## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Daniel A. Steck

Oregon Center for Optics and Department of Physics, University of Oregon

Copyright © 2006, by Daniel Adam Steck. All rights reserved. This material may be distributed only subject to the terms and conditions set forth in the Open Publication License, v1.0 or later (the latest version is presently available at http://www.opencontent.org/openpub/). Distribution of substantively modiﬁed versions of this document is prohibited without the explicit permission of the copyright holder. Distribution of the work or derivative of the work in any standard (paper) book form is prohibited unless prior permission is obtained from the copyright holder. Original revision posted 16 June 2006. This is revision 1.4.5, 30 March 2010. Cite this document as: Daniel A. Steck, Classical and Modern Optics, available online at http://steck.us/teaching (revision 1.4.5, 30 March 2010). Author contact information: Daniel Steck Department of Physics 1274 University of Oregon Eugene, Oregon 97403-1274 dsteck@uoregon.edu

Contents

1 Review of Linear Algebra 1.1 Deﬁnitions . . . . . . . . . . . 1.2 Linear Transformations . . . 1.3 Matrix Arithmetic . . . . . . 1.4 Eigenvalues and Eigenvectors 1.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 9 9 10 11 12 13 13 13 14 16 17 21 21 22 23 24 25 27 35 35 37 37 39 39 41 43 45 45 46 46 47 48 49 50 50 52 53

2 Ray Optics 2.1 Introduction . . . . . . . . . . . . . . . . . . 2.2 Ray Optics and Fermat’s Principle . . . . . 2.3 Fermat’s Principle: Examples . . . . . . . . 2.4 Paraxial Rays . . . . . . . . . . . . . . . . . 2.5 Matrix Optics . . . . . . . . . . . . . . . . . 2.6 Composite Systems . . . . . . . . . . . . . . 2.6.1 Example: Thin Lens . . . . . . . . . 2.7 Resonator Stability . . . . . . . . . . . . . . 2.7.1 Stability Condition . . . . . . . . . . 2.7.2 Periodic Motion . . . . . . . . . . . 2.7.3 Resonator Stability: Standard Form 2.8 Exercises . . . . . . . . . . . . . . . . . . .

3 Fourier Analysis 3.1 Periodic Functions: Fourier Series . . . . . . . . . . . . . 3.1.1 Example: Rectiﬁed Sine Wave . . . . . . . . . . 3.2 Aperiodic Functions: Fourier Transform . . . . . . . . . 3.2.1 Example: Fourier Transform of a Gaussian Pulse 3.3 The Fourier Transform in Optics . . . . . . . . . . . . . 3.4 Delta Function . . . . . . . . . . . . . . . . . . . . . . . 3.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Review of Electromagnetic Theory 4.1 Maxwell Equations in Vacuum . . . . . . . . . . . . . 4.2 Intensity . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Maxwell Equations in Media . . . . . . . . . . . . . . 4.4 Simple Dielectric Media . . . . . . . . . . . . . . . . . 4.5 Monochromatic Waves and Complex Notation . . . . . 4.6 Intensity in Complex Notation . . . . . . . . . . . . . 4.6.1 Complex Notation for Simple Dielectric Media 4.7 Plane Waves . . . . . . . . . . . . . . . . . . . . . . . 4.8 Vector Plane Waves . . . . . . . . . . . . . . . . . . . 4.8.1 Wave Impedance . . . . . . . . . . . . . . . . . . . . . . . . . . .

4

Contents

4.9

Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

53 57 57 58 59 60 60 62 62 63 63 65 65 66 67 68 69 70 70 71 72 72 73 73 74 74 74 76 77 77 83 83 84 85 85 86 86 87 87 88 88 89 90 90 92 93 93 93 94 95

5 Interference 5.1 Superposition of Two Plane Waves . . . . . 5.2 Mach-Zehnder Interferometer . . . . . . . . 5.3 Stokes Relations . . . . . . . . . . . . . . . 5.4 Mach-Zehnder Interferometer: Applications 5.5 Michelson Interferometer . . . . . . . . . . . 5.6 Sagnac Interferometer . . . . . . . . . . . . 5.7 Interference of Two Tilted Plane Waves . . 5.8 Multiple-Wave Interference . . . . . . . . . 5.9 Exercises . . . . . . . . . . . . . . . . . . .

6 Gaussian Beams 6.1 Paraxial Wave Equation . . . . . . . . . . . . . . . . . . . . . . 6.2 Gaussian Beams . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.1 Amplitude Factor . . . . . . . . . . . . . . . . . . . . . . 6.2.2 Longitudinal Phase Factor . . . . . . . . . . . . . . . . . 6.2.3 Radial Phase Factor . . . . . . . . . . . . . . . . . . . . 6.3 Speciﬁcation of Gaussian Beams . . . . . . . . . . . . . . . . . 6.4 Vector Gaussian Beams . . . . . . . . . . . . . . . . . . . . . . 6.5 ABCD Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.1 Free-Space Propagation . . . . . . . . . . . . . . . . . . 6.5.2 Thin Optic . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.3 Cascaded Optics . . . . . . . . . . . . . . . . . . . . . . 6.5.4 Factorization of a General Matrix . . . . . . . . . . . . 6.5.5 “Deeper Meaning” of the ABCD Law . . . . . . . . . . 6.5.6 Example: Focusing of a Gaussian Beam by a Thin Lens 6.5.7 Example: Minimum Spot Size by Lens Focusing . . . . 6.6 Hermite–Gaussian Beams . . . . . . . . . . . . . . . . . . . . . 6.6.1 Doughnut Mode . . . . . . . . . . . . . . . . . . . . . . 6.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Fabry–Perot Cavities 7.1 Resonance Condition . . . . . . . . . . . . . . . . 7.2 Broadening of the Resonances: Cavity Damping . 7.2.1 Standard Form . . . . . . . . . . . . . . . 7.2.2 Maximum and Minimum Intensity . . . . 7.2.3 Width of the Resonances . . . . . . . . . 7.2.4 Survival Probability . . . . . . . . . . . . 7.2.5 Photon Lifetime . . . . . . . . . . . . . . 7.2.6 Q Factor . . . . . . . . . . . . . . . . . . 7.2.7 Example: Finesse and Q . . . . . . . . . . 7.3 Cavity Transmission . . . . . . . . . . . . . . . . 7.3.1 Reﬂected Intensity . . . . . . . . . . . . . 7.3.2 Intracavity Buildup . . . . . . . . . . . . 7.4 Optical Spectrum Analyzer . . . . . . . . . . . . 7.5 Spherical-Mirror Cavities: Gaussian Modes . . . 7.5.1 Physical Modes . . . . . . . . . . . . . . . 7.5.2 Symmetric Cavities . . . . . . . . . . . . . 7.5.3 Special Cavities . . . . . . . . . . . . . . . 7.5.4 Resonance Frequencies . . . . . . . . . . . 7.5.5 Algebraic Digression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Contents

5

7.6 7.7

Spherical-Mirror Cavities: Hermite–Gaussian Modes . . . . . . . . . . . . . . . . . . . . . . . 7.6.1 Confocal Cavity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

95 96 96 101 101 101 102 104 105 105 106 106 108 108 108 110 110 110 112 113 115 115 118 119 120 122 123 125 125 127 129 130 133 133 135 136 139 139 140 141 142 143 143 143 144 144 145 146 147

8 Polarization 8.1 Vector Plane Waves . . . . . . . . . 8.2 Polarization Ellipse . . . . . . . . . . 8.2.1 Simple Cases . . . . . . . . . 8.3 Polarization States: Jones Vectors . 8.3.1 Vector Properties . . . . . . . 8.4 Polarization Devices: Jones Matrices 8.4.1 Linear Polarizer . . . . . . . 8.4.2 Wave Retarder . . . . . . . . 8.4.3 Polarization Rotator . . . . . 8.4.4 Cascaded Systems . . . . . . 8.5 Coordinate Transformations . . . . . 8.6 Normal Modes . . . . . . . . . . . . 8.7 Polarization Materials . . . . . . . . 8.7.1 Birefringence . . . . . . . . . 8.7.2 Optical Activity . . . . . . . 8.8 Exercises . . . . . . . . . . . . . . .

9 Fresnel Relations 9.1 Optical Waves at a Dielectric Interface . . . . . 9.1.1 Phase Changes and the Brewster Angle 9.2 Reﬂectance and Transmittance . . . . . . . . . 9.3 Internal Reﬂection . . . . . . . . . . . . . . . . 9.3.1 Phase Shifts . . . . . . . . . . . . . . . . 9.4 Air-Glass Interface: Sample Numbers . . . . . . 9.5 Reﬂection at a Dielectric-Conductor Interface . 9.5.1 Propagation in a Conducting Medium . 9.5.2 Inductive Heating . . . . . . . . . . . . 9.5.3 Fresnel Relations . . . . . . . . . . . . . 9.6 Exercises . . . . . . . . . . . . . . . . . . . . .

10 Thin Films 10.1 Reﬂection-Summation Model . . . . . . . . . . . . . . . . . . 10.1.1 Example: Single Glass Plate as a Fabry–Perot Etalon 10.2 Thin Films: Matrix Formalism . . . . . . . . . . . . . . . . . 10.3 Optical Coating Design . . . . . . . . . . . . . . . . . . . . . 10.3.1 Single-Layer Antireﬂection Coating . . . . . . . . . . . 10.3.2 Two-Layer Antireﬂection Coating . . . . . . . . . . . . 10.3.3 High Reﬂector: Quarter-Wave Stack . . . . . . . . . . 10.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Fourier Analysis II: Convolution 11.1 Spatial Fourier Transforms . . . . . . . . . . . . 11.2 Convolution . . . . . . . . . . . . . . . . . . . . . 11.2.1 Example: Convolution of Box Functions . 11.3 Convolution Theorem . . . . . . . . . . . . . . . 11.3.1 Example: Convolution of Two Gaussians . 11.4 Application: Error Analysis . . . . . . . . . . . . 11.5 Application: Central Limit Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . 12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . .2 Central Dark-Ground Method . .3 Far-Field Limit . .1 Fourier Transforms in Multiple Dimensions .4. .6. . . . . . . . .2 Bragg Regime . . . . . . .1 Diﬀraction Amplitudes: Bessel Functions . . . 12. .7 Exercises . . . . . . . . . . . .1 Convolution Revisited . . . . . . . . . . . . . . . . . . . . . . . .3 Fraunhofer Diﬀraction . . . . . . . . . . .5. . . . . . . . . 12. . . . . . . . . . 12. . . . . . . . . . . . . . . . . . . . 12. . . . . . . . .1. . . . . . . 12. . . . . . . . . . .3. .5. . . . . . . . . . . . 12. . . . 148 149 149 151 152 155 155 155 155 156 156 157 157 158 158 159 160 161 162 162 162 163 163 164 165 168 170 170 171 171 174 176 177 177 178 179 179 183 184 186 188 188 189 195 195 196 196 14 Coherence 199 14. . . . . 12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Holography . . . . . . . . . . . . . . . . . . . . . . 12. . . . .3 Hologram of a Plane Wave and Oﬀ-Axis Holography . . . . . . . . . . . . . . . .3.4 Setup: Oﬀ-Axis Reﬂection Hologram .5. . 13. . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . .2 Thin Lens as a Fourier Transform Computer . . . . . . . . .3 Schlieren Method . . . . . . . . . . . . Grating . . . . . . . . . . . . . . . . 200 14.2 Impulse Response . . . . . . . . .2 Optical Wiener–Khinchin Theorem . . . . . . . .5. . . . . . . .4 Example: Fresnel Diﬀraction from a Slit . . . . . . . . . . . . . . . . . . . . . . . . . 12. . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . . 12.7 Application: Spectral Transmission . . . . . . .2 Central Limit Theorem Application: Standard Deviation of the 11. . . . . . . 12. . . . . . . 12. . . . . . . .3. . . . . . . . . .2 Wave Propagation .4 Paraxial Propagation . . Mean . . .1 Raman–Nath Regime . . . . . . .2. . . . . . . . . . . . . . . . . .1 Wiener–Khinchin Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . .3 Borderline . . . . . . 12. . . 12. . . . . . . . .5. . . . . . . . . 12. . . . . . . . . . . . . 11. . . . . . . . . . . . . . . . 12. . . . . . . . . . . . . . . . . . . . 12. . . . . . . . . . . . . . . . . . . . . . . . . . . . 13. . . . . . . . . . . . . . . . . . 12. . . . . . .2 Visualization of Phase Objects . . . . . . .5. . . 12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Example: Single-Frequency Hologram . . .6. . . .3 Fourier-Transform Recipe .2 Film Holograms . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . 13 Acousto-Optic Diffraction 13. . . . . . . . . . . . . . . . . . 12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . 13.5. . . . . . . . .1 Central Limit Theorem Application: Random Walk . . . . . . . . . . . . . . .6. . . . . . 12. . . . . . 11. . . . . . . . .5. . . . . . . . 13. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . .4 Example: Diﬀraction from a Sinusoidal Intensity-Mask 12.2 Frequency Shifts . . . . . . . . . . . . . . . . . . . .6. . 12. . . . . . . . . . . . 12. . . . . . . 12. . . . . .5 Example: Diﬀraction from an Arbitrary Grating .1. . . . . . . . . . . . . . . . . . . . . . . . . .1 Zernike Phase-Contrast Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . 13. . . . 201 . . . . . . . . . . . . . . . . . . . . . . . .4 Numerical Examples .2. .2 Decomposition . . . . . .2. . . . . . . . . . . . .4 Fresnel Diﬀraction . . . . . . . . . . .1 Eﬃciency .1 Fingerprints of Propagation . . .2. . . . .1. . . . . . . . . . . . . . . . . 12. . . . . . . . . . . . . . . . . . .1 Far-ﬁeld Propagation . . . . . . . . . . . . . . . . . . . . 12. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Application: Impulse Response and Green’s Functions . . . . .5 Nonparaxial Propagation and the Diﬀraction Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13. . . . . . . . . .1 Spatial Filtering of a Gaussian Beam . . . . . . . . . . 12 Fourier Optics 12. . . . 12.3 Momentum Conservation . . . . . . . . . . . .5 Spatial Filters . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . 11. . . . . . . . . . . 13. . .3 Example: Diﬀraction from a Double Slit . . . . . . . . . . . . . . . . . . . .4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Contents 11. . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Example: TeO2 Modulator (Bragg Regime) 13. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . .8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12. . . . . . . . . . . . . .

. . . . . .2. . . . . . . . .1 Example: Lorentzian Absorption . . . . . . 15. . . . . . . . . . . . . . . .4. . . . . . . . . . . .2 Q-Switching . . . . . . . . 15. . .5. . . . . . 15. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 Laser Output: Pulsed . . . .4 Pumping Schemes . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Laser Spiking . .5 14. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15. . . . 15. . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . .4 Mode Locking . . . . . . . . . . . .1 Application: FTIR Spectroscopy . . . . . . . . . . . . . . . . . . . . 15. . . . . . . .3. . . . . . . . .3 Einstein Rate Equations . .1 Refractive Index . .1 Causality and the Kramers–Kronig Relations . . . . .3 Optical Resonator . . . . . . . . . . . . . . . . . . . . . 15.4 Relations Between the Einstein Coeﬃcients . . . . . . . . . . . . . . . . . . . . . .7. . . . . . and Uncertainty Interference Between Two Partially Coherent Sources Exercises . . . . 15. . . . . . . . . . . . . . . . .2. . . . . . . . . . . 15. . . . . . . .7. . . . . . . . . . . . . . . . . . . 15.1. . . . 15. . . . . . . . . .4 14. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . 15. . . . .1 Overview . . . . . . . . . . . . . . . . .2 Fundamental Light–Atom Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Broadband Light . . . . . . . 15. .6. . . . . . . . . . . . . . . . . . . . . . 16 Dispersion and Wave Propagation 16. 15. . . . . .1. . . . . . . . . .2. . . . . . . . . . . . . . . . . .3 Molecules . . . . . . .2 Absorption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Nearly Monochromatic Light . . 15. . . . . . . . .8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Four-Level Laser . . . . . . 15. . . . . .1. . . 15. . . . . . . . .1. . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . .3 14. . . . . . .3. . .7. . . . . . . . . .2. . . . . . . . . . . . . . . . . . . 15. . . . . . . . . . . . . . . . . . . . . . . . 15. . . .5 Line Shape and Spectral Distributions . . . .1. . . .1. . . . . . . . 202 202 203 204 206 206 207 207 207 207 208 208 208 209 209 209 210 211 211 212 213 214 214 215 215 216 216 216 217 217 217 217 219 219 221 222 224 225 225 226 227 227 228 229 230 231 239 239 241 241 242 15 Laser Physics 15. . . . . . . . . .3. .3 Cavity Dumper . . . . . .5 Gain Coeﬃcient . . . . . . . . . . . . . . . . .5. . . . . . . . . . . . . .3 Light Ampliﬁcation . . . . . . . . . . . . .5 A Less-Simple Model of Laser Oscillation: Steady-State Oscillation 15. . . . .2 Atoms Embedded in Transparent Solids . Coherence Time. . 15. . . . . . . . . . . . . . . . . . . . . . . 15. . . . . . . . Coherence Length. . . . . . 15. . . . . . 15. . . . . . . . . .1. . . .2 Gain Media . . . . . . . . . . . . . . . . . . .2. . .2. . . . 15. . . . . . . . 15. . . . . . . . . . . .1 Optimum Output . . . . . .3. . . .6 14.1. . . . . . . . . . . . . . . . . . . . . . . . 15. . . . . . .1 Laser Pumps . . . . . . . . . .1 Three-Level Laser . . . . . . . . . . . . . . . . 15. . . . Visibility . . . . . . . . . . . . . . . . . Measures . . . . . . . 15.2. . . . . . . . . .2 Quantum Eﬃciency . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Quantization . . . . . . . . . . . . . . . . . . 16. . . . 15. . 16. . . . . . 15. . . . . . . . . . . . . . . . . . . . . . . . . .1 Gain Coeﬃcient . . .1. . . . . . . . 15. . . . . . . . . . . . . .2 Threshold Behavior and Single-Mode Operation . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . .1. . 15. . . . . . . . . . . . . . . . . . . . . . .1. .3. . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Spontaneous Emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15. . . . . . . 15. . . .4 A Simple Model of Laser Oscillation: Threshold Behavior . . 14. .2 Example: Monochromatic Light . . . . . . . . . . . . . . . . . . .5.Contents 7 14. . . 15. .1 Gain in a Medium of Finite Length . . . . . .4.4 Combined Eﬀects .1. . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . .4 Semiconductor Lasers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6. . . . . . . . . . . . . .6 Laser Output: CW . . . . . . . . . . . . . . . . . . . . . . . . . .0. . . . . .1 DC Component . . . . . . . . . . . . . . . . . . . . . . . . .2 Light–Atom Interactions . . . . . . . 15. . . . . .2. . 15. . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . .1 Gas-Phase Atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7. . . .1 Stimulated Emission . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . .

. . . . 17.3 Laser Cooling: Optical Molasses . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . .3 Slow and Fast Light . . . . . . . . . . . 16.2 Damping Coeﬃcient . . . . . . . . . . . . .2. 17. . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . 17. . .3. .2. . . . 16. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Pulse Spreading . . . 17. .1. . . . . . . . . . . . . . . . . . 16. . . . . . . . . 243 243 243 245 247 248 249 250 253 253 254 255 255 257 258 258 259 260 261 262 264 265 266 266 268 270 270 17 Classical Light–Atom Interactions 17. . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . 17. . 16. . . . . . . . . . . . . . . .1 Connection to Dielectric Media . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . 17. . . . . .2 Magneto-Optical Trap . . . . . . . . . . . .2. . .1 Dipole Force . . .3. . . . . . . . . . . . . . . .3. . .3. . . 17. . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . 17. . . . . . . . . . . . .1 Oscillator Strength . . . .1 Phase Velocity . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8 Contents 16. . . . 17. . . . . . . . 17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17. . . . . . . . . . . . . 17.2 Conducting Media: Plasma Model . . 17. . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . .2 Pulse Propagation and Group Velocity .3. . . . . . . . . . . . .4 Exercises .1 Doppler Cooling Limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16. . . . . . . . . . . .2. . . . . . . . . . .1 Dipole Radiation . . . . . . . . . . . . . . .2. 17. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . .4 Scattering Force . . . . .2 Damping: Lorentz Model . . . . . . 17. . . . . . . . . . . . . . . . . . . . . . . . . .1 Quantum Coherence: Slow Light 16. . . . . . .3. . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . .2 Fast Light . . . . . . . . . . . . . . . . . . . . . . . .2 Group Velocity . . . . . . . . . . . . . . . . .1 Dipole Potential: Standard Form . . . . . . . . . . . . . . .2 Conductor with Damping: Drude Model . . . 16. . . . . . . . . . .3 Atom Optics: Mechanical Eﬀects of Light on Atoms 17. . . . . . . 17. . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . .1 Polarizability . . . . .3. . . . . . . . . .3 Photon Scattering Rate . .4 Exercises . . . .3. . . . .2. . . . . . . . . . . . . . . . . . . .2 Radiation Pressure . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . .

x + (y + z) = (x + y) + z 2. speciﬁcally an m × n matrix. we can also write out all the elements of a matrix.1 Deﬁnitions Before tackling optics formalism directly. We will denote a matrix by a boldface letter. For example. a complex-valued matrix is a function mapping m × n −→ C. 1. and we will denote elements of the matrix (i.2 Linear Transformations Why are matrices so important? They deﬁne linear transformations between vector spaces. we need to deﬁne some more fundamental mathematical concepts. A vector space V is a set of elements (vectors) with closed operations + and · that satisfy the following axioms: 1. Of course. we will spend a bit of time reviewing some of the mathematical concepts that we will need. . results of the function) by Aij . as a function mapping m × n −→ R. • The cartesian product A × B of two sets A and B is the set of all ordered pairs a. • A function f : A −→ B (“f maps A into B”) is a subset of A × B such that if a. Now we can deﬁne a matrix. functions of the two integer indices.e. (1. the fundamental object in linear algebra. this is for a real matrix. n}. f maps each element in A to a unique element in B. which is slightly unconventional for set theory (but suﬃcient for our purposes). • A set is a collection of objects. speciﬁcally the basics of linear algebra. We write a ∈ A to say that the element a belongs to the set A. c ∈ f then b = c.. . a 2 × 2 matrix would look like A11 A12 . while the second refers to the column. b such that a ∈ A and b ∈ B. . • A bit of notation. such as A. Again. b ∈ f and a.1) A21 A22 where note that by convention the ﬁrst index refers to the row.z∈V . ∃e∈V (x + e = x) ∀x. we are writing real numbers. Of course.y.Chapter 1 Review of Linear Algebra 1. Obviously to get started we will need to deﬁne what we mean by a matrix. This is a naive deﬁnition that can have problems. But in order to do so in a rigorous way. That is. but it will suﬃce for our purposes. where i ∈ m and j ∈ n. . or elements. We will use the integer n as a shorthand for the set {1.

We will most commonly think of ﬁnite-dimensional vectors as (n × 1)-dimensional matrices (singlecolumn matrices).y∈V ∀a. linear approximations are much easier to understand and give insight into the more complicated case. and satisﬁes AA−1 = A−1 A = In .10 Chapter 1. • The inverse matrix of an n × n matrix A is denoted A−1 .x∈V All of these properties make intuitive sense for the “usual” ﬁnite-dimensional vectors in Rn .e. (a + b) · x = (a · x) + (b · x) 7. 1. 1 · x = x ∀x∈V ∀a. ∀x∈V ∃x−1 ∈V (x + x−1 = e) 4. Review of Linear Algebra 3.x. A and B must have the same dimension.b∈R. (ab) · x = a · (b · x) 8. a · (x + y) = (a · x) + (a · y) 6.3 Matrix Arithmetic We will now deﬁne a few fundamental mathematical operations with matrices. • The transpose matrix of a matrix A is given by (AT )ij = Aji . and a matrix for which the inverse exists is said to be invertible or nonsingular.. Many simple problems that you study in physics are linear. where δij is the Kronecker delta (δij = 1 if i = j and 0 otherwise). Even in the much more diﬃcult nonlinear cases. for this to work. Clearly if A is m × n then AT is n × m.y∈V ∀a∈R. i. if A is m × n then B must have dimension n × p for some integer p. L(x + y) = L(x) + L(y) 2. Clearly. Linear transformations are of great importance in physics because they are tractable.b∈R. • A matrix C is the sum of matrices A and B (C = A + B) if Cij = Aij + Bij for all i and j.2) for all i and j. and they can be readily solved. • The identity matrix In is an n × n matrix deﬁned by (In )ij = δij . Clearly.x∈V ∀x. but the point is that more general objects such as functions and derivative operators can live in vector spaces as well. A linear transformation L of a vector space V into another vector space W is a function L : V −→ W such that: 1.y∈V The point is that any linear transformation between ﬁnite-dimensional vector spaces can be represented by a matrix (exercise: show this). for this to work. L(ax) = aL(x) ∀a∈R. x + y = y + x ∀x. matrices can deﬁne any linear transformation between ﬁnite-dimensional vector spaces. The inverse matrix is not guaranteed to exist. • A matrix C is the product of matrices A and B (C = AB) if Cij = k Aik Bkj (1. and the product C will have dimension m × p. .x∈V 5.

