability to unambiguously teleport the qubit value, in spite of the fact that the basis states are nonorthogonal, allows gates to be implemented for all values of |α|. Single Qubit Gates The Phase Rotation Gate (Z θ ) can be achieved using the resource state |Z θ i = exp[iθ ]|α, αi + exp[−iθ ]|α, −αi + exp[iθ ]| − α, αi + exp[−iθ ]| − α, −αi
(25)
whilst the Hadamard gate (H) can be achieved using the resource state |hadi = |α, αi + |α, −αi + | − α, αi − | − α, −αi.
(26)
It is straightforward to show that if a Bell-state measurement is made between an arbitrary qubit state |σ i and the first qubit of |Z θ i (|hadi) then the operation Z θ |σ i (H |σ i) is performed, where, dependent on the outcome of the Bell-measurement, a bit-flip correction, a phase-flip correction, or both may be necessary. If the outcome is zero photons in both arms, the gate fails and the qubit is erased. The resource states |Z θ i and |hadi can be produced non-deterministically from cat state resources, linear optics and photon counting. Controlled-Sign Gate To complete the universal gate set the CZ gate can be achieved using the resource state |C Z i = |α, α, α, αi + |α, α, −α, −αi + | − α, −α, α, αi − | − α, −α, −α, −αi.
(27)
If a Bell-state measurement is made between an arbitrary qubit state |σ i and the first qubit of |C Z i and a Bell-state measurement is made between an arbitrary qubit state |φi and the last qubit of |C Z i then the operation C Z |σ i|φi is performed, where again, dependent on the outcomes of the Bell-measurements, bit-flip and/or phase-flip corrections may be necessary. If the outcome is zero photons in both arms of either of the Bell measurements the gate fails and the respective qubit is erased. The resource state |C Z i can be produced non-deterministically from cat state resources, linear optics and photon counting.
234
Optical Quantum Computation
Correction of Phase-flips After each gate it was noted that bit flip and/or phase flip corrections may be necessary. Bit flips can be implemented easily by simply delaying a qubit with respect to the local oscillator. However phase-flips are more difficult and require further teleportation. However Jeong and Ralph (2007) have pointed out that only active correction of bit-flips is necessary. This CSQC scheme requires only small amplitude cats and remains significantly more resource efficient than dual-rail schemes, needing around 12 cats per gate on average. Of course this assumes that cat states and dual-rail Bell states are of approximately equal difficulty to produce. Progress in producing both these key resource states will be discussed in Section 7.2.
4. CLUSTER STATES So far we have been implicitly discussing quantum computation in terms of the circuit model, familiar from classical computation, in which qubits are prepared in some fiducial state, acted on sequentially to produce logic operations, and then measured in their computational bases to obtain the answer to the computation. Raussendorf and Briegel (2001) have suggested an alternative way of performing quantum computing, distinct from the usual circuit model, called cluster-state quantum computation. It is based on measurement induced quantum evolution and so is sometimes referred to as ‘one-way’ quantum computation. In principle, cluster state computation can be carried out on any physical platform. However, the emphasis in many optical quantum computation architectures on measurement induced nonlinearities and off-line resources turns out to be particularly compatible with the cluster state approach. In Raussendorf and Briegel’s protocol, a large entangled state of a particular form – called a cluster state – is constructed first. Quantum computation is then carried out by making a series of measurements in diagonal (X ), and phase-rotated-diagonal (Z −θ X Z θ ) bases on the cluster state. For example any evolution of a single qubit can be simulated by: (1) preparing a string of qubits all in the |+i state; (2) linking each nearest neighbour by CZ gates (this forms a linear cluster state); and (3) measuring the single qubits in the string in sequence. The measurement basis chosen for each qubit depends on the single qubit unitaries one wishes to simulate and the result of the measurement of the preceding qubit. In particular, a qubit measurement in the basis |R1(θ )i = |0i + e±iθ |1i, |R2(θ )i = −|0i + e±iθ |1i simulates the unitary evolution H Z θ on the adjoining qubit. The ‘+’ (‘−’) sign in the phase factor is chosen if the outcome of the previous measurement was R1 (R2). The last remaining qubit in the chain is the output of the evolution and can be measured in the computational
Cluster States
235
readout
computation
FIGURE 8 Conceptual diagram representing cluster state quantum computing. Each circle represents a qubit, and the lines connecting them represent entanglement in the form of CZ operations. In this 2D square cluster, the computation starts with a measurement on the first column of qubits (black arrow) – each qubit is measured separately (i.e. the measurement is separable). Based on the outcomes, feedforward is used to change the measurement bases for measurements on the next column. Measurement on the final column yields the outcome of the computation
basis. An arbitrary single qubit unitary can be simulated using a fourqubit cluster state and three measurements. By joining linear chains with CZ gates to create 2-dimensional cluster states, two qubit gates can be built into the cluster, enabling universal quantum computation (see Figure 8). The first suggestion that measurement-based quantum computation could help to reduce the resources in an optical system was made by Yoran and Reznik (2003). Subsequently Nielsen (2004) adapted the complete cluster state approach to LOQC. He showed that cluster states could be efficiently built up using the KLM teleportation gates. This follows from the fact that the cluster states are able to recover from computational basis measurements in a similar (but not identical) way to that of the parity states. The application of the fusion techniques we described for parity states in Section 3.2.2, that were in fact initially developed by Browne and Rudolph (2005) for cluster state production, further reduces the resource overhead. In this approach ‘mini-cluster’ states are built up non-deterministically and then fused on to the main cluster in a similar way to that already described for parity states. This is perhaps the most efficient of the linear optical dualrail schemes, requiring approximately 60 Bell pairs per two-qubit gate,
236
Optical Quantum Computation
though the exact meaning of ‘per gate’ in the cluster state paradigm is not as obvious as in circuit models. Work has also been done on cluster state construction methods based on percolation approaches which can dramatically reduce the amount of conditional optical operations required (Kieling, Rudolph, & Eisert, 2007). In principle, optical demonstrations of one-way quantum computation using dual-rail encoding have now been achieved, as will be discussed in Section 6.1.
5. FAULT TOLERANCE The discussions up to this point have only considered errors that occur due to the physics of the fundamental interactions, such as teleportation gate failures. However, in any realistic implementation, there will be additional experimental imperfections in the devices, that may lead to additional problems. Typically, such non-ideal interactions will lead to random errors being introduced. Even if these errors are small, when large scale quantum processing is considered we have to worry about their propagation during gate operations. If uncorrected, such errors would grow uncontrollably and make the computation useless. The answer to this problem is fault tolerant error correction (Shor, 1995; Steane, 1996). In the following, after briefly reviewing the basic principles of error correction, we will look at particular schemes that have been developed and evaluated for several of the optical protocols we have discussed. The idea of error correction is self-explanatory, though the description of its application to quantum systems requires some care. Classically we might consider using a redundancy code such that (for example) 0 → 0, 0, 0 and 1 → 1, 1, 1. If a bit flip occurs on one of the bits we might end up with 0, 1, 0 or 1, 0, 1, but we can recover the original bit value by taking a majority vote. At first it may seem that such a code cannot be used for quantum mechanical systems because: (1) the nocloning theorem (Wootters & Zurek, 1982) means we can not make copies of an unknown qubit and: (2) taking the majority vote is a measurement that will collapse our quantum superposition. It turns out however, that a quantum analog is possible. An example of a quantum redundant encoding is α|0i + β|1i → α|000i + β|111i where we have created an entangled state rather than copies. It is then possible, using two CNOT gates and two ancillas, to identify an error without collapsing the state, by reading out the parity of pairs of qubits. For example a bit-flip error might result in the state α|001i + β|110i. The parity of the first two qubits will be zero whilst the parity of the second two qubits will be one, thus unambiguously identifying that an error has occured on the last qubit. Because we are measuring the parity, not the qubit value, the superposition is not collapsed. Such codes can be expanded to cope
Fault Tolerance
237
with the possibility of more than one error occurring between correction attempts and to cope with multiple types of errors. Of course the CNOT gates being used to detect and correct the errors may themselves be faulty. An error correction code is said to be fault tolerant if error propagation can be prevented even if the components used to do the error correction introduce errors themselves. Typically this is only possible if the error rate per operation is below some level known as the fault tolerant threshold.
5.1. Loss Tolerance Presently the dominant source of errors in optical quantum processing is photon loss – in components, detectors and sources. As a result, initial codes for error correction in optical quantum computing were aimed specifically against loss. KLM estimated a threshold of about 1% for loss tolerance, i.e. fault tolerance where the only error considered is loss. Remaining with the original KLM gate approach, Silva, Roetteler, and Zalka (2005) were able to show that the loss threshold might lie as high as 11%. Using the parity state approach and assuming that sources and detectors each had an equal loss of x%, Ralph, Hayes, and Gilchrist (2005) numerically obtained a loss threshold of x = 17%. A roughly equivalent value was obtained by Varnava, Browne, and Rudolph (2008) for loss tolerance of cluster states. Readers with some knowledge of fault tolerant codes will notice that these threshold values are very high. The reason for this is that all the codes so far mentioned are dual-rail codes for which loss constitutes a locatable error. That is, as dual-rail codes are particle-like codes, and loss destroys particles, so loss acts to remove the qubit from the qubit space. Finding out if all your qubits (photons) are present can be achieved without encoding via a quantum non-demolition measurement (QND) (Gleyzes, Kuhr, Guerlin, Bernu, Deleglise, Hoff, Brune, Raimond, & Haroche, 2007). Thus an error – a photon missing – is locatable without encoding. By contrast, a bit flip leaves the qubit in the qubit space and can only be found by using codes such as the one discussed at the beginning of this section. Such errors are described as being unlocatable. Coding is still required in order to recover from locatable errors, but the easier detection of such errors makes thresholds considerably higher than for unlocatable errors. For single-rail codes, loss tends to produce unlocatable errors. As a result, targeting loss for single-rail systems holds no particular advantage. Unfortunately, the loss-only codes so far described have been shown to be incompatible with more general codes. Specifically, these loss codes tend to amplify unlocatable errors such as bit flips, strongly decreasing the effective thresholds for their correction (Rohde, Ralph, & Munro, 2007) and hence making them impractical in the presence of even very small rates of unlocatable errors. Although it can be shown that there is no
238
Optical Quantum Computation
fundamental reason why codes that optimally correct locatable errors must increase the rate of unlocatable errors (Haselgrove & Rohde, 2008), until now no explicit examples of such codes have been demonstrated. Instead, codes have been developed that simultaneously correct both types of error, though with loss thresholds significantly lower than for the loss-only codes. In the following we will discuss such a code, developed specifically for optical systems.
5.2. Telecorrection Telecorrection is a modified version of the Steane error correction protocol (Steane, 1996) developed by Dawson, Haselgrove, and Nielsen (2006). Although developed in the context of dual-rail optical cluster states (see Section 4), the code is well suited to optical schemes in general and has been applied to several different protocols, enabling a consistent comparison between them. Logical qubits are encoded across seven physical qubits according to the Steane code. A feature of this code is that logical Clifford gate operations can be implemented by applying the desired gate at the physical qubit level on (or between, for two-qubit gates) the individual physical qubits making up the Steane logical qubit(s). The Hadamard, Pauli and CZ gates are examples of Clifford gates. In order to implement a universal set of gates on the logical qubits, at least one non-Clifford gate will be required (e.g. Z θ ). Such logical non-Clifford gates are more complicated to apply, requiring teleportation with a specially prepared resource state. Rounds of error correction are also applied via teleportation, where now the resource state is the telecorrector state. The logical circuit for creating the telecorrector state is depicted inside the dotted box of Figure 9(a) with the physical qubit level circuit for creating the Steane code logical zero states, that are fed into the logical circuit, depicted in Figure 9(b). Notice that there are several places in the preparation procedure for the telecorrector state at which measurements are made. These correspond to the syndrome measurements in the original Steane protocol (Steane, 1996). These measurements are made in the process of creating the telecorrector state, prior to its interaction with the logical qubits. If unwanted measurement outcomes occur, the telecorrector state can be rejected and a new one prepared without affecting the logical qubits. The logical circuit for implementing the teleportation is shown in Figure 9(a). Numerical simulations show (Dawson, Haselgrove, & Nielsen, 2006) that if the physical error rates of the components used to construct the telecorrector state are below a certain threshold, then the teleported logical qubit will have a smaller error rate than the input qubit, thus implementing a round of error correction. The threshold
Fault Tolerance
(a)
239
input |+〉 ⊗n |0〉
|0〉 |+〉 ⊗n
out put |0〉
|0〉
(b)
|+〉 |+〉 |+〉 |+〉 |+〉 |+〉 |+〉 |+〉 |+〉 |+〉 |+〉
H H H H
FIGURE 9 Schematic of (a) telecorrector circuit at the level of Steane code logical qubits and (b) ancilla circuit for fault-tolerantly producing the |0i ancilla state from unencoded qubits
level depends on the relative distribution of errors between locatable and unlocatable, with higher thresholds obtained when the errors are predominately locatable. The trade-off is shown in Figure 10 for a generic system.
5.3. Thresholds and Resource Counts for Optical Schemes Converting the generic error rate thresholds shown in Figure 10 into threshold levels for actual physical parameters requires the creation of physical models for particular gate implementations and their analysis with respect to error rates given particular levels of imperfection. Once such models are in place, it is possible to estimate the number of resources required to create the telecorrector state and hence the resources required for one round of error correction. This process has now been carried out (at least as a function of loss) for four different optical schemes, allowing a direct comparison between them. LOQC Cluster States Dawson, Haselgrove, and Nielsen (2006) applied their telecorrector code to the linear optical cluster state protocol developed by Nielsen (2004) and Browne and Rudolph (2005). They obtained numerical thresholds
240
Optical Quantum Computation
–3
Unlocated error rate, p
1.2
x10
1 0.8 0.6 0.4 0.2 0
0
0.02
0.04
0.06 0.08 0.1 0.12 Located error rate, q
0.14
0.16
0.18
FIGURE 10 Threshold graph for seven-qubit Steane code based telecorrection, plotted as trade-off between locatable and unlocatable errors. Fault tolerant operation is possible for parameters in the region below the solid line. Circles indicate the parameter values for which the simulations were run. Reproduced from Dawson, Haselgrove, and Nielsen (2006)
as a function of the physical loss rates and depolarization rates of their dual-rail photons, and estimated the number of operations and hence the number of photonic Bell pairs required to implement a round of error correction. As well as results specific to optical cluster states, results for generic codes were also presented that could be used to estimate thresholds for more general platforms. Non-linear Zeno Gates Leung and Ralph (2007) used the generic telecorrector code (Dawson, Haselgrove, & Nielsen, 2006) to estimate thresholds for a modified version of the nonlinear Zeno protocol developed by Franson, Jacobs, and Pittman (2004). The thresholds were estimated as a function of the physical loss rates in the nonlinear medium and the detectors and the level of mode matching of the optics. Resources can also be estimated from the number of operations in the generic code. CSQC Lund, Ralph, and Haselgrove (2008) generalized the CSQC protocol developed by Ralph, Gilchrist, Milburn, Munro, and Glancy (2003) and applied the telecorrector code. They obtained numerical thresholds as a function of the physical loss rates of the source, detectors and memory. They obtained the number of operations required and hence the average number of cat state resources consumed in one round of error correction.