Then P−1 AP is a diagonal matrix. we obtain an eigenvalue equation. .. and eigenvectors typically represent physically important elements of vector spaces (typically providing a physically signiﬁcant basis for a vector space). consider a system of coupled oscillators (mechanical or otherwise). When coupled together. 2.. can be obtained from (1. symmetric matrices with distinct eigenvalues) and normalized. . . While this serves as a formal deﬁnition of the determinant. . The eigenvector corresponding to an eigenvalue λ can then be found by solving the homogenous linear system (A−λIn )·x = 0. . However. . .4 Eigenvalues and Eigenvectors 11 Finally.. n) by exchanging pairs of numbers an even number of times).. .3) where the permutation symbol εj1 j2 . it is −1 if (j1 .e. As a simple example. it is not possible to write any one of them as a linear combination of the others). .4 Eigenvalues and Eigenvectors Let A be an n × n matrix. Anjn . jn ) is an even permutation of (1.6) Upon making the ansatz x(t) = x(0)e−iωt . and thus the diagonal matrix is PT AP.. and is 0 otherwise (i. if any number is repeated). note that the eigenvalue condition above implies that (A − λIn ) · x = 0. How do we ﬁnd the eigenvalues and eigenvectors? First. we will deﬁne the determinant of a matrix. (1.. it is unwieldy except for small matrices. (1. . If the eigenvectors are furthermore mutually orthogonal (as is the case for real. (1. 2. where the determinant is given by det A C B D = AD − BC. We want to consider cases where A · x = λx. then we have PT = P−1 (i. n) (i.e. If there are λ and x that satisfy this relation then λ is said to be an eigenvalue of A with corresponding eigenvector x. so that A − λIn is a singular matrix. we have the condition that det(A − λIn ) = 0. while the eigenvectors represent how diﬀerent oscillators move together to make each distinct “mode” of oscillation corresponding to each frequency.9) Here. . .. 1. ¨ (1. . . For a 2 × 2 matrix A. Let A be an n × n matrix. In fact the diagonal elements are the eigenvalues. . the eigenvalues of A represent the distinct frequencies of oscillation for the coupled system.5) where x is an n-dimensional vector and λ ∈ R. Mathematically. then they form a “nice basis” for the vector space in the sense that in this basis.jn is +1 if (j1 .8) which yields the characteristic polynomial in λ. j2 .. If the eigenvectors are linearly independent (i. A · x = −ω 2 x. .4) The determinant has the important property that it is nonzero if and only if the matrix is nonsingular.e. real-valued matrix to a real number.jn εj1 j2 . (1. . Each uncoupled oscillator would satisfy an equation of the form x = −ω 2 x.1. . . The eigenvalues are the roots of the characteristic polynomial. n).. Why should we consider eigenproblems? Eigenvalues typically represent physically important values of physical quantities. Thus..e. the linear transformation represented by A is now represented by a diagonal matrix. Then the determinant is det(A) := j1 j2 . something which maps a square. 2. P is an orthogonal matrix). . .7) Thus. jn ) is an odd permutation of (1. we will be most concerned with 2 × 2 matrices in optics. deﬁned as the sum over the diagonal elements. let P be a matrix such that the columns are eigenvectors of A. Tr(A) is the trace of the matrix. (1.jn A1j1 A2j2 . (1. . j2 . the characteristic polynomial is simple (it is handy to know this): λ2 − Tr(A)λ + det(A) = 0. the system of oscillators would more generally satisfy a matrix ¨ equation of the form x = A · x. .

Let M and N be vector spaces of dimension M and N . A(N ) is invariant under cyclic permutation of the product. show that there is a matrix A such that A · x = φ(x) for every vector x ∈ M. Let A be an n × n matrix. Tr [AB] = Tr [BA] . Further. the characteristic polynomial is given by λ2 − Tr(A)λ + det(A) = 0.12 Chapter 1. .2. respectively. . Show that det(A) = (−1)n det(−A). and let φ : M −→ N be a linear transformation. . Problem 1.3.5 Exercises Problem 1. Show that φ can be represented by an N × M matrix.2 = Tr(A) ± 2 Tr(A) 2 2 − det(A). that is.4. . Show that if A is a 2 × 2 matrix. A(2) . show explicitly that if det(A) = 1. .1. Problem 1. Let A and B be n × n matrices. Problem 1. Prove that the trace of the product is order-invariant. Review of Linear Algebra 1. and thus that the eigenvalues are given by λ1. (b) Prove that the trace of the product of n × n matrices A(1) . then λ2 = 1/λ1 . (a) The trace of a square matrix A is deﬁned by Tr [A] := j Ajj .

where c is the vacuum speed of light (deﬁned to be exactly 2. and thus we can characterize them completely by their index of refraction. a single ray could represent a beam of light. The “mathematical” deﬁnition is simply that the ray is a path. The speed of light in a medium of refractive index n is simply c/n. but typically you would use many rays to model light propagation (e. B r(s) A Then the optical path length for this path is the length of the path. The only eﬀect that we require of the refractive index is that it changes the speed of light. it also misses many phenomena. called Fermat’s Principle. which we will denote by n. slowly diverging beam of light. with n = 1 corresponding to vacuum. The more intuitive “deﬁnition” is that a ray represents the center of a thin. Usually. meaning that the refractive index is uniform throughout the medium. parameterized by the variable s.Chapter 2 Ray Optics 2. but weighted by the local refractive index. To keep things simple. Mathematically. The optical rays propagate in optical media.997 924 58 × 108 m/s).1 Introduction Ray optics.. Again. many media also have refractive indices n(r) that vary spatially.g. (2. we will assume that the media are lossless. we will start by noting that the fundamental assumption in ray optics is that light travels in the form of rays. But even when we get to wave optics. is the simplest theory of optics. which we will get to very shortly. 2. Then Fermat’s Principle states that optical rays traverse paths that satisfy δℓ = 0.2) . But ﬁrst. geometrical optics will still provide a lot of basic intuition that we need to understand the more complex behavior of optical waves. Consider a path inside an optical medium between points A and B. (Fermat’s Principle) (2. And while many media are uniform. as mentioned above. Now to the fundamental principle of ray optics. While this theory gets a lot of things right. to model the performance of an imaging system). or geometrical optics.2 Ray Optics and Fermat’s Principle Ray optics boils down to a single statement.1) This quantity is proportional to the time light takes to traverse the path: ∆t = ℓ/c. n ≥ 1. we can deﬁne the optical path length functional as B ℓ[r] := A n(r) ds.

but we won’t really go much more into this here. 1. This quantity is thus minimized for a straight line connecting A and B. the condition δℓ = 0 yields a minimum for ℓ. this isn’t necessarily the case.6) . and this is one of the motivations for Fermat’s Principle. Consider a nearby point P ′ .14 Chapter 2. What the statement δℓ = 0 means is that nearby paths have the same path length. Often. Formally.3) at least to ﬁrst order in ǫ. Obviously δℓ = 0 can never yield a true global maximum. = a2 ε−b + + (b − ε)2 a2 ε+b . Ray Optics The “δ” here is like a derivative. Often. Given an stationary path. The length of AP ′ A′ is ℓ = n ∂ ∂ε ℓ n a2 + (b − ε)2 + a2 + (b + ε)2 . 2. A (-a. there is always a nearby path that is slightly longer. Hence Fermat’s Principle is often referred to as a “principle of least time. -b) P ¢ (0. B dl = 0 r(s) A In this case the optical path length is simply B B A ℓ=n A ds = nd. e) P We can guess from symmetry that AP A′ is the minimum-length path (of the paths that bounce oﬀ the mirror). Plane Mirror. where δr is an arbitrary path satisfying δr(A) = δr(B) = 0. but for functions. where ℓ is a minimum along one direction but a maximum along another. (2. the stationary condition turns out to be an inﬂection point or more commonly a saddle point. (2. However. But let’s prove it. it is for this reason that Fermat’s Principle is called a variational principle. It turns out that variational principles are extremely important in many areas of physics. as happens in standard calculus. + (b + ε)2 (2.” However.3 Fermat’s Principle: Examples We will now consider some applications of Fermat’s Principle. We know experimentally that light travels in straight lines. The proper mathematical framework for all this is the calculus of variations.5) Diﬀerentiating with respect to the perturbation ε. (2. this means that ℓ[r + ǫδr] − ℓ[r] = 0. Homogenous Medium.4) where d is the regular length of the path. b) qi qr A¢ (-a. 2.

. n¡ n™ C (c. b) q2 q1 A (0. the ray splits. The path length is ℓ = n1 a2 + b 2 + n 2 (c − a)2 + (d − b)2 . 4. n2 (2. Thus. The other. we arrive at the extremal condition ∂ℓ n1 b − = √ ∂b a2 + b 2 (c − a)2 + (d − b)2 n2 (d − b) = 0.7) thus arriving at the Reﬂection Law for optical rays.12) such that if θ1 > θc .10) If n1 > n2 then there is a critical angle θc given by n1 sin θc = 1. All the light is instead reﬂected. (2.e. Note that sin θ2 = n1 sin θ1 ≤ 1. The reﬂected ray behaves according to the law of reﬂection that we just derived. then there is no possible transmitted ray. δ(ABC) = 0.2.9) Using the angles marked in the diagram.3 Fermat’s Principle: Examples 15 This quantity vanishes for ε = 0. refracted ray is a little diﬀerent. and this phenomenon is called total internal reﬂection. and so the point P represents the extremal path.. we take both A and C to be ﬁxed. we can rewrite this condition as n1 sin θ1 − n2 sin θ2 = 0. since we really need full electromagnetism to show this from ﬁrst principles. 0) At a planar interface between two optical media of diﬀerent refractive index. We have to assume this as an experimental fact at this point. We will assume it crosses the refractive interface at point B. (2. and now we will compute where the point B must be according to Fermat’s Principle.8) Diﬀerentiating with respect to the moveable coordinate b of B. Refractive Interface. 3.11) (Snell’s Law) (2. which is simply Snell’s Law. Spherical Mirror. n2 (2. Assume that the refracted ray begins at point A and ends at point C (i. d) qr B (a. (Law of Reﬂection) (2. we conclude that θi = θr .

we know that AP A′ is constant. a spherical mirror focuses rays from an object at the center point back onto itself.e. Ray Optics All rays from the center to the outer edge and back have the same optical path length ℓ (i.4 Paraxial Rays Now we will set up a formalism for keeping track of optical rays more precisely. Elliptical Mirror.. which is the focus of the parabola. 5. In the other direction. We will represent a ray by a vector.16 Chapter 2. all incoming parallel rays orthogonal to the directrix are concentrated to the focus A. a parabolic mirror collimates all rays starting at A. A¢¢ A P A¢ directrix We can deﬁne a parabola as the set of all points {P : AP = P A′ where P A′ ⊥ directrix}. Fermat’s Principle tells us immediately that since for any point P on the ellipse. Thus. and thus rays starting at the point A will end at the point A′ . 2. which keeps track of the position and direction of the ray with respect to the optical axis. Thus an elliptical mirror images an object at A to the other focus A′ . So AP A′′ is constant for all rays. P A A¢ We can deﬁne an ellipse as the set of all points {P : AP A′ = d} for some constant distance d. the minimum ℓ for reﬂected rays). The points A and A′ are the foci of the ellipse. 6. where P A = A′′ P . Thus. y y q oà y ¢ optical axis z . Parabolic Mirror.

Then we can expand the function to lowest order in y1 and y1 : ′ y2 = f1 (y1 . y¡ ¢ q¡ oà y¡ OA y™ optical system ¢ q™ oà y™ In the most general case. 2. y1 ). we can write ′ y2 = f1 (y1 . The higher order corrections that we are neglected are treated in aberration theory. y1 ) = ∂f1 ∂y1 y1 + ′ y1 =y1 =0 ∂f1 ′ ∂y1 ′ ′ y1 + quadratic terms in y1 . (2. (2. which we will not treat here. y1 ).2. which justiﬁes our interchangeable use of θ and y ). ′ y1 + ′ y1 =y1 =0 Then we can rewrite this expansion in matrix ∂f1 ∂f1 ′ y2 ∂y1 ∂y1 = ′ y2 ∂f2 ∂f2 ∂y1 ′ ∂y1 form: y1 ′ y1 ′ + quadratic terms in y1 .14) or in vector form. y1 ) ∂f2 = ∂y1 ∂f2 y1 + ′ ∂y1 y1 =y ′ =0 1 (2.5 Matrix Optics Recall from before that the most general linear transformation of a two-dimensional vector is a 2 × 2 matrix.5 Matrix Optics 17 In the paraxial approximation that we will get to below. y1 . θα . so that θα ≈ ′ ′ sin θα ≈ tan θα = yα . We have written a matrix above as an expansion of a more general transformation. so we will write vectors interchangeably with the angle and slope of the ray: y y ≈ . y1 (2. y1 ′ y1 =y1 =0 ′ y2 = ′ f2 (y1 .17) ′ y1 =y1 =0 In the paraxial approximation. the ray must always stay close to the optical axis. ′ ′ Assume that yα and yα are small. ′ ′ y2 = f2 (y1 . This approximation is valid for small yα and yα (or equivalently. but for the general paraxial case we will use the notation y2 ′ y2 = A C B D y1 ′ y1 . we will see that the angle θ is approximately equivalent to the slope y ′ of the ray.18) . (2. y2 ′ y2 =f y1 ′ y1 .16) quadratic terms in ′ y1 .15) where the vector function f models the optical system. we will ignore the quadratic terms and model the optical system using only ′ linear transformations. (2. Note that for this approximation to be valid. In this sense.13) θ y′ We want to be able to compute the change in the ray vector for any optical system. we will model an optical system as a transformation of ray vectors.

Thus. Thus. however. We will now derive the fundamental matrices.20) ′ By comparison to y2 = Ay1 + By1 . Thin Lens. y¡ y¡¢ d y™ y™¢ OA M= 1 d 0 1 One of the simplest cases is propagation in free space over a distance d.21) 0 1 2. Since ′ the slope is y1 . so y2 = y1 . A = 1 and B = 0. the ray does not propagate over any distance. The ray is deﬂected. The ray is continuous. (2.2 do y¡¢ y™¢ di 1 M= 1 − f 0 1 Because the lens is thin. the single parameter that completely characterizes a thin lens. and f < 0 for a concave lens.” “ray matrix. The sign convention for the focal length is that f > 0 for a convex lens. such that it satisﬁes the thin lens law: 1 1 1 + = . y 1. (2. and f is the focal length. (2. we can conclude that C = 0 and D = 1.19) ′ ′ By comparison to the matrix equation y2 = Cy1 + Dy1 . di is the image distance. The ray travels in a straight line.18 Chapter 2. Free-Space Propagation. we can conclude that A = 1 and B = d. (2.” or “raytransfer matrix”. 1. so the angle does not change ′ ′ y2 = y1 . as shown here: convex lens (f > 0): or concave lens (f < 0): or . Ray Optics The matrix representing the optical system is referred to as an “ABCD matrix. the free-space matrix is simply 1 d M= . the position changes according to ′ y2 = y1 + dy1 .22) do di f Here do is the object distance.

do (2. 4. then the mirror is really equivalent to nothing. q™ = q¡ q¡ OA y¡¢ y™¢ = y¡¢ OA M= 1 0 0 1 Again. (2. this is a thin optic. and we can write the ray matrix for a thin lens as 1 0 . Thus C = −1/f and D = 1. f (2. where the (z) optical axis always points in the general direction of the ray travel.5 Matrix Optics 19 The line drawings shown to the right in the above ﬁgure are common schematic representations of convex and concave lenses in diagrams. q¡ j -q™ (-R) y¡ = y™ 1 M= 2 OA R concave (shown): R convex: R 0 1 <0 >0 .25) M= 1 1 − f 3. Spherical Mirror. Plane Mirror. Thus. (2. The reﬂection law says that θ2 = θ1 . respectively.2. as shown schematically in the above sketch. so y2 = y1 .24) where we used y2 = y1 and the thin lens law to eliminate y2 . Thus the ray matrix is simply the identity matrix: 1 0 M= . we can write the initial slope as ′ y1 = y1 . we can take the object and image distances to be where the ray crosses the axis before and after the lens. what is the eﬀect of a planar mirror? Really it is just to reverse the direction of propagation.23) and similarly we can write ′ y2 = − y2 y1 = − = −y1 di di 1 1 − f do =− y1 ′ + y1 . To arrive at the rest of the ray-matrix elements. If we adopt an “unfolded” convention.26) 0 1 So if the ray matrix is the identity.

We can also write the angle with the radius line as y1 ϕ= .29) Comparing this matrix to the thin-lens matrix.e. (2. we can write the ray matrix as M= 1 2 R 0 1 . 5. We will also mark the angle shown as (−θ2 ) because the ray.31) n¡ n™ 1 0 0 n1 n2 M= 6. We can rewrite this as θ2 = θ1 − 2ϕ = θ1 + Thus. Ray Optics For the spherical mirror. we use the sign convention that R < 0 for a concave mirror (as shown here) and R > 0 for a convex mirror. n¡ R n™ 1 M = (n2 − n1 ) − n2 R 0 n1 n2 convex (shown): R > 0 concave: R < 0 .28) (2. as it is drawn.30) 2y1 . a spherical mirror is equivalent to a thin lens with a focal length f =− with.20 Chapter 2. the reversal of the optical axis.27) −R The Law of Reﬂection implies that the angles on either side of the radius line are equal: ϕ − θ1 = −θ2 − ϕ. the mirror is a thin optic). 2 (2. points downward (compare to the plane mirror sketch).. Thus we will use (−R) > 0 in the ﬁgure. R . of course. Again. R (2. Planar Refractive Interface. (2. we see that in the paraxial approximation. y2 = y1 in the paraxial approximation (i. Spherical Refractive Interface.

(2. so it must be the rightmost in the product.32) = M1 ′ ′ y0 y1 We can repeat this for the second optical element: y2 ′ y2 = M2 y1 ′ y1 = M2 M1 y0 ′ y0 .6 Composite Systems Now we can consider more general optical systems.6. (2. Note the right-to-left ordering of the product.36) M= (n2 − 1) 1 1 − R2 R1 .35) 2. we can start on the ﬁrst component by applying the ﬁrst matrix: y0 y1 . (2. n¡ = 1 R¡ The composite matrix is thus 1 0 1 M = (n1 − n2 ) n2 (n2 − n1 ) − − n1 R2 n1 n2 R1 If we take n1 = 1. (2. this simpliﬁes to n™ > 1 R™ 1 0 1 1 n1 = (n2 − n1 ) − n1 R2 R1 n2 1 0 1 0 1 .37) . y0 y′ 0 y1 y′ 1 y2 y′ 2 yn−1 y′ n−1 yn y′ n OA M¡ M™ M£ Mñ Mcomposite = Mn Mn−1 · · · M2 M1 When propagating the ray through the composite system. or composite optical systems made up of the more basic optical elements. the ray matrix of a composite systems is simply the product of the individual ray matrices.34) So. Since the lens is thin. (2.1 Example: Thin Lens We can regard a thin lens as a composition of two cascaded refractive interfaces. we can arrive at the transformation for the entire system: yn ′ yn = Mn Mn−1 · · · M2 M1 y0 ′ y0 =: Mcomposite y0 ′ y0 .6 Composite Systems 21 2. (2. we assume that there is no distance between the interfaces.33) Iterating this procedure.2. Mcomposite = Mn Mn−1 · · · M2 M1 M1 acts ﬁrst on the input ray.

let’s look at a resonator composed of two spherical mirrors separated by a distance d: R™ R¡ d It is easier to analyze this if we “unwrap” the system into an equivalent waveguide of lenses as follows: f™ = -R™/2 f¡ = -R¡/2 d d This is just the “unit cell” of the waveguide. which means that 1 = −(n2 − 1) f and thus a negative focal length.39) and thus a positive focal length. or optical systems that trap light rays.22 Chapter 2. we can equate the C matrix entry and write 1 = (n2 − 1) f 1 1 − R2 R1 . which follows the path of the ray through the cavity/waveguide: ﬁrst (the rightmost matrix) is left-to-right propagation of distance d.2 /2|. The matrix for one round trip (or the waveguide unit cell) is the product of two free-space propagation matrices and two thin-lens (spherical mirror) matrices: 1 0 1 0 A B 1 d 1 d . (convex lens) (2. where light often needs to pass through a gain medium many times.. For a convex lens. i. for 2 concave mirrors. For a concave lens.40) 2. 1 1 (2. f1. Such things are very important for the operation of lasers. . as we’ll get to later.7 Resonator Stability We want to consider resonators. then reﬂection oﬀ the right mirror. we have R1 ≥ 0 and R2 ≤ 0.38) which is known as the Lensmaker’s Formula.e.41) M= = 1 1 − − 0 1 0 1 C D f1 f2 Note the proper order of multiplication. As a basic example. As we have drawn it. we have R1 ≤ 0 and R2 ≥ 0. which means that 1 = (n2 − 1) f 1 1 + |R2 | |R1 | 1 1 + |R2 | |R1 | =⇒ f > 0. which repeats over and over again for each round-trip of the ray in the cavity. Ray Optics If we compare this to the standard thin-lens matrix. or for interferometry. (Lensmaker’s Formula) (2.2 = |R1. We have exploited the equivalence of spherical mirrors and thin lenses here. (convex lens) (2. =⇒ f < 0. The sign conventions work out as follows.

+ − + − (2.44) For a ray matrix. M and after n passes.46) where β := Tr(M)/2 = (A + D)/2.45) where n1 is the refractive index at the input of the optical system and n2 is the refractive index at the output. given by λ± = β ± β 2 − 1.7 Resonator Stability 23 right-to-left propagation. Recall that the characteristic polynomial for a 2 × 2 matrix is λ2 − Tr(M)λ + det(M) = 0.1 Stability Condition The question we want to ask now is. and the vectors [y± y± ]T are the eigenvectors corresponding to λ± : M Then after one round trip in the cavity.48) ′ for some constants α± . we get d d 1− d 2− f2 f2 M= (2. (2. Thus. is the cavity stable? Consider the ray after n round trips in the cavity: yn ′ yn = A C B D n y0 ′ y0 (2.43) To answer this question. Multiplying this all out. the characteristic polynomial becomes λ2 − Tr(M)λ + 1 = 0. 2. does the resonator conﬁne the ray? In other words.50) = α+ λn + + α− λn − Let’s simplify things a bit by only considering the positions yn . The eigenvalues of M are the roots of this polynomial.2. In particular.51) . yn ′ yn = Mn y0 ′ y0 y0 ′ y0 y± ′ y± = λ± y± ′ y± . (2. the input and output are exactly the same place. there is a general result that states that det(M) = n1 . Then we can write yn = (α+ y+ )λn + (α− y− )λn =: γ+ λn + γ− λn . For the matrix describing a single pass through a resonator.47) (2. thus n1 = n2 and det(M) = 1. (2. So how does this help? Remember that we can decompose an arbitrary vector into eigenvectors.49) = α+ λ+ y+ ′ y+ + α− λ− y+ ′ y+ y− ′ y− .42) d d d 1 d 1 1− 1− − + − − f1 f2 f1 f2 f1 f2 f1 for the cavity round-trip matrix. (2. and reﬂection oﬀ the left mirror (the leftmost matrix in the product). for the initial condition vector we can write y0 ′ y0 = α+ y+ ′ y+ + α− y− ′ y− (2. n2 (2.52) y− ′ y− . we can diagonalize the matrix. (2.7.