Fault Tolerance
241
106 Cluster (2006) resources (bell pairs per operation)
105 104 103 102
Zeno (2007)
Parity (2008)
Coherent (2008)
10
10–4
10–3
10–2
threshold for loss
FIGURE 11 Resources versus threshold for various optical implementations of the telecorrector-Steane code. The code is fully fault tolerant and thus can correct all noise types however, for the purpose of this graph, it is assumed that loss is the dominant noise type and the threshold with respect to it is plotted. Resources are the number of Bell pairs (cat states for the coherent scheme) needed per physical operation when running the fault tolerant code (typically one round of 1st level error correction requires 1000 operations). The different points are based on: Cluster (Dawson, Haselgrove, & Nielsen, 2006); Zeno (Leung & Ralph, 2007); Coherent (Lund, Ralph, & Haselgrove, 2008); and Parity (Hayes, Haselgrove, Gilchrist, & Ralph, 2009)
LOQC Parity States Hayes, Haselgrove, Gilchrist, and Ralph (2009) applied the telecorrector code to the LOQC parity state protocol (Gilchrist, Hayes, & Ralph, 2007; Hayes, Gilchrist, Myers, & Ralph, 2004). They obtained numerical thresholds as a function of the physical loss rates and depolarization rates of their dual-rail photons. They obtained the number of operations required and hence the average number of Bell state resources consumed in one round of error correction. In Figure 11 we show a comparison of these results, showing resources versus threshold (assuming the dominant error source is loss) for the four protcols. An interesting trade-off between threshold and resources emerges from the plot. Choosing the best architecture then depends on whether the technological bottlenecks lying ahead favour higher thresholds or lower resource requirements. Based on the present analysis the approaches offering the best compromise for medium scale quantum computing (100s of logical operations) appear to be the CSQC scheme and the Parity State scheme. For large-scale quantum computing, the overheads for all implementations, optical or otherwise, still remain very high.
242
Optical Quantum Computation
6. EXPERIMENTAL DEMONSTRATIONS There have been a wide range of experimental demonstrations of optical quantum logic, applications and resource generation. The majority of optical experiments have used linear optics methods, with either the KLM approach or the use of exisiting entanglement (e.g. cluster states) to add the required nonlinearity. These linear optical schemes can be divided into those where the qubits are encoded in particles, and those with a field encoding.
6.1. Linear Optics, Particle Encodings In the particle encoding, a qubit is typically encoded onto the state of a single photon. This is a dual-rail qubit – the logical state corresponds to the occupation of one or another mode by the photon. Although path, frequency and other encodings have been used, the most common approach is to represent the logical states by orthogonal polarization states of the photon’s electric field. Polarization is especially practical because logical modes are both degenerate and co-propagating. This means that most sources of phase noise in an optical circuit (e.g. vibrations of components) are common-mode, and the qubit is robust against dephasing. It has been noted that polarization is preserved over long distances in space—light from the Crab Nebula, some 6500 light years away, is still significantly polarized in some regions of the spectrum (Oort & Walraven, 1956). In dielectric media such as glass optical fibres, the polarization state is subject to changes in birefringence (or refractive index and dispersion for other encodings). As the interactions with these media are essentially non-resonant at optical frequencies, changes due to locally fluctuating electric and magnetic fields are small and the predominant effect is from thermal changes in the material, something that happens on a time scale much longer than the time taken for the qubits to propagate through the device. Therefore, even in dielectric media, phase noise on optical (and especially common-mode polarization) qubits are generally negligible for a single shot. Additionally, linear optical transformations on a single qubit are generally easy and can be performed with high precision (Peters, Altepeter, Jeffrey, Branning, & Kwiat, 2003). Polarization transformations, for example, can be achieved with fidelities of >99%, using waveplates. Transformations in path encodings may be achieved using a combination of beam splitters and phase shifters, and in some cases using the extra polarization degree of freedom can assist in implementing operations with high fidelity. For example, high visibility interferences of over 99% can be achieved in polarizing interferometers. The standard polarization encoding requires, not surprisingly, the ability to make single photon states of definite polarization. Since the late
Experimental Demonstrations
243
1980s, the solution of choice has been parametric down conversion in a χ (2) medium (Ghosh & Mandel, 1987). Weak degenerate parametric down conversion results in the spontaneous conversion of single pump photons at the harmonic frequency into pairs of photons at the fundamental. If ¨ the down conversion is spatially non-degenerate then, in the Schrodinger picture, initial vacuum inputs are transformed according to |0ia |0ib → (|0ia |0ib + χ 0 |1ia |1ib + χ 02 |2ia |2ib + . . .)
(28)
where χ 0 is an effective non-linear interaction strength, proportional to the pump power. If we now allow χ 0 to be very small (which is not hard to arrange experimentally) then the state produced is given, to an excellent approximation, by |ψiab = |0ia |0ib + χ 0 |1ia |1ib .
(29)
If we postselect only those events from the detection record in which two photons are detected ‘simultaneously’, or in coincidence (within some preset time window), then we will only record the part of the state which is due to the pairs of photons. Thus by using the combination of parametric down-conversion, the polarization degree of freedom and postselection, we can perform two-qubit experiments. Experiments carried out this way are sometimes referred to as coincidence experiments and we will discuss various examples in later sections. However, note that this source is still spontaneous, i.e. successful events are rare, random and we do not know if they have occurred until after the fact. Although three- and four-qubit experiments have been achieved by a simple generalization of the techniques just outlined, the cost is an exponential drop in the probability of success. Therefore, experiments carried out in coincidence can demonstrate the basic physics of particular systems, and prove the principles behind optical quantum information techniques, but are not intrinsically scalable to large-scale quantum information processing. Progress in producing sources without this drawback is discussed in the next section. The strength of linear optics in implementing single qubit gates is balanced by the main challenge of linear optical quantum computing, which is implementing two-qubit entangling gates. The key to implementing these with linear optics is the measurement-induced nonlinearity of the KLM and related schemes, or the pre-existing entanglement of the cluster states schemes.2 We will now consider some 2 Measurement-induced nonlinearity is required in these schemes too, for building the cluster initially.
244
Optical Quantum Computation
η
|11〉
C
(2 η –1)|11〉
FIGURE 12 Partial Hong–Ou–Mandel interference between two single photons on a beam splitter of reflectivity η. The formula shows the state transformation in the |11i subspace, i.e. conditional on getting one photon at each of the outputs. The probability of this event occurring decreases as η → 21
of the experimental demonstrations of two-qubit gates using linear optics in a photonic encoding, as well as some of the small-scale applications of these gates. Ralph, Langford, Bell, and White (2002), and independently Hofmann and Takeuchi (2002), proposed a simplification of the KLM scheme in which simple two-photon gates could be implemented nondeterministically with two photonic qubits and no ancillas. This scheme provided for a powerful proof-of-principle of measurement-induced nonlinearity gates without the need for complicated architectural overheads. Although, being coincidence gates they are not immediately scalable to large systems, nevertheless in many cases it is possible to chain logic gates together – prior to photon detection – making this scheme suitable for small circuit demonstrations. The basic operation of the scheme is as follows: a partial Hong–Ou–Mandel interference (Hong, Ou, & Mandel, 1987), on a beam splitter of reflectivity η can be configured such that the output state is a superposition of three possibilities. One of these possibilities is a phase flip of the |11i number state (Figure 12). The remaining two possibilities correspond to final states with two photons in one mode – these lie outside the qubit Hilbert space we are interested in. If the number state |11i also represents the logical |11i state, then the device can implement the nonlinearity required for an entangling two-qubit gate. Specifically, a beam splitter with an intensity reflectivity of η = 1/3 acts upon a state with one photon in each q mode (see Figure 12) to produce the logical state transformation |11i → − 13 |11i, conditional on the output containing one photon per mode (Ralph, Langford, Bell, & White, 2002). Embedding this effect in a larger polarization interferometer (Figure 13) produces the full
Experimental Demonstrations
245
0
Control in
Control out
1 1 Target in
Target out
0
η = 1/3 BS
PBS
HOM interference
FIGURE 13 Conceptual diagram of a partial Hong–Ou–Mandel interference embedded in a polarizing interferometer. The phase flip only occurs on the logical |11i term, leading to the CZ action of the circuit, conditional upon getting one photon in each of the control and target inputs. The beam splitters in the top and bottom arms balance the probability amplitudes for all logical basis states
controlled-SIGN (CZ) logic r
1 |H H i 3 r 1 |H V i → |H V i 3 r 1 |V H i → |V H i 3 r 1 |V V i → − |V V i 3 |H H i →
where only outputs in the two-qubit Hilbert space are considered – this is the effect of the measurement-induced nonlinearity. (We have used the notation |H (V )i ≡ |0(1)i for a horizontally (vertically) polarized photon). The cost of this truncation of Hilbert space is a non-unit probability of operation; P = 1/9 for this particular scheme. An optical circuit implementing this scheme was constructed using passively stable classical interferometers to ensure a robust experimental design (O’Brien, Pryde, White, Ralph, & Branning, 2003), and achieved a high gate fidelity (O’Brien, Pryde, Gilchrist, James, Langford, Ralph, & White, 2004). It is evident that the key element of this scheme is high-quality nonclassical and classical interference between modes of the optical circuit. Using SPDC sources and free space optics, nonclassical
246
Optical Quantum Computation
interferences of ≈96% (relative to the maximum achievable for the given beam splitter reflectivity) are typically achievable, with classical interferences of >98% common. In the original demonstrations (O’Brien, Pryde, White, Ralph, & Branning, 2003; O’Brien, Pryde, Gilchrist, James, Langford, Ralph, & White, 2004), slightly lower interference visibilities led to gate fidelities of 0.89–0.95. Subsequent improvements in circuit design (described below) have led to entangling gate fidelities of up to 0.98 (Lanyon, Barbieri, Almeida, Jennewein, Ralph, Resch, Pryde, O’Brien, Gilchrist, & White, 2009). How are these gate fidelities determined? The technique of quantum process tomography was developed in order to characterize quantum processes such as quantum gates (Chuang & Nielsen, 1997; Poyatos, Cirac, & Zoller, 1997; White, Gilchrist, Pryde, O’Brien, Bremner, & Langford, 2007). The basic idea of process tomography is to prepare, one at a time, a range of input states that span the space of allowed density matrices, and to perform measurements spanning the measurement space on identical copies of each input state. In so doing, information is obtained about how all possible input states map onto all possible output states. Full tomography provides a complete description of the process in the computational Hilbert space, so not only is it useful for determining gate fidelities but it is also useful for determining the errors that are introduced by imperfect gate realizations. However, while the information obtained from tomography provides a mathematical description of the errors in the process, parallel modelling is generally required to obtain a physical insight into the errors in the quantum circuit (Rohde, Pryde, O’Brien, & Ralph, 2005). The fidelity of these linear optics gates has been improved by simplifications of the optical circuits and by better control of the optical modes. For example, the classical interferences that are required in a circuit like that in Figure 13 can be removed by using partiallypolarizing beam splitters (Kiesel, Schmid, Weber, Ursin, & Weinfurter, 2005; Langford, Weinhold, Prevedel, Resch, Gilchrist, O’Brien, Pryde, & White, 2005; Okamoto, Hofmann, Takeuchi, & Sasaki, 2005). In this case, the polarization modes need not be separated for the nonclassical interference to work, because the beam splitter that enables the HOM works on only one polarization state. Not only does this allow the classical interference visibility to approach unity (relative to the maximum for the beam splitter reflectivity), it also significantly simplifies the optical setup. The tradeoff is that the HOM condition is now set by the physical reflectivity of the beam splitter, rather than being tuneable by a wave plate setting as it is in polarizing interferometers – it is therefore susceptible to manufacturing errors. However, it has been noted that the gate fidelity is relatively insensitive to imperfections in the BS reflectivity (Ralph,
Experimental Demonstrations
247
Langford, Bell, & White, 2002). Additionally, the use of guided-mode optics at the input and output of the gates has improved nonclassical mode matching, a theme that will be explored further in Section 7.4 below. Internal ancilla gates, where additional photons are used in the gate operation itself (more in line with the original KLM suggestion) have also been demonstrated. The most popular of these is the design of Pittman, Jacobs, and Franson (2001) (also Franson, Donegan, Fitch, Jacobs, and Pittman (2002)), which is based on the teleportation-based quantum computing scheme of Gottesman and Chuang (1999) (see Section 3.2.1), and which uses a maximally-entangled photon pair as the ancilla resource. By harnessing the pre-exisiting entanglement in the ancilla, the gate is able to achieve a higher success probability of 1/4, with the trade-off of requiring both single photons and an entangled pair as inputs. In priniciple, the internal ancillas allow the gate to operate in a heralded fashion without detection of photons in the output modes, by the registration of particular photon number detections in the ancilla output ports. In practice, this requires high-efficiency number-resolving photon counters (see Section 7.1), which are not yet available. A simplified version of this gate, with the entangled ancilla replaced by a single photon, was demonstrated by Pittman, Fitch, Jacobs, and Franson (2003). Versions of the gate with entangled ancillas were demonstrated by Gasparoni, Pan, Walther, Rudolph, and Zeilinger (2004) and Zhao, Zhang, Chen, Zhang, Du, Yang, and Pan (2005). Although process tomography was not used to quantify the performance of these gates, they demonstrated logical basis fidelities of ∼80%. Later, a version of the gate was demonstrated that replaced the entangled ancilla with a nondeterministic Bell measurement (Bao, Chen, Zhang, Yang, Zhang, Yang, & Pan, 2007) – this demonstration used the method of Hofmann (2005) to bound the gate fidelity in the range 0.78–0.88. As noted in Section 3.2, one of the challenges of the KLM approach that applies to all of these gates is the unfavourable overheads in implementing deterministic gates from non-deterministic ones, even though the scheme is computationally ‘efficient’. Several approaches have been described to help overcome this problem, amongst them the cluster state and parity state methods. In the cluster state method, a highly entangled state of many qubits is built up as a resource. Single qubit measurements on the state, coupled with feedforward operations onto the remaining qubits, suffice to perform universal quantum computing (see Section 4). Demonstrations of the cluster state method have been performed for cluster states of up to six photons (Chen, Li, Qiang, Chen, Goebel, Chen, Mair, & Pan, 2007; Kiesel, Schmid, Weber, Toth, Guhne, Ursin, & Weinfurter, 2005; Lu, Zhou, Guehne, Gao, Zhang, Yuan, Goebel, Yang, & Pan, 2007; Prevedel, Tame,
248
Optical Quantum Computation
Stefanov, Paternostro, Kim, & Zeilinger, 2007; Tokunaga, Kuwashiro, Yamamoto, Koashi, & Imoto, 2008; Walther, Resch, Rudolph, Schenck, Weinfurter, Vedral, Aspelmeyer, & Zeilinger, 2005), and also with cluster states of multiple qubits using more than one qubit per photon (by using multiple degrees of freedom on a single photon, for example – Vallone, Pomarico, Mataloni, De Martini, and Berardi (2007)). Although this latter approach is not scalable, it may provide a platform for small-scale tests of information protocols beyond what can be achieved with single-qubit-per-photon encodings alone. As well as demonstrating the measurement operations on cluster states, feedforward operations have also been demonstrated (Prevedel, Walther, Tiefenbacher, Bohl, Kaltenbaek, Jennewein, & Zeilinger, 2007; Vallone, Pomarico, De Martini, & Mataloni, 2008), with fidelities of >95% and feedforward operation times of order 150 ns – mostly limited by the speed of the detection electronics and the switching electronics, which has to switch the ∼kV supply for Pockels cells. In future generations, one imagines using fibre or waveguide electro-optic modulators which can have sub-nanosecond switching times. Experimentally demonstrating that the measurement and feedforward operations actually work is an important proof-of-principle verification of the scheme. Equally important is demonstrating the manufacture of cluster states. The most efficient scheme presently known is that of Browne and Rudolph (2005), which uses polarizing beam splitters and photon detection to forge links between entangled states of photons. Once these links have been established, the resulting state is larger than any of the input states. Present demonstrations of cluster state generation have not used this scheme in its fullest sense, instead relying on postselection of certain measurement results in quantum interference between entangled pairs (e.g. Kiesel, Schmid, Weber, Toth, Guhne, Ursin, and Weinfurter, 2005; Lu, Zhou, Guehne, Gao, Zhang, Yuan, Goebel, Yang, and Pan, 2007; Walther, Resch, Rudolph, Schenck, Weinfurter, Vedral, Aspelmeyer, and Zeilinger, 2005). Because both techniques rely on the same interference effect, the postselected technique is a good representation of the full model for cluster state generation. The standard cluster state growth technique requires the use of high-efficiency sources and detectors, which are presently still under development – see Sections 7.1 and 7.2.2 below. Yet another class of linear optics gates uses pre-existing entanglement in a second degree of freedom to power the two-qubit operation. An example of this is the gate of Sanaka, Kawahara, and Kuga (2002), in which pre-existing time-energy entanglement from a spontaneous parametric down conversion source may be converted into polarization entanglement using an optical interferometer. By controlling the initial polarization states of the photons, a controlled-NOT operation between
Experimental Demonstrations
249
the polarization qubits was performed. This scheme is expandable, in principle, to multi-qubit gates (Gong, Guo, & Ralph, 2008) or small scale circuits by generating multi-qubit entanglement in the other degree of freedom.
6.2. Circuits and Protocols Using Linear Optics Gates A variety of gates and protocols have been demonstrated using the various linear optics techniques discussed so far. These include: realizations of two-qubit entangling gates such as the CNOT (Gasparoni, Pan, Walther, Rudolph, & Zeilinger, 2004; O’Brien, Pryde, White, Ralph, & Branning, 2003; Pittman, Fitch, Jacobs, & Franson, 2003; Zhao, Zhang, Chen, Zhang, Du, Yang, & Pan, 2005) or CZ (Kiesel, Schmid, Weber, Ursin, & Weinfurter, 2005; Langford, Weinhold, Prevedel, Resch, Gilchrist, O’Brien, Pryde, & White, 2005; Okamoto, Hofmann, Takeuchi, & Sasaki, 2005); basic error encoding and detection (Lu, Gao, Zhang, Zhou, Yang, & Pan, 2008; O’Brien, Pryde, White, & Ralph, 2005; Pittman, Jacobs, & Franson, 2005a; Prevedel, Tame, Stefanov, Paternostro, Kim, & Zeilinger, 2007); realizations of simple quantum algorithms such as Deutsch’s algorithm (Tame, Prevedel, Paternostro, Boehi, Kim, & Zeilinger, 2007); Grover’s algorithm for small numbers of qubits (Chen, Li, Qiang, Chen, Goebel, Chen, Mair, & Pan, 2007; Kwiat, Mitchell, Schwindt, & White, 2000; Walther, Resch, Rudolph, Schenck, Weinfurter, Vedral, Aspelmeyer, & Zeilinger, 2005); a compiled version of Shor’s algorithm for factoring 15 into its prime factors (Lanyon, Weinhold, Langford, Barbieri, James, Gilchrist, & White, 2007; Lu, Browne, Yang, & Pan, 2007); a demonstration of the quantum phase estimation algorithm (QPEA) (Lanyon, Whitfield, Gillett, Goggin, Almeida, Kassal, Biamonte, Mohseni, Powell, Barbieri, Aspuru-Guzik, & White, in press); using polarization qubits, for applications in quantum chemistry (Aspuru-Guzik, Dutoi, Love, & Head-Gordon, 2005);3 simulations of anyonic statistics and braiding (Lu, Gao, Guehne, Zhou, Chen, & Pan, 2009; Pachos, Wieczorek, Schmid, Kiesel, Pohlner, & Weinfurter, 2007); investigations of quantum computing operations with no entanglement (Lanyon, Barbieri, Almeida, & White, 2008); quantum games (Prevedel, Stefanov, Walther, & Zeilinger, 2007); counterfactual quantum computation (Hosten, Rakher, Barreiro, Peters, & Kwiat, 2006); demonstration of more complex logic gates (such as the three-qubit Toffoli gate and the controlled-arbitrary-unitary gate – Lanyon, Barbieri, Almeida, Jennewein, Ralph, Resch, Pryde, O’Brien, Gilchrist, and White (2009)); and more.
3 The QPEA was also modified to perform photonic estimation of a classical optical phase (Higgins, Berry, Bartlett, Wiseman, & Pryde, 2007).
250
Optical Quantum Computation
Additionally, linear optical CNOT gates and cluster states have enabled experimental investigations of fundamental quantum physics (Pryde, O’Brien, White, Ralph, & Wiseman, 2005; Pryde, O’Brien, White, & Bartlett, 2005; Walther, Aspelmeyer, Resch, & Zeilinger, 2005) and new quantum measurements (including quantum nondemolition measurements and arbitrary strength measurements on flying qubits (Pryde, O’Brien, White, Bartlett, & Ralph, 2004; Ralph, Bartlett, O’Brien, Pryde, & Wiseman, 2006), and entangling measurements (Langford, Weinhold, Prevedel, Resch, Gilchrist, O’Brien, Pryde, & White, 2005; Walther & Zeilinger, 2005)). These results have demonstrated that optics is a suitable technology for performing quantum information processing applications. The outstanding challenge is to conquer the various challenges to scalability both in terms of resource usage to overcome non-determinism, and obtaining fault tolerance. The current line of experimental research is to take a two-pronged approach: increasing the number of qubits and the sophistication of the optical circuits and algorithms and: improving the performance of auxiliary components (see Section 7). As the performance of these components improves, their incorporation into LOQC circuits will increase.
6.3. Linear Optics, Field Encodings Our discussion so far has only considered photonic LOQC. Although not yet as advanced, development has also been progressing in the area of linear optical quantum computing using field encodings. In particular, there has been significant emphasis on coherent state quantum computing, largely focussing on the generation of the CSS (coherent state superposition) resource states. Several groups have approximate CSS resources (see Section 7.2.3), based on subtraction of a photon from a squeezed vacuum state to generate a close approximation of an ‘odd cat’ – a CSS with only odd photon number terms in the number-state expansion. These types of CSS have been demonstrated for both continuous states of light (NeergaardNielsen, Nielsen, Hettich, Molmer, & Polzik, 2006; Wakui, Takahashi, Furusawa, & Sasaki, 2007) and for fields in single short (typically ∼ picosecond) pulses (Ourjoumtsev, Tualle-Brouri, Laurat, & Grangier, 2006). Using these approaches, fidelities of ∼70% (and upwards) with the ideal cat state have been achieved for states with an amplitude of ∼1. Larger, squeezed cat states, with effective amplitudes of ∼1.6 have been created using conditional homodyne detection of a two-photon state (Ourjoumtsev, Jeong, Tualle-Brouri, & Grangier, 2007). Other interesting states such as ‘even cats’ have also been demonstrated (Takahashi,
Experimental Demonstrations
251
Wakui, Suzuki, Takeoka, Hayasaka, Furusawa, & Sasaki, 2008). Although actual gate operations employing these resource states have not yet been performed, preparation of necessary entangled states, such as the Bell-cat state and a state operationally equivalent to |Z θ i (see Section 3.3.2) has been demonstrated (Ourjoumtsev, Ferreyrol, Tualle-Brouri, & Grangier, 2009). Cluster-state quantum computing has also been explored in the continuous-variables regime, with preliminary experiments based on use of QND-like beam splitter circuits for building cluster states (Yoshikawa, Miwa, Huck, Andersen, van Loock, & Furusawa, 2008; Yukawa, Ukai, van Loock, & Furusawa, 2008), as well as plans to build frequency-comb cluster states using strongly squeezed light in multiple frequency modes (Menicucci, Flammia, & Pfister, 2008; Pysher, Bloomer, Pfister, Kaleva, Roberts, & Battle, 2008). This latter field is in its early days. Although perhaps not the presently preferred scheme for optical quantum computing, the ability to make single-rail qubits in the |photon numberi ∈ {|0i, |1i} subspace has also been demonstrated. Again, the generation of these field-encoded qubits progresses using a combination of techniques from continuous and discrete quantum optics – squeezing and photon counting (Babichev, Brezger, & Lvovsky, 2004; Lvovsky & Mlynek, 2002). However, the full range of quantum operations has not yet been demonstrated with these states.
6.4. Nonlinear Optical Quantum Computing Experiments By the early twenty-first century, there have been very few demonstrations of optical quantum computing elements based on nonlinear interactions. Generally speaking, the approach employed for the existing experiments has been either to amplify the nonlinearity associated with a single atom (e.g. using a cavity), or to employ some weak nonlinearity amplified by the use of many atoms. Perhaps the best-known experiment in the former category is that of Turchette, Hood, Lange, Mabuchi, and Kimble (1995), who used an atom in a cavity to demonstrate a conditional birefringent phase shift of ∼ 16◦ per intracavity photon, with a weak coherent state probe. This experiment was a highly impressive feat of experimental optical quantum science, but it has proved challenging to increase the phase shift to larger values due to practical issues. Recently, the group of Vu˘ckovi´c has demonstrated similar cavity QED experiments using the compact configuration of quantum dots in photonic crystal cavities (Englund, Fushman, Faraon, & Vu˘ckovi´c, 2009). Using this technique, phase shifts of up to 0.16π radians have been observed (Fushman, Englund, Faraon, Stoltz, Petroff, & Vu˘ckovi´c, 2008) at the single-photon level, also using a weak coherent beam. A host of other experiments have been performed addressing the nonlinearity available
252
Optical Quantum Computation
in the strong coupling regime. Some of these include: the observation of the photon blockade effect (Birnbaum, Boca, Miller, Boozer, Northup, & Kimble, 2005); strong coupling with microtoroid cavities (Aoki, Dayan, Wilcut, Bowen, Parkins, Kippenberg, Vahala, & Kimble, 2006); two-photon dressed states (Schuster, Kubanek, Fuhrmanek, Puppe, Pinkse, Murr, & Rempe, 2008); and strong coupling at microwave frequencies (Brune, Hagley, Dreyer, Maitre, Maali, Wunderlich, Raimondand, & Haroche, 1996; Gleyzes, Kuhr, Guerlin, Bernu, Deleglise, Hoff, Brune, Raimond, & Haroche, 2007). Another proposed technique for generating large nonlinearities is electromagnetically induced transparency, where the application of a pump field can be used to cause a phase shift on another optical field (Schmidt & Imamoglu, 1996). Although such nonlinear phase shifts have not been demonstrated with single photons, promising lowfield demonstrations have been made (Kang & Zhu, 2003). Even weak nonlinearities may be able to help realize quantum computer gates (Nemoto & Munro, 2004). Recently, other nonlinear ideas have been explored. Franson et al. have shown theoretically that it is possible to use the two-photon absorption in a suitable atomic ensemble to provide a weak, Zeno-like measurement that suppresses the known failure mode of photonic linear optics approaches – moving out of the computational code space and generating states with two photons in the same mode (see Section 3.1.2). Although logic has not yet been demonstrated using this scheme, several indicative experiments have been performed, including low-intensity two-photon absorption experiments using rubidium vapour and guidedmode optics (Hendrickson, Pittman, & Franson, 2009).
7. AUXILIARY COMPONENTS Demonstrations of optical quantum logic, and small algorithms, can be achieved with the technology that is currently available. The ability to scale to larger problems will require efficient production and detection of single photon (or other highly quantum) states – a major challenge. In this section, we discuss the progress towards high-efficiency sources and detectors, as well as some of the other technologies which will likely play a part in larger-scale optical quantum computing implementations.4 Figure 14 provides a graphical overview of how some of the different components may fit together to form part of a large-scale optical quantum computer. 4 There is a large and growing number of experimental results in this category, and it is not possible to provide exhaustive references – we have chosen a representative sample.
Auxiliary Components
253
high-fidelity detectors
photon count
homodyne
measurement
input state |ψ〉
high-fidelity sources
linear optical circuit
feedforward
quantum memory
output state |φ 〉
ancilla
FIGURE 14 Components of a measurement-induced-nonlinearity-based optical quantum computer. High fidelity sources of single photons, Bell pairs or CSS states are required for input states and ancillas, and high-fidelity detectors are required for state conditioning and final detection. A quantum memory allows states to be stored while other operations are occurring, and feedforward is used to update quantum states based on the results of previous detections. Photons interact interferometrically in a linear optics circuit. The architecture shown is conceptual, and a full-scale quantum computer would involve many repetitions of blocks such as these
7.1. Detectors The silicon avalanche photodiodes that serve as detectors in most present-day experiments have a quantum efficiency of about 50–75%, depending on wavelength and the manufacture of the individual device. Furthermore, they act as ‘click’ or ‘no-click detectors’, meaning that a positive signal indicates the detection of at least one photon, but with no ability to assign photon numbers from a click. Although the exact detector requirements vary from scheme to scheme, OQC protocols generally require that the efficiency be close to unity (see Section 5 for a discussion of just how close), and with at least some number-resolving capability (e.g. the ability to distinguish 0, 1 and 2 photon events from one another). Other desirable properties include high speed operation (in order to detect many photons per second) and low dark counts (in order to distinguish real events from thermal noise in the detector). Although the commonly used detectors do not perform especially favourably on the first two criteria, there are new detectors that are beginning to do so. In terms of quantum efficiency, the best developed single photon counting detector is the tungsten transition-edge superconducting (TES) sensor of Lita, Miller, and Nam (2008). It has reached a measured quantum efficiency of ∼95%, and it also possesses the capability to resolve photon numbers up to 7 per pulse. TES devices work by using a small, thermally
254
Optical Quantum Computation
isolated region of superconducting material that is temperature-stabilized such that the conduction is right at the boundary of superconducting and normal behaviour – a region of high thermal sensitivity. When photons are absorbed by the metal, the heat changes its resistance in quantized steps, allowing for the photon number resolving capability. Because the device is bolomteric, the quantum efficiency is very high as long as the photon is absorbed and gives up its heat. Other high-efficiency, number resolving devices are also under development. Examples include the visible-light photon counter (Kim, Takeuchi, Yamamoto, & Hogue, 1999; Takeuchi, Kim, Yamamoto, & Hogue, 1999), which has been demonstrated to have efficiencies of >90%, as well as photon-number-resolving capability. Additionally, a wide variety of other technologies are under development, including other superconductor schemes (e.g. Miki, Fujiwara, Sasaki, Baek, Miller, Hadfield, Nam, and Wang, 2008; Rosfjord, Yang, Dauler, Kerman, Anant, Voronov, Gol’tsman, and Berggren, 2006), single-photon upconversion schemes (Albota & Wong, 2004; Roussev, Langrock, Kurz, & Fejer, 2004; Thew, Zbinden, & Gisin, 2008; VanDevender & Kwiat, 2007), and clever uses of existing technologies with new back-end electronics (Fujiwara & Sasaki, 2006; Kardynal, Yuan, & Shields, 2008). Each of these schemes address one or both of the issues of detector quantum efficiency and number-resolving capability. Continued development is required to achieve high efficiency, number resolving capability, low dark counts and high speed, all in one device, but progress in detector technology is highly encouraging. The detectors just described are not yet commercially available, and in general are restricted to a few laboratories and projects. In parallel, a number of ideas are being implemented to increase the flexibility of existing avalanche photodiodes. Number resolving detectors can be simulated by multiplexing in time (Achilles, Silberhorn, Sliwa, Banaszek, Walmsley, Fitch, Jacobs, Pittman, & Franson, 2004) or space. Photons are split into different time bins or spatial modes using beamsplitters, and although this procedure is non-deterministic, the probability of having more than one photon per channel is low if the number of channels significantly exceeds the number of photons to be discriminated. Such detection schemes are usually employed with stardard single photon counting modules (SPCMs) based on avalanche photodiodes – devices that do not have a high intrinsic detection efficiency. If the singleunit quantum efficiency is η, then the chance of detecting N photons (assuming we are in the limit where there is no more than one per channel) in a multichannel device is ∼ η N . With η = 0.5 and N = 6, this overall efficiency is less than 2%. Nevertheless, such schemes make possible experiments that may otherwise be out of reach, and in
Auxiliary Components
255
some cases, the use of tomographic detector characterization schemes (Achilles, Silberhorn, & Walmsley, 2006; Lundeen, Feito, ColdenstrodtRonge, Pregnell, Silberhorn, Ralph, Eisert, Plenio, & Walmsley, 2009) may enable initial demonstrations even with significantly imperfect devices.