(2. We can do this by ﬁnding the eigenvectors and decomposing the initial ray vector.60) or |A + D| ≤ 2 for the stability condition explicitly in terms of the matrix elements. Then clearly λ± are real.57) for some constants ymax and φ0 . the existence of a single periodic ray implies that all rays for the optical system are . each pass through the cavity simply increments the phase of a harmonic oscillation by some ﬁxed amount φ. We can already see that yn will stay bounded as n increases. Ray Optics where we are introducing the new constants γ± := α± λ± for notational convenience.62) If s is the smallest integer for which this is true. the solution grows exponentially as yn ∼ λn . We can also see that |λ+ | > 1 and |λ− | < 1 (in fact. Let’s see this more explicitly: deﬁne φ := cos−1 β β = cos φ. Now let’s reexamine the solution: yn = γ+ λn + γ− λn .58) Solving these two equations leads to γ± = y1 − y0 λ∓ λ± − λ∓ (2. or.54) So |λ± | = 1. so we can write λ± = β ± i 1 − β 2 . This is the unstable case. Note that in the paraxial approximation. for generic initial conditions (i.56) Then λ± = exp(±iφ). We can also write |Tr(M)| ≤ 2 (2. |β| ≤ 1. + since the solution runs away to inﬁnity. 1 − β 2 = sin φ.53) + − The ﬁrst term grows exponentially with n. then s is called the period of the ray.e. As a side note. λ+ λ− = 1. since the argument of the radical is positive: β 2 − 1 > 0. (2. Then β 2 − 1 < 0.. (2.7. (2.55) (2.61) 2. since det(M) = λ+ λ− = 1). for example. Let’s consider the |β| > 1 case ﬁrst. Note also that both eigenvalues have unit modulus: |λ± |2 = λ± λ∗ = β ± i ± 1 − β2 β∓i 1 − β 2 = β 2 + (1 − β 2 ) = 1. Thus the stability condition for the ray to remain bounded in the long term is simply β ≤ 1. in both cases we can determine the constants of the motion from the initial condition. Now let’s consider the other case. There are two possibilities that we have to consider: either |β| ≤ 1 or |β| > 1. where s is some integer: ym+s = ym for all m.24 Chapter 2. In other words. and λn = exp(±inφ).59) as a compact formula for the coeﬃcients. Thus. (2. and clearly now the eigenvalues are complex.2 Periodic Motion A condition more restrictive than the stability condition is the periodic ray condition. and thus it follows that |λn | = 1. Thus we can write the solution as ± yn = γ+ einφ + γ− e−inφ = ymax sin(nφ + φ0 ) (2. y1 = γ+ λ+ + γ− λ− . ± so this is the stable case. (2. γ+ = 0). which states that the ray repeats itself exactly after s passes through the cavity. we can equivalently use y0 and y1 : y0 = γ+ + γ− . while the second damps exponentially away. which can be obtained from γ± noting that yn must be real-valued.

62) as γ+ einφ+isφ + γ− e−inφ−isφ = γ+ einφ + γ− e−inφ . This holds for arbitrary coeﬃcients if the phases diﬀer by exact multiples of 2π. g™ (2. Then we can sketch the stability diagram according to this inequality.66) 4 which.2 := 1− d 2f1. Obviously for a ray to be periodic the resonator must be stable.7.e.7 Resonator Stability 25 periodic (except in trivial cases).2.3 Resonator Stability: Standard Form For the two-mirror resonator that we started out with.64) (2. In terms of these parameters.69) unstable 4 1 1 stable -1 3 2 -1 1 g¡ . sφ = 2πq for some integer q.2 .65) (2. the stability condition is particularly simple: 0 ≤ g1 g2 ≤ 1. More generally. β = cos for integers q and s for periodic motion to occur. though. because the matrix Ms collapses to the identity.63) 2. (2. (2.. after inserting the matrix elements from Eq. just depending on the initial condition. 2πq s (2. we can rewrite the stability condition as d d 1− ≤ 1.68) where the rightmost expression applies to the original two-mirror resonator rather than the equivalent lens waveguide. we can write 0≤ A+D+2 ≤ 1.67) 0≤ 1− 2f1 2f2 It is conventional to deﬁne the stability parameters g1.42) and a bit of algebra. (2. Starting with |(A + D)/2| ≤ 1. in nonlinear systems both periodic and nonperiodic rays are possible in the same system. Thus. (2. We can explore this a bit further mathematically.2 = 1+ d R1. i. We can thus rewrite the periodic-ray condition (2. we can write out the stability condition more explicitly.

70) Recalling that we deﬁned the phase angle φ = cos−1 β. (2. so these are all marginally stable cases. g1. . Hence every ray is periodic with period 1 for a planar resonator. (2. g1. We can compute β = 2g1 g2 − 1 = 1.2 = ∞ (Planar resonator) Case 2. we see that yn = y0 for all n. (2. as we will see when we discuss optical spectrum analyzers. leading to “ﬁgure-eight” orbits in the cavity. we ﬁnd φ = 0 for this case.57). There are four special cases marked in the diagram that we should consider. From Eq. All four cases are on the borders of the stability regions. From Eq. g1 = g2 = 1 =⇒ R1. For the confocal resonator. this is an artifact of the eigenvalue formalism. so each ray repeats itself after 2 passes through the cavity. g2 = 1 =⇒ R1 = d. the light regions are stable. or half of a spherical resonator of length 2d) R The resonator!planar is special in the following sense. The confocal resonator is one of the most important resonators.26 Chapter 2.2 = 0 =⇒ R1. Case 1. Actually.2 = −1 =⇒ R1. the position reverses itself (this happens for the angle as well). so β = 2g1 g2 − 1 = −1 and φ = cos−1 (−1) = π.2 = −d/2 (Spherical resonator/symmetrical concentric) R R Case 4.71) Thus. g1 = 0. R2 = ∞ (Confocal–planar resonator. the solutions can increase ′ linearly rather than exponentially: yn = y0 + ny0 .57).2 = −d (Confocal resonator) R Case 3. every ray is periodic with period 2. d = −R. (2. we see that yn = (−1)yn−1 = (−1)n y0 . On a single pass. in this marginally stable (“parabolic”) case. Ray Optics The shaded regions correspond to unstable resonators.

1. Suppose that a ray in air (n = 1) is incident on a planar window of uniform thickness T and refractive index n. cos θ n cos θ′ . Problem 2. A corner-cube reﬂector is an arrangement of 3 planar mirrors to form. The beam enters each prism at normal incidence to the front surface. An anamorphic prism pair is used to expand or shrink a beam in one dimension without deﬂecting its angle.3. as in this photograph of the back side of a mounted corner-cube prism. As a model of the beam. as shown here. Problem 2. (a) Write down an expression that relates the deﬂection angle θ after the ﬁrst prism to α and n.8 Exercises 27 2.2. How deep does it appear to be. a q a The pair consists of two idential prisms with wedge angle α and refractive index n. Commercial corner cubes often use internal reﬂections to form the mirror surfaces. The reﬂecting surfaces face the interior of the cube. from the point of view of an observer above the pool? The refractive index of water is n = 1. A ﬁsh is 1 m beneath the surface of a pool of water. show that the transmitted beam is parallel to the incident beam but has a horizontal displacement given by δ = T sin θ 1 − where n sin θ′ = sin θ. it is useful to consider parallel rays as shown.8 Exercises Problem 2. Show that any ray entering the corner cube is reﬂected such that the exiting ray is parallel (but opposite) to the incident ray.2. the corner of a cube.33. Problem 2.4. . If the angle of incidence is θ. You may assume that the incident ray’s direction is such that it reﬂects from all 3 surfaces. appropriately enough.

with a > 0. If the focal length is f = 1 m in air. Ray Optics (b) By how much is the beam reduced after the ﬁrst prism? Write your answer in terms of n and α only. −b). . (a) Derive the ABCD matrix for a refractive spherical boundary: n¡ R n™ 1 M = (n2 − n1 ) − n2 R 0 n1 n2 convex (shown): R > 0 concave: R < 0 1 Adapted from H. Introduction to Hamiltonian Optics (Dover. Problem 2. where kc := 2(a2 a . Buchdahl. what is the focal length in water? Assume the lens is made of fused silica (n = 1. (b) Show that AP A′ is a minimum when k < kc and a maximum when k > kc . Show that the optical path length of AP ′ A′ is an extremum when P ′ coincides with P . A thin lens is submerged in water (n = 1.33). Problem 2. b) and (a.46). (c) The locus of all points B such that ABA′ = AP A′ is clearly an ellipse with foci A and A′ . Show that the equation describing this ellipse is given by y2 (x − a)2 + 2 = 1. b) P¢ P x A¢ (a. 1993). A.28 Chapter 2. y x = ky™ A (a. Argue that this result is consistent with the result of part (b). we can ﬁx the endpoints of the reﬂected ray to have coordinates (a.7. + b2 ) By symmetry of the reﬂection at the center. γ 2 − b2 γ where 2γ := ABA′ . (c) By how much is the beam reduced after the second prism? Problem 2.6. -b) (a) Consider a point P ′ a small distance (with vertical component ε) from P .5. Consider1 a reﬂection oﬀ the center P of a parabolic mirror (x = ky 2 ) as shown.

y.8. Prove that det(M) = n1 n2 (as you saw in Problem 6(c)). Problem 2. write down the Lagrangian. and note that z is also ignorable in the sense of being completely determined by x. the height of the ray does not change across the boundary. f d¡ (b) Show that applying the thin-lens law. Show that the conjugate momentum p for the position y is n dy/ds. followed by a thin lens. independent of the input angle y1 . as shown in the Figure. Let M be the ray matrix for an arbitrary optical system where the input and output refractive indices are n1 and n2 . and then write down the Hamiltonian. followed by more free-space propagation. Take the coordinate z to be the “time” variable and the coordinate y to be the position coordinate. and s. where ds2 = dx2 + dy 2 + dz 2 and n(x.8 Exercises 29 The convention is that R > 0 for a convex surface (as shown here) and R < 0 for a concave surface. d™ 1 1 1 + = . so x is an ignorable coordinate.9.2. y. (a) Derive the ray-transfer matrix for free-space propagation. Let’s consider the two-dimensional case. z) ds = 0. respectively. . which exploits the formal equivalence of ray optics to classical Hamiltonian mechanics. (b) Derive the ABCD matrix for a refractive planar boundary: n¡ n™ 1 0 0 n1 n2 M= (c) Find the determinants of the ABCD matrices in parts (a) and (b). using the following outline. Compute the magniﬁcation y2 /y1 . Then for the optical case. Compare to the action principle for classical mechanics. Problem 2. y. all parallel incident rays are focused to a single point in the output plane. Note that in the paraxial approximation. (c) Show that if d2 = f . d1 d2 f all rays originating from a single point y1 in the input plane reach the output plane at the single point ′ y2 . Recall the action principle (Fermat’s principle) for ray optics: δ n(x. z) is a refractive-index proﬁle that models the optical system.

q. separated by a distance d. why is the optical power a natural way to characterize a thin lens? Problem 2. d1 d f1 f1 d 1 f1 1 1 = − + . p. Show that the combination acts as a thin lens with a focal length given by 1 1 1 + . Recall that the action functional is given by the integral t2 S[L] := t1 L(q. Using this result. and compute the determinant of M. Show that the eﬀective focal length feﬀ of two lenses having focal lengths f1 and f2 . A very brief review of variational principles in classical mechanics may help. = M′ p1 p2 which of course is valid in the paraxial approximation (where it is also true that s ≈ z). Because y and p are canonical variables and M′ represents “time evolution” of a Hamiltonian system. (a) Suppose that two thin lenses of focal length f1 and f2 are placed in contact. M′ represents a canonical transformation and in particular M′ is a symplectic matrix. with free-space propagation of distances d1 and d2 before and after the single lens.11. Ray Optics Now consider the following transformation relating the canonical coordinates before and after the optical system. where f is the focal length of the lens. This is essentially the content of Liouville’s theorem. Typically the lens power is measured in diopters. feﬀ f1 f2 f1 f2 Note that this system is no longer a thin lens. a lens with a 100 mm focal length has a power of 10 diopters).30 Chapter 2. t) := q α pα − L(q. t) dt. ˙ ˙ where the conjugate momenta are deﬁned by pα := ∂L/∂ q α . where 1 f2 1 1 = − + .. respectively.10. which implies that det(M′ ) = 1. ˙ where the Lagrangian L is typically of the form L = T (q) − V (q) in particle mechanics. t). is given by 1 1 d 1 = + − .e. y1 y2 . which for our purposes implies the Euler-Lagrange equation d ∂L ∂L − α =0 dt ∂ q α ˙ ∂q under the condition that the endpoints of the variation are ﬁxed (δq(t1 ) = δq(t2 ) = 0). ˙ Problem 2. The variational ˙ principle (Hamilton’s principle) is δS[L] = 0. so for this to work out you should show that the eﬀect of the two-lens optical system is equivalent to that of a single lens of focal length feﬀ . q. The Hamiltonian is given by a Legendre transformation of the Lagrangian via H(q. Based on your answer for part (a). deﬁned as 1/f where f is measured in meters (i. d2 d f2 f2 d . transform to the standard (noncanonical) variables y and y ′ . = f f1 f2 (b) The optical power of a lens is deﬁned as 1/f .

51432) and F2 ﬂint glass (nF = 1. we can deﬁne the Abbé v-constant by nd − 1 vd := . Problem 2. λd = 587. The length of the telescope satisﬁes L = fo + fe . and λF = 486. A refracting telescope is an optical system consisting of two thin lenses with a ﬁxed space in between. nC = 1. and F2 for the second section in the shape of a plano-concave lens. δλ vd (λC − λF ) . This eﬀect is called chromatic aberration. and F lines. R£ = • objective lens fo ocular lens fe L (a) Construct the ABCD matrix for propagation through a telescope (from left to right in this diagram). Because of dispersion.3 nm.63208. achromatic doublet with the following materials: BK7 borosilicate crown glass (nF = 1. the index of refraction varies slightly with the wavelength of light.61989. assume a simple linear model of the refractive-index variation: n(λ) ≈ n(λ0 ) + dn dλ (λ − λ0 ). the Fraunhofer C.61503). the radii of curvature at the cemented interface should match. and goes through the ocular lens (eyepiece lens) of focal length fe .13. independent of the initial ray position y1 . propagates over a distance L.12.51673. incoming rays with angle θ1 from the optical axis exit the system at angle −(fo /fe)θ1 . nC = 1. named after Fraunhofer’s catalog of the dark features in the solar spectrum.2. λ=λ0 where λ0 is some wavelength in the center of the region of interest. Make the thin-lens approximation and choose the two radii of curvature to achieve a lens with f = 100 mm over the visible spectrum.8 Exercises 31 Problem 2.52238. nd = 1. That is. nF − nC in terms of which we can write the refractive index as n(λ) ≈ 1 + (nd − 1) 1 − where δλ := λ − λd . Design a thin. given by λC = 656.6 nm. . Use BK7 for the ﬁrst section in the shape of a biconvex lens. nd = 1. Light ﬁrst enters the objective lens of focal length fo . For this problem. (b) Show that a telescope produces angular magniﬁcation. The dispersion of an optical glass is often characterized by its refractive indices at three special wavelengths.1 nm. In terms of the three indices. and thus the focal length of a thin lens varies slightly with optical wavelength. d. F2 BK7 R¡ R™ Obviously. as shown. A common technique to correct for this aberration is to cement two lenses together of diﬀerent materials to form an achromatic doublet or achromat.

show that a bundle of rays of diameter d traveling parallel to the optical axis has diameter (fe /fo )d when exiting the eyepiece. as shown here. Consider a two-mirror resonator. (e) Clearly. One mirror is concave (R < 0) and the other is ﬂat. where both focal lengths are positive. Problem 2.16. and that after much labor you ﬁnd that the laser operates in the ranges d < 50 cm and 100 cm < d < 150 cm. a telescope with positive angular magniﬁcation (i. traveling to the right. (d) Sketch a Keplerian telescope. Also sketch a ray that is initially not parallel to the optical axis. (a) Consider a cavity consisting of two planar mirrors and identical thin lenses of focal length f .e. R¡ R™ d Suppose also that you can vary the length d of the cavity. What are the numerical values of R1 and R2 ? Keep in mind that gas lasers have relatively low gain per pass. Draw in the parallel rays corresponding to part (c).32 Chapter 2. Which of the two lenses can have negative focal length if the telescope produces a magniﬁed (not reduced) image? Problem 2.15. Problem 2. and thus proper laser operation requires that the light makes many round trips inside the resonator before leaking out.e. with unknown and possibly diﬀerent radii of curvature R1 and R2 . use the thin lens law to justify the minus sign in the angular magniﬁcation of part (b). R d (a) What is the round-trip ray matrix for this resonator? Derive the matrix for a ray starting just to the right of the curved mirror. Ray Optics (c) Show that a telescope can also act as a beam reducer (or expander): i. Sketch an example ray that illustrates your answer. Such a refracting telescope is known as a Galilean telescope. (b) For what range of d is the cavity stable? (c) Derive the ray matrix for two round trips for the special case R = −2d. a telescope that produces an upright image) has exactly one lens with negative focal length so that (−fo /fe ) > 0.. regularly spaced as shown. spherical mirrors in a gas laser resonator. Suppose you have two convex. f f f f f f f f d/2 d d d d d d d d/2 ..14.

Problem 2. what is the range of d for which the cavity is stable? (b) Write down the eigenvalues of the round-trip ray matrix for this cavity. Suppose that a thin lens of focal length f is placed inside the cavity against one of the mirrors. Consider a symmetric cavity consisting of two planar mirrors separated by 1 m.8 Exercises 33 For a given set of lenses.2.17. For what range of f is the cavity stable? .

.

a function deﬁned on a bounded domain). but we can make things a bit easier to generalize if we deﬁne negative ones too: a−n := an .6) In each term of the form exp(−inωt) we can make the transformation n −→ −n to simplify the sums: f (t) = ∞ n=−∞ an e−inωt + i ∞ n=−∞ bn e−inωt . we will only use those harmonic functions that are also periodic with period T : f (t) = a0 + 2 ∞ n=1 an cos(nωt) + 2 ∞ n=1 bn sin(nωt). T (3. (Fourier series) (3. Let’s rewrite the Fourier series: f (t) = a0 + b0 + ∞ n=1 (3. The basic point is that we can think of the harmonic functions (sines and cosines) as being fundamental building blocks for functions.1 Periodic Functions: Fourier Series Before getting into deﬁning the Fourier transform.2) (3. (3.4) for all nonnegative n (in particular. Since we know that f (t) is periodic. b0 = 0 by our deﬁnition).5) an einωt + e−inωt − i ∞ n=1 bn einωt − e−inωt . Then for any integer n (positive and negative). The frequency corresponding to the period is given by ω = 2πν = 2π . So let’s try to build up f (t) out of harmonic functions.7) . b−n := −bn (3. for all t. it is helpful to motivate it by ﬁrst considering the simpler Fourier series for a periodic function (or equivalently. Suppose f (t) is a periodic function with period T . so that it satisﬁes f (t) = f (t + T ).1) where ν is the frequency and ω is the angular frequency. (3. This deﬁnition allows us to write things a bit more compactly. we can deﬁne the complex Fourier coeﬃcient by cn := an + ibn .3) This expansion uses only expansion coeﬃcients for positive n.Chapter 3 Fourier Analysis 3.

though. However. (3.10) takes the value 2π/ω = T if n = n′ .36 Chapter 3. whereas the conjugate function exp(iωt) = exp[−i(−ω)t] corresponds to a “negative” frequency −ω. Of course. If we deﬁne s = n − n′ . We can rewrite this relation in the more meaningful form 1 T −in′ ωt ∗ −inωt (3. (complex Fourier series) (3. Now consider the inner product of a basis vector with f (t): e−in ωt . but the inﬁnite-dimensional.3) in that the −n second series can also represent complex-valued functions.3): an = 1 (cn + c∗ ) = n 2 1 bn = (cn − c∗ ) = n 2i 1 (cn + c−n ) 2 1 (cn − c−n ). continuous version instead of the ﬁnite. e T 0 where δnn′ is the Kronecker delta (δnn′ = 1 if n = n′ and 0 otherwise). The convention is that the harmonic function of the form exp(−iωt) corresponds to a “positive” frequency ω. since there is a removable singularity here.8) Now we can see that this Fourier series is a sum over complex harmonic functions with frequency ωn = nω.14) . It might sound strange to talk about positive and negative frequencies. then s→0 lim e−i2πs − 1 s = lim s→0 [1 − i2πs + O(s2 )] − 1 s = −2πi. Fourier Analysis Using the complex coeﬃcients.11) so that the integral (3. This is the orthogonality relation for the harmonic functions. ′ 2π (3. given the cn coeﬃcients.13) So we are still doing linear algebra.9) But how do we obtain the cn coeﬃcients in the ﬁrst place? First. the series (3. discrete version with matrices that we reviewed in the beginning. f2 := 1 T T 0 ∗ f1 (t)f2 (t) dt.12) e dt = δnn′ . and the orthogonality relation is a special case of the inner product deﬁned by the same integral: f1 . We need to be more careful with the n = n′ case.10) 0 where we deﬁned x := ωt. 2i (3.8) is more general than the original series (3. The harmonic functions are basis vectors in a vector space of functions. observe that for some integer n′ . (3. we can obtain the coeﬃcients in the original Fourier series (3. (3. this simpliﬁes greatly: f (t) = ∞ n=−∞ cn e−inωt . we can evaluate the integral T 0 e−inωt ein ωt dt = 0 ′ T e−i(n−n )ωt dt 2π 0 ′ = 1 ω e−i(n−n )x dx ′ ′ 1 = −iω =0 e−i(n−n )x n − n′ if n = n . but for a real function f (t) the positive frequency “contributes as much as” its negative counterpart in the sense that the coeﬃcients must obey the constraint cn = c∗ . f = ′ 1 T T 0 e−in ωt f (t) dt = ′ cn T n=−∞ ∞ T 0 e−in ωt ′ ∗ e−inωt dt = ∞ n=−∞ cn δnn′ = cn′ .

we can write the series as 1 T | sin ωt|ein(2ω)t dt T 0 1 T 1 iωt e − e−iωt ein(2ω)t dt T 0 2i T 1 ei(2n+1)ωt − ei(2n−1)ωt dt 2iT 0 π 1 ei(2n+1)x − ei(2n−1)x dx 2πi 0 1 ei(2n+1)π − 1 ei(2n−1)π − 1 dx − 2πi i(2n + 1) i(2n − 1) 1 1 2 − = . but with a slightly diﬀerent normalization: f (t) = ∞ n=−∞ cn e−inωt −→ f (t) = 1 2π ∞ −∞ ˜ f (ω)e−iωt dω. It can be shown for “reasonable” functions that N n=−N cn e−inωt −→ f (t) as N −→ ∞ (3. since there is much more “information” in the function than in the periodic case. so we must use a representation where ∆ω −→ 0. π(2n + 1) π(2n − 1) π(1 − 4n2 ) f (t) = 2 e−i2nωt . (inverse Fourier transform) (3.19) . when designing a device to double the frequency of an input signal.2 Aperiodic Functions: Fourier Transform 37 Thus. then we were using a discrete (countable) set of harmonic functions with a frequency spacing given by ∆ω = 2π/T . letting T = π/ω. cn = = = = = = Thus.17) (3. Thus. and none of the initial frequency. we need a continuous spectrum to represent an aperiodic function. This is a useful feature to keep in mind. 3. we can also now calculate the an and bn coeﬃcients in terms of the cn .1. the rectiﬁed sine wave contains only the even harmonics of the original pure harmonic wave. Recall that if T is the period. Note that because of the “rectiﬁcation. An aperiodic function corresponds to T −→ ∞. so the eﬀective frequency of this function is 2ω.” this function has a period of π/ω.3.2 Aperiodic Functions: Fourier Transform We can also use the same harmonic functions e−iωt to build up aperiodic functions. we can use the orthonormal properties of basis functions to project out the coeﬃcients of exp(−inωt): cn = 1 T T 0 einωt f (t) dt. Note that cn = c−n = c∗ n in this example because f (t) is real and even. (3. for example.1 Example: Rectiﬁed Sine Wave As a simple example.15) Of course. Thus. So let’s deﬁne the Fourier transform as a generalization of the Fourier series.16) at each point t except possibly on a set of zero measure. π(1 − 4n2 ) n=−∞ ∞ (3. whereas in the more general case of a real function we would only expect the less restrictive case cn = c∗ . let’s compute the Fourier series for the function | sin ωt|.18) As reﬂected in the initial setup of this problem. −n 3. as we noted above.