7.2. Sources High-efficiency, high-fidelity sources are another key requirement of building an optical quantum computer. Depending on the protocol, this may be an array of single photon sources, Bell-pair sources or CSS sources. In each case, the desired performance is that the source produces exactly the required state every time. Of course, a slight amount of imperfection should be tolerable, and it is an ongoing problem in the theory of error correction for optical schemes to determine how high these tolerances can be (see Section 5). For this subsection, we will simply express the goals in terms of their absolute ideal values. 7.2.1. Single Photon Sources The original KLM proposal for LOQC harnessed the idea of the ideal single photon source, a device that, when triggered, produced one and only one photon into a desired mode with unit efficiency. Such a source does not yet exist. However, its development is the subject of a large and ongoing effort, encompassing several technological approaches. The initial step in developing a single photon source is identifying a quantum process that produces the single photon state, as opposed to some other common state such as a thermal state of light. Natural candidates include single atoms (Darquie, Jones, Dingjan, Beugnon, Bergamini, Sortais, Messin, Browaeys, & Grangier, 2005; Kuhn, Hennrich, & Rempe, 2002; McKeever, Boca, Boozer, Miller, Buck, Kuzmich, & Kimble, 2004), single ions (Diedrich & Walther, 1987; Keller, Lange, Hayasaka, Lange, & Walther, 2004), single colour centres (Beveratos, Kuhn, Brouri, Gacoin, Poizat, & Grangier, 2002; Brouri, Beveratos, Poizat, & Grangier, 2000; Kurtsiefer, Mayer, Zarda, & Weinfurter, 2000), single semiconductor quantum dots (Englund, Fushman, Faraon, & Vu˘ckovi´c, 2009; Michler, Kiraz, Becher, Schoenfeld, Petroff, Zhang, Hu, & Imamoglu, 2000; Moreau, Robert, Gerard, Abram, Manin, & Thierry-Mieg, 2001; Noda, Chutinan, & Imada, 2000; Santori, Fattal, Vu˘ckovi´c, Solomon, & Yamamoto, 2002; Santori, Pelton, Solomon, Dale, & Yamamoto, 2001; Yuan, Kardynal, Stevenson, Shields, Lobo, Cooper, Beattie, Ritchie, & Pepper, 2002) and possibly other naturally quantized individual systems which will only emit one quantum of radiation after excitation. Additionally, ensemble or bulk-material approaches, conditioned on certain herald events, can be used as a source of single photons, albeit not a triggered one. Examples in this latter category include
256
Optical Quantum Computation
heralded spontaneous parametric downconversion (Hong & Mandel, 1986; Pittman, Jacobs, & Franson, 2002b, 2005b) and four-wave mixing (Fulconis, Alibart, Wadsworth, & Rarity, 2007; Li, Voss, Sharping, & Kumar, 2005), and the so-called DLCZ ensemble technique (Kuzmich, Bowen, Boozer, Boca, Chou, Duan, & Kimble, 2003; Laurat, Riedmatten, Felinto, Chou, Schomburg, & Kimble, 2006). Since producing more than one photon at a time is an obvious form of error, the quantum optical second-order correlation function g (2) – which can be used to identify states with more than one photon produced – is used to characterize the sources. The best single photon sources reported have g (2) (0) on the order of a percent, e.g. Vu˘ckovi´c, Fattal, Santori, and Solomon (2003), which is substantially lower than the value g 2 (0) = 1 expected for a coherent state, for example. In many prototype single photon sources, much of the remaining 2-event signal may be attributed to contributions from the single photon and either an environmental background photon, or a thermal dark count in the detector. A more challenging problem for single photon sources is the issue of collection efficiency. Even if only one photon is generated, it must be reliably inserted into the desired spatial mode at each and every trigger event. In practice, this is remarkably difficult. Sources based on spontaneous emission suffer from the problem that their natural emission is into 4π steradians, and collecting a sizeable fraction (ideally all) of this emission is challenging. A potential solution is to place the emitter in an optical cavity. Many cavity types have been employed, including Bragg stacks and photonic crystals for quantum dots, Fabry–Perot cavities for atoms and ions, and whispering-gallery-mode resonators in a variety of systems. Although these can, in principle, increase the collection efficiency to close to unity (Noda, Fujita, & Asano, 2007), high degrees of coupling to the cavity (potentially close to 100%, e.g. Englund, Faraon, Zhang, Yamamoto, and Vu˘ckovi´c (2007)) have not yet lead to high outcoupling efficiencies in practice. In the case of inefficient collection, one is faced with the problem that, instead of having a single photon at each and every trigger event, the vacuum state is sometimes present instead. Heralded schemes such as spontaneous parametric downconversion and the DLCZ scheme are designed for more directional emission, so that the collection efficiency problem, in principle, is greatly reduced. Using these schemes, inferred single photon collection efficiencies of 83% (Pittman, Jacobs, & Franson, 2005b) and 50% (Laurat, Riedmatten, Felinto, Chou, Schomburg, & Kimble, 2006) have been measured respectively. The final, and equally critical, criterion for single photon sources is the indistinguishability of the photons produced, from both a single source or different elements in an array of SPSs. This indistinguishability is
Auxiliary Components
257
critical for achieving high-fidelity nonclassical interference, which is a key component of nearly all optical schemes. In some cases, moderateto high-visibility quantum interference has already been measured for two photons from independent sources (e.g. Beugnon, Jones, Dingjan, Darquie, Messin, Browaeys, and Grangier, 2006; Kaltenbaek, Blauensteiner, Zukowski, Aspelmeyer, and Zeilinger, 2006; Maunz, Moehring, Olmschenk, Younge, Matsukevich, and Monroe, 2007; Mosley, Lundeen, Smith, Wasylczyk, U’Ren, Silberhorn, and Walmsley, 2008; Sanaka, Pawlis, Ladd, Lischka, and Yamamoto, 2009). Distinguishability can be introduced in several ways – key examples include: the timing jitter (uncertain emission time) of spontaneous emission sources; different centre frequencies of independent sources; spectral distinguishability introduced through mixture from non-transform limited sources; and the natural entanglement in energy-time of SPDC sources. It should be noted, however, that solutions to all of these problems exist in principle, and some have been demonstrated in practice. For instance, with careful spectral engineering it is possible to remove the frequency entanglement in SPDC sources without requiring the filtering that would reduce efficiency (Mosley, Lundeen, Smith, Wasylczyk, U’Ren, Silberhorn, & Walmsley, 2008). 7.2.2. Bell-pair Sources Some optical quantum computing schemes, such as certain cluster-state proposals (e.g. Browne and Rudolph, 2005), require maximally-entangled photonic qubit pairs (Bell pairs) rather than single photon sources. Having access to pre-existing entanglement of this kind simplifies the operations required in implementing LOQC. Generating triggered, or even heralded, entangled pairs is difficult. One tends to think of suitably phase-matched SPDC as naturally producing entangled pairs (Kwiat, Mattle, Weinfurter, Zeilinger, Sergienko, & Shih, 1995), but these are generated randomly and there is far from any guarantee that one and only one pair will be produced at each pump pulse – rather, the pairs are Poisson-distributed. Nevertheless, SPDC can produce entanglement with high fidelity, conditional on a pair being present – such sources are useful in coincidence experiments. A number of sources have been developed that produce highly entangled states with high flux (Altepeter, Jeffrey, & Kwiat, 2005; Fedrizzi, Herbst, Poppe, Jennewein, & Zeilinger, 2007). The most successful schemes to date for generating triggered entangled pairs are based on semiconductor quantum dots. This involves the production, through pumping, of a biexciton state which can decay via several indistinguishable paths to produce a pair of polarizationentangled photons (Akopian, Lindner, Poem, Berlatzky, Avron, Gershoni,
258
Optical Quantum Computation
Gerardot, & Petroff, 2006; Young, Stevenson, Atkinson, Cooper, Ritchie, & Shields, 2006). The challenge of this approach is keeping the emitted photons from being distinguishable by other means, such as by the slight difference in frequency that arises due to splittings of the intermediate exciton levels. As with other photon sources, collection efficiency is still problematic, although similar solutions to those proposed for SPSs should also be viable. Recently, a triggered Bell pair source using an atom in a cavity has been demonstrated, with the output state demonstrating a fidelity of 0.9 with ¨ ¨ the singlet Bell state (Weber, Specht, Muller, Bochmann, Mucke, Moehring, & Rempe, 2009). 7.2.3. CSS sources For CSQC, the required optical resources are coherent states and coherent state superpositions (CSSs). The former is easily generated on demand by using an appropriately stabilized laser or above-threshold parametric oscillator. The latter, being a fragile non-Gaussian state, is, not surprisingly, somewhat more difficult to generate. The generation of CSSs has already been covered in some detail in Section 6.3 above. There are several areas for active research in developing CSS states. One limitation of present generation of sources is that they are heralded, rather than triggered. The CSS is generated when the photon detector fires. However, because of both the inefficiency of photon counters and the uncertainty in splitting off a photon using the beam splitter, this event does not occur on every pulse (for example). Another problem to be overcome with CSS sources is the loss of fidelity due to inefficiency in the circuit. This introduces vacuum noise that decreases the fidelity of the state. 7.2.4. From Heralded to Triggered Sources Many of the schemes for generating single photon (or other resource) states for optical quantum computing suffer from being nondeterministic, either in principle or in practice. In the latter case, this is often a consequence of the fact that the outcoupling efficiency is not 100%. As long as the schemes are heralded, however, this does not have to be a show-stopper. Several groups are working on techniques to turn heralded sources into triggered, on-demand sources. The basic idea of most of these approaches is to use an array of parallel heralded sources and switching circuitry to bring the probability of getting a single event very close to unity (Migdall, Branning, & Castelletto, 2002; Pittman, Jacobs, & Franson, 2002b). Because the probability of not getting any events decreases exponentially with the number of sources in parallel, the scheme can be very efficient in principle. In practice, sources
Auxiliary Components
259
of error, such as dark counts in the heralding detectors, switch loss and timing considerations, need to be addressed, and work is continuing in these areas.
7.3. Memory At times during a quantum computation, quantum information must be stored in memory. A simple example of this in LOQC is the requirement to hold a qubit while a feedforward operation on the state is triggered and implemented. The basic requirements of a memory are: that it has high (ideally unit) fidelity, that the hold time be suitably long, and that the information can be switched in and out on demand. Photons – and other states useful for optical quantum computing – have spatiotemporal structure, and so it is necessary to design a memory with sufficient bandwidth and physical dimensions to accommodate the optical modes of interest. The simplest memory is the optical delay line, which consists of a free-space or fibre path length to store (or delay) the light by a fixed amount (Pittman, Jacobs, & Franson, 2002a; Prevedel, Walther, Tiefenbacher, Bohl, Kaltenbaek, Jennewein, & Zeilinger, 2007). While this is a simple, potentially high-bandwidth, and relatively highfidelity solution, the basic delay line is also inflexible in that the memory is not switchable – for maximum flexibility one wishes to be able to decide the length of time that the light is held in memory after it is stored. A more sophisticated version of the delay line is the storage loop, in which light is sent around a loop such that once on each pass, it passes a switch which can be flipped to switch it back into the circuit (Pittman & Franson, 2002). In this case, the storage time is not arbitrary, but is rather a multiple of the loop’s single round trip time. One of the major practical challenges of this type of memory is the loss through the switch, even when in the passive condition, which is experienced with each pass, making the chance of retaining a photon for many loop cycles small. Perhaps the most flexible memory, in principle at least, is that promised by atomic ensembles. A variety of techniques have been proposed to store light in these systems, but the feature common to all is that the optical state is transferred into a quantum state of a collection of atoms – often a spin-wave in the ensemble – which is later switched (on demand) back to the optical state. Although several schemes exist for using atomic ensembles, perhaps the best-known are ‘stopped light’ techniques5 using electromagneticallyinduced transparency (EIT) (Appel, Figueroa, Korystov, Lobino, & Lvovsky, 2008; Chaneliere, Matsukevich, Jenkins, Lan, Kennedy, &
5 The slow light phenomenon can also be used, in principle, to make an optical quantum delay line.
260
Optical Quantum Computation
Kuzmich, 2005; Choi, Deng, Laurat, & Kimble, 2008; Eisaman, Andre, Massou, Fleischhauer, Zibrov, & Lukin, 2005; Hetet, Buchler, Gloeckl, Hsu, Akulshin, Bachor, & Lam, 2008), and techniques known as GEM (gradient echo memory) and CRIB (coherent reversible inhomogeneous broadening) (Hetet, Longdell, Alexander, Lam, & Sellars, 2008; Tittel, ¨ Moiseev, & Sellars, 2008). Afzelius, Cone, Chaneli´ere, Kroll, EIT stopped light uses the slow light medium created by a narrow transparency window to compress the light into the medium, and an adiabatic switching of the control field captures the wavefunction in the atomic state of the ensemble. GEM and CRIB work by absorptively creating an atomic coherence stored amongst the atoms in the medium. By globally manipulating the atomic coherence – specifically, by switching detunings – the optical quantum state can be recovered. The interaction of off-resonant light with spin-polarized ensembles has also been used to demonstrate quantum optical storage (Julsgaard, Sherson, Cirac, Fiurasek, & Polzik, 2004). What is common to these techniques, in general, is that they are low bandwidth. For example, the EIT typically produces a delay for light in a ∼MHz bandwidth, a similar bandwidth as for a preliminary CRIB demonstration (Hetet, Longdell, Alexander, Lam, & Sellars, 2008). This should be contrasted with the typical terahertz bandwidth of SPDC. However, as has been noted, SPDC may not be the ultimate choice for an SPS, and even if it is, it is possible to produce narrow bandwidth SPDC sources which might be more suitable for working with this type of memory (e.g. Haase, Piro, Eschner, and Mitchell, 2009). Additionally, improvements in materials development (such as in the solid-state systems) should allow for wider memory bandwidths. Two requirements for a memory are that one can achieve a suitably long delay time, and that the fidelity of the output state, with respect to the input, is high. This latter criterion is sometimes broken up into the storage efficiency and the conditional fidelity. In the case of a single photon, for example, this would be the probability of getting a photon out for each photon sent in, and the fidelity of that output photon’s state with the input state. Storage efficiencies of ∼15% been achieved in quantum memories (Appel, Figueroa, Korystov, Lobino, & Lvovsky, 2008; Choi, Deng, Laurat, & Kimble, 2008; Hetet, Longdell, Alexander, Lam, & Sellars, 2008), with the possibility of making this much higher – indeed, approaching unity in principle. Delay times are largely limited by coherence properties of the atomic media used for storage, and range from ∼ns to ∼ µs, with the possibility of adapting second-timescale classical storage (Longdell, Fraval, Sellars, & Manson, 2005) to quantum memories in the future.
Auxiliary Components
261
7.4. Integrated Optics In the long term, no one expects a large-scale optical quantum computer, comprising many thousands of gates, to be constructed from bulk components laid out on optical tables in a laboratory. As well as the alignment instability, such a quantum computer would take up an enormous amount of space! A more reasonable proposition is to have hardwired optical circuits, possibly on the micro-scale, using some form of integrated optics. As well as the obvious benefit in size and stability, such circuits should also simplify the mode-matching operations that are central to achieving high-quality classical and non-classical interference. The two most obvious forms of guided optics are fibre optics and integrated optics – the latter is usually considered to mean planar waveguides or something written into or onto bulk material. Although there have been quite a number of examples of optical quantum information protocols in optical fibre, especially in regard to quantum key distribution (Gisin, Ribordy, Tittel, & Zbinden, 2002), there have not been many predominantly fibre implementations of quantum computing gates. Some notable demonstrations include ‘plug and play’ fibre and free space modules (Pittman, Fitch, Jacobs, & Franson, 2003) and full fibre implementations (Clark, Fulconis, Rarity, Wadsworth, & O’Brien, 2008) that include a fibre-based photon source. A recent development has been the use of integrated optics for quantum logic. The O’Brien group has demonstrated the nondeterministic CNOT gate of Ralph, Langford, Bell, and White (2002) using planar waveguide circuits (Politi, Cryan, Rarity, Yu, & O’Brien, 2008). One of the properties of standard planar waveguides is that single-polarization operation is exceedingly preferred, so that a dual-rail spatial mode encoding is employed, where superpositions are between the occupancy of two spatial paths defined by the waveguides. The key elements of such an encoding are beam splitters, phase shifters, and interferometers built from these components. To date, the ability to construct and utilize beam splitters with suitable ratios has been demonstrated, as has the ability to make tunable phase shifters using resistive heating of one of the interferometer arms. A future requirement will be fast phase shifters for feedforward operations (Prevedel, Walther, Tiefenbacher, Bohl, Kaltenbaek, Jennewein, & Zeilinger, 2007). The widespread use of electrooptic modulators in telecommunications waveguide technology suggests that this may be achievable in quantum optics devices as well. Another promising technique for building optical waveguide circuits is the direct write technique, in which a high-power laser is used to write a circular waveguide below the surface of some material, generally a glass (Marshall, Ams, & Withford, 2006). The advantage of this technique is that it can be used to generate three-dimensional circuits, which allows optical
262
Optical Quantum Computation
paths to easily cross. Preliminary demonstrations with this technology have achieved high visibility nonclassical interference for two-photon and multi-photon events (Marshall, Politi, Matthews, Dekker, Ams, Withford, & O’Brien, 2009). An outstanding problem for the application of waveguide technologies is the insertion loss of the waveguide circuits, and of specialized components (such as electro-optic devices). The first problem may be solved by introducing sources and detectors directly into the waveguide environment, and the development of fibre-based sources is already a step in that direction. Additionally, work is in progress towards the implementation of adaptive optics waveguide coupling schemes (Kumar, Kwiat, Migdall, Nam, Vu˘ckovi´c, & Wong, 2004), in which the mode pattern from an SPDC source may be converted into the mode pattern corresponding to waveguide propagation.
8. CONCLUSION AND SUMMARY In this chapter, we have examined the potential of optics for realizing a quantum computer. Quantum information can be encoded on light and can be carried, manipulated, entangled and measured with high precision. Both photonic encodings – where the information is carried on a binary degree of freedom such as polarization – and field encodings – with information encoded on different field states – are possible. As a sign of their utility in quantum information science, both types of encoding have been successfully deployed in quantum key distribution protocols (Poppe, Peev, & Maurhart, 2008). Both linear and nonlinear strategies have been considered for implementing universal gate sets. In the non-linear approaches, very strong optical non-linearities are proposed for use in-line, to interact optical qubits with one another. In the linear approaches, off-line nonlinearities, in the form of state production and measurement, are used to add measurement-induced non-linearity into a linear optical network. Although proposals exist for all combinations of strategies and encodings, the most advanced experimental gate demonstrations so far have been performed using the linear approach and photonic encoding. It is possible, in principle, to obtain fault-tolerant operation with both particle and field encodings. Threshold estimates have been generated for several linear strategies, as well as one nonlinear one, using a generalized Steane code. A general trend can be identified whereby higher thresholds tend to lead to higher resource overheads and vice versa, suggesting a trade-off between the precision of the physical operations and the number required. A new, high threshold, efficient fault tolerant protocol based on 3-D cluster states has recently been developed by Raussendorf
References
263
and Harrington (2007) that appears particularly well suited to optical architectures (Devitt, Fowler, Stephens, Greentree, Hollenberg, Munro, & Nemoto, 2009). Future work is likely to focus on the optimal encoding and gate types for these kinds of protocols. A large-scale optical quantum computer will require specialized sources, detectors, switches and circuits, optical networks and memories. Each of these is presently under development, with many different approaches being pursued. While some of these optical technologies have already demonstrated performance levels close to those required, significant improvements are still necessary in some areas to make largescale quantum computing possible. Encouragingly, progress seems to be rapid and there are no in-principle reasons why the required performance levels cannot be achieved. There are other directions in optics-related quantum computing that could not be covered in a chapter of this length. For example, hybrid optical/atomic or optical/solid-state systems may be contenders for quantum information processing applications. Optical ‘flying qubits’ could act as a data bus, solving the connectivity problem in atomic or solid-state quantum computer architectures. Alternatively we might use the ‘standing qubits’ as memory, whilst processing the quantum information optically. An example of experimental progress in this direction is the generation of entanglement between distant ions using an optical quantum bus and LOQC processing (Matsukevich, Maunz, Moehring, Olmschenk, & Monroe, 2008). Of the physical systems being considered for quantum computation, optics is perhaps the best understood in terms of the physics of the interactions and decoherence mechanisms. Optics has also demonstrated an outstanding precision in operations demonstrated to date. The gap between the theoretical requirements and experimental demonstrations, while large, has been shrinking consistently due to both advances in technology and improved protocols. We can conclude that there appear to be no fundamental barriers to optical quantum computation and although formidable practical barriers remain, current research suggests an optimistic outlook for overcoming them.
ACKNOWLEDGMENTS This work was supported by the Australian Research Council and the IARPA funded US Army Research Office Contract No. W911NF-05-0397.
REFERENCES Achilles, D., Silberhorn, C., Sliwa, C., Banaszek, K., Walmsley, I. A., Fitch, M. J., Jacobs, B. C., Pittman, T. B., & Franson, J. D. (2004). Jornal of Modern Optics, 51, 1499.
264
Optical Quantum Computation
Achilles, D., Silberhorn, C., & Walmsley, I. A. (2006). Physical Review Letters, 97, 043602. Akopian, N., Lindner, N. H., Poem, E., Berlatzky, Y., Avron, J., Gershoni, D., Gerardot, B. D., & Petroff, P. M. (2006). Physical Review Letters, 96, 130501. Albota, M. A., & Wong, F. N. C. (2004). Optics Letters, 29, 1449. Altepeter, J. B., Jeffrey, E. R., & Kwiat, P. G. (2005). Optics Express, 13, 8951. Aoki, T., Dayan, B., Wilcut, E., Bowen, W. P., Parkins, A. S., Kippenberg, T. J., Vahala, K. J., & Kimble, H. J. (2006). Nature, 443, 671. Appel, J., Figueroa, E., Korystov, D., Lobino, M., & Lvovsky, A. I. (2008). Physical Review Letters, 100, 093602. Aspuru-Guzik, A., Dutoi, A. D., Love, P. J., & Head-Gordon, M. (2005). Science, 309, 1704. Babichev, S. A., Brezger, B., & Lvovsky, A. I. (2004). Physical Review Letters, 92, 047903. Bachor, H.-A., & Ralph, T. C. (2004). A guide to experiments in quantum optics (2nd ed.). Weinheim: Wiley-VCH. Bao, X. H., Chen, T. Y., Zhang, Q., Yang, J., Zhang, H., Yang, T., & Pan, J. W. (2007). Physical Review Letters, 98, 170502. Braunstein, S. L., & Pati, A. (2003). Continuous variable quantum information. The Netherlands: Kluwer Academic Publishers. Braunstein, S. L., & van Loock, P. (2005). Reviews of Modern Physics, 77, 513. Bennett, C. H., Brassard, G., Crepeau, C., Jozsa, R., Peres, A., & Wootters, W. K. (1993). Physical Review Letters, 70, 1895. Beugnon, J., Jones, M. P. A., Dingjan, J., Darquie, B., Messin, G., Browaeys, A., & Grangier, P. (2006). Nature, 440, 779. Beveratos, A., Kuhn, S., Brouri, R., Gacoin, T., Poizat, J. P., & Grangier, P. (2002). European Physical Journal D, 18, 191. Birnbaum, K. M., Boca, A., Miller, R., Boozer, A. D., Northup, T. E., & Kimble, H. J. (2005). Nature, 436, 87. Brouri, R., Beveratos, A., Poizat, J. P., & Grangier, P. (2000). Optics Letters, 25, 1294. Browne, D. E., & Rudolph, T. (2005). Physical Review Letters, 95, 010501. Brune, M., Hagley, E., Dreyer, J., Maitre, X., Maali, A., Wunderlich, C., Raimondand, J. M., & Haroche, S. (1996). Physical Review Letters, 77, 4887. Bullock, S. S., OLeary, D. P., & Brennen, G. K. (2005). Physical Review Letters, 94, 230502. Chaneliere, T., Matsukevich, D. N., Jenkins, S. D., Lan, S. Y., Kennedy, T. A. B., & Kuzmich, A. (2005). Nature, 438, 833. Chen, K., Li., C. M., Qiang, Z., Chen, Y. A., Goebel, A., Chen, S., Mair, A., & Pan, J. W. (2007). Physical Review Letters, 99, 120503. Choi, K. S., Deng, H., Laurat, J., & Kimble, H. J. (2008). Nature, 452, 67. Chuang, I. L., & Nielsen, M. A. (1997). Journal of Modern Optics, 44, 2455. Clark, A. S., Fulconis, J., Rarity, J. G., Wadsworth, W. J., & O’Brien, J. L. 2008, arXiv:0802.1676. Cochrane, P., Milburn, G. J., & Munro, W. J. (1998). Physical Review A, 59, 2631. Darquie, B., Jones, M. P. A., Dingjan, J., Beugnon, J., Bergamini, S., Sortais, Y., Messin, G., Browaeys, A., & Grangier, P. (2005). Science, 309, 454. Dawson, C. M., Haselgrove, H. L., & Nielsen, M. A. (2006). Physical Review A, 73, 052306. Deutsch, D. (1985). Proceedings of the Royal Society of London A, 400, 97. Devitt, S. J., Fowler, A. G., Stephens, A. M., Greentree, A. D., Hollenberg, L. C. L., Munro, W. J., & Nemoto, K. (2009). New Journal of Physics, 11, 083032. Diedrich, F., & Walther, H. (1987). Physical Review Letters, 58, 203. Dirac, P. A. M. (1958). The principles of quantum mechanics (4th ed.). London: Oxford University Press. Dodd, J. L., Ralph, T. C., & Milburn, G. J. (2003). Physical Review A, 68, 042328. Duan, L.-M., & Kimble, H. J. (2004). Physical Review Letters, 92, 127902. Eisaman, M. D., Andre, A., Massou, F., Fleischhauer, M., Zibrov, A. S., & Lukin, M. D. (2005). Nature, 438, 837. Englund, D., Faraon, A., Zhang, B., Yamamoto, Y., & Vu˘ckovi´c, J. (2007). Optics Express, 15, 5550.
References
265
Englund, D., Fushman, I., Faraon, A., & Vu˘ckovi´c, J. (2009). Photonics and Nanostructures, Fundamentals and Applications, 7, 56. van Enk, S. J., & Hirota, O. (2001). Physical Review A, 64, 022313. Fedrizzi, A., Herbst, T., Poppe, A., Jennewein, T., & Zeilinger, A. (2007). Optics Express, 15, 15377. Feynman, R. P. (1986). Foundations of Physics, 16, 507. Franson, J. D., Donegan, M. M., Fitch, M. J., Jacobs, B. C., & Pittman, T. B. (2002). Physical Review Letters, 89, 137901. Franson, J. D., Jacobs, B. C., & Pittman, T. B. (2004). Physical Review A, 70, 062302. Fujiwara, M., & Sasaki, M. (2006). Optics Letters, 31, 691. Fulconis, J., Alibart, O., Wadsworth, W. J., & Rarity, J. G. (2007). New Journal of Physics, 9, 276. Fushman, I., Englund, D., Faraon, A., Stoltz, N., Petroff, P., & Vu˘ckovi´c, J. (2008). Science, 320, 769. Gasparoni, S., Pan, J. W., Walther, P., Rudolph, T., & Zeilinger, A. (2004). Physical Review Letters, 93, 020504. Gaebel, T., Domhan, M., Popa, I., Wittmann, C., Neumann, P., Jelezko, F., Rabeau, J. R., Stavrias, N., Greentree, A. D., Prawer, S., Meijer, J., Twamley, J., Hemmer, P. R., & Wrachtrup, J. (2006). Nature Physics, 2, 408. Ghosh, R., & Mandel, L. (1987). Physical Review Letters, 59, 1903. Gilchrist, A., Hayes, A. J. F., & Ralph, T. C. (2007). Physical Review A, 75, 052328. Gisin, N., Ribordy, G. G., Tittel, W., & Zbinden, H. (2002). Reviews of Modern Physics, 74, 145. Glancy, S., & Knill, E. (2006). Physical Review A, 73, 012325. Glauber, R. J. (1962). Physical Review, 131, 2766. Gleyzes, S., Kuhr, S., Guerlin, C., Bernu, J., Deleglise, S., Hoff, U. B., Brune, M., Raimond, J.-M., & Haroche, S. (2007). Nature, 446, 297. Gong, Y. X., Guo, G., & Ralph, T. C. (2008). Physical Review A, 78, 012305. Gottesman, D., & Chuang, I. L. (1999). Nature, 402, 390. Gottesman, D., Kitaev, A., & Preskill, J. (2001). Physical Review A, 64, 012310. Grover, L. K. (1997). Physical Review Letters, 79, 325. Haase, A., Piro, N., Eschner, J., & Mitchell, M. W. (2009). Optics Letters, 34, 55. Haeffner, H., Roos, C. F., & Blatt, R. (2008). Physics Reports, 469, 155. Haselgrove, H. L., & Rohde, P. P. (2008). Quantum Information and Computing, 8, 399. Hayes, A. J. F., Gilchrist, A., Myers, C. R., & Ralph, T. C. (2004). Journal of Optics B, 6, 533. Hayes, A. J. F., Haselgrove, H. L., Gilchrist, A., & Ralph, T. C. (2009). arXiv:0908.3932. Hendrickson, S. M., Pittman, T. B., & Franson, J. D. (2009). Journal of the Optical Society of America B-Optical Physics, 26, 267. Hetet, G., Buchler, B. C., Gloeckl, O., Hsu, M. T. L., Akulshin, A. M., Bachor, H. A., & Lam, P. K. (2008). Optics Express, 16, 7369–7381. Hetet, G., Longdell, J. J., Alexander, A. L., Lam, P. K., & Sellars, M. J. (2008). Physical Review Letters, 100, 023601. Higgins, B. L., Berry, D. W., Bartlett, S. D., Wiseman, H. M., & Pryde, G. J. (2007). Nature, 450, 393. Hofmann, H. F., & Takeuchi, S. (2002). Physical Review A, 66, 024308. Hofmann, H. F. (2005). Physical Review Letters, 94, 160504. Hong, C. K., & Mandel, L. (1986). Physical Review Letters, 56, 58. Hong, C. K., Ou, Z. Y., & Mandel, L. (1987). Physical Review Letters, 59, 2044. Hosten, O., Rakher, M. T., Barreiro, J. T., Peters, N. A., & Kwiat, P. G. (2006). Nature, 439, 949. Huntington, E. H., & Ralph, T. C. (2004). Physical Review A, 69, 042318. Ivanovic, I. D. (1987). Physics Letters A, 123, 257. Jeong, H., Kim, M. S., & Lee, J. (2001). Physical Review A, 64, 052308. Jeong, H., & Ralph, T. C. (2007). In N. Cerf, G. Leuchs, & E. S. Polzik (Eds.), Quantum information with continuous variables of atoms and light. London: Imperial College Press. Julsgaard, B., Sherson, J., Cirac, J. I., Fiurasek, J., & Polzik, E. S. (2004). Nature, 432, 482.