as we will see shortly). e. when is it possible to have ˜ a Fourier transform? As with the Fourier series. continuous limit. since we are ﬁnding f (t) from its Fourier transform. where instead of ω we can use the standard frequency ν = ω/2π.21) 2π −∞ −∞ ˜ The functions f (t) and f (ω) are said to be a Fourier transform pair. we can do a projection to ﬁnd the amplitudes (Fourier transform). functions useful in physics. The other usual ˜ nomenclature is that f (ω)/2π is the Fourier transform of f (t). if f (t) is a real function.g.20) The pair of equations (3. but it is interesting to ask.22) so that the positive and negative frequencies contribute equally in amplitude to form a real-valued function. to model wave phenomena. We haven’t been too concerned with rigor here. f has no inﬁnite discontinuities. F [g(t)] = g (ω) g ˜ for all α. ˜ ˜ Again. ∞ −∞ −∞ −∞ f (t) dt exists. If we denote the Fourier transform by the symbol “F . (3. ˜ f (t) = F −1 [f (ω)]. The analogue of the Fourier-series projection is cn = 1 T T 0 ˜ einωt f (t) dt −→ f (ω) = ∞ −∞ f (t)eiωt dt.25) . tend to be reasonable in precisely this sense.19) is one of the most important tools in physics. let’s reemphasize the connection to linear algebra: the Fourier transform is a linear transformation between vector spaces of functions.19) and (3. In the same way as before. Naturally. f has only a ﬁnite number of discontinuities and a ﬁnite number of maxima and minima in any ﬁnite interval. (3.23) ¯ ˜ Here. β ∈ R. then one possible set of suﬃcient conditions is as follows: 1. 3. The above mathematical operation is the inverse Fourier transform. f (t) = (3. then the Fourier transform satisﬁes f (ω) = −f ∗ (−ω). Note that there is an important alternate convention. f (ω)/2π is the amplitude of the frequency component e−iωt (the normalization coeﬃcient depends on how we deﬁne the density of the continuum of basis functions. (3. we have deﬁned an alternate Fourier transform by f (ν) := f (ω/2π). f (ω) = f (t)eiωt dt. in which case f (t) = 2Re 1 2π ∞ −∞ ˜ f (ω)e−iωt dω . linearity of the Fourier transform means that ˜ ˜ F [αf (t) + βg(t)] = αf (ω) + β˜(ω) iﬀ F [f (t)] = f (ω). (Fourier transform) (3. f (ν) = f (t)ei2πνt dt. Again. This changes the density of the basis functions in a way that makes the normalizations more symmetric: ∞ ∞ ¯ ¯ f (t) = f (ν)e−i2πνt dν. this is like a matrix transformation.” we can write the above deﬁnitions in the compact form ˜ f (ω) = F [f (t)].24) Then again. and their relationship is of fundamental importance in understanding linear systems in physics. If f (t) is deﬁned on the real line. so they deserve to be written again: ∞ ∞ 1 ˜ ˜ f (ω)e−iωt dω. but in the inﬁnite. Fourier Analysis ˜ Thus. 2.38 Chapter 3. the f (t) must be a “reasonable” function for f (ω) to exist. Finally.. (3. It is a bit easier to remember this form to work out where the (1/2π) goes in the ω form of the transform equations.

2. f (t) = Ae−αt .3 The Fourier Transform in Optics 39 3.20).28) Ae−at 2 +c dt = Aec π =A a ω2 π exp − α 4α . and brieﬂy discuss examples of situations where they come up. Hence we see that the Fourier transform of a Gaussian is a Gaussian.31) Since the standard deviation is a measure of the width of a function (more precisely the square root of the variance of the function). the width of the original pulse decreases. (3. . (3. but the width of the transform increases. ˜ f (ω) = ∞ −∞ (3. Equating powers of t. δω (3. The central eﬀect behind Fourier optics is that within the paraxial approximation.30) 2σ 2πσ −∞ √ by taking the standard deviation σ to be 1/ 2α.33) ˜ where δt is the “width” of f (t) and δω is the width of f (ω). That is.1 Example: Fourier Transform of a Gaussian Pulse One of the most useful Fourier transforms that can be easily calculated is of the Gaussian pulse. (3. we need to perform a mathematical trick. ˜ f (ω) = ∞ −∞ 2 (3. let’s rewrite the argument in the exponent in the form a(t − b)2 + c = at2 − 2abt + c − ab2.3. (3.3 The Fourier Transform in Optics We will spend essentially the rest of the course using the Fourier transform to understand wave optics. This reﬂects a general property of Fourier transforms. 2α the standard deviation of the Fourier transform is σω = √ 2α.29) where we evaluated the integral by comparison to the standard normalized form of the Gaussian (you should memorize this. we can see that as a increases. that δt ∼ 1 . we ﬁnd the new coeﬃcients: t2: a = α −iω 2α 2 ω t0: c − ab2 = 0 =⇒ c = − . completing the square in the exponent. But ﬁrst let’s review a few of the physically important Fourier transforms that you should know. Notice that while the standard deviation of the original Gaussian is 1 σt = √ .27) To evaluate the integral. by the way). as we will see. 4α t1: −2ab = iω =⇒ b = Substituting the new form of the exponent and letting t −→ t + b. a thin lens acts as a Fourier-transform computer.26) Ae−αt 2 +iωt dt. The major use of the Fourier transform. 3. is in Fourier optics. Using the Fourier transform deﬁnition (3. This is precisely the same uncertainty principle that is a fundamental principle in quantum mechanics.32) (3. ∞ x2 1 √ exp − 2 dx = 1.

Some of the most important and useful ones are summarized here. given a scalar electric ﬁeld E(x) one focal length before the lens.37) (3. Fourier Analysis E(x) F[E(x)] f f That is. If Ne is the number of atoms in the excited state. 1. the ﬁeld after the lens is related to F [E(x)]. The Fourier transform of a constant is a delta function. The Gaussian function is the most localized function that has this property. The Fourier transform of a square pulse is a sinc function. The Fourier transform of a Gaussian is a Gaussian. F e−|t| = 2 1 + ω2 (3. 5. (3. It turns out that the Fourier transform of the time dependence of the emission gives the radiation spectrum for spontaneous emission. In other words.40 Chapter 3. F 1 = δ(ω) 2π (3. But Fourier transforms show up everywhere in optics as well as in the rest of physics. For now it suﬃces to think of the delta function as a unit-area pulse in the limit as δt tends to zero. 2. We will see that Gaussian beams are electromagnetic ﬁeld modes of a spherical-mirror resonator. You should memorize these. A simple way to see this is that the modes must be Fourier transform of itself in order to resonate (repeat itself) in the cavity. 3.36) The far-ﬁeld diﬀraction pattern of a uniformly illuminated slit is a sinc function for this reason. F e−t 2 /2 = e−ω 2 /2 (3.34) Recall that the spherical mirrors of resonators act as lenses. the decay law is Ne (t) = Ne (0) exp(−Γt). . F [δ(t)] = 1 (3. The Fourier transform of a delta function is a constant. For atoms obeying the exponential decay law. the emission spectrum is Lorentzian. a fact that is related to the Fourier-transform property of a lens as mentioned above. this relation is the extreme limit of the uncertainty principle. F [rect(t)] = sinc(ω/2) := sin(ω/2)/(ω/2) The rectangular function is deﬁned by if 1 1/2 if rect(t) := 0 if |t| < 1/2 |t| = 1/2 |t| > 1/2.39) This is the opposite extreme of the uncertainty principle: an arbitrarily short pulse has an arbitrarily broad spectrum.38) We haven’t yet deﬁned the delta function but we will do so shortly. The Fourier transform of an exponential is a Lorentzian. 4.35) Atoms decay exponentially due to spontaneous emission.

we often write δ(t) = lim hn (t) n− →∞ (sloppy!) (3.45) Proof. the integral and the limit don’t commute.. simply peaked around t = 0). In normalized form. The hn are normalized: ∞ −∞ hn (t) dt = 1 for all n.3.47) Proof. But as physicists.42) We will deﬁne the delta function δ(t) such that ∞ −∞ δ(t)f (t) dt := lim ∞ −∞ n− →∞ hn (t)f (t) dt. 1 f (t) = f (0) + f ′ (0)t + f ′′ (0)t2 + · · · 2 (3. Now we will review and derive a few properties of the delta function.44) as a shorthand for the above (rigorous) deﬁnition. ∞ −∞ δ(t) dt = lim ∞ −∞ n−→∞ hn (t) dt = 1. so it is useful to think of δ(t) as a “unit-area function that is zero everywhere but ∞ at t = 0.40) (3.g. For example. i.4 Delta Function As we just mentioned.” as long as you are careful about it. a delta function is an idealized limit of a very short pulse. the delta function only makes sense as part of the argument of an integral. 1. (3. Assuming f has a Taylor expansion about t = 0. The list that follows is by no means exhaustive. 2. 3.48) .41) hn (t)f (t) dt. Projection property of δ(t): ∞ −∞ δ(t)f (t) dt = f (0).. But this is a sloppy deﬁnition.46) 2. The hn are “reasonable” (e. (3. Similarly.4 Delta Function 41 3. Since the functions hn (t) are normalized. To do this in a mathematically well-deﬁned way. This is because limn−→∞ hn (t) does not exist. Thus. (3.43) Notice that the order of the integral sign and the limit are important. Then consider the integral ∞ −∞ . (3. The width of hn converges to zero as n −→ ∞. (3.e. note that limn−→∞ hn (t) = 0 for all t = 0. we can use Gaussian functions. the functions become hn (t) = n exp −πn2 t2 . δ(t) is normalized: ∞ −∞ δ(t) dt = 1 (3. we need to consider a sequence of functions hn (t) that have the following properties: 1. we can write 1 t2 hn (t) = √ exp − 2 2σn 2πσn √ and if we choose σn = 1/ 2πn.

52) The proof of this follows trivially from the previous property and the deﬁnition (3. and the application of the inverse transform integral (3. the harmonic functions can form an orthonormal set. we see that in the form exp(−iωt)/2π [or alternately. 3. but expanding about t = a. the integral representation (3.53) This important representation of the delta function follows from the inversion of the previous property. for showing the orthogonality of the harmonic basis functions. 5. 4. since δ(t) is normalized. Fourier transform of δ(t): F [δ(t − a)] = eiωa . . exp(i2πνt)]. and then shift t −→ t + a to evaluate the integrals.53) of δ(t) is useful. we can deﬁne two suitable functions f (t) and g(t) to be orthogonal if ∞ −∞ (3.49) for the Gaussian hn (t)’s deﬁned above. (3. we can write ∞ −∞ hn (t)f (t) dt = f (0) + f ′′ (0) +O 2πn2 1 n4 . (3. Also.19). Fourier Analysis Noting that ∞ −∞ hn (t)tm dt = n ∞ −∞ tm exp(−πn2 t2 ) dt = 1 + (−1)m Γ 2 1+m 2 π −(m+1)/2 n−m (3. exp(−iωt) and exp(−iω ′ t) are orthogonal because ∞ −∞ exp(iωt) exp(−iω ′ t) dt = 2πδ(ω − ω ′ ).54) f ∗ (t)g(t) dt = 0.55) For harmonic functions. (3. (3. leaving just f (0). (3.50) As n −→ ∞.20) of the Fourier transform.56) which is zero if ω = ω ′ . This ﬁnal property above. As in the periodic case. Integral representation of δ(t): δ(t − a) = 1 2π ∞ −∞ e−iω(t−a) dω. all the higher order terms vanish. (3. Shifted projection property of δ(t): ∞ −∞ δ(t − a)f (t) dt = f (a).42 Chapter 3.51) This can be proved by repeating the argument for the unshifted projection. for example. F −1 [eiωa ] = δ(t − a).

|t| < a/2 0 elsewhere.2.1 A common feature of truncated Fourier series is that relatively few terms are needed to make a fairly good approximant. (c) We said that the functions fN (t) converge to f (t). Fourier Analysis (Cambridge.) As a simple model. Compute the Fourier transform f (ω) of the rectangular-pulse function f (t) = 1. i.60) nice discussion of this example appears in T. the width of f (ω) increases. 3.57) x>0 x=0 x < 0.. Körner. where the Gibbs phenomenon is absent.5 Exercises ˜ Problem 3. given by f(t) = sgn[sin(2πt)] where sgn(x) is the sign of x. This is called the Gibbs phenomenon. |t| < T /2 f (t + T ) = f (t − T ) elsewhere. 21. One application of Fourier series is in synthesizing waveforms or images with relatively little information. (This is even the case for continuous functions. suppose you want to synthesize a triangle wave. and 101 to check how good the ﬁnite approximations are.58) (b) Make plots of the truncated Fourier series. 1A 1 − 4|t|/T. But the Gibbs phenomenon says that for any N . From your plots.3. 1988). For example. (3. Problem 3. (3. (a) Compute the Fourier series for a square wave of unit period. f (t) := using as few harmonics as possible.5 Exercises 43 3. the initial convergence of the Fourier series is rapid. (3. many terms may be necessary. the JPEG image compression algorithm achieves much of its compression by removing harmonic components.1. W. Explain how these two statements are consistent. ˜ Show from your answer that as the pulse width a decreases. and electronic organs can use relatively few harmonics (on the order of 10) to make a passable imitation of a musical instrument. estimate the magnitude of the Gibbs overshoot.e. (a) Compute the Fourier series for the triangle wave as deﬁned above. You should notice a ringing behavior near the discontinuities. Mathematically. but the convergence rate becomes very slow by the time many terms have accumulated. . Problem 3. there is always some t such that fN (t) is at least some minumum distance away from f (t). where the approximant fN (t) overshoots the actual function f (t). 5. and persists at an essentially constant (nonzero) value even for arbitrarily large N . but for a high accuracy approximant. fN (t) = n=−N cn e−inωt . sgn(x) := 1 if 0 if −1 if N (3.59) along with the exact function for N = 1.3.

4.5.44 Chapter 3. (3.61) and let’s deﬁne the fractional error of the partial sum to be normalized root-mean-square error: T εN := 0 |fN (t) − f (t)|2 dt T 0 . Let f (t) be a function with Fourier transform f (ω). Show that the Fourier transform of ˜(ω + α) + f (ω − α)]/2. N fN (t) = n=−N cn e−inωt . Fourier Analysis (b) Consider the partial sum.62) |f (t)| dt 2 How large should N be in the partial sum to obtain an accuracy of 10%? 1%? 0.01%? (A computer may be of great help in solving this problem.1%? 0. Prove the following form of the Poisson sum rule: ∞ n=−∞ δ(t − n) = ∞ n=−∞ cos(2πnt). (3.) ˜ Problem 3. ˜ f (t) cos(αt) is [f Problem 3. .

and µ0 := 4π × 10−7 N A2 (4.1) where E(r. ǫ0 ≃ 8. Maxwell’s equations form the basis of electromagnetism. t) is the magnetic ﬁeld. Rather. ∂t Now we will use the vector identity ∇ × (∇ × A) = ∇(∇ · A) − ∇2 A. Actually.1 Maxwell Equations in Vacuum Electromagnetism is the fundamental theory that underlies all of wave optics. t) is the electric ﬁeld. we won’t use Maxwell’s equations very often in the form we wrote.4) for the vector derivative operator that we used in writing down the Maxwell equations. To do this. we will use them to derive the wave equation.85 × 10−12 is the electric permittivity of the vacuum.6) . so before proceeding with this more general theory of optics. we ﬁrst apply (∇×) to Faraday’s law: ∂ (4.2) is the magnetic permeability of the vacuum. we will stop and review the basics.Chapter 4 Review of Electromagnetic Theory 4. Also recall the notation ∇ := x ˆ ∂ ∂ ∂ +y ˆ +z ˆ ∂x ∂y ∂z (4. H(r. We will write Maxwell’s equations in free space as ∇·E= 0 ∇·H= 0 ∂H ∇ × E = −µ0 ∂t ∂E ∇ × H = ǫ0 ∂t (Gauss’ law in free space) (no magnetic monopoles) (Faraday’s law) (Ampére’s law with Maxwell’s correction in free space) (4. (4.5) ∇ × (∇ × E) = −µ0 (∇ × H).3) F m (4. the fundamental equation of optics.

10) Finally. Review of Electromagnetic Theory which holds for any vector A to evaluate the left-hand side of this equation. note that it is orthogonal to both the electric and magnetic ﬁeld vectors. More commonly. I := |S| .13) where the average is taken over time scales long compared to an optical cycle (a few fs) but short compared to other time scales of interest. and we will use Ampére’s law to evaluate the right-hand side: ∇(∇ · E) − ∇2 E = −µ0 ∂ ∂t ǫ0 ∂E ∂t . (4.12) The Poynting vector points in the direction of energy ﬂow. The intensity is deﬁned as simply the time-averaged magnitude of the Poynting vector.2 Intensity One of the most important features of electromagnetic waves (time-dependent solutions of the Maxwell or wave equations) is that they transport energy. given by c0 := 2.3 Maxwell Equations in Media Maxwell’s equations take on a slightly diﬀerent form for electromagnetic ﬁelds in a medium: ∇·D=0 ∇·B=0 ∂B ∇×E=− ∂t ∂D ∇×H= . deﬁned by S := E × H. µ0 c0 (4.11) (4.9) is the speed of light in vacuum.14) . 1 ∂2E = 0. the exact value of ǫ0 is deﬁned by ǫ0 := 1 2.46 Chapter 4. Since µ0 and c0 are both exactly deﬁned quantities. ∂t (4. we note that the magnetic ﬁeld H satisﬁes the same wave equation as the electric ﬁeld.8) 1 c0 = √ ǫ0 µ0 (4. energy transport is quantiﬁed by the intensity of the electromagnetic wave.997 924 58 × 108 m/s. (4. 4. (4. as we can see from Gauss’ law. 2 c0 ∂t2 (wave equation) (4. 4. Then cleaning things up a bit. This is actually now a deﬁned quantity. we arrive at the standard form of the wave equation: ∇2 E − Here. The ﬂow of energy is described by the Poynting vector.7) The ﬁrst term vanishes.

Spatial nondispersivity (locality): P(r) is a function of only E(r). In optics. E dielectric medium P Thus. These equations are still valid only in a source-free region (i. Similarly the relation between the H and B ﬁelds is related to the magnetic properties of the medium: B = µ0 H + M. deﬁned as the magnetic dipole moment per unit volume. the simple dielectric has a polarization that can be written as a simple proportionality: P = ǫ0 χE. (4.16) here. 1. though some materials can be fabricated with permanent polarization (electrets). Nondispersivity: the speed of light does not depend on the frequency of the electromagnetic waves. free of charges and currents). (4. but here we will only treat the simplest possible such media. Thus. and B is the magnetic ﬂux density. r. 5. so D = ǫ0 E and B = µ0 H. The simple dielectric medium has a number of constraints on its properties. The only way for this to work is for P E. 2. but suﬃcient for many purposes. never E(r′ ) for r′ = r. as well as space and time: P(r. This is always an idealization. t). we can think of the polarization as a function of the electric ﬁeld.17) . To satisfy all these conditions. we usually deal with dielectric materials. Isotropy: P(E) of orientation. deﬁned to be the electric dipole moment per unit volume. since it amounts to saying that P(t) is a function of only E(t). P is the polarization density. never E(t′ ) for t′ < t. this means that there is no hysteresis. t) = P(E(r. 4. t). In free space.4.e.14). Homogeneity: P(E) is independent the position r inside the medium. For our purposes here it will be useful to think of the electric ﬁeld E as something externally applied to the medium.15) Here. M is the magnetization density. The relation between the E and D ﬁelds is determined by the electric properties of the medium: D = ǫ0 E + P.4 Simple Dielectric Media We will now discuss dielectric media in more detail. Usually the polarization is induced by the electric ﬁeld. (4. (4. Linearity: P ∝ E. in which case we recover the free-space Maxwell equations (4. Sometimes the magnetization is induced by an external magnetic ﬁeld. 4. but permanent magnets are relatively commonplace. In the time domain. which greatly simplify things.4 Simple Dielectric Media 47 Here. so there is no “preferred direction” in the medium. from now on we will assume M = 0. 3.. while the polarization P is the response of the medium to the ﬁeld.1) from Eqs. so that the superposition principle still holds in the medium. P = M = 0. so that B = µ0 H. D is the electric ﬂux density or electric displacement.

Note that this equation only holds in cases like simple media where the diﬀerent vector components don’t get mixed. Review of Electromagnetic Theory The constant χ is the electric susceptibility. These solutions represent functions that do not change shape but propagate rightward (x − ct) or leftward (x + ct) with speed c. 2 2 (4.18) where ǫ := ǫ0 (1 + χ) is the electric permittivity of the medium. Thus. which have a harmonic time dependence with a single angular frequency ω: E(r. As in the Fourier analysis case. and the ratio ǫ =1+χ ǫ0 (4. which in general mixes diﬀerent ﬁeld components except in a special basis.15) as D = ǫE.26) . t) = E(r) cos(ωt + φ). (4.48 Chapter 4. which competely characterizes the simple medium. For example.25) Here. we ﬁnd that the equation has the same form as before.22) which gives the speed of light in the dielectric medium in terms of the index of refraction.24) where E now stands for any component of E or H. Thus. H(r. we can rewrite Eq. we can write a scalar wave equation. you have to be more careful when analyzing a refractive interface. ∇2 E − 1 ∂2E = 0. deﬁned as the square root of the dielectric constant: ǫ n := . Finally.5 Monochromatic Waves and Complex Notation In general. t) = H(r) cos(ωt + φ′ ).23) ǫ0 Thus we see how the dielectric properties of the medium relate to the refractive index that we used in ray optics. each component of E and H separately satisﬁes the wave equation.20) (4. t) = E(x ± ct) is a solution to the scalar wave equation. c 2 ∂t2 (scalar wave equation) (4. we can study monochromatic waves.21) (H also satisﬁes this equation) except that c0 is replaced by c := c0 . There are many possible functions of this form. (4. any function of the form E(x. n (speed of light in dielectric medium) (4.and negative-frequency parts: E(r. c 2 ∂t2 (wave equation for dielectric medium) (4. but motivated by our review of Fourier analysis.19) is the dielectric constant. (4. we can instead use complex harmonic functions and deﬁne the complex monochromatic ﬁelds E(±) by breaking up the real ﬁeld into its positive. 4. t) = E(r) eiφ iωt e−iφ −iωt e + E(r) e =: E(+) (r)e−iωt + E(−) (r)eiωt . If we rederive the wave equation using the Maxwell equations for a dielectric. note that in a simple medium (as well as in vacuum). (4. ∇2 E − 1 ∂2E = 0. φ and φ′ are constant phase oﬀsets.