266
Optical Quantum Computation
Kaltenbaek, R., Blauensteiner, B., Zukowski, M., Aspelmeyer, M., & Zeilinger, A. (2006). Physical Review Letters, 96, 240502. Kang, H. S., & Zhu, Y. F. (2003). Physical Review Letters, 91, 093601. Kardynal, B. E., Yuan, Z. L., & Shields, A. J. (2008). Nature Photonics, 2, 425. Keller, M., Lange, B., Hayasaka, K., Lange, W., & Walther, H. (2004). Nature, 431, 1075. Kieling, K., Rudolph, T., & Eisert, J. (2007). Physical Review Letters, 99, 130501. Kiesel, N., Schmid, C., Weber, U., Toth, G., Guhne, O., Ursin, R., & Weinfurter, H. (2005). Physical Review Letters, 95, 210502. Kiesel, N., Schmid, C., Weber, U., Ursin, R., & Weinfurter, H. (2005). Physical Review Letters, 95, 210505. Kim, J. S., Takeuchi, S., Yamamoto, Y., & Hogue, H. H. (1999). Applied Physics Letters, 74, 902. Knill, E., Laflamme, R., & Milburn, G. J. (2001). Nature, 409, 46. Knill, E. (2003). Physical Review A, 68, 064303. Kok, P., Munro, W. J., Nemoto, K., Ralph, T. C., Dowling, J. P., & Milburn, G. J. (2007). Reviews of Modern Physics, 79, 135. Kuhn, A., Hennrich, M., & Rempe, G. (2002). Physical Review Letters, 89, 067901. Kumar, P., Kwiat, P., Migdall, A. L., Nam, S. W., Vu˘ckovi´c, J., & Wong, F. N. C. (2004). Quantum Information Processing, 3, 215. Kurtsiefer, C., Mayer, S., Zarda, P., & Weinfurter, H. (2000). Physical Review Letters, 85, 290. Kuzmich, A., Bowen, W. P., Boozer, A. D., Boca, A., Chou, C. W., Duan, L. M., & Kimble, H. J. (2003). Nature, 423, 731. Kwiat, P. G., Mitchell, J. R., Schwindt, P. D. D., & White, A. G. (2000). Journal of Modern Optics, 47, 257. Kwiat, P. G., Mattle, K., Weinfurter, H., Zeilinger, A., Sergienko, A. V., & Shih, Y. H. (1995). Physical Review Letters, 75, 4337. Langford, N. K., Weinhold, T. J., Prevedel, R., Resch, K. J., Gilchrist, A., O’Brien, J. L., Pryde, G. J., & White, A. G. (2005). Physical Review Letters, 95, 210504. Lanyon, B. P., Weinhold, T. J., Langford, N. K., Barbieri, M., James, D. F. V., Gilchrist, A., & White, A. G. (2007). Physical Review Letters, 99, 250505. Lanyon, B. P., Barbieri, M., Almeida, M. P., & White, A. G. (2008). Physical Review Letters, 101, 200501. Lanyon, B. P., Barbieri, M., Almeida, M. P., Jennewein, T., Ralph, T. C., Resch, K. J., Pryde, G. J., O’Brien, J. L., Gilchrist, A., & White, A. G. (2009). Nature Physics, 5, 134. Lanyon, B. P., Whitfield, J. D., Gillett, G. G., Goggin, M. E., Almeida, M. P., Kassal, I., Biamonte, J. D., Mohseni, M., Powell, B. J., Barbieri, M., Aspuru-Guzik, A., & White, A. G. (2009). Nature Chemistry (in press). arXiv:0905.0887. Laurat, J., de Riedmatten, H., Felinto, D., Chou, C., Schomburg, E. W., & Kimble, H. J. (2006). Optics Express, 14, 6912. Leung, P. M., & Ralph, T. C. (2006). Physical Review A, 74, 062325. Leung, P. M., & Ralph, T. C. (2007). New Journal of Physics, 9, 224. Li, X. Y., Voss, P. L., Sharping, J. E., & Kumar, P. (2005). Physical Review Letters, 94, 053601. Lita, A. E., Miller, A. J., & Nam, S. W. (2008). Optics Express, 16, 3032. Lloyd, S., & Braunstein, S. L. (1999). Physical Review Letters, 82, 1784. Longdell, J. J., Fraval, E., Sellars, M. J., & Manson, N. B. (2005). Physical Review Letters, 95, 063601. Lu, C., Gao, W., Guehne, O., Zhou, X., Chen, Z., & Pan, J. W. (2009). Physical Review Letters, 102, 030502. Lu, C., Browne, D. E., Yang, T., & Pan, J. W. (2007). Physical Review Letters, 99, 250504. Lu, C., Zhou, X., Guehne, O., Gao, W., Zhang, J., Yuan, Z., Goebel, A., Yang, T., & Pan, J. W. (2007). Nature Physics, 3, 91. Lu, C., Gao, W., Zhang, L., Zhou, X., Yang, T., & Pan, J. W. (2008). Proceedings of the National Academy of Sciences of the United States of America, 105, 11050. Lund, A. P., & Ralph, T. C. (2002). Physical Review A, 66, 032307. Lund, A. P., Ralph, T. C., & Haselgrove, H. L. (2008). Physical Review Letters, 100, 030503.
References
267
Lundeen, J. S., Feito, A., Coldenstrodt-Ronge, H., Pregnell, K. L., Silberhorn, C., Ralph, T. C., Eisert, J., Plenio, M. B., & Walmsley, I. A. (2009). Nature Physics, 5, 27. ¨ Lutkenhaus, N., Calsamiglia, J., & Suominen, K.-A. (1999). Physical Review A, 59, 3295. Lvovsky, A. I., & Mlynek, J. (2002). Physical Review Letters, 88, 250401. Marshall, G. D., Politi, A., Matthews, J. C. F., Dekker, P., Ams, M., Withford, M. J., & O’Brien, J. L. 2009, arXiv:0902.4357. Marshall, G. D., Ams, M., & Withford, M. J. (2006). Optics Letters, 31, 2690. Matsukevich, D. N., Maunz, P., Moehring, D. L., Olmschenk, S., & Monroe, C. (2008). Physical Review Letters, 100, 150404. Maunz, P., Moehring, D. L., Olmschenk, S., Younge, K. C., Matsukevich, D. N., & Monroe, C. (2007). Nature Physics, 3, 538. Mckeever, J., Boca, A., Boozer, A. D., Miller, R., Buck, J. R., Kuzmich, A., & Kimble, H. J. (2004). Science, 303, 1992. Menicucci, N. C., van Loock, P., Gu, M., Weedbrook, C., Ralph, T. C., & Nielsen, M. A. (2006). Physical Review Letters, 97, 110501. Menicucci, N. C., Flammia, S. T., & Pfister, O. (2008). Physical Review Letters, 101, 130501. Michler, P., Kiraz, A., Becher, C., Schoenfeld, W. V., Petroff, P. M., Zhang, L. D., Hu, E., & Imamoglu, A. (2000). Science, 290, 2282. Migdall, A. L., Branning, D., & Castelletto, S. (2002). Physical Review A, 66, 053805. Miki, S., Fujiwara, M., Sasaki, M., Baek, B., Miller, A. J., Hadfield, R. H., Nam, S. W., & Wang, Z. (2008). Applied Physics Letters, 92, 061116. Milburn, G. J. (1989). Physical Review Letters, 62, 2124. Misra, B., & Sudarshan, E. C. G. (1977). Journal of Mathematical Physics, 18, 756. Mølmer, K. (1997). Physical Review A, 55, 3195. Moreau, E., Robert, I., Gerard, J. M., Abram, I., Manin, L., & Thierry-Mieg, V. (2001). Applied Physics Letters, 79, 2865. Mosley, P. J., Lundeen, J. S., Smith, B. J., Wasylczyk, P., U’Ren, A. B., Silberhorn, C., & Walmsley, I. A. (2008). Physical Review Letters, 100, 133601. Neergaard-Nielsen, J. S., Nielsen, B. M., Hettich, C., Molmer, K., & Polzik, E. S. (2006). Physical Review Letters, 97, 083604. Nemoto, K., & Munro, W. J. (2004). Physical Review Letters, 93, 250502. Nielsen, M., & Chuang, I. (2000). Quantum computation and quantum information. Cambridge UK: Cambridge University Press. Nielsen, M. (2004). Physical Review Letters, 93, 040503. Noda, S., Chutinan, A., & Imada, M. (2000). Nature, 407, 608. Noda, S., Fujita, M., & Asano, T. (2007). Nature Photonics, 1, 449. O’Brien, J. L., Pryde, G. J., White, A. G., Ralph, T. C., & Branning, D. (2003). Nature, 426, 264. O’Brien, J. L., Pryde, G. J., White, A. G., & Ralph, T. C. (2005). Physical Review A, 71, 060303. O’Brien, J. L., Pryde, G. J., Gilchrist, A., James, D. F. V., Langford, N. K., Ralph, T. C., & White, A. G. (2004). Physical Review Letters, 93, 080502. Okamoto, R., Hofmann, H. F., Takeuchi, S., & Sasaki, K. (2005). Physical Review Letters, 95, 210506. Oort, J. H., & Walraven, T. (1956). Bulletin of the Astronomical Institutes of the Netherlands, 12, 285. Ourjoumtsev, A., Tualle-Brouri, R., Laurat, J., & Grangier, P. (2006). Science, 312, 83. Ourjoumtsev, A., Jeong, H., Tualle-Brouri, R., & Grangier, P. (2007). Nature, 448, 784. Ourjoumtsev, A., Ferreyrol, F., Tualle-Brouri, R., & Grangier, P. (2009). Nature Physics, 5, 189. Pachos, J. K., Wieczorek, W., Schmid, C., Kiesel, N., Pohlner, R., & Weinfurter, H. 2007, arXiv:0710.0895. Peters, N., Altepeter, J., Jeffrey, E., Branning, D., & Kwiat, P. G. (2003). Quantum Information & Computation, 3, 503. Pittman, T. B., Fitch, M. J., Jacobs, B. C., & Franson, J. D. (2003). Physical Review A, 68, 032316. Pittman, T. B., & Franson, J. D. (2002). Physical Review A, 66, 062302. Pittman, T. B., Jacobs, B. C., & Franson, J. D. (2001). Physical Review A, 64, 062311.
268
Optical Quantum Computation
Pittman, T. B., Jacobs, B. C., & Franson, J. D. (2002a). Physical Review A, 66, 052305. Pittman, T. B., Jacobs, B. C., & Franson, J. D. (2002b). Physical Review A, 66, 042303. Pittman, T. B., Jacobs, B. C., & Franson, J. D. (2005a). Physical Review A, 71, 052332. Pittman, T. B., Jacobs, B. C., & Franson, J. D. (2005b). Optics Communications, 246, 545. Politi, A., Cryan, M. J., Rarity, J. G., Yu, S., & O’Brien, J. L. (2008). Science, 320, 646. Poppe, A., Peev, M., & Maurhart, O. (2008). International Journal of Quantum Information, 6, 209. Poyatos, J. F., Cirac, J. I., & Zoller, P. (1997). Physical Review Letters, 78, 390. Prevedel, R., Tame, M. S., Stefanov, A., Paternostro, M., Kim, M. S., & Zeilinger, A. (2007). Physical Review Letters, 99, 250503. Prevedel, R., Stefanov, A., Walther, P., & Zeilinger, A. (2007). New Journal of Physics, 9, 205. Prevedel, R., Walther, P., Tiefenbacher, F., Bohl, P., Kaltenbaek, R., Jennewein, T., & Zeilinger, A. (2007). Nature, 445, 65. Pryde, G. J., O’Brien, J. L., White, A. G., Ralph, T. C., & Wiseman, H. M. (2005). Physical Review Letters, 94, 220405. Pryde, G. J., O’Brien, J. L., White, A. G., Bartlett, S. D., & Ralph, T. C. (2004). Physical Review Letters, 92, 190402. Pryde, G. J., O’Brien, J. L., White, A. G., & Bartlett, S. D. (2005). Physical Review Letters, 94, 220406. Pysher, M., Bloomer, R., Pfister, O., Kaleva, C. M., Roberts, T. D., & Battle, P. (2008). arXiv:0809.1611v1. Ralph, T. C. (2006). Reports on Progress in Physics, 69, 853. Ralph, T. C., Bartlett, S. D., O’Brien, J. L., Pryde, G. J., & Wiseman, H. M. (2006). Physical Review A, 73, 012113. Ralph, T. C., Gilchrist, A., Milburn, G. J., Munro, W. J., & Glancy, S. (2003). Physical Review A, 68, 042319. Ralph, T. C., Hayes, A. J. F., & Gilchrist, A. (2005). Physical Review Letters, 95, 100501. Ralph, T. C., Langford, N. K., Bell, T. B., & White, A. G. (2002). Physical Review A, 65, 062324. Ralph, T. C., Munro, W. J., & Milburn, G. J. (2002). Proceedings of SPIE, 4917, 1. Ralph, T. C., White, A. G., Munro, W. J., & Milburn, G. J. (2002). Physical Review A, 65, 012314. Raussendorf, R., & Briegel, H. J. (2001). Physical Review Letters, 86, 5188. Raussendorf, R., & Harrington, J. (2007). Physical Review Letters, 98, 190504. Reck, M., Zeilinger, A., Bernstein, H. J., & Bertani, P. (1994). Physical Review Letters, 73, 58. Rohde, P. P., Pryde, G. J., O’Brien, J. L., & Ralph, T. C. (2005). Physical Review A, 72, 032306. Rohde, P. P., Ralph, T. C., & Munro, W. J. (2007). Physical Review A, 75, 010302. Rosfjord, K. M., Yang, J. K. W., Dauler, E. A., Kerman, A. J., Anant, V., Voronov, B. M., Gol’tsman, G. N., & Berggren, K. K. (2006). Optics Express, 14, 527. Roussev, R. V., Langrock, C., Kurz, J. R., & Fejer, M. M. (2004). Optics Letters, 29, 1518. Sakurai, J. J. (1985). Modern quantum optics. California: Addison-Wesley. Sanaka, K., Kawahara, K., & Kuga, T. (2002). Physical Review A, 66, 040301. Sanaka, K., Pawlis, A., Ladd, T., Lischka, K., & Yamamoto, Y. 2009, arXiv:0903.1849v1. Santori, C., Fattal, D., Vu˘ckovi´c, J., Solomon, G. S., & Yamamoto, Y. (2002). Nature, 419, 594. Santori, C., Pelton, M., Solomon, G., Dale, Y., & Yamamoto, E. (2001). Physical Review Letters, 86, 1502. Schmidt, H., & Imamoglu, A. (1996). Optics Letters, 21, 1936. Schoelkopf, R. J., & Girvin, S. M. (2008). Nature, 451, 664. Schumacher, B. (1995). Physical Review A, 51, 2738. Schuster, I., Kubanek, A., Fuhrmanek, A., Puppe, T., Pinkse, P. W. H., Murr, K., & Rempe, G. (2008). Nature Physics, 4, 382. Shapiro, J. H. (2006). Physical Review A, 73, 062305. Shor, P. W. (1994). Proceedings of the 35th annual symposium on the foundations of computer science. Los Alamitos California: IEEE Computer Society Press. Shor, P. (1995). Physical Review A, 52, 2493. Silva, M., Roetteler, M., & Zalka, C. (2005). Physical Review A, 72, 032307.
References
269
Steane, A. M. (1996). Phys Rev Lett, 77, 793. Stucki, D., Gisin, N., Guinnard, O., Ribordy, G., & Zbinden, H. (2002). New Journal of Physics, 4, 41. Takahashi, H., Wakui, K., Suzuki, S., Takeoka, M., Hayasaka, K., Furusawa, A., & Sasaki, M. (2008). Physical Review Letters, 101, 233605. Takeuchi, S., Kim, J., Yamamoto, Y., & Hogue, H. H. (1999). Applied Physics Letters, 74, 1063. Tame, M. S., Prevedel, R., Paternostro, M., Boehi, P., Kim, M. S., & Zeilinger, A. (2007). Physical Review Letters, 98, 140501. Thew, R. T., Zbinden, H., & Gisin, N. (2008). Applied Physics Letters, 93, 071104. ¨ S., Moiseev, S. A., & Sellars, M. 2008, Tittel, W., Afzelius, M., Cone, R. L., Chaneli´ere, T., Kroll, arXiv:0810.0172. Tokunaga, Y., Kuwashiro, S., Yamamoto, T., Koashi, M., & Imoto, N. (2008). Physical Review Letters, 100, 210501. Turchette, Q. A., Hood, C. J., Lange, W., Mabuchi, H., & Kimble, H. J. (1995). Physical Review Letters, 75, 4710. Uskov, D. B., Kaplan, L., Smith, A. M., Huver, S. D., & Dowling, J. P. (2009). Physical Review A, 79, 042326. Vallone, G., Pomarico, E., De Martini, F., & Mataloni, P. (2008). Physical Review Letters, 100, 160502. Vallone, G., Pomarico, E., Mataloni, P., De Martini, F., & Berardi, V. (2007). Physical Review Letters, 98, 180502. Varnava, M., Browne, D. E., & Rudolph, T. (2008). Physical Review Letters, 100, 060502. VanDevender, A. P., & Kwiat, P. G. (2007). Journal of the Optical Society of America B-Optical Physics, 24, 295. Vu˘ckovi´c, J., Fattal, D., Santori, C., & Solomon, G. S. (2003). Applied Physics Letters, 82, 3596. Wakui, K., Takahashi, H., Furusawa, A., & Sasaki, M. (2007). Optics Express, 15, 3568. Walls, D. F., & Milburn, G. J. (1994). Quantum optics. Berlin: Springer-Verlag. Walther, P., Resch, K. J., Rudolph, T., Schenck, E., Weinfurter, H., Vedral, V., Aspelmeyer, M., & Zeilinger, A. (2005). Nature, 434, 169. Walther, P., & Zeilinger, A. (2005). Physical Review A, 72, 010302. Walther, P., Aspelmeyer, M., Resch, K. J., & Zeilinger, A. (2005). Physical Review Letters, 95, 020403. ¨ ¨ Weber, B., Specht, H. P., Muller, T., Bochmann, J., Mucke, M., Moehring, D. L., & Rempe, G. (2009). Physical Review Letters, 102, 030501. White, A. G., Gilchrist, A., Pryde, G. J., O’Brien, J. L., Bremner, M. J., & Langford, N. K. (2007). Journal of the Optical Society of America B-Optical Physics, 24, 172. Wootters, W. K., & Zurek, W. H. (1982). Nature, 299, 802. Yoshikawa, J., Miwa, Y., Huck, A., Andersen, U. L., van Loock, P., & Furusawa, A. (2008). Physical Review Letters, 101, 250501. Young, R. J., Stevenson, R. M., Atkinson, P., Cooper, K., Ritchie, D. A., & Shields, A. J. (2006). New Journal of Physics, 8, 29. Yoran, N., & Reznik, B. (2003). Physical Review Letters, 91, 037903. Yuan, Z. L., Kardynal, B. E., Stevenson, R. M., Shields, A. J., Lobo, C. J., Cooper, K., Beattie, N. S., Ritchie, D. A., & Pepper, M. (2002). Science, 295, 102. Yukawa, M., Ukai, R., van Loock, P., & Furusawa, A. (2008). Physical Review A, 78, 012301. Zhao, Z., Zhang, A. N., Chen, Y. A., Zhang, H., Du, J. F., Yang, T., & Pan, J. W. (2005). Physical Review Letters, 94, 030501.
CHAPTER
5 Science and Engineering of Photonic Crystals Kazuaki Sakoda a and Joseph W. Haus b a Quantum Dot Research Center, National Institute for Materials
Science, Tsukuba, Ibaraki 305-0044, Japan b Electro-Optics Program, University of Dayton, Dayton, OH 45469-0245, USA Contents
Introduction Pursuit of the 3D PBG 2.1. Photonic Band Calculations 2.2. Realization of the 3D PBG 3. Eigenmode Symmetry and Transmittance 4. Defect Modes and Waveguides 4.1. Defect Modes 4.2. PC Waveguides 5. PC Slab 5.1. Fabrication of PC Slabs 5.2. Dispersion Relation 5.3. Diffraction by Periodic Structures 5.4. Air-Bridge PC Slab 6. Cavity QED with PC 7. Band Edge Enhancement 8. Concluding Remarks References
1. 2.
271 273 273 279 284 288 288 290 290 290 292 296 299 303 306 310 313
1. INTRODUCTION Diffraction is one of the most well-known yet important properties of electromagnetic waves. In the frequency range of X-rays, diffraction by natural crystals is a powerful tool for the analysis of their structures at the atomic level. The interaction of X-rays with materials is, however, so c 2009 Elsevier B.V. Progress in Optics, Volume 54 ISSN 0079-6638, DOI 10.1016/S0079-6638(10)05410-7 All rights reserved.
271
272
Science and Engineering of Photonic Crystals
small that any diffraction may usually be treated as a perturbation. On the other hand, diffraction at optical and lower frequencies can be stronger by several orders of magnitude due to the large refractive index of matter in these frequency ranges, and thus, the treatment of such diffraction as a perturbation often fails to describe fundamental aspects of the phenomena that we actually observe. In the late 1960s, ordered aqueous suspensions of polystyrene particles, which are now called colloidal crystals, attracted much attention (Hachisu Kobayashi, & Kose, 1972; Hiltner & Krieger, 1969; Williams, Crandall, & Wojtwitz, 1976). The average distance between adjacent particles was of the order of hundreds of nanometers, so, the colloidal crystals displayed beautiful colours due to the Bragg reflection of visible light. Because the contrast of the refractive index between water and polystyrene was not so small, self-consistent treatment of diffracted and transmitted optical waves was necessary to describe their propagation in the colloidal crystal. This problem was solved by Kazuo Ohtaka in 1979 (Ohtaka, 1979). He extended the theoretical framework of electronic wave propagation in a lattice composed of atoms with spherical potential, which is known as the Korringa–Kohn–Rostoker (KKR) formalism (Kohn & Rostoker, 1954; Korringa, 1947), to electromagnetic wave propagation, and described the energy band of photons. His theory was an extension of the scalar field of electrons to the vector field of photons. It was also an extension of the single-particle scattering theory by Mie (1908) and Debye (1909) to the case of a regular array of spherical scatterers. At that time, however, the refractive-index contrast was not so large, and the revolutionary idea of the complete photonic band gap (PBG) was not conceived. In his historic paper in 1987 (Yablonovitch, 1987), Eli Yablonovitch mentioned for the first time the possibility of the three-dimensional (3D) complete PBG and discussed what it might bring to optics. The one-dimensional (1D) PBG, that is, the stop band realized by dielectric multilayers was quite a common feature of periodic systems. Yablonovitch’s idea was to extend this 1D property to the three dimensions by introducing 3D periodic modulation of the refractive index, and later he succeeded in realizing the complete PBG (Yablonovitch, Gmitter, & Leung, 1991). In Yablonovitch (1987), he proposed several applications of PBGs such as suppression of spontaneous emission and lasing with cavity modes, which were also achieved later by fabricating photonic crystals (PCs) using sophisticated lithography technologies. In the same year, Sajeev John discussed the strong localization of photons in disordered PCs based on the anticipation of PBGs or pseudo gaps at least by periodic arrays of dielectric scatterers (John, 1987).
Pursuit of the 3D PBG
273
These pioneering works stimulated the imagination of researchers tremendously and led to an explosion of successive research activities. In this article, we highlight some achievements of this enthusiastic research.
2. PURSUIT OF THE 3D PBG 2.1. Photonic Band Calculations In the first stage of the history of photonic crystals (PCs), realization of the three-dimensional (3D) photonic band gap (PBG) was the central issue of the research. To attack this problem, researchers had to develop an efficient tool to calculate the dispersion relation of the electromagnetic field in periodic systems, and found a very general framework using the plane-wave expansion method (Ho, Chan, & Soukoulis, 1990; Leung & Liu, 1990; Zhang & Satpathy, 1990). We present some details of this method here. The electromagnetic field is governed by the Maxwell equations. To derive the wave equation in PCs without net charge or current, it is sufficient to consider two of the four Maxwell equations: ∂ B(r, t), ∂t ∂ ∇ × H(r, t) = D(r, t), ∂t
∇ × E(r, t) = −
(1) (2)
where the standard notations of four relevant vector fields are used, that is, E for the electric field, B for the magnetic induction, H for the magnetic field, and D for the electric displacement. r and t are space and time coordinates. The remaining two equations, ∇ · D(r, t) = 0, ∇ · B(r, t) = 0,
(3) (4)
are satisfied automatically. For the linear electromagnetic properties of the PCs, such as the dispersion relation of the eigenmodes, we assume that D and B are proportional to E and H, respectively: D(r, t) = ε0 ε(r)E(r, t), B(r, t) = µ0 H(r, t).
(5) (6)
Here ε0 and µ0 are permittivity and permeability in free space, respectively. ε(r) is the position-dependent dielectric constant of the PC. In Equation (6), we assumed that the PC is composed of nonmagnetic materials and put its permeability equal to that of free space.
274
Science and Engineering of Photonic Crystals
For simplicity, we also assumed that the dielectric constant is independent of the frequency. To derive the wave equation, we eliminate the electric field from Equations (1) and (2): 1 ∂2 1 ∇ × H = − 2 2 H, ∇× ε(r) c ∂t
(7)
√ where c = 1/ ε0 µ0 is the speed of light in free space. To calculate the dispersion relation, we assume a harmonic temporal oscillation of H: H(r, t) = H0 (r)e−iωt ,
(8)
and we obtain the eigen equation: L H H0 (r) ≡ ∇ ×
1 ω2 ∇ × H0 (r) = 2 H0 (r). ε(r) c
(9)
When the dielectric constant is real, we can prove that the differential operator L H defined by the first equality in Equation (9) is an Hermitian operator. By this property, and the fact that H is transverse, the following calculation becomes efficient as we will see. To calculate the dispersion relation, we also assume that the PC has perfect periodicity, that is, we assume ε(r + ai ) = ε(r)
(i = 1, 2, 3),
(10)
where a1 , a2 , and a3 are the elementary lattice vectors. We introduce the elementary reciprocal lattice vectors bi (i = 1, 2, 3) that satisfy the following equations: ai · b j = 2π δi j ,
(11)
where δi j is Kronecker’s delta. To show them explicitly, 2π a2 × a3 , V0 2π b2 = a3 × a1 , V0 2π b3 = a1 × a2 , V0 b1 =
(12) (13) (14)
Pursuit of the 3D PBG
275
where V0 is the volume of the unit cell: V0 = a1 · (a2 × a3 ).
(15)
From the periodicity of Equation (10), 1/ε can be expressed by a Fourier series: X 1 = κ(G)eiG·r , ε(r) G
(16)
where the summation is taken over all reciprocal lattice vectors G, which are given by G = l1 b1 + l2 b2 + l3 b3
(l1 , l2 , l3 are integers).
By the inverse Fourier transform, we can derive Z 1 1 −iG·r κ(G) = dr e V0 V0 ε(r)
(17)
(18)
and can prove κ(−G) = κ ∗ (G).
(19)
By the Bloch theorem derived from the periodicity of Equation (10) (Sakoda, 2004), the magnetic field is given by the product of a periodic vector function and a plane wave: H0 (r) = vk (r)eik·r , vk (r + ai ) = vk (r).
(20) (21)
Here k is a wave vector in the first Brillouin zone. From the periodicity of vk , we can express it by a Fourier series. So, Hk is given by Hk (r) =
X
Vk (G)ei(k+G)·r .
(22)
G
Substituting Equation (22) into Equation (9), we obtain −
X G0
ω2 κ G − G0 (k + G) × (k + G0 ) × Vk (G0 ) = 2k Vk (G), c
where ωk is the eigen angular frequency for k.
(23)
276
Science and Engineering of Photonic Crystals
Substituting Equations (6) and (22) into Equation (4), we obtain X (k + G) · Vk (G)ei(k+G)·r = 0.
(24)
G
Because this equation holds for all r, the following relation must be satisfied for each G: (k + G) · Vk (G) = 0.
(25)
So, vector Vk (G) is perpendicular to k + G. We denote two unit vectors perpendicular to k + G by eG1 and eG2 and assume, without loss of generality, that three unit vectors,
k+G = eG3 , eG1 , eG2 , |k + G|
(26)
form a mutually perpendicular right-hand system. Then we have the following relations: eG1 × eG2 = eG3 ,
(27)
eG2 × eG3 = eG1 ,
(28)
eG3 × eG1 = eG2 .
(29)
Vk (G) can be expressed by a linear combination of eG1 and eG2 alone: Vk (G) = vkG1 eG1 + vkG2 eG2 .
(30)
Substituting Equation (30) into Equation (23) and using Equations (28) and (29), we obtain −|k + G|
X
0 0 κ(G − G0 )|k + G0 |eG3 × vkG 1 eG0 2 − vkG 2 eG0 1
G0
=
ωk2 G1 vk eG1 + vkG2 eG2 . 2 c
(31)
For three vectors, A, B, and C, we generally have A · (B × C) = B · (C × A) = C · (A × B).
(32)
Pursuit of the 3D PBG
277
By using these relations, we obtain eG1 · eG3 × eG0 2 = eG0 2 · (eG1 × eG3 ) = −eG2 · eG0 2 .
(33)
Similarly we have eG2 · eG3 × eG0 2 = eG1 · eG0 2 , eG1 · eG3 × eG0 1 = −eG2 · eG0 1 , eG2 · eG3 × eG0 1 = eG1 · eG0 1 .
(34) (35) (36)
Now we define a matrix Mk by Mk (G, G0 ) = |k + G||k + G0 |κ(G − G0 )
eG2 · eG0 2 , −eG2 · eG0 1 . −eG1 · eG0 2 , eG1 · eG0 1
(37)
Taking the inner product of Equation (31) with eG1 or eG2 , we obtain 2 XX
ij
G0 j
Mk (G, G0 )vk
G0 j=1
=
2 ωkn v Gi . c2 k
(38)
Since we can verify easily that ij
ji∗
Mk (G, G0 ) = Mk (G0 , G),
(39)
Mk is an Hermitian matrix. So, the calculation of the electromagnetic eigenmodes has been reduced to the diagonalization of the Hermitian matrix. The reason why we have the Hermitian matrix here is that the differential operator L H in Equation (9) is an Hermitian operator. Since there are efficient software packages and subroutines for the diagonalization of Hermitian matrices, we can obtain the eigen frequency easily. As for the eigen functions, we can show their orthogonality from that of the eigen vectors of the Hermitian matrix. First we note that the whole set of the normalized eigen vectors {Vkn }, where n is a suffix denoting the nth eigenmode, is an orthonormal set: 2 XX G i=1
Gi∗ Gi vkn · vkn 0 = δnn 0 .