ω) (and so on for H(+) . t) = 1 2π ∞ −∞ E(+) (r. keeping track of only E (+) will greatly simplify electromagnetism calculations.c. Again. D.29) (4. we can always write the physical ﬁeld as E (+) with its conjugate: E(r. (4. and thereby obtain the general solution as an inverse Fourier transform.31) where we can identify E(r.). and because ∂ −iωt e = −iωe−iωt . Note that all of this complex notation stuﬀ works out because the wave and Maxwell equations are all linear. = 2Re E(+) (r)e−iωt . the monochromatic Maxwell equation establishes the forms of the harmonic basis functions. 4.30) with D(+) = ǫ0 E(+) + P(+) and B(+) = µ0 H(+) . −ω)(+) .and negativefrequency parts.and negative-frequency parts) satisfy these equations separately.4. this follows from the linearity of the ﬁeld equations along with the harmonic time dependence of the harmonic ﬁelds. The imaginary part is just an extra piece that helps to simplify calculations. ω)(−) = E(r. we are implicitly taking a Fourier transform of the general ﬁeld.6 Intensity in Complex Notation Quantities that are not linear in the ﬁelds are a bit more subtle. which doesn’t do any harm so long as you get rid of it at the end.32) +E (−) +E (−) ×H (+) +E (+) ×H (+) −i2ωt e ×H (−) i2ωt e . In particular. Of course. Thus. positive. etc. the real and imaginary parts (equivalently. . Still.6 Intensity in Complex Notation 49 Recall that we are deﬁning E(±) to go with e∓iωt since by convention e−iωt corresponds to the positive frequency ω and eiωt = e−i(−ω)t corresponds to the negative frequency (−ω). the Maxwell equations take on a simpler form for complex monochromatic ﬁelds. t) = E(+) (r)e−iωt + c. We can write the solutions to the monochromatic Maxwell equations as E(+) (r. ∂t we can always make the formal identiﬁcation ∂ ≡ −iω ∂t for a monochromatic ﬁeld. B(+) . the Poynting vector and intensity are relatively simple: S=E×H =E (+) = E(+) e−iωt + E(−) eiωt × H(+) e−iωt + H(−) eiωt ×H (−) (4. as is required to get a real (physical) ﬁeld. for monochromatic waves. In other words. In fact. E(r. Then the Maxwell equations become ∇ · D(+) ∇ · B(+) ∇ × E(+) ∇ × H(+) =0 =0 = iωB(+) = −iωD(+) . ω)e−iωt dt. so they are equivalent to the regular Maxwell equations. (4. Because the time dependence is of the simple form e−iωt . B. in writing down the Maxwell equations for monochromatic ﬁelds. We can then build up arbitrary solutions out of these basis functions. (4. P. That is. The physical ﬁeld is just the sum of the positive.27) Since all of the time dependence in E (+) is of the form e−iωt . we can do the same for all of the other ﬁelds H. But notice that these parts are complex conjugates. D(+) .28) (4.

The optical intensity then is simply the magnitude of the Poynting vector. (4. (+) (+) (+) (4. so all the components of the E and H ﬁelds satisfy the scalar Helmholtz equation ∇2 + k 2 E = 0.36) √ where k = ω/c = ω ǫµ0 = nk0 is the wave number in the medium. so S = E(+) × H(−) + E(−) × H(+) = E(+) × H(−) + c.33) 4. Iterating the diﬀerentiation once.6. ˆ ∂x (4.50 Chapter 4. As for the time dependence of the monochromatic ﬁeld. Recall that this expression has an implied harmonic time dependence: E (+) (r. ky .c. we can rewrite the Maxwell equations for a monochromatic ﬁeld in a simple dielectric as ∇ · E(+) ∇ · H(+) ∇ × E(+) ∇ × H(+) =0 =0 = iωµ0 H(+) = −iωǫE(+) . where we must keep in mind that a solution to the Helmholtz equation has an implied time dependence of e−iωt . (scalar Helmholtz equation) (4. The plane wave satisﬁes the scalar Helmholtz equation.c. 4.8).35) This is essentially the same set as the original Maxwell equations. This is the time-independent version of the wave equation (4. We can see this by observing that x ˆ ∂ (+) E = xikx E (+) .1 Complex Notation for Simple Dielectric Media Since D = ǫE and B = µ0 H for a simple dielectric medium. ∇E (+) = ikE (+) . E0 is a complex constant scalar and k is the wave vector. we time-average the 2ω terms. kz ). . Combining all three components. but with ∂/∂t −→ −iω. = 2Re E(+) × H(−) but now in terms of the complex ﬁeld components. and k0 = ω/c0 is the wave number in vacuum. In terms of the components (kx . We can deﬁne the plane wave as the monochromatic ﬁeld E (+) (r) = E0 eik·r .34) (4. I = E(+) × H(−) + c.41) (4.39) with similar relations holding for y and z. (4. (4. ∇2 E (+) = −k · kE (+) = −k 2 E (+) .37) (plane wave) (4.7 Plane Waves The plane wave is one of the simplest and most important solutions of the scalar wave equation. Review of Electromagnetic Theory For intensity calculations.38) Here. = 2Re E(+) × H(−) . t) = E0 ei(k·r−ωt) . the inner product k · r is simply kx x + ky y + kz z.40) . we can make the formal identiﬁcation ∂/∂x ≡ ikx for the plane wave.

Thus. so that k = kˆ. Fixed t: l z Fixed z: T = 2p/w t .45) in space. wave fronts k l Let’s simplify a bit by orienting the z-axis along k.49) where again c = ω/k is the speed of light. we recover the Helmholtz equation. = 2|E0 | cos(kz − ωt + φ) In standard form. t) = E0 cos[k(z − ct) + φ)] (4. Note that we can also write the plane wave in the form E(r.42) then we can write iφ Since e =e i(φ+2π) . Then the plane wave has the form z E (+) (r) = E0 eikz .c. t) = E (+) (r)e−iωt + c.4. (+) (4. t) = E0 cos(kz − ωt + φ). The corresponding real wave is E(r.47) (real plane wave) (4.43) for integer q. (+) (+) (+) (+) (4. Now let’s examine the phase of the plane wave. If we deﬁne φ to be the phase of the complex amplitude (+) E0 . (4. the wave repeats itself every 2π in phase. (+) (+) E0 =: |E0 |eiφ . the points of stationary phase—constant (z − ct)—move with phase velocity c.44) E (+) (r) = |E0 |ei(k·r+φ) .7 Plane Waves 51 Thus. the real plane wave is E(r.48) where E0 := 2|E0 |. separated by the wavelength (4. These surfaces of constant k · r are planes perpendicular to k. As time evolves.46) (4.c. the wave fronts are given by k · r = kx x + ky y + kz z = 2πq + φ λ := 2π k (4. where we can identify the wave number we deﬁned before with the magnitude of the wave vector k. = |E0 |ei(kz−ωt+φ) + c.

the spatial quantities are multiplied by the refractive index. t) = H(+) (r)e−iωt + c.c. as before. k = ω µ0 ǫ = = c c0 (4. Using the Maxwell equations from Eqs. For these two equations to be consistent. whereas the time quantities are unchanged: c = c0 /n λ = λ0 /n (4. Taking the magnitudes of the above two Maxwell equations. and sometimes it is insuﬃcient to treat the electromagnetic ﬁeld as a scalar.52 Chapter 4. . ∇ × H(+) = −iωǫE(+) ∇ × E(+) = iωµ0 H(+) . Thus we can write E(r.c. = H0 ei(k·r−ωt) + c. H0 .50) k = nk0 ω = ω0 .54) (4. we require ωǫ k .c. (4. plane waves are often called Transverse Electromagnetic Waves (TEM waves). We can also guess that this form holds for the magnetic ﬁeld. Here. (4. t) = E(+) (r)e−iωt + c.35). the electric ﬁeld is a vector.56) (4. (+) (4. as compared to the vacuum. = E0 ei(k·r−ωt) + c. and E0 are all mutually orthogonal.8 Vector Plane Waves In the general case.. since each of its components satisﬁes the scalar wave equation: H(r.c. the subscripted quantities refer to the vacuum. 4. Review of Electromagnetic Theory In a medium. we see that k.51) and since each component of this vector ﬁeld separately satisﬁes the wave equation. we ﬁnd ωǫ E0 k k H0 = E0 .52) By inspecting this pair of equations..53) (+) (4. = k ωµ0 which leads to nω ω √ = nk0 . Since E0 and H0 are orthogonal to k.55) where again E0 := 2|E0 | and similarly H0 := 2|H0 |. ωµ0 H0 = (+) (+) (4.57) which is the same as the condition for the wave to satisfy the Helmholtz equation. we can use the identiﬁcation ∇ ≡ ik for the plane wave to write k × H(+) = −ωǫE(+) k × E(+) = ωµ0 H(+) . this ﬁeld satisﬁes the same wave equation.

59) n where 1 µ0 = µ0 c0 = ≈ 377 Ω (4.3 W/cm2 and an electric ﬁeld of 15 V/cm. = E0 H0 (+) (−) ˆ + c. k = η (+) 2 2 E0 ˆ E ˆ + c.9 Exercises Problem 4.1. the intensity at the surface is the optical power per unit area: W 2. (2. so we can deﬁne η to be the impedance of the medium. (a) Prove the vector identity for an arbitrary vector A. Making the simplistic estimate I = P/d2 . for a plane wave. 4η 2η (4. Assuming a light conversion eﬃciency of ∼ 2. a typical light bulb uses 100 W of electric power. we ﬁnd an intensity of 0.4.63) 2 E0 . as we will see later.65) at the surface. has much higher intensity and electric ﬁeld even for a much more modest power.62) in terms of the complex ﬁeld |E (+) |.1 Wave Impedance Now if we consider the ratio of the electric and magnetic ﬁelds. we can also write the intensity I= 2|E (+) | 2 η (4. k = |E0 |2 ˆ + c.c. A laser.9 Exercises 53 4.4 .c. values comparable to the light bulb despite a much smaller power rating.8.c. 4. Returning to the Poynting vector. Thus.60) η0 := ǫ0 ǫ 0 c0 is the impedance of the vacuum. we can write the intensity in terms of the real amplitude E0 as I= Alternately. A low-power He-Ne laser might typically have P ∼ 2 mW of output power and a beam diameter of d ∼ 0.64) (4.c. 2η (4. As an example. ǫ0 n (4. (4.5% and a diameter of ∼ 5 cm.61) ˆ where k := k/k. We can also write η0 η= .5 W = 0. ωµ0 µ0 c0 E0 = = = H0 k n µ0 1 =: η.58) the constant η has the dimensions of impedance.8 mm.5 cm)2 cm2 The electric ﬁeld is thus E0 = 2ηI = 17 V cm (4. S = E(+) × H(−) + c. (Hint: try doing this ﬁrst for only one Cartesian component of this equation and then argue that the other components follow in the same way. k = 0 k.) ∇ × (∇ × A) = ∇(∇ · A) − ∇2 A .

(d) Using a similar argument. t) = −ˆ y E0 cos(kx − ωt + φ0 ). E(r. show that ω/k is the phase velocity (the velocity at which the wave peaks travel) of the wave. Be explicit about the dependence on all the parameters in the real wave. Problem 4. t). t) = z E0 ei(kx−ωt) . Show that the wave cannot be polarized along the k direction.3. Problem 4. t) = E0 cos(k · r − ωt + φ0 ). ˆ where E0 and φ0 are constants and z is the unit vector along the z-direction. and thus show that the solution is periodic in time with period T (i. t) in free space satisﬁes the wave equation 1 ∂2 ∇2 H − 2 2 H = 0. ˆ (a) Show explicitly that this solution satisﬁes the wave equation provided ω/k = c. where T = 2π/ω.2. do not make any assumption about the form of the magnetic ﬁeld. 4 (a) Show that the time-averaged energy density is given by w = . For a plane-wave solution of the wave equation.4. Recall the wave equation for the electric ﬁeld in a simple. c0 ∂t which is the same equation satisﬁed by the electric ﬁeld E(r. linear dielectric. we can write the constant E0 as eE0 . (c) Show that the wave is spatially periodic with period λ. Here. Review of Electromagnetic Theory (b) Use this identity and Maxwell’s equations to show that the magnetic ﬁeld H(r. 1 2 ǫE0 + µ0 H02 . where λ = 2π/k. where the unit vector e is the polarization and E0 is a scalar ˆ ˆ constant. E(r. linear dielectric is 1 w= ǫE · E + µ0 H · H 2 for real ﬁeld E and H. the time required for the wave to repeat itself at ﬁxed position is T ).e.54 Chapter 4. c0 is the speed of light in vacuum. ˆ with E0 (+) (+) a complex constant. Problem 4. ∇2 E − Consider the particular solution. t) = z E0 cos(kx − ωt + φ0 ). Consider a plane wave in complex notation of the form E(+) (r. Prove this directly.. (b) Show explicitly that the solution is invariant under transformations of the form t −→ t + T . The energy density (energy stored per unit volume) of an electromagnetic ﬁeld in a simple. c2 ∂t2 (f) Write down the complex form E(+) of the wave in terms of the above real wave. (e) Show that the corresponding solution for the magnetic ﬁeld is H(r. µ0 c 1 ∂2 E = 0.

giving numerical values for any parameters you use.4. Don’t worry too much about arithmetic.9 Exercises 55 where E0 = 2|E0 |. Problem 4. just reduce your numerical answers to reasonably simple fractions. Assume the laser output is a plane wave in free space with intensity 1 W/cm2 . that S = cwˆ. etc. Write down expressions for the electric and magnetic ﬁelds for the laser beam. A laser with λ = 200π nm points along the −x-direction. With the interpretation that the Poynting x vector represents an energy ﬂux density. powers. and the intensity I for the real plane wave of part (b). (d) Show. What is the signiﬁcance of the extra factor of 1/2 in this expression? (Hint: it’s not the deﬁnition of E0 . and its electric ﬁeld is polarized in the z-direction. for the real plane wave of part (b). (c) Calculate w. argue that this means that the electromagnetic energy of a plane wave propagates with velocity c.) (b) What is the real wave corresponding to the complex wave given above? Be explicit about the parameter dependence. the Poynting vector S. (+) .5.

.

Both have the same frequency. which ranges from −1 to 1.2 | 2 .7) I1.8) . we will consider the simplest case of adding two monochromatic (+) (+) waves E1 (r) and E2 (r).2 .2 .2 = we ﬁnd that I = I1 + I2 + = I1 + I2 + 2 I1 I2 ei(φ2 −φ1 ) + c. and thus an implicit time dependence of e−iωt .c. U1. (5.1) (5. η (+) (+) (+) (5.2 := 2 (+) E (r).2) To simplify our notation a bit. A useful and simple special case is where the intensities are equal: I1 = I2 =: I0 .63)that the intensity of each component wave is given by I1.5) Writing out explicitly the phases of U1.1 Superposition of Two Plane Waves Interference occurs when multiple waves are added together. Clearly the combined ﬁeld E (+) (r) is also monochromatic. Recall from Eq. (5.2 = 2|E1. η 1.Chapter 5 Interference 5. and it is one of the most obvious eﬀects not accounted for by geometrical optics. Putting these √ √ √ √ extreme values into this expression.2 (r).2 eiφ1. (4. deﬁned by U1.2 . (5. we can introduce the ﬁelds U1.2 (5. (5. the intensity of the superposition is simply ∗ ∗ I = |U |2 = |U1 + U2 |2 = |U1 |2 + |U2 |2 + U1 U2 + U1 U2 . we see that the intensity ranges from ( I1 − I2 )2 to ( I1 + I2 )2 .2 = U1. (5.6) Notice that all the phase dependence is in the factor cos(φ2 − φ1 ). The superposition of the two waves is just the sum E (+) (r) = E1 (r) + E2 (r). First.3) so that the intensity is simply the square of the U ﬁeld without any extra coeﬃcients: 2 I1. I1 I2 cos(φ2 − φ1 ).4) In this notation.

where φ1 = . 2 λ (5. and this situation is fully destructive interference. Comparing to the intensity range above.12) I+ = √ + √ 2 2 2 At the second port. (5.f¡ 5. we see that the case of equal intensities ¯ is the case of maximum intensity variation. (5. while allowing the rest to pass). for the wave in path 2 just before the second beam splitter I0 iφ2 2πnd2 U2 = e . and this situation is fully constructive interference. (5. 2 I0 U2 U1 = [1 + cos(φ2 − φ1 )] . where φ2 = . so that half the intensity goes into either arm of the interferometer. I I = 2I0 [1 + cos(φ2 − φ1 )] . the intensity attains its minimum value I = 0.13) .9) Imax = 4Iº I = 2Iº Iº -2p -p 0 p 2p 3p f™ . shown here.10) where d1 is the length of path 1 and λ is the optical wavelength.58 Chapter 5. The intensity I averaged over all phases is 2I0 . mirror path 2 beam splitter Iº/2 input beam Iº/2 Iº beam splitter path 1 mirror Iò output ports I– A beam of intensity I0 is split into two components by a beam splitter (a beam splitter is simply a device that deﬂects part of the intensity of an optical beam. the intensity attains its maximum value I = 4I0 . as we might expect on average for adding two beams of intensity I0 from energy conservation. The wave in path 1. Similarly. At the ﬁrst port. One of the simplest interferometers is the Mach-Zehnder interferometer. If φ2 − φ1 = π. For simplicity we will assume a 50% lossless beam splitter. 2 (5. just before the second beam splitter.11) 2 λ Then the intensities of the two output ports are given by squaring the sums of the intensities. but with an extra minus sign: U1 U2 I− = √ − √ 2 2 2 = I0 [1 − cos(φ2 − φ1 )] . The two components are then recombined on the second beam splitter and the intensities I+ and I− can be monitored at their respective output ports.2 Mach-Zehnder Interferometer An interferometer is a device that uses the interference eﬀect to do something useful. we also add the ﬁelds. Interference In this case the intensity becomes If φ2 = φ1 . is U1 = 2πnd1 I0 iφ1 e .

So we can write this ﬁeld as rrU0 + t′ tU0 . From geometrical optics. so we can write this ﬁeld as trU0 + r′ tU0 . since the total output intensity is I+ + I− = I0 . Uº rUº tUº Now let’s consider all the reﬂections and transmissions from the two incident beams that form the two outgoing beams in the time-reversed case. Uº rUº tUº In standard terminology.5. then the amplitude of the reﬂected ﬁeld is rU0 . which is the same as the input intensity.g. r is the ﬁeld reﬂection coeﬃcient and t is the ﬁeld transmission coeﬃcient. and the amplitude of the transmitted ﬁeld is tU0 . 5. although we don’t know how much of the light will go in each direction. there really is none due to complete destructive interference.3 Stokes Relations The Stokes relations follow elegantly from the time-reversal invariance of electromagnetism.14) Since time appears only as a second derivative. rrUº + t¢tUº rUº trUº + r¢tUº tUº . On the top side. Similarly. we know that part of the beam will transmit (refract) and part will reﬂect. 2 c0 ∂t2 (5. Note that in principle this is possible and it works for the purposes of this derivation. We can see this invariance directly from the electromagnetic wave equation: ∇2 E − 1 ∂ 2E = 0. ﬁrst of all. The primed coeﬃcients r′ and t′ denote the reﬂection and transmission coeﬃcients for a ray incident from below. the outgoing beam on the bottom comprises a transmission of the rU0 beam and a reﬂection of the tU0 beam. This wouldn’t have worked out if we omitted the extra minus sign. the outgoing beam comprises a reﬂection of the rU0 and a transmission from the tU0 beam. But we can do a better job of justifying it by considering an important time-reversal argument. this happens by constructive interference. the equation is invariant under the replacement t −→ −t. it is often very diﬃcult to implement ideal time-reversed situations such as this. they should combine to form a single beam that travels away from the interface. Time-reversal invariance states that this should also work in reverse: if two waves of amplitude rU0 and tU0 are incident from opposite sides of the same interface. In the other opposite location where you might expect to see a beam going away from the interface. a dielectric interface).. we can see that energy conservation works in this case. which are possibly diﬀerent from r and t. In practice. Of course. So we’ll just deﬁne coeﬃcients r and t. Consider a beam of light at an interface between two media (e. such that if U0 is the amplitude of the incident ﬁeld.3 Stokes Relations 59 Why the extra minus sign in I− ? Well.

5. Since r2 = 0. and the bottom beam must vanish. the reﬂection coeﬃcient must be −r. so for the top beam we require that r2 U0 + t′ tU0 = U0 . 5. This new conﬁguration is the Michelson interferometer. Simplifying these two equations a bit. where the shape of the optical fringes reﬂects the spatial index proﬁle of the object.16) (5. A physical beam splitter is typically something like a slab of glass.5 for the 50% splitters.60 Chapter 5. The other arm acts as a “phase reference” for the phase-shifted beam.17) (5. but it works out that they are equal. Or you can visualize a phase object such as a ﬂame or a gas ﬂow. For example. so trU0 + r′ tU0 = 0. we arrive at the Stokes relations r′ = −r t′ t = 1 − r 2 (Stokes relations) (5. you could use this technique to measure the index of refraction of gas in a cell by measuring the phase shift as a function of the gas pressure. the beam in the other arm of the interferometer impinges on the same interface from within the beam splitter. we have that t′ t = 1 − r2 = 0.15) in standard form.5 Michelson Interferometer The Mach-Zehnder interferometer can be “folded” so that the two beam splitters coincide and can be replaced by a single splitter. We can’t work out exactly what happens to the transmission coeﬃcients from the Stokes relations.5. . Thus. rUº Uº tUº Uº t¢Uº r¢Uºoo=o -rUº If the reﬂection coeﬃcient is r for the beam that impinges on the beam splitter from the outside.4 Mach-Zehnder Interferometer: Applications Since the Mach-Zehnder interferometer has two spatially separated but otherwise equivalent paths. where the splitting occurs at one interface and the no splitting occurs at the other (often this is accomplished with special dielectric coatings). Interference But the last two diagrams must be consistent with each other. Now we can use the Stokes relations to analyze what happens at a physical beam splitter. for the other arm. it is typically used to measure an optical phase shift (path-length change) due to an object in one arm. so there is no sign change between t and t′ .

detector d™ fixed mirror d¡ free-falling corner cube This method is still the state-of-the-art method for sensitive measurements of g. which uses some of the largest and most impressive Michelson interferometers built thus far.5. The strain sensitivities achieved so far are better than 10−21 . .5 Michelson Interferometer 61 mirror d¡ input beam beamsplitter d™ mirror output port Note that in this diagram. For example. Assuming the same refractive index for both arms. One example is the interferometric gravimeter. corresponding to detecting the displacement of a macroscopic mirror to a precision of much less than a nuclear diameter. The interferometer would show a fringe shift if it moved through the aether (because it was stationary on Earth) depending on its orientation with respect to its velocity. Yet another example is the LIGO (Laser Interferometric Gravitational Wave Observatory) experiment. but often in this conﬁguration you would only monitor the other output port. which would deﬁne a “rest frame” for light. The corner cube acts as a freely falling mass. with many technical reﬁnements to detect displacements of mirrors due to gravitational waves.” the fundamental medium through which light propagates. Just as in the Mach-Zehnder case. and by monitoring the output it is possible to get a real-time position measurement of the cube’s position. the conclusion being that there is no preferred rest frame for light propagation. the input port is also one of the output ports. the Michelson interferometer measures the path-length diﬀerence between the two arms. the output intensity is Iout = I0 1 + cos 2 2πn(d2 − d1 ) λ . one arm can act as a reference distance. The idea here was to search for an “aether. which is a Michelson interferometer where one of the mirrors is actually a corner-cube reﬂector (a prism that retroreﬂects the incident beam in the same direction independent of small changes in its orientation).18) This interferometer is very useful for measuring lengths. No eﬀect was observed. (5. The arms of the LIGO interferometers are 4 km long. Another important application of the Michelson interferometer was its original version and namesake in the Michelson-Morley experiment. while the output changes depending on how far the other “probe” mirror moves.

(5.22) .9). Interference 5. if the interferometer rotates. U1 = I0 eikz . 5. the interference pattern varies sinusoidally with spatial period λ 2π = . the Sagnac interferometer is a very eﬀective rotation sensor. k sin θ sin θ (5. giving an eﬀective phase shift between the two beams. which uses this interferometer as the cavity for a laser for sensitive detection of rotation. so really there are not two distinct “arms” of this interferometer.19) and a second wave of equal intensity propagating at an angle θ with respect to the ﬁrst. Rather. there is no sense of distance measurement in this system.20) U¡ z I(x) Iº q Iº U™ z=0 For a screen at z = 0.6 Sagnac Interferometer The Sagnac interferometer is another variation on the Mach-Zehnder conﬁguration. Since there is no path-length diﬀerence. The mirror causes the beams to continue around. where the second beam splitter is replaced by a mirror. U2 = I0 ei[k(cos θ)z+k(sin θ)x] . so from Eq. Consider a plane wave propagating in the z-direction. the phase diﬀerence between the two waves is just (k sin θ)x. x (5. One common instrument that uses this eﬀect is the ring-laser gyroscope.21) Thus.62 Chapter 5. (5.7 Interference of Two Tilted Plane Waves A simple but useful eﬀect to understand is the interference pattern between two plane waves with slightly diﬀerent angles. However. Thus. (5. the symmetry between the two beams is broken due to the Doppler eﬀect. the intensity on the screen is Iscreen = 2I0 {1 + cos[(k sin θ)x]} . two beams traverse the same ring but in opposite directions.