(40)
Science and Engineering of Photonic Crystals
278
We denote the volume of the PC by V , on which we impose the periodic boundary condition. Then we can show the orthogonality of the eigen functions as follows: Z V
drH∗kn (r) · Hkn 0 (r)
Z dr
= V
=V
XX
0
V∗kn (G) · Vkn 0 (G0 )ei(G −G)·r
G0 ∗ Vkn (G) · Vkn 0 (G) G
X G
= V δnn 0 .
(41)
For k 6= k0 , Z V
drH∗kn (r) · H∗k0 n 0 (r)
Z
0
= V
drv∗kn (r) · vk0 n 0 (r)ei(k −k)·r .
(42)
Because both k and k0 are wave vectors in the first Brillouin zone, k0 − k is not equal to any reciprocal lattice vector. On the other hand, v∗kn · vk0 n 0 is a periodic function, and it can be expressed by a Fourier series with reciprocal lattice vectors. Thus the integrand of Equation (42) is a spatially oscillating function and the integral is exactly equal to zero. Finally from Equations (41) and (42), we obtain the orthogonality: Z V
drH∗kn (r) · Hk0 n 0 (r) = V δkk0 δnn 0 .
(43)
As for the electric field, the eigen equation, which is derived from Equations (1) and (2) by eliminating the magnetic field, is given by L E E0 (r) ≡
1 ω2 ∇ × {∇ × E0 (r)} = 2 E0 (r). ε(r) c
(44)
In contrast to the case of the magnetic field, the differential operator L E , defined by the first equality in Equation (44), is not Hermitian. So, the eigen functions of the electric field do not give an orthogonal complete set. However, the following weighted orthonormality holds (Dowling & Bowden, 1992; Glauber & Lewenstein, 1991; Sakoda & Ohtaka, 1996): Z V
drε(r)E∗kn (r) · Ek0 n 0 (r) = V δkk0 δnn 0 .
(45)
Pursuit of the 3D PBG
279
2.2. Realization of the 3D PBG Before we show examples of 3D PBGs, we briefly view some essential points of PBG formation. For this purpose, let’s consider a 1D PC. When a weak 1D periodic modulation is added to a uniform dielectric structure, the original eigenmodes of the electromagnetic field, i.e. the plane waves, are slightly modified and give new eigenmodes. We denote the lattice constant and the elementary reciprocal lattice constant by a and G(= 2π/a), respectively. When we neglect the second and higher order terms in the Fourier expansion of 1/ε(x) assuming that the periodic modulation is weak, then we obtain 1 = κ0 + κ1 eiGx + κ−1 e−iGx , ε(x)
(46)
where κ0 and κ±1 are the zeroth and the first order components, respectively. The dispersion relation of the original eigenmode, which is the plane wave with wave vector k, is simply given by ω = vk,
(47)
√ with the light velocity in the uniform material, v = c κ0 . Due to the periodic modulation of Equation (46), the original eigenmodes with wave vectors k and k ± G are mixed and give new eigenmodes in the first-order approximation. The mixing is large when the eigen frequencies of the two original modes are close. This condition is fulfilled when k ≈ ±π/a, i.e. for those wave vectors close to the Brillouin zone boundary. For example, let us assume that k is close to and smaller than π/a and introduce h = k − π/a. Then we can prove in the first-order approximation that the dispersion relation of the periodic structure around the zone boundary is given by πc p ac ω± (h) ≈ κ0 ± |κ1 | ± √ a π |κ1 | κ0
κ02
|κ1 |2 − 4
! h2.
(48)
This splitting of the eigen frequency, which generates the PBG, is illustrated in Figure 1, where solid and broken lines represent the dispersion relations of the modulated and uniform systems, respectively. To intuitively understand the mixing of the original eigenmodes and the splitting of the eigen frequency, the dispersion relations of the original system are drawn for various ks shifted by multiples of G. Around the crossing points of the broken lines, relatively strong mixing and splitting takes place because of the matching of the wave vectors and closeness
280
Science and Engineering of Photonic Crystals ω
ω = vk
– 2π a
–
π a
O
π a
2π a
k
FIGURE 1 Formation of 1D PBGs
of the eigen frequencies. In addition to the lowest PBG described by Equation (48), there is an infinite number of PBGs created by higherorder perturbations and higher-order Fourier components of the periodic dielectric constant. Now let’s go back to the case of 3D PBGs. The lowest PBG of the 3D PCs should appear around the Brillouin zone boundary like the 1D case. To create a complete 3D PBG, there has to be a common frequency range for PBGs in all propagation directions. As we saw above by the perturbation calculation, the centre frequency (ωc ) of the PBG can be given approximately by ωc ≈ v|kb |,
(49)
where kb is a wave vector on the Brillouin zone boundary. So, to create a common PBG for all propagation directions, it is desirable that |kb | does not depend so much on the direction of the wave vector. This means that the distance between the Brillouin zone boundary and the origin of the k space should remain as constant as possible, irrespective of the propagation direction, which is equivalent to saying that the shape of the Brillouin zone should be as spherical as possible. This condition is approximately satisfied by the Brillouin zone of the face-centred cubic (fcc) lattice. Figure 2 shows the first Brillouin zone of the fcc and simple
Pursuit of the 3D PBG
281
kz kz X
S U X
Δ
L Q
Γ Σ kx
R
Δ ky
Z
K
Z W
S
W T
Γ
Σ
ky
M
kx U
2π /a
FIGURE 2 The first Brillouin zone of the fcc (left) and simple cubic (right) lattices. Greek and Roman letters denote highly symmetric points and points on paths connecting them
cubic lattices, from which we can understand that the fcc Brillouin zone is much closer to a sphere. According to this idea, Yablonovitch and Gmitter (1989) fabricated an fcc lattice of air spheres composed in a uniform dielectric material with ε = 12.3, and measured its transmittance in the microwave frequency range. The lattice constant of the specimen was 12.7 mm. The transmission spectra measured for various incident angles seemed to show a common opaque frequency range, from which they concluded that they had realized a complete PBG. This was a reasonable judgment at the time when the accurate full-vectorial calculation based on the plane-wave expansion was not yet available. But soon after this report, Leung and Liu (1990), Zhang and Satpathy (1990), and Ho, Chan, and Soukoulis (1990) showed by the full-vectorial calculation that the degeneracy of eigenmodes on the W point of the Brillouin zone inhibits the presence of the PBG for the fcc lattice. In addition, Ho, Chan, and Soukoulis found that, by introducing the second atom to the unit cell to form a diamond lattice and thus slightly reducing the symmetry of the structure, the degeneracy on the W point is removed and a complete PBG can be obtained at large air fractions (see Figure 3). This remarkable theoretical finding led to the realization of the complete PBG. Yablonovitch, Gmitter, and Leung (1991) removed the degeneracy on the W point by introducing a non-symmetric atom to the unit cell of the fcc lattice and eventually achieved the complete PBG in the microwave range. Their specimen was fabricated by drilling a dielectric block from three different directions, and is now called the Yablonovite (see Figure 4).
Science and Engineering of Photonic Crystals
282
0.7 0.6
fcc
15
Frequency
Frequency (GHz)
20
10 5
Diamond
0.5 0.4 0.3 0.2 0.1
0 X U
L
Γ Wavevector
X
0.0 X U
W K
L
Γ Wavevector
(a)
X
W K
(b)
FIGURE 3 Band diagrams of (a) the fcc lattice and (b) the diamond lattice (after Ho, Chan, and Soukoulis (1990))
(a) X
W K3 U1 K1
L1
35°
120° 120°
X
L1 L1
U3
X 35°
° 120
U3
(b) FREQUENCY (in units of c/a)
35°
L3
U3
0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Γ
L3
W
K1
FIGURE 4 Fabrication of Yablonovite (left) and its photonic band structure (right). The dark area in the band diagram denotes the PBG in each direction of the k space. It is wide enough to realize a complete PBG, that is, the PBG frequency range common to all directions (after Yablonovitch, Gmitter, and Leung (1991))
Later it was also found that the complete PBG can be realized with several other types of PCs such as the woodpile structure (Fleming & Lin, 1999; Fleming, Lin, El-Kady, Biswas, & Ho, 2002; Lin, Fleming, Hetherington, Smith, Biswas, Ho, Sigalas, Zubrzycki, Kurtz, & Bur, 1998; ¨ Noda, Tomoda, Yamamoto, & Chutinan, 2000; Ozbay, Abeyta, Tuttle,
Pursuit of the 3D PBG
(a)
(b)
(c)
(d) I
283
Si II
Z [111] m
X [110] Y [112]
4n
56
FIGURE 5 Structures that realize the complete PBG. (a) woodpile structure made of tungsten (after Fleming, Lin, El-Kady, Biswas, and Ho (2002)), (b) inverse opal made of silicon (after Blanco, Chomski, Grabtchak, Ibisate, John, Leonard, Lopez, Meseguer, Miguez, Mondia, Ozin, Toader, and van Driel (2000)), (c) spiral structure (after Toader and John (2001)), and (d) 3D layered structure (after Qi, Lidorikis, Rakich, Johnson, Joannopoulos, Ippen, and Smith (2004))
¨ Tringides, Biswas, Chan, Soukoulis, & Ho, 1994; Ozbay, Michel, Tuttle, Biswas, Sigalas, & Ho, 1994), inverse opal (Blanco, Chomski, Grabtchak, Ibisate, John, Leonard, Lopez, Meseguer, Miguez, Mondia, Ozin, Toader, & van Driel, 2000; Busch & John, 1998, 1999; Norris & Vlasov, 2001; ¨ uer, ¨ Soz Haus, & Inguva, 1992) with a sufficiently large contrast of the dielectric constant, 3D spiral structure (Toader & John, 2001), and 3D layered structure (Qi, Lidorikis, Rakich, Johnson, Joannopoulos, Ippen, & Smith, 2004) (see Figure 5). Because there is the scaling law between the wavelength of the PBG and the lattice constant of the PC, we can realize 3D PBGs in the visible and infrared ranges by downsizing those structures from their size in the microwave range. This effort has actually been done and 3D PBGs in the near infrared region have been realized (Fleming & Lin, 1999; Fleming, Lin, El-Kady, Biswas, & Ho, 2002; Lin, Fleming, Hetherington, Smith, Biswas, Ho, Sigalas, Zubrzycki, Kurtz, & Bur, 1998; Noda, Tomoda, Yamamoto, & Chutinan, 2000; Qi, Lidorikis, Rakich,
284
Science and Engineering of Photonic Crystals
Johnson, Joannopoulos, Ippen, & Smith, 2004). Sometimes those PCs were fabricated with metals such as tungsten and gold (Fleming, Lin, El-Kady, Biswas, & Ho, 2002). They are capable of emitting infrared radiation by heating, and their spectra are controllable by the design of their dispersion and PBGs. Although the 3D PCs with the complete PBGs in the near infrared range have thus been realized, they are still not easy to fabricate, even with sophisticated modern lithography technologies such as electron beam (EB) lithography and focused ion beam (FIB) milling. This problem was partially solved by introducing PC slabs, which will be described in Section 5.
3. EIGENMODE SYMMETRY AND TRANSMITTANCE To check the presence of the photonic band gaps (PBGs) and more generally to observe the dispersion curves, transmission spectra of the specimens have been intensively studied. In contrast to the optical frequency range where simple experiments can only measure the transmission intensity, not only the intensity but also the phase of the transmitted wave can be measured in the microwave and terahertz ranges. Such experiments revealed the dispersion relation of photonic crystals (PCs) (McCall, Platzman, Dalichaouch, Smith, & Schultz, 1991; Meade, Brommer, Rappe, & Joannopoulos, 1992; Robertson, Arjavalingam, Meade, Brommer, Rappe, & Joannopoulos, 1992, 1993). When we examine the transmission spectra, we have to note the presence of inactive eigenmodes that do not contribute to the transmission due to the mismatching of spatial symmetry between the eigenmodes and the incident wave, which was found by Robertson et al. for two-dimensional (2D) PCs (Robertson, Arjavalingam, Meade, Brommer, Rappe, & Joannopoulos, 1992, 1993). Later a systematic theoretical study of the spatial symmetry of the eigenmodes was done, based on group theory, and all inactive eigenmodes were identified (Chap. 3 of Sakoda, 2004; Lopez-Tejeira, Ochiai, Sakoda, and Sanchez-Dehesa, 2002; Ohtaka and Tanabe, 1996; Sakoda, 1995a, 1997; Yuan, Haus, and Sakoda, 1998). In this section, we describe some essential points to note when interpreting experimental data of transmittance. The first experimental study of the inactive mode was reported by Robertson et al. for 2D PCs in the microwave range (Robertson, Arjavalingam, Meade, Brommer, Rappe, & Joannopoulos, 1992, 1993). Since the eigenmode of the 2D PC can be reduced to a scalar field, as we shall see below, the symmetry mismatching between the eigenmode and incident field is easier to understand than the 3D case, so let us start with the 2D PCs.
Eigenmode Symmetry and Transmittance
285
150 BAND 5
125
BAND 4 BAND 3
100
BAND 2
FREQUENCY (GHz)
75 50
BAND 1
25
(a) 0 150 BAND 4
125
BAND 3
100
BAND 2
75 BAND 1
50 25 SAMPLE
0 0.0
(b) 4.2
8.4
12.6
16.8
WAVEVECTOR (CM–1)
FIGURE 6 (left) Structure of the 2D specimen composed of a square array of dielectric rods, (right) calculated dispersion curves (dashed line) in the (1,0) direction and observed data (solid circles and squares) for the E polarization (upper panel) and for the H polarization (lower panel) (after Robertson, Arjavalingam, Meade, Brommer, Rappe, and Joannopoulos (1992))
The left panel of Figure 6 shows the structure of the specimen that they studied. It is composed of a square array of circular dielectric rods with a dielectric constant of 8.9 and a diameter of 0.74 mm. The lattice constant, a, of the PC is 1.87 mm. The average dielectric constant, ε, calculated from the volume fraction of the rod in the unit cell is 1.97. So, the important frequency range of the lowest PBG, which may be roughly estimated by √ c/2a ε, is around 57 GHz. The calculated dispersion curves and observed data are shown in the right panel of Figure 6. The agreement between the calculation and experiment is quite good except for the fact that two dispersion curves among nine were not observed in the experiment. This mystery was brought about by the mismatching between the spatial symmetry of the incident wave and internal eigenmodes. Before we describe this phenomenon in detail, let us examine the electromagnetic field of 2D PCs whose dielectric constant has a periodic variation in the x y plane and is uniform in the z direction. We deal with those waves with wave vectors lying in the x y plane. From these two conditions, all derivatives with respect to z vanish in Equations (1) and (2).
286
Science and Engineering of Photonic Crystals
Then we find that the six components of these two vector equations are decoupled into the following two groups:
E z , Hx , Hy ,
Hz , E x , E y .
(50)
The first one is called the E polarization, and the second one is called the H polarization. Their eigen equations are obtained by eliminating x and y components from Equations (1) and (2) and assuming a harmonic temporal oscillation: ) ( 1 ∂2 ∂2 ω2 − + E = E z : E polarization, z ε ∂x2 ∂ y2 c2 ∂ 1 ∂ ∂ 1 ∂ ω2 − + Hz = 2 Hz : H polarization. ∂x ε ∂x ∂y ε ∂y c
(51) (52)
These eigen equations can be solved by the plane wave expansion method as before, since the Bloch theorem holds for this case due to the periodicity of ε in the x y plane. The dispersion curves shown on the right-hand side of Figure 6 were obtained by this calculation. The upper panel is for the E polarization and the lower panel is for the H polarization. These two panels show that the third band was not observed for the E polarization and the fourth band was not observed for the H polarization. Now, let us go back to the description of the inactive modes. Figure 7 shows the field distribution (E z ) of four bands of the E polarization. We can see that the third band has an antisymmetric field distribution about the x axis. On the other hand, t