.23) In general.. This corresponds to an interference pattern with tall. this is a very complicated problem. (5. The Young double slit experiment is another example of a simple interferometer.8 Multiple-Wave Interference A useful generalization of the two-beam interference theme is the interference due to N waves: U = U1 + U2 + · · · + UN . n = 1. narrow peaks of height N 2 I0 with N − 2 sub-lobes in between.g. Two slits in a planar wall are illuminated by the same light source. so that Un = Then I0 hn−1 . 1 − eiN φ . for holographically fabricating diﬀraction gratings).9 Exercises Problem 5. . but we can easily treat the special case of equal amplitudes and constant phase diﬀerences: Un = I0 ei(m−1)φ .5. . because the path lengths of the two beams to the screen vary slightly with the propagation direction of the waves. sin2 (φ/2) (5. 1 − eiφ (5. Obviously this result is a useful “interferometer” in its own right: this gives you an accurate method for measuring angles or printing accurate periodic patterns lithographically (e. 2π 2p 4p No-o2 minor lobes 6p f } 0 I dφ = N I0 .24) The algebra simpliﬁes if we deﬁne h := eiφ . .8 Multiple-Wave Interference 63 This is useful as a simple model of misalignment in a two-beam interferometer. (5. which doesn’t change anything. planar screen (parallel to the ﬁrst wall). we multiplied through by a factor of e−iN φ/2 /e−iφ/2 inside the magnitude. ™ I Imax = N Iº 2π N I = NIº 0 The corresponding average intensity is 1 ¯ I= 2π again as we expect from energy conservation.1. 5.27) Here. 2. N.26) = I0 sin2 (N φ/2) .25) 1 − hN = 1−h The corresponding intensity is the squared magnitude of this expression: U= I0 1 + h + h2 + · · · + hN −1 = I0 I = I0 e−iN φ/2 − eiN φ/2 e−iφ/2 − eiφ/2 2 I0 (5. . Interference fringes appear in the light on a distant.28) 5. (5.

You measure the fringe spacing (distance between dark bands) on the screen to be 1. Because the screen is distant. gas cell port 1 The cell is initially evacuated. Problem 5. Assume that the incident light is well described by a plane wave.2. (b) Suppose that the two slits are 0. What is the refractive index of the gas at the ﬁnal pressure? . Describe the interference pattern on a screen at the output of the interferometer.64 Chapter 5. What is the wavelength of the light? (c) What type of laser could plausibly be the source of the light in part (b)? Problem 5. Suppose that one of the retroreﬂecting mirrors has a small misalignment of angle δθ. you observe that port 1 goes dark then becomes bright again a total of 10 complete cycles.266 cm. As the cell is pressurized. The input light is λ = 200π nm. you can assume the two waves propagating to any particular point on the screen are approximately parallel. Also describe how the pattern changes as either mirror moves.3. Consider a Mach-Zehnder interferometer with a gas cell of length 10 cm in one arm as shown. and all of the interferometer light exits output port 1.5 mm apart and diﬀract monochromatic light onto a screen 10 m away. Consider a Michelson interferometer illuminated by laser light. Interference (a) Sketch the setup and show that the angular distance between successive dark bands (areas of destructive interference) is ∆θ = λ/a. relating aspects of the interference pattern to δθ as appropriate. where λ is the wavelength of the light and a is the separation between slits.

we ﬁnd ∇2 + i2k T where ∇2 := T ∂2 ∂ + 2 ∂z ∂z ψ = 0. Thus.3) is the transverse Laplacian. the envelope ψ should be smooth on the scale of λ.5) (6. ψ is the envelope and eikz is the carrier wave. For a scalar electric ﬁeld propagating along the z-direction.4) (6.e. If ψ is smooth (i. ∂z λ which we can rewrite as We can similarly write ∂ψ ≪ kψ.1 Paraxial Wave Equation Just as paraxial rays greatly simpliﬁed ray optics by allowing us to use linear propagation equations. varies slowly) on the scale of λ.Chapter 6 Gaussian Beams 6. (6. A paraxial wave is one where the curves normal to the wave fronts are paraxial rays..6) . then ψ ∂ψ ≪ . Plugging this into the wave (Helmholtz) equation. paraxial waves are much simpler to treat than waves in the general case. E (+) (r) = ψ(r)eikz . For a paraxial wave. ∂z ∂ψ ∂2ψ ≪k . we can factor out the plane-wave part of the paraxial wave.1) Here. (6. OA z Paraxial waves propagate (more or less) along the optical axis. (∇2 + k 2 )E (+) = 0. so they should be fairly close to a plane wave. 2 ∂z ∂z (6.2) ∂2 ∂2 + 2 ∂x2 ∂y (6.

2). t): 2 − 2m ∇2 ψ + V (x.0 (for “transverse electromagnetic”. y)ψ = i T ∂ψ . (+) we have used polar coordinates (r.” not an “omega. It also turns out that the outputs of sphericalmirror resonators and lasers are often Gaussian beams.8) We can see this equivalence by rewriting this equation as ∇2 − T 2m 2 V (x. The Gaussian beam is just like the Gaussian wave packet in quantum mechanics. .13) is the radius of curvature of the wave fronts. as we will see. In this expression.7) It is interesting to note that this is exactly the same as the two-dimensional Schrödinger equation in quantum mechanics for the wave function ψ(x.12) (6. being just complex enough to be interesting but simple enough to be tractable. at least in the paraxial approximation. so why would we want to study something more complicated like a Gaussian beam? The Gaussian beam is the simplest model of a directed beam that satisﬁes Maxwell’s equations. z) with r = x2 + y 2 . Gaussian Beams and thus we can neglect the ∂ 2 /∂z 2 term in Eq. and the resulting wave equation is the paraxial wave equation: ∇2 + i2k T ∂ ∂z ψ = 0. (6. at least until we study Fourier optics): E (+) (r) = E0 (+) w0 r2 exp ikz − i tan−1 exp − 2 w(z) w (z) z z0 exp ik r2 . These are useful models.10) This is the Gaussian or TEM0.” by the way). which we will come to later). which is of the form cos(kx − ωt). and R(z) := z 1 + z0 z 2 1+ z z0 2 (6.9) and then making the formal identiﬁcations z −→ t and k −→ m/ . w0 := λz0 π (6. E0 is as usual an overall ﬁeld-amplitude constant. y) + i2m ∂ ∂t ψ = 0. The potential V (x. ∂t (Schrödinger wave equation) (6.66 Chapter 6. though in a somewhat subtle way—the obvious correspondence is to the Schrödinger equation in free space.11) is the beam waist parameter or beam radius. (paraxial wave equation) (6. z0 is a constant called the Rayleigh length (or “Rayleigh range”). as for the plane wave) beam (the latter name comes from the hierarchy of Hermite–Gaussian beams.2 Gaussian Beams The simplest waves are the plane wave. 6. y. w(z) := w0 is the beam waist as a function of z. Let’s start by just writing it down (we will take it for granted that this satisﬁes the paraxial wave equation. and the spherical wave. This is called the slowly varying envelope approximation. (6. (that’s a “double-u. 2R(z) (6. which is of the form cos(kR − ωt)/R. y) turns out to be related to the refractive index proﬁle.

A circle of radius r = 1. Near the focus at z = 0. In this sense.14). note that the width of the Gaussian is w(z). we have deﬁned I0 := 2|E0 |2 /η as the maximum intensity of the Gaussian beam (which occurs at the center r = 0).5w(z) contains 99% of the power of a Gaussian beam.16) P = πw0 . There are several important features to understand here: Intensity Proﬁle. the hyperbola approaches its asymptotes.17) (+) It is useful to consider some similar integrals over a disc of ﬁnite radius. Note that the beam waist w(z) traces out a hyperbolic curve in z. 2R(z) (+) (amplitude factor) (longitudinal phase factor) (radial phase factor) (6. (6. thus.18) for the regions near to and far away from the focus. At large distances from the focus. which completely describes the intensity proﬁle of the Gaussian beam. hence the name “Gaussian beam. this is the most useful factor for understanding the behavior of the Gaussian beam. 2 0 This result holds independent of z. z) = 2r2 2P . z) 2πr dr = (6. For example. Thus we can rewrite the intensity proﬁle as I(r.2 Gaussian Beams 67 This pretty much looks like a big mess. 6.14) We will now analyze each factor separately. but the other two factors make no contribution: 2 w0 2r2 I(r. The intensity is just the squared modulus of Eq. First. which reduces to the minimum value w0 at z = 0. (6. let’s try rewriting the Gaussian beam solution in a way that’s partitioned into three smaller factors: E (+) (r) = E0 w0 r2 exp − 2 w(z) w (z) z × exp ikz − i tan−1 z0 r2 × exp ik .6. . Beam Divergence. as a measure of how “large” is a Gaussian beam. the intensity falls oﬀ in the radial direction like a Gaussian function. Clearly. w(z) ∼ w0 w(z) ∼ w0 z0 for |z| ≪ z0 z for |z| ≫ z0 (6.2. respectively. a circle of radius r = w(z) contains 86% of the optical power of a Gaussian beam. exp − 2 2 (z) πw w (z) (6.1 Amplitude Factor The ﬁrst factor in Eq. The Rayleigh length z0 marks the crossover between these two regimes. the waist achieves its minimum value w0 .14) is the amplitude factor. we just have to break it down and analyze each piece to see what is going on. right? It’s not nearly as bad as it looks. Total Power. z) = I0 .” Also. given by (w0 /z0 )z.15) exp − 2 w(z) w (z) Here. The total power of the Gaussian beam is given by integrating the intensity over a transverse plane: ∞ I0 2 I(r. (6.

the best focus occurs at z = 0. 145 (1891). the beam propagates in the form of a cone of half angle θ0 . we have θ0 ≈ λ w0 = . C.20) Thus. At a distance of one Rayleigh length z0 from the focus.19) zº qº qºz -qºz λ w0 = . λ (6. whereas a larger beam (large w0 ) diverges less. in the far ﬁeld we have I(r = 0. Gaussian Beams w(z) wº -zº -w(z) From this we see that in the far ﬁeld.21) so the Gaussian beam is consistent with the inverse-square intensity law in the far ﬁeld (as is also the case for the spherical wave). 6e série 24. Gouy. z0 marks the transition from focused behavior to far-ﬁeld divergence. “Sur une propriété nouvelle des ondes lumineuses. R. so the “depth of focus” for a Gaussian beam is 2z0 .2. 1251 (1890). a tightly focused beam (small w0 ) diverges more quickly. As for the intensity. 6. As we just mentioned. where tan θ0 := In the paraxial regime.22) w(±z0 ) = 2 w0 . but 1 C. which is that larger apertures produce smaller depths of focus.23) From this relation we can see explicitly that tightly focused beams have a smaller depth of focus.” Comptes Rendus Hebdomadaires des Séances de l’Académie des Sciences (Paris) 110. Depth of Focus.2 Longitudinal Phase Factor Recall that the longitudinal phase factor has the form exp ikz − i tan−1 z z0 . This is related to a general principle in imaging optics. the beam waist increases by 41% compared to the focal value: √ (6. The longitudinal phase is dominated by the ikz term. The second term is called the Gouy phase shift1 and represents a small departure from planarity. . “Sur la Propagation Anomale Des Ondes. z0 πw0 (6. Again.” Annales de Chimie et de Physique. z0 πw0 (6. R. So we can say that the beam is roughly focused between ±z0 . Gouy. (6.68 Chapter 6.24) The ﬁrst term in the phase is simply the phase of a plane wave ikz propagating in the same direction and with the same optical frequency as the Gaussian beam. (6. where again 2z0 = 2 2πw0 . z) ≃ I0 z0 z 2 .

26) This factor gives the dependence of the phase on r. since the spacing between wave fronts is slightly larger than λ. z ≈ R in the paraxial approximation. 2R(z) z0 z 2 (6.29) E r2 = 0 exp ik|z| + ik R 2|z| ≃ r2 E0 . it implies that the phase velocity of a Gaussian beam is slightly larger than c.6. we see that near and far away from the focus.2. (6. E (+) (r) = √ E E0 2 2 eikR = 0 eik r +z . In the paraxial case. As we will see. which represents only a small correction.30) R(z) ∼ z Again.25) . It represents a phase retardation compared to the plane wave. 6.2 Gaussian Beams 69 don’t go underestimating the importance of the Gouy term. R R (+) (+) (6. R(z) in the Gaussian beam represents the radius of curvature of the wave fronts. we need to compare this phase to the phase of a spherical wave.3 Radial Phase Factor Recall that the radial phase factor is of the form exp ik where R(z) = z 1 + r2 . and the fact that the radius of curvature changes with z for the Gaussian beam. (6. R(z) ∼ 2 z0 z for |z| ≪ z0 for |z| ≫ z0 (6. the Gouy phase. we see that in the paraxial approximation. the Gouy phase is important in computing the resonant frequencies of optical resonators. exp ik|z| + ik R 2R (+) Comparing this phase to Eq. The tan−1 form implies a monotonically increasing retardation. which accounts for the fact that the spherical wave propagates outward from R = 0.27) To understand this. . amounting to a total of π change in phase over all z. the spherical wave and Gaussian beam have the same form except for the absolute value of z. 2R(z) (6. Also. Thus the entire phase factor is exp ikz − i tan−1 z z0 + ik r2 .27). whereas the longitudinal phase factor simple gave the on-axis phase behavior. Gouy eﬀects are generic to focusing-beam-type solutions to the wave equation. so E (+) (r) = ≃ r2 E0 exp ik|z| 1 + 2 R z E0 r2 exp ik|z| 1 + 2 R 2z (+) (+) (+) (6. Examining the radius of curvature. so the Gaussian beam approaches a spherical wave for large |z|.28) where R = |r|. Thus. r ≪ z.

6. But in particular. From a Fourier-transform point of view. This allows the wave to map back on itself. relatively little information is needed to completely specify it.4 Vector Gaussian Beams We have thus far only treated Gaussian beams as solutions to the scalar wave equation. the wave-front curvature decreases as for a spherical wave. the curvature is maximum) by diﬀerentiating R: d dR = dz dz z+ 2 z0 z =1− 2 z0 . recall that for a confocal cavity. In fact. the tradeoﬀ between tighter focus and larger divergence angle is precisely the manifestation of the uncertainty principle in the Gaussian beam. or it is suﬃcient to know w(z) and R(z) at some distance z. the ﬁeld is mostly transverse. it is suﬃcient to know w0 and R(z) at some distance z. The maximum radius of curvature is R(±z0 ) = 2z0 . For example. ˆ z − iz0 (6.. Optical beams are nearly transverse (TEM).3 Speciﬁcation of Gaussian Beams Despite the complex form of the Gaussian beam. and the optical wavelength λ all the other parameters are uniquely ﬁxed. By construction of the Gaussian beam. Gaussian Beams We can also see where R(z) is minimum (i.2 . Thus.32) Note that the polarization is mostly along the x-direction in the region where the paraxial approximation holds.e. The divergence is set by the width of the Fourier transform of the wave packet. the boundary conditions imposed by the cavity mirrors require that the curvature of the spherical mirrors and the curvature of the wave fronts match. . which we know is ﬁxed via the uncertainty principle. It turns out that for the full vector electromagnetic ﬁeld case. the wave fronts become ﬂat (as we could guess from symmetry considerations). That is. the only diﬀerence is in the overall amplitude. 2 z (6. so maximum curvature occurs at the Rayleigh length. this is because a Gaussian wave packet is uniquely determined by its center and width. dR/dz = 0 when z = ±z0 . Hence. 6. z0 is also referred to as the confocal parameter.2 | has wave fronts that exactly match the mirror curvatures at z = ±z0 . if we know where z = 0 is located. we can obtain the vector ﬁeld from the scalar ﬁeld by the simple replacement E0 (+) −→ E0 (+) −ˆ + x x z . R(z) ™ ~ozºo/z ~oz 2zº zº z In a cavity.31) Thus. since |x| ≪ |z + iz0 |. the length d of the cavity is the radius of curvature of both mirrors R1. Alternately. the value of w0 . and farther out. This is one reason why Gaussian beams can exist in resonators. Closer to the focus.70 Chapter 6. supposing we have a Gaussian beam propagating along the z-direction and polarized in the x-direction. A Gaussian beam with 2z0 = |R1. the center is zero.

The Gouy phase ends up being just the phase angle of 1/q. In the paraxial approximation. which happens for a Gaussian beam if w0 ≫ λ.38) 1 + (z/z0 )2 . we have used the notation q0 := q(0) = −iz0 .0 for the Gaussian beam. (6. Et + kEz = 0. Here is the other nice thing about q: it transforms through an optical system according to a simple law related to the geometrical optics ABCD matrix for the system. z + iz0 z z0 1 1 = 2 = 2 = z2 + z 2 + i z2 + z 2 . Thus. (6.39) Again.5 ABCD Law 71 but the transverse property is not completely necessary (as it is for the plane wave). (6. Note that q determines the Rayleigh length and the location of the beam focus. justifying the label of TEM0. Writing out this condition. ∂Ex ∂Ey ∂Ez + + = 0. (ABCD Law) (6.34) (6.6.5 ABCD Law The ABCD Law is a quick method for propagating Gaussian beams. while q also represents the width and curvature of the beam. the ﬁeld is mostly transverse.37) It is worth reiterating that the longitudinal coordinate z is measured with respect to the beam focus. It is convenient to deﬁne a complex q parameter by q(z) := z − iz0 . 6. so that ∇T · E + ∂Ez /∂z = 0. We can see that this simpliﬁes things greatly by rewriting the Gaussian beam in a compact form in terms of q: ikr2 (+) q0 exp(ikz). the ABCD law reads Aq1 + B q2 = . Putting in approximate numbers. so q completely determines the Gaussian beam geometry. In terms of the matrix elements. we can compute the inverse of q. 2w0 2 2 Ex + Ey and 2w0 is the beam diameter. we can write 1/q in the 1 iλ 1 = + .40) exp E (+) (r) = E0 q(z) 2q(z) Here. w(z) = w0 (6. as long as ∇ · E = 0. We’ll begin by just stating it. q z − iz0 z + z0 0 0 Recalling that w0 = more useful form λz0 /π.41) Cq1 + D . q(z) R(z) πw2 (z) (6. and therefore determines the entire geometry of the Gaussian beam. the ﬁeld is transverse to good approximation.33) ∂x ∂y ∂z The ﬁrst two terms represent a transverse divergence. (6.35) where Et is the transverse ﬁeld 1 λ Ez ∼ ∼ . q determines both the spot size and the radius of curvature at a given location. Alternately.36) So as long as the beam intensity falls oﬀ on a length scale large compared to λ. Et 2kw0 4πw0 (6. and R(z) = z[1 + (z0 /z)2 ]. and then we’ll justify it later.

R1. (6. For the (paraxial) rays normal to the wave fronts.1 Free-Space Propagation The matrix for propagation over distance d in free space is M= Thus. . we ﬁnd D 1 =C+ .2 Thin Optic Recall that for a thin optic.2 (6. Gaussian Beams where q1 and q2 are the q parameters before and after the optical system.50) (6. C + D/q1 1 = . (6. by repeating the argument with the replacement d −→ d/n. we ﬁnd D 1 =C+ q2 q1 =⇒ q2 = q1 . The real part states that z2 = z1 + d. applying the ABCD law gives q2 = 1 0 d 1 . In ray optics. respectively. Thus. the matrix aﬀects the transmitted angle. (6.48) Putting this into Eq. 6.5.43) Aq1 + B = q1 + d.2 . 6.46) C D This condition is also equivalent to the condition that the Gaussian beam waist does not change across the optic (w02 = w01 ).72 Chapter 6.45) The imaginary part of this equation states that the Rayleigh length is unchanged. z2 − iz02 = (z1 + d) − iz01 .47) In the paraxial approximation.5. ′ ′ y2 = Cy1 + Dy1 .44) Writing out the real and imaginary parts explicitly.49) so we recover the ABCD law in this restricted case. and then use them to argue that the ABCD law is generally valid.42) Let’s check this for a couple of simple examples. and thus the shape of the beam is unaﬀected. the Gaussian beam is equivalent to spherical a spherical wave of radius R1 or R2 before or after the optic. or in alternate form. z02 = z01 . the position of the ray does not change (y2 = y1 ).2 ≈ y1. that is.47). so that power density on each side of the optic is consistent. R2 R1 Using 1/R = 1/q − iλ/πw2 and D = det(M) = n1 /n2 = λ2 /λ1 . q2 A + B/q1 (ABCD Law) (6. Note that the argument carries through for a dielectric medium of refractive index n. (6. we can write ′ y1. the ABCD law gives a sensible result. Thus. as we expect for a propagation of the same distance. (6. respectively. Cq1 + D (6. the position along the beam is shifted by d. the matrix for a thin optic has the general form 1 0 M= . Cq1 + D (6.

thus by the cascading rule it works for the matrix M. 1 (6.5. .56) We have already veriﬁed that the ABCD law works for any of the component matrices. the ABCD law works for cascaded systems. (6. Consider a system comprising (1) a propagation in free space over distance d1 . Consider a general ray matrix. (6.5 ABCD Law 73 6. We now know that the ABCD law works for any cascade of propagations and thin optics. the ABCD law should then work for any optical system. It is possible to verify directly by substitution that if A2 q2 + B2 A1 q1 + B1 q3 = and q2 = .6. We can argue this more precisely without so much hand-waving. as well as for a thin optic. (6. 6. we can always in principle break any medium up into a cascade of inﬁnitesimal propagations and weak.51) C2 q2 + D2 C1 q1 + D1 then q3 = Aq1 + B Cq1 + D (6. Even for continuous media. (2) a thin lens of focal length f . The ray matrix is d2 1 0 1 0 1 1 d1 .52) provided that the new matrix elements are obtained by matrix multiplication in the proper order: A C B D = A2 C2 B2 D2 A1 C1 B1 D1 .54) C D We can write down an explicit factorization of this matrix in terms of propagation and thin-optic matrices. In principle. (3) a planar refractive interface to a homogenous medium of index n. thin optics. however.55) M′ = n 1 − 0 1 0 n 0 1 f Then these two systems are equivalent (M = M′ ) provided that n = det(M) det(M) n =− C C D − det(M) D −f = d1 = C C (A − 1)det(M) d2 = f (1 − A) = C f =− (6.4 Factorization of a General Matrix We showed that the ABCD law works in the case of propagation in free space (and thus also a homogenous medium). A B M= .3 Cascaded Optics As in ray optics.5. we can handle Gaussian-beam propagation through any composite system via the composite ray matrix. and ﬁnally (4) a propagation over distance d2 in the medium.53) Thus.

But the important point is that Gaussian beam optics is really just ray optics. at least within the paraxial approximation. even for a classical initial state. As we will see. is unchanged—again. Gaussian beams do not show any manifestly wave-like behavior. Showing this mathematically really requires the Wigner transform. one would not expect a beam in an unstable resonator to remain Gaussian. wave phenomena are absent in the paraxial approximation. For example. Nonlinear systems dynamically generate quantum eﬀects. 6. in some sense. free-space propagation and the harmonic oscillator). The ray matrix is 1 0 . The setup is the same as for the last example. let’s consider the focusing of a Gaussian beam by a lens.. this is equivalent to the relation between quantum and classical mechanics for linear systems. nonlinear terms become important).6 Example: Focusing of a Gaussian Beam by a Thin Lens As a more useful example of the ABCD law in action. q2 A + B/q1 f z01 (6.5. As soon as the paraxial approximation breaks down (i. Again. However. (6. Gaussian Beams 6.58) = q1 R1 πw2 (z) z01 where we recall the initial Rayleigh length is z01 = πw2 (z)/λ.5. 2wº¡ 2wº™ beam waist d beam waist . as we expect for any thin optic. the beam will become non-Gaussian.5. you can attribute any quantum eﬀects in a linear system to the initial condition. 1 (6. Applying the ABCD law. a bundle of geometrical rays with a Gaussian density proﬁle (in both y and y ′ . this transformation rule applies in more general situations as long as the paraxial approximation holds.e. All the diﬀraction-type eﬀects can be mimicked by an appropriate ray ensemble. within the paraxial approximation. 6.59) Thus.74 Chapter 6.7 Example: Minimum Spot Size by Lens Focusing We can do a similar analysis to compute the minimum spot size obtained by a focusing lens for a Gaussian beam. which means that the Rayleigh length. But let’s focus on the Gaussian beam for concreteness. in a way that satisﬁes the uncertainty principle) follows the Gaussian beam exactly. so that the initial radius of curvature R1 = ∞. so we won’t get into this here. and z02 = z01 . For linear systems (i.e. We will consider the special case where the lens is placed at the waist of the beam to be focused. and thus the beam size.57) 1 − f and the initial beam parameter is 1 iλ i 1 = + . since paraxial wave propagation is equivalent to quantum mechanics. the radius of curvature at the output is R2 = f . 1 C + D/q1 1 i = =− + . quantum and classical mechanics turn out to be equivalent. That is. not to the time evolution.5 “Deeper Meaning” of the ABCD Law It seems amazing that a solution to the wave equation can be propagated using classical rules.. but we will also consider a free-space propagation over a distance d after the lens. The Gaussian beam stays Gaussian through any optical system as long as the paraxial approximation is valid. But really what this is saying is that.

π (realistic minimum spot size) (6. πw01 π fd f2 ≈ . so R2 = ∞ and thus the denominator of this expression must vanish: 2 − (f − d) + df 2 /z01 = 0 =⇒ d = f 2 . let’s consider f = 0. which is sensible when trying to focus to a small spot in view of the reciprocal relationship between the spot size w0 and the divergence angle θ0 .66) (6.5 · w01 .61) 2 −(f − d) + df 2 /z01 (f − d)(f /z01 ) + df /z01 1 = .5 m.64) Note that the focus is not located a distance f away as in classical optics—the focus occurs at a slightly smaller distance. .68) recommended by the Melles-Griot catalog. A simple approximate correction to account for aberrations2 is to multiply this result by 4/3: w02 ≈ 2 As 4 λ(f /#). and w01 = 1 cm. z01 −→ ∞ (i. the eﬀect is very small (only a 1 ppm reduction in the focal distance). with D denoting the diameter of the lens. 1 + f 2 /z01 (6. Then d ≈ f and z02 ≈ Thus. w02 ≈ 3 λf ≈ λ(f /#). the plane-wave limit).e. .67) where f /# (the “f -number”) is given by f /D. The imaginary part of Eq. Since z01 ≫ f . z01 z01 (ideal minimum spot size) (6. (6. 2 λ = 2π × 10−7 m (close to a He-Ne laser line). (6. −(f − d) + df 2 /z01 But plane 2 is at the beam waist.65) We will consider the limit z01 ≫ f . where we can ignore aberrations. For some sample numbers. We only recover the geometrical result d = f in the limit of a large input beam. This calculation is for an ideal lens. Now we can come to the main point.63) R2 = 2 . This isn’t really a large eﬀect. the system matrix is 1 d 0 1 1 1 − f d f = 1 1 − f 0 1− d 1 M= Applying the ABCD law. Thus.+i . q2 A + B/q1 f − d + idf /z01 (6. we are assuming a diameter of 2 · 1. In writing down this expression. a beam with a ﬁnite z01 (or w01 ) is still “like” a converging beam at the focus. which passes 99% of the total incident beam power. C + D/q1 −1 + if /z01 1 = = .60) Separating the real and imaginary parts.39).62) gives z02 = 2 (f − d)2 − (f d/z01 )2 πw02 . the minimum spot size. z01 = πw01 /λ = 500 m. In this case. (6. = λ (f − d)(f /z01 ) + f d/z01 (6.5 ABCD Law 75 Thus. Roughly speaking. respectively.62) From Eq.. we can equate the real and imaginary parts of this expression with 1/R2 and 1/z02 .6. 2 − (f d/z )2 q2 (f − d) (f − d)2 − (f d/z01 )2 01 (6. (f − d)2 − (f d/z01 )2 (6.

69) These are also referred to as TEMl. Alternately. except for the overall scale. (2) the normalization factor just before the Hermite (+) polynomials. all the same transformation rules apply to the Hermite–Gaussian beams. any beam can be represented as a linear combination of Hermite–Gaussian modes. They have the same far-ﬁeld divergence characteristics as for the Gaussian beam (except for the diﬀerent spatial proﬁle). the Hermite–Gaussian beams have intensity patterns that stay the same as they propagate in space. and (3) the factor of (1 + l + m) in the Gouy phase. and its spatial evolution is a manifestation of the interferences between the component modes.m modes. we see that for l = m = 0 we recover the Gaussian beam. the Hermite–Gaussian modes form a complete set—that is. The Hermite polynomial Hn (z) is a polynomial of order n.m mode has l dark bands across the x-direction and m dark bands across the y-direction.m (r) = (+) E0 w0 w(z) √ √ 1 2x 2y Hm H l+m l!m! l w(z) 2 w(z) z z0 r2 . there is an important connection with quantum mechanics. chosen so that the total beam power for a given ﬁeld E0 is the same for all l and m. The higher-order modes are wider for the same value of w0 . In general. Like the Gaussian beam. the intensity pattern has a grid of l + 1 bright spots in the x-direction and m + 1 in the y-direction. the intensity pattern of the TEMl.m mode) For our purposes here. Thus. exp ik 2R(z) (6. (6. Gaussian Beams 6.0 TEM0.70) (explicit formula) (recurrence relation) r2 × exp − 2 exp ikz − i(1 + l + m) tan−1 w (z) (TEMl. We can write them out explicitly as H0 (z) = 1 H1 (z) = 2z H2 (z) = 2(2z 2 − 1) n 2 2 d e−z Hn (z) = (−1)n ez dz n Hn+1 (z) = 2zHn (z) − 2nHn−1 (z). including the ABCD law. However. it is most important to know that Hn (z) is an nth order polynomial with n zeros on the real axis. The diﬀerences compared to the Gaussian beam are: (1) the presence of the Hermite polynomials Hn (z). The Hermite–Gaussian modes form one possible set of solutions (+) El. as is consistent with our earlier notation. The Hermite–Gaussian beams are equivalent to the eigenfunctions of the quantum harmonic oscillator.76 Chapter 6.6 Hermite–Gaussian Beams If we don’t require axial symmetry. Again. TEM0. we can obtain more general solutions to the paraxial wave equation.2 .1 TEM0. Note that since H0 (z) = 1.

This beam has an axially symmetric intensity pattern. This beam is called the doughnut mode or TEM0.0 beams. A Gaussian beam is incident on an optical power meter.1 + TEM1.6. it is worth mentioning one more beam.1 ∗ mode.6 6. the equal superposition of TEM0. This mode is important in describing the output of lasers as the dominant component for slightly distorted (but still axially symmetric) Gaussian beams. TEM0.3 TEM1. A common technique in the laboratory for measuring the beam waist parameter of a Gaussian beam is illustrated in the diagram.6. where the intensity vanishes at the origin.1 TEM2.0 TEM1.0 TEM2.7 Exercises 77 TEM0.0 6.1 Doughnut Mode Before ﬁnishing.7 Exercises Problem 6.4 TEM5. which .1 and TEM1.1.

Gaussian Beams registers the total power of the incident beam. Note that we are suppressing the x-dependence for simplicity. (a) What are the parameters? (b) State (general) conditions on the parameters that guarantee the paraxial approximation is valid. z) = (0.e...78 Chapter 6.3. The measured beam radius at the output coupler (mirror) is w = 0. Gaussian) beam from a continuous-wave Ti:sapphire laser (λ = 850 nm). and the beam radius in the laser rod is w = 330 µm. Then the beam radius W (z) at the knife-edge location z along the beam is given by w(z) = α|x10% − x90% |. (a) What is the optical distance between the output coupler and the laser rod? Assume that the length of the rod is negligible (it is actually around 20 mm) and that the focus of the beam occurs in the center of the rod. z0 ).2.4. ℓlens (y) = d − 2f where d is an arbitrary length constant and y is as usual the normal distance from the optical axis.e. but it isn’t diﬃcult to extend the argument to the full three-dimensional case. A knife edge can be translated in the transverse direction to block part of the beam (i. (b) A spherical wave centered on the optical axis propagating inward has the form E (+) (R) = E0 exp(−ikR).7 mm. . Problem 6. Use the following procedure. Calculate the numerical value of α. R (+) where R2 = y 2 + (z − z0 )2 is the radial distance from the center of the wave at (y. Suppose you have a 500 mW. Also recall that the Gaussian beam is speciﬁed completely in terms of relatively few parameters. Recall that the Gaussian (TEM0. a thin lens converts a plane wave into a spherical wave. and then the position x90% where the power meter reads 90% of the total beam power.75 for sapphire. then the parts of the beam with x < xknife is blocked from reaching the power meter). (b) How far will this beam propagate (past the output coupler) before the spot size is 1 cm? (c) What is the radius of curvature of the phase front at the distance you calculated in (b)? (d) What is the amplitude of the electric ﬁeld at the center of the focus? Assume a refractive index of n = 1. where α is some constant factor. The “10-90” rule is to measure the knife edge position x10% where the power meter reads 10% of the total beam power. Problem 6. if the position of the knife edge is xknife . (a) Use Fermat’s principle to show that a thin lens of focal length f introduces an optical delay with an equivalent path length given by y2 .0 ) beam is a solution to the wave equation in the paraxial approximation. x z knife edge power meter Problem 6. Show that within the paraxial approximation. TEM0.0 (i.

7 Exercises 79 Take z0 = f . What is w0 ? . (6.7. A Gaussian beam passes through a thin lens. 2R(z) Problem 6. (6. (a) At what distance from the lens does the new focus occur? (b) What is the spot size at the new focus? Give numbers for both an ideal and a realistic lens. Verify that the compact form for the Gaussian beam in terms of the complex beam parameter. and R2 is the radius of curvature of the output beam at the lens. make the paraxial approximation (expanding to lowest order in y) and show that the ﬁeld at z = 0 can be written E (+) (R) = E0 f (+) exp(−ikf ) exp −ik y2 2f . R1 is the radius of curvature of the input beam at the lens.5. E (+) (r) = E0 (+) w0 r2 exp ikz − i tan−1 exp − 2 w(z) w (z) z z0 exp ik r2 .10). The lens is placed at the focus of the original Gaussian beam.40).8. Eq. ikr2 (+) q0 E (+) (r) = E0 exp(ikz). Show that R2 = R1 f . A Gaussian beam with w0 = 1 cm is focused by a thin lens of focal length f = 2 cm.) Problem 6.6. f − R1 where f is the focal length of the lens. Eq. Assume an optical wavelength of λ = 1.6. exp q(z) 2q(z) is equivalent to the standard form for the Gaussian beam. Problem 6.0 µm. (Use the ABCD Law. (c) What is the far-ﬁeld expansion angle? Problem 6. (a) What is the Hermite–Gaussian beam pictured below? (b) Suppose the horizontal distance between the vertical dark bands is 100 µm at the focus. (c) What is the phase shift corresponding to ℓlens (y)? Write down a plane wave that matches the spherical wave provided the lens is placed at z = 0.

80 Chapter 6.10.1 (ρ → ∞) for the TEM0.0 (ρ → ∞) w0 for the TEM0. Note: you will need to evaluate an integral of the form x2n+1 e−bx dx.org/wiki/Pinhole_camera) as of 25 February 2006. aperture screen plane wave 3 “Pinhole d camera.0 mode. but also is blurred by diﬀraction if the aperture is too small.” Wikipedia entry (http://en.1 mode. ∂b 2b 2b (b) For each mode.0 .1 (ρ → ∞) a 0 ρ w0 2 1+2 ρ w0 2 ρ w0 2 1+2 ρ w0 2 +2 ρ w0 4 for the TEM1.1 mode. show that the fraction of the total power contained in a disc of radius ρ is 2 P0. Consider an aperture illuminated by a plane wave. since the integral is just a Gaussian: a 0 xe−bx dx = 2 1 exp(−bx2 ) 2b a = 0 1 1 − exp(−ba2 ) . (a) For the TEM0. and TEM1. a 0 x3 e−bx dx = − 2 ∂ ∂b a 0 xe−bx dx = − 2 1 ∂ 1 1 − exp(−ba2 ) = 2 1 − exp(−ba2 )(1 + ba2 ) .wikipedia. P0. Gaussian Beams Problem 6.0 (ρ) ρ = 1 − exp −2 P0.1 modes. The idea is to compute the optimal size of the pinhole to get the best image resolution: the image on the screen of a distant point source must be at least the size of the aperture. 2b Other cases may be evaluated recursively. and P1. 2 For example. The light passing through the aperture is viewed on a screen a distance d away. Lord Rayleigh3 √ found an optimal pinhole diameter of 1.1 (ρ) = 1 − exp −2 P0. what radius contains 99% of the total beam power? Problem 6. noting that a 0 x2n+1 e−bx dx = − 2 ∂ ∂b a 0 x2n−1 e−bx dx. The optimal diameter can be chosen as a compromise between the two eﬀects.1 (ρ) = 1 − exp −2 P1.1 . .9 λd for a uniform circular aperture. 2 This is easy for n = 0.9. TEM0. Here we will consider a simple model for the pinhole camera.

so consider what the ﬁeld looks like there. The set of all Hermite–Gaussian modes with the same beam parameter w0 form a complete set. A monochromatic Gaussian beam (with beam waist parameter w0 ) is focused onto a thin. (+) (a) (15 points) Noting that the Hermite–Gaussian modes are all orthogonal. where α is some constant. (a) What is the Hermite–Gaussian beam pictured below? (b) Suppose this beam has the same fractional power contained in a circular disc of radius a as a Gaussian beam. Now what is the size of the aperture that gives the smallest spot on the screen? Problem 6. whose eﬀect on the incident ﬁeld is Eout = α(Ein )2 .m . (i. explain how you would compute the coeﬃcients cl. Problem 6.m El.13. w0 ) to emphasize the dependence on w0 . Eout = tEin ). w0 ) or evaluate any complicated integrals. Hint: The nonlinear optical element only acts at z = 0.. the output ﬁeld has a component with optical frequency 2ω. (Hint: what do plane waves correspond to in geometrical optics?) (b) Now answer the same question in wave optics. so that a is the “radius” of the aperture.m mode by (+) El. Compare (qualitatively) the far-ﬁeld divergence of the two beams.) .m (r.6.m (x. w0 ). this means that an arbitrary electric ﬁeld proﬁle E (+) (r) at z = 0 can be written as a superposition of all the modes at z = 0: E (+) (x. y.12. Describe completely how the geometry (both near. what is the size of the aperture that gives the smallest spot on the screen? Explain.e. y) = ∞ l.m (r. Assume that the aperture has a Gaussian transmission coeﬀcient for the electric ﬁeld of t = exp − r2 a2 .11.and far-ﬁeld) of the 2ω output ﬁeld diﬀers from the input ﬁeld. nonlinear optical element. Problem 6. Denoting the TEMl.m=0 cl.7 Exercises 81 (a) If the incident light is treated in geometrical optics. Show that if the optical frequency of the input beam is ω. (Set up any necessary equations but don’t refer to the explicit form of (+) El.

m=0 Find the ﬂaw in the following argument: each mode El. you can compute the ﬁeld at any location along the z-axis by E (+) (x.m (x. y. which is computed for a ray starting at location zi . Assume that a Gaussian beam is the characteristic mode of the cavity. every ﬁeld. and let q = q(zi ) denote the complex beam parameter for the beam at z = zi .m (r. (a) (15 points) Show that q= A−D 2C −i 1 − C2 A+D 2C 2 . (b) (10 points) Show that the radius of curvature at z = zi is R(zi ) = − and the beam radius at z = zi is w(zi ) = π 1− λB A+D 2 2 2B A−D . but this clearly has a diﬀerent far-ﬁeld divergence angle than θ0 for the constituent beams. (+) (+) l. w0 ). z. (For examply. z) = ∞ cl.) Hint: how is the far-ﬁeld divergence angle for a Gaussian beam deﬁned? How would modifying the deﬁnition give a diﬀerent answer? Problem 6.14. you can apply this argument to a Gaussian of size 15w0 . y.82 Chapter 6. has a far-ﬁeld divergence angle of θ0 . Gaussian Beams (b) (10 points) Once you know these coeﬃcients. Hint: what is the general condition for a wave to be the resonant mode of a cavity? . Suppose that an optical cavity is described by the round-trip ray matrix M(zi ) = A C B D . no matter what its proﬁle (or size) at z = 0.m El. thus. w0 ) has the same far-ﬁeld divergence angle (θ0 = tan−1 (w0 /z0 )).

Interference is a central feature of resonators with waves. Otherwise. d We can also rewrite this resonance condition in terms of related quantities: λq = 2d . and eventually lead to destructive interference. where the mirrors are perfect reﬂectors. In this case.Chapter 7 Fabry–Perot Cavities Now let’s return to the subject of resonators. the round-trip accumulated phase must be some integer multiple of 2π: 2kd = 2πq (q ∈ Z). d In order for a plane wave to exist in the cavity.3) The corresponding spectrum is a set of regularly spaced. and thus resonators are also interferometers.7 in the context of ray optics. of the cavity are given by πq kq = .1) Thus. Now we will perform a better analysis. the allowed stationary waves.2) (7. the phase will precess on each successive round trip. it will constructively interfere with itself. Fabry–Perot cavity. d (7. 7. Thus.1 Resonance Condition We will start with the simplest case of a planar cavity of length d. it must return to exactly the same phase after one round trip through the cavity. I fsr = c/2d nœo–o¡ nœ nœoòo¡ n . 2d ωq = πcq . or an optical spectrum analyzer. q νq = cq . which we treated earlier in Section 2. (7. which includes the wave nature of the light. the conventional two-mirror resonator is referred to as a Fabry–Perot interferometer. or modes. arbitrarily sharp peaks. Fabry–Perot etalon. Because of this.

and right-going waves: E (+) (r) = E01 eikz + E02 e−ikz .6) (7.10) Here. as any nonzero phase angle can be absorbed into the phase factor as an additional eﬀective length. (7. λ (7. we can write E (+) (r) = E0 (+) E01 (+) (+) (+) (+) (+) (+) (7. (7.12) Obviously. the phase of the plane wave changes by ∆φ = 2kd = so that (+) 4πd .7) sin(kz + φ0 ). (+) (+) (7. where φ0 is determined by the complex phase of and The wave must vanish at the mirrors. Fabry–Perot Cavities The frequency spacing is a special quantity called the free spectral range. The losses damp the amplitude of the cavity. which is a superposition of left. The total wave is E (+) = E0 (+) + E1 (+) + E2 (+) + · · · = E0 (+) 1 + rei2kd + (rei2kd )2 + · · · = E0 .13) . The wave inside the cavity is a standing-wave mode. An alternative to this constructive-interference approach to the resonance condition is to view the reﬂectors as imposing boundary conditions on the wave.2 Broadening of the Resonances: Cavity Damping Real cavities involve some loss of the light on each round trip. 7. Then the right-hand mirror is at z = d: E (+) (z = d) = 0 =⇒ sin(kd) = 0 =⇒ kd = qπ. we may assume that r is real. (+) E02 . the amplitudes of the two waves are equal.9) (7.4) where n is the refractive index of the medium ﬁlling the cavity. Thus.8) This is the same resonance condition as before. 2d 2nd (7. which we can view as perfect conductors. the damping leads to broadening of the resonances.84 Chapter 7. Suppose that the amplitude of the wave is reduced by a factor of r on each round trip: (+) |En+1 | = r|En |. Just as in the case of a driven mechanical oscillator. 1 − rei2kd (+) (7.11) (+) En+1 = rei2kd En . On one round trip.5) (7. so that |E01 | = |E02 | =: E0 . deﬁned as FSR = νq+1 − νq = c c0 = . En is the electric ﬁeld amplitude of the plane wave on the nth round trip through the cavity. If the left-hand mirror is at z = 0. E (+) (z = 0) = 0 =⇒ φ0 = 0. For ideal mirrors.

1 Standard Form In standard form. which gives the condition πν sin2 = 0 =⇒ ν = q(FSR). This occurs when the sinusoidal term vanishes.16) √ π r . so the intensity is periodic in d with period λ/2. 2 2F 2 sin (kd) π I0 . (7. the cavity attains the maximum intensity Imax . .14) 7. 7. we can write the resonator intensity as I= 1+ where the maximum intensity is Imax := and the ﬁnesse. (7. 2F 1+ π The minimum intensity only goes to zero in the limit of large ﬁnesse. The minimum intensity occurs when the denominator is maximum.2 Maximum and Minimum Intensity The intensity is maximized when the denominator is smallest.19) FSR This is just the resonance condition for the lossless cavity. deﬁned by Imax . (1 − r)2 + 4r sin2 (kd) (7.2 Broadening of the Resonances: Cavity Damping 85 The total intensity is I= = = = = = where I0 = 2|E0 |2 /η. In terms of the free spectral range c/2d. The only dependence of the intensity on the cavity length d is in the sin function. Thus the sinusoidal term becomes unity.17) 1−r is an important parameter that characterizes the lossy nature of the cavity.18) so that in frequency the intensity is also periodic. (+) I0 |1 − rei2kd |2 I0 |1 − r cos(2kd) − i sin(2kd)|2 I0 [1 − r cos(2kd)]2 + r2 sin2 (2kd) I0 2 cos2 (2kd) + r2 sin2 (2kd) 1 − 2r cos(2kd) + r I0 2 − 2r cos(2kd) 1+r I0 .7.20) Imin = 2.2. and Imax (7.2. we can write the intensity as F := I= 1+ Imax 2F π 2 sin 2 πν FSR . (7. At these points. where the period is the free spectral range. (1 − r)2 (7.15) (7.

2. showing the transition from broad resonances to sharply deﬁned peaks. To generalize the treatment of losses. (7.3 Width of the Resonances As a way to characterize the width of the resonances. when the resonances are well resolved. we can deﬁne the survival probability Ps . The mirrors can lead to intensity loss due to partial transmission. all other parameters are determined by these two.86 Chapter 7. I Imax Imaxoo/2 fsr/F Imin (qo-o1)(fsr) qo(fsr) (qo+o1)(fsr) n The next diagram compares the cavity intensities for several diﬀerent ﬁnesses. In this case. We can summarize the characteristics of the Fabry-Perot cavity in the diagram. In this limit. the width condition is sin2 which is satisﬁed by the frequencies πν FSR FSR = π 2F 2 . we can thus write the full width of the resonance at half maximum as δνFWHM = FSR F . we can write the half-maximum intensity frequencies as ν=± ν≈± FSR 2F .21) π .2 .4 Survival Probability Losses in a cavity can come from a number of sources. which is the fraction of the intensity that stays in the cavity after one round trip. absorption of the light in the reﬂective coatings.22) sin−1 π 2F The width of the resonances is really only sensible concept in the limit of large ﬁnesse. we will now ﬁnd the frequencies such that the intensity falls to Imax /2. A medium inside the cavity can also absorb or scatter the light.23) In this approximation. (7.24) Notice that in the limit of arbitrarily large ﬁnesse. (7. for example. Fabry–Perot Cavities 7. recalling that two parameters (FSR and F ) are necessary to completely specify the cavity behavior.2. (7. perfectly deﬁned resonances as in the lossless cavity. and scatter due to surface roughness. losses of beams around the mirror edges. I F=2 F = 10 F = 50 (qo-o1)(fsr) qo(fsr) (qo+o1)(fsr) n 7. then the survival probability is just Ps = R1 R2 . If the two mirrors have reﬂectances (intensity reﬂection coeﬃcients) R1. δνFWHM −→ 0 while Imax diverges—the resonances become δ functions.25) . (7.

which are of the order of a few hundred THz.26) F = 1 − Ps I0 √ (7. to other objects in the cavity. (FSR)(1 − Ps ) (7. In this limit we can expand to lowest √ √ √ √ 1/4 order in (1 − Ps ).34) For a “good” oscillator—that is. (7. the ﬁnesse plays the same role as Q for optical cavities. Ps is close to unity.31) 1 − Ps τp ≈ 1 1 F √ . . one that does not dissipate energy very rapidly—the Q factor is large because the denominator in the deﬁnition is small.5 Photon Lifetime The name “survival probability” more appropriately describes the fate of a photon after one round trip through the cavity. A cavity is an oscillator.6 Q Factor The Q factor represents the “quality” of any oscillator. energy loss per oscillation cycle (general oscillator) (7. (7. we can write 1/4 πPs √ (7. In terms of the survival probability.2. so Ps is also close to unity. but generally produces more manageable numbers. (1 − Ps )2 .2.7. In fact.29) and so √ For a “good” resonator.27) Imax := for the maximum intensity. which amounts to setting Ps ≈ 1 and (1 − Ps ) = (1 − Ps )(1 + Ps ) ≈ 2(1 − Ps ). = = 2π(FSR) 2πδνFWHM 2(FSR)(1 − Ps ) 1 2π (7.2 Broadening of the Resonances: Cavity Damping 87 as a more convenient expression for the ﬁnesse and Additional factors can model other losses due. but at optical frequencies. The time for one round trip is τrt = 2d/c = so the “lifetime” of a photon inside the cavity is τp = 1 . Thus. Thus. we can already see that optical Q factors are very large.32) Thus we arrive at an “uncertainty relation” τp δνFWHM = for optical cavities between frequency and time. The probability that the photon is lost is simply 1 − Ps . 7. π √ F ≈ (7.33) 7. The general deﬁnition is Q := 2π · stored energy . Thus.30) 1 FSR (7. a photon lasts for 1 1 − Ps round trips on average.28) . for example.

J. just the right distance away.1 7.7 Example: Finesse and Q Consider a planar cavity where both mirrors reﬂect 99% of the incident intensity. A typical optical frequency is 300 THz.35) We can also write this in terms of the cavity line width.992 . a really large number compared to the ﬁnesse. This leads to a cavity line width of δνFWHM = (FSR)/F ≈ 3 MHz.” Optics Letters 17. (A related handy number to remember for the speed of light is c0 ≈ 1 ft/ns. Again. A laser beam impinging on the front mirror will mostly bounce oﬀ. The total circulating intensity I within the cavity is the intensity propagating to the right of and away from mirror 1. 7. cavity Q factors are more reasonable. The order is thus q = νq /(FSR) = 3 × 105 . since the cavity length often far exceeds the optical wavelength. (7.3 Cavity Transmission If you think about it in a certain way. For a cavity length d = 15 cm. (7. and R. Imagine a mirror that is nearly perfectly reﬂecting. respectively. Lalezari. Rempe. doesn’t it? Such is the power of interference. the round-trip time is τrt = 2d/c = 1 ns. the Q factor for optical cavities is typically huge compared to the ﬁnesse. We now need to be a bit more precise about the intensities and electric ﬁelds in and around the cavity.33): Q= νq . Q is huge because the optical frequency is so high. The photon lifetime is τp = F /2π(FSR) ≈ 50 ns. using the uncertainty relation in Eq.38) The order q is typically a very large integer for optical cavities. so that Q = νq /δνFWHM = 108 .2. The corresponding wavelength is λ = c0 /νq = 1 µm. Thompson. or 1/νq τp . and so δνFWHM ∼ 1 MHz would yield a Q of 103 . Thus.24). In the microwave. relative to the amount of energy in the cavity. only transmitting a small fraction of the intensity. 363 (1992). the energy loss per oscillation cycle is just the optical period multiplied by this rate. 1A . Then Ps = 0. δνFWHM (7.88 Chapter 7. we see that Q and F diﬀer by a factor of the resonance order q: Q = qF . Kimble. For example. Let’s see how this works out. in the microwave νq ∼ 1 GHz. Putting this into the above deﬁnition. ﬁnesse of 1. R. We will refer to the left. a Fabry–Perot cavity is an amazing device. we see that Q = 2πνq τp . which is in the near-infrared (approximately the wavelength of a Nd:YAG laser). the rate at which energy leaves the cavity is 1/τp . J. Makes your brain tingle just thinking about it. so microwave Q factors can be large as well. That’s why the ﬁnesse is a preferred quantity in the optical regime—the numbers are much easier to deal with. (7. H. the fraction of energy lost per round trip is 1 − Ps . (7.9 × 106 was reported by G. But note that some microwave devices can have extremely narrow line widths. F = FSR δνFWHM . Since νq = q(FSR).36) We can compare this to the form of the ﬁnesse from Eq. Thus. “Measurement of ultralow losses in an optical interferometer. (optical cavity) (7. Following the same logic as in the last section.) Then the free spectral range is FSR = 1/τrt ≈ 1 GHz.37) and see that Q and F are the same except that a diﬀerent (smaller) oscillator frequency is used for the ﬁnesse.and right-hand mirrors as mirrors 1 and 2. Fabry–Perot Cavities For an optical cavity. all the laser power can instead transmit through the mirror. But by putting a second mirror behind that mirror. and so F ≈ 300. The best “supercavities” in the optical have ﬁnesses in the range of 105 to 106 at present.

45) |t′ t2 |2 1 (1 − r1 r2 )2 (7. 1 Thus we can deﬁne the cavity transmission coeﬃcient Tcav by Tcav := Iout I = |t′ t2 |2 = 1 Iin I0 Tcav.18): I= 1+ Imax 2F π 2 sin2 πν FSR . From the way we deﬁned I.39) where again Imax 2 2 Ps = r1 r2 . Then the maximum cavity transmission becomes |t′ t|2 .3 Cavity Transmission 89 I E (+) E (+) t¡E (+) 1 r¡E (+) r™ E (+) 2 t™ E (+) The circulating intensity is related to the intensity of the initial wave before the resonant buildup by Eq. at least in steady state after the circulating ﬁeld builds up.44) Tcav. Eq.1 Reﬂected Intensity So what happened to the input light that supposedly reﬂects oﬀ of the input mirror? It must work out that it disappears via destructive interference.3. (7. so for two identical mirrors.41) sin 2 πν FSR . a resonant Fabry-Perot cavity transmits all the input light. then r1 = r2 =: r and t1 = t2 =: t. the part transmitted through the front mirror is also reduced by the reﬂection coeﬃcient of the second mirror: Iout (front) = |t1 |2 |r2 |2 I = 2 |r2 |2 |1 − r1 |2 I.43) Thus. The output (transmitted) intensity is related to the intracavity intensity by Iout = |t2 |2 I.42) where Tcav. 7. (7.max := is the maximum fraction of the input intensity that the cavity can transmit.7. (7. So here’s the statement that we started with: if the two mirrors are identical. (7.46) . states that t′ t = 1 − r2 .40) (7. while the initial intensity I0 is similarly related to the input intensity: I0 = |t′ |2 Iin .17).max = (1 − r2 )2 But one of the Stokes relations. ′ |2 |t1 (7. and we can now show this explicitly.max = 1.max 1+ 2F π 2 √ = I0 /(1 − Ps )2 . Tcav. (7. Part of the intracavity wave transmits through the front mirror as well. when the two mirrors are identical. (7. In terms of the ﬁeld reﬂection coeﬃcients from the inside of the cavity. (5.

The width δνFWHM is the resolution of the analyzer.32). But in terms of the ﬁeld. as pictured at the beginning of this chapter. so Again. due to the other Stokes relation r′ = −r.and right-going waves and thus a standing wave. this reduction is small. That is. In the case of identical mirrors. (7. ≈ 1 − Ps 2π 2(1 − Ps ) (7. we used the Stokes relation t′ t = 1 − r2 to arrive at the last expression.4 Optical Spectrum Analyzer In the good-cavity regime. there is a π phase shift of the reﬂected beam compared to the transmitted beam. I= Iout Iout F √ ≈ Iout . since the output intensity for a symmetric cavity is just what is lost on each round trip. the cavity intensity is a sinusoidal pattern with maximum intensity 4I ≈ 2F /π and period λ/2. then.49) τrt 2π Thus. which we can rewrite as F τp ≈ . we can conveniently identify F /2π as a “buildup factor. (7. however.” since light entering a resonant cavity makes this many round trips inside the cavity before leaving it. properly represent the circulating ﬁeld in a ring cavity. The total intracavity intensity Icav for a linear cavity is thus Icav (z) = √ ikz √ −ikz Ie − Ie 2 = 2I[1 − cos(2kz)] = 4I sin2 (kz). since it cannot distinguish frequencies within this narrow range. I = Imax = I0 /(1 − r2 )2 .going waves. (7. (7. is a superposition of left. Iin = Iout . In the good-cavity limit. this is only the intensity of the rightwardpropagating ﬁeld in the cavity. Thus. That’s how all the power can transmit through the resonant cavity and allow for energy conservation in steady state. a ﬁeld probe placed in the cavity would see more than I.51) Thus.47) Iout (front) = |t′ |2 r2 I0 = r2 Iin . Thus the Fabry–Perot cavity is extremely useful in the laboratory as an optical spectrum analyzer—a device that can measure the spectrum of a input light source. 7.50) Also for a symmetric cavity. |r|2 |1 − r2 |2 I. each of intensity I—a standing-wave pattern. and the total ﬁeld is an interference pattern of left. the left-going wave is reduced in amplitude because of reﬂection from the righthand (output) mirror. We now also see the reason for the terminology .2 Intracavity Buildup Notice that these eﬀects require that on resonance. But the ﬁeld inside a linear cavity. so the cavity circulating intensity is a factor of F /2π larger than the input in the good-cavity limit.and right. so we chose the proper phase of the left-going ﬁeld. We can also see this directly from the intensities. The intensity I does. these two beams destructively interfere in steady state. where the intensity only circulates in one direction. |t′ |2 Iout (front) = (7.90 Chapter 7. Note that the way we deﬁned the circulating intensity I. a Fabry–Perot cavity transmits light only at well-deﬁned frequencies. Recall that for consistency with the boundaries that the ﬁeld must vanish at z = 0 and d. Fabry–Perot Cavities and on resonance. 7.48) This is exactly the same intensity as the part of the input beam that reﬂects oﬀ the ﬁrst mirror. For the linear cavity. This is most easily seen from Eq.3. the cavity circulating intensity must be much larger than the input or output intensity in the regime of large ﬁnesse.

This is what a Fabry–Perot optical spectrum analyzer looks like in the laboratory. it is very useful to note that small changes in any cavity parameter translates into linear changes in other parameters.4 Optical Spectrum Analyzer 91 free spectral range for the spacing between cavity resonances. In general. Most useful spectrum-analyzer-type problems can be solved with similar relations. a Fabry–Perot spectrum analyzer can be scanned or tuned by changing the cavity length d. This is actually a confocal cavity. which gives a total displacement of 1. The other mirror is attached to the right-hand side of the main body on a tubular piezo stack. which we’ll get to later. The cavity length is d = 5 cm giving FSR = 1. δkq δd δνq δλq =− .2% reﬂectivity.53) = ≈− λ d νq kq at least to lowest order in the perturbation. Very often.7. so the change in free spectral range is a second-order eﬀect (a factor of ∼ 105 smaller for typical optical cavities). These relations follow from the fact that these quantities are all proportional or inversely proportional.52) where we have expanded to lowest order in δd. the corresponding change in the resonance frequencies is given by δνq = qc δd qc qcδd − ≈ − 2 = −νq . .3 µm for a ±15 V control signal. 2(d + δd) 2d 2d d (7. The mirror cover on the right contains a photodiode to monitor the transmitted light. Note that by a similar argument. so that F = 200. The barrel is used to adjust the cavity length to match precisely the mirror curvatures.5 GHz. This can be accomplished by mounting one of the mirrors on a piezoelectric stack. (7. so the parameters work out a little diﬀerently than for the planar cavity. On the left you can see one of the mirrors mounted on a threaded barrel. since this is the longest range in frequency that the cavity can measure without repetition or “wrapping” of the spectrum. The mirrors are coated for 99. If the length cavity is changed by an amount δd. Here is the same cavity disassembled. the free spectral range changes by the much smaller amount [−(FSR)δd/d].

5 Spherical-Mirror Cavities: Gaussian Modes In the more general case that we analyzed in ray optics. Thus. Plane waves are no longer compatible with the mirrors. Consider the spherical-mirror cavity shown here. resonators are composed of spherical mirrors. the length condition is z1 + z2 = d.92 Chapter 7. the waves must vanish at the mirror boundary. Matching the wavefront curvature to the curvature of mirror 1. Fabry–Perot Cavities 7. respectively. We will try to “ﬁt” a Gaussian beam to it. z2 (7.54) (7. R(z1 ) = R1 At the second mirror. Instead. they are a natural choice for cavity modes that match the spherical boundary conditions. we must turn to Gaussian beams. R(z2 ) = −R2 =⇒ −R2 = z2 + =⇒ R1 = z1 + 2 z0 . As in the planar case. R¡ focus R™ z¡ zo=o0 d z™ z As usual the Gaussian beam focus occurs at z = 0. Because Gaussian beams have spherical wave fronts.55) 2 z0 . z1 (7. and they are separated by a total distance d.56) . and the mirrors of curvature radii R1 and R2 are placed at z1 and z2 .

2 < 0 but R(z) ≷ 0 for z ≷ 0. (7.2 = 1 + (7. (7. 7. −d 2 d z2 = 2 d 2|R| − 1.2 ) = λd/π d |R| d 2− |R| . Also.59) which is just the same as the stability condition from ray optics.5. . z0 = 2 d z1 = The corresponding beam waist at the focus is 2 w0 = (7. Thus.57) We have determined the location of z = 0 relative to the mirrors along with z0 .61) and the beam waist at the mirrors is 2 w1. If z0 is imaginary. the Rayleigh length z0 must be real. we ﬁnd that this condition becomes 0 ≤ g1 g2 ≤ 1. Let’s look at the symmetric ones again in the context of Gaussian-beam modes. R1.58) d .2 Inserting the expression from Eqs.3 Special Cavities Recall the special resonator casesfrom Section 2. These three equations have the solution −d(R2 + d) R1 + R2 + 2d d(R1 + d) z2 = R1 + R2 + 2d −d(R1 + d)(R2 + d)(R1 + R2 + d) 2 .2 Symmetric Cavities In the special case of a symmetric resonator. R1.7.3. By symmetry we can guess that the focus occurs at the center of the cavity.57) and performing much algebra. 7.5. so we have completely determined the Gaussian beam geometry.5 Spherical-Mirror Cavities: Gaussian Modes 93 Note the diﬀerence in minus sign in the two cases: for concave mirrors.5. and thus the Gaussian amplitude factor exp[−r2 /w2 (z)] diverges for large r. the solutions (7.60) λd 2π 2|R| − 1. the physical mode condition is 2 z0 ≥ 0.57) simplify greatly.7. (7. w2 (z) becomes negative at large z.2 = w2 (z1.62) 7. R1 = R2 =: R. z0 = (R1 + R2 + 2d)2 z1 = (7. g1. d (7.1 Physical Modes For a physical (bounded) mode.

2 = −d/2 (Spherical resonator/symmetrical concentric) R R Here z0 −→ 0. Again.67) . 7.27) that the total phase of a Gaussian beam is φ(r.64) For constructive interference. z2 ) − φ(0. (7. Thus. (7.2 = −d (Confocal resonator) R √ For the confocal resonator. this is the reason that z0 is also called the confocal parameter.94 Chapter 7.4 Resonance Frequencies Recall from Eq. this phase change must be ∆φrt = 2πq for some integer q. z0 = d/2. which is what we expect in the limit of a spherical boundary. and w1. This last relation implies that each mirror is a distance z0 from the focus. so we recover our analysis of the planar cavity. we will only consider the plane-wave and Gouy phases: φ(0. Thus.2 = 2w0 . and w1.2 = ∞ (Planar resonator) In this case. g1.2 −→ ∞. the focal length of both mirrors is f = −R/2 = d/2 = z0 . z) = kz − tan−1 z z0 +k r2 .5. 2R(z) (7. Case 2. This is the spherical-wave limit of the Gaussian-beam mode. z0 −→ ∞. g1 = g2 = 1 =⇒ R1. This is the plane-wave limit of the Gaussian beam. w0 −→ ∞.2 −→ ∞. z1 )] = 2k(z2 − z1 ) − tan−1 z2 z0 − tan−1 z1 z0 . z) = kz − tan−1 The round-trip phase is ∆φrt = 2 [φ(0. νq = (FSR) q + 1 tan−1 π z2 z0 − tan−1 z1 z0 . g1.2 = 0 =⇒ R1.66) Rewriting this in terms of frequency using k = 2πν/c. (6. Case 3. and w1.2 = −1 =⇒ R1. (7. Fabry–Perot Cavities Case 1. so it suﬃces to consider the on-axis phase φ(0. z). w0 = λd/2π. so the cavity length is 2z0 . Alternately. w0 −→ 0. kq = π 1 q+ tan−1 d d z2 z0 − tan−1 z1 z0 . (7.65) z z0 .63) We already matched the shape of the phase fronts to the mirrors.

71) for the TEMl. The Gouy-phase term (with the cos−1 ) is again a geometry-dependent term that is independent of q. First. . Since q is very large at optical frequencies.57) in terms of the stability parameters. Repeating the above argument shows that only the Gouy term is modiﬁed. Note that there are a number of degeneracies among the modes due to the axial symmetry of the resonator.69) In terms of these parameters.72) Now the Gouy term represents a frequency shift that depends on the cavity geometry and the order of the Hermite–Gaussian mode. with the result −g2 (1 − g1 ) g1 + g2 − 2g1 g2 g1 (1 − g2 ) z2 = g1 + g2 − 2g1 g2 g1 g2 (1 − g1 g2 ) 2 z0 = . since they have diﬀerent transverse intensity patterns. the cavity modes must be labeled by three indices as shown.2 = 1 + d/R1.5. the frequency shift is not necessarily small on the scale of the free spectral range.68) where again g1. The shift vanishes for in the planar and spherical-concentric cavities. as we can see by examination of Eq. (7. z) = kz − (1 + l + m) tan−1 z z0 (7. (6.2 .57).6 Spherical-Mirror Cavities: Hermite–Gaussian Modes 95 Using Eqs. Thus.6 Spherical-Mirror Cavities: Hermite–Gaussian Modes The same idea applies to the higher-order TEMl. we can write down a mess of stuﬀ that is diﬃcult to simplify because of the signs of the factors under the radical signs. so that νl. The terminology is that modes corresponding to diﬀerent q values are the longitudinal modes. Through a rather long sequence of algebraic steps. However.67) to Eq.m.q = (FSR) q + 1 √ (1 + l + m) cos−1 g1 g2 . but for the confocal limit.m mode.70) after a bit of algebraic manipulation. (7. this terms leads to an overall shift of the spectrum that is small compared to the optical frequency νq . νq = (FSR) q + √ 1 cos−1 g1 g2 . which are similar in structure to the plane-wave resonances. π (7. 7.68)? A (very) brief outline of the algebra is as follows.m modes of the cavities. (g1 + g2 − 2g1 g2 )2 z1 = (7. (7. the Gouy frequency shift is exactly half of the free spectral range. write the beam-parameter solutions of Eqs. since the wave fronts oﬀ all the higher-order modes are essentially the same as for the Gaussian. (7. the contribution of the Gouy phase reduces to a startlingly simple expression. The main diﬀerence is that the Gouy term in the on-axis phase becomes φl.5 Algebraic Digression How does one go from Eq.69). 7. π (7. Then it is relatively easy to show that cos tan−1 z2 z0 − tan−1 z1 z0 = 2 z0 + z1 z2 2 (z0 2 2 z1 ) (z0 2 z2 ) + + = √ g1 g2 (7. the signs are easier to handle since g1 g2 > 0 for a stable cavity.7. The modes corresponding to diﬀerent values of l and m are the diﬀerent transverse modes.m (0.

so since cos−1 (0) = π/2.” Notice also that the ﬁnesse eﬀectively changes. Notice. we recover the spectrum for a planar cavity.m. which is a ring cavity with an optical switch as shown.2.1. And that’s why laboratory spectrum analyzers are often confocal cavities: the modes are degenerate. 91) a confocal cavity? It would seem that a cavity with spherical mirrors would have a more complicated mode structure than the planar cavity and thus more diﬃcult to use as an optical spectrum analyzer.96 Chapter 7. (a) What is the free spectral range if the cavity is completely empty? (b) What is the free spectral range if the cavity is otherwise empty but a 1 mm glass plate (n = 1. Recall the typical trajectory of the confocal cavity that makes 4 passes per round trip. Thus. Consider a cavity constructed from two planar mirrors separated by 1 cm. g1 g2 = 0. A confocal cavity operated in this manner is said to be operated in “ring mode. the input beam to a confocal resonator is usually slightly misaligned to ensure that the “in-between” modes are populated as well.6.5) is placed in the cavity? Ignore any reﬂections from the surfaces of the glass plate.q (confocal) = (FSR) q + (1 + l + m) .74) c c0 = = . where s = 2q + 1 + l + m and FSR (7. not FSRconfocal. Thus.1 Confocal Cavity Why was the laboratory cavity in the photographs (see p. that this argument only works if several transverse modes are excited by the input source. . If only the TEM0. so the relevant free spectral range is FSRconfocal . (7. FSRconfocal δνFWHM = F 2 (7.73) νl. though. Fabry–Perot Cavities 7.0 mode is present. 2 We can write this in terms of an eﬀective index s as νs (confocal) = (FSRconfocal )s. the mode structure is simple for this case as well. then the free spectral range is back to the old value. The eﬀective confocal free-spectral range only works if modes operate in the ring-trajectory manner shown. That’s because longitudinal modes are still spaced by FSR.75) 2 4d 4nd is an eﬀective free-spectral range for the confocal cavity. For a symmetric confocal cavity. the oﬀ-axis modes travel twice as far before repeating themselves. the mode frequencies become 1 (7. Fconfocal = for a confocal cavity in ring mode.7 Exercises Problem 7. since the line width is unaﬀected but the free spectral range is diﬀerent. in the laboratory. One method for boosting the intensity of continuous-wave (cw) light is to use a cavity dumper.76) 7. Thus. FSRconfocal := That’s why the cavity is eﬀectively twice as long as for the planar cavity. however. As it turns out. Problem 7. but with a length of 2d.

(a) Here is a measured transmission spectrum. cavity transmission (arb. units) Dd (arb. Consider a symmetric confocal optical spectrum analyzer operating at a nominal wavelength of 780 nm. (c) Suppose the input power is 1 W and that the circulating power is allowed to build up to its steady-state value. and the circulating light builds up in the cavity to a large intensity. deﬂecting the circulating intensity out of the cavity.3. (d) What is the maximum repetition rate of the cavity dumper if a 50 W output pulse is desired? Problem 7. Then the optical switch is activated. The nominal cavity length is 5 cm.7. What are the duration and power of the output pulse when the switch is activated? Assume the switch couples all the light out of the cavity.5) of length 2 cm.8% for all the mirrors and an intensity loss of 0. (b) What are the survival probability.2% per pass due to the switch. units) and this is the same set of data but with the horizontal axis zoomed by a factor of 20: . (a) What are the round-trip time and free spectral range? Assume the beam path is a rectangle of dimensions 12 × 20 cm.7 Exercises 97 optical switch The cavity dumper operates as follows: the input light is turned on. and photon lifetime of the cavity? Assume an intensity reﬂection coeﬃcient of 99. Also model the optical switch as a block of glass (n = 1. The cavity transmission is monitored as the cavity length d is scanned. which can be changed precisely (on a sub-λ scale) using a piezo stack attached to one of the mirrors. ﬁnesse.

units) Dd (arb. (Note: in the ﬁrst scan. such that the cavity is slightly longer than required to be confocal. keeping in mind that this cavity is confocal.) (b) The cavity length is changed by a relatively large amount (compared to λ. units. it is best to average the two distances to get an accurate value. The measured spectrum now looks more complicated: cavity transmission (arb. units) Dd (arb. say by a threaded barrel adjuster. 20¥ zoom) Compute the ﬁnesse and resolution δνFWHM of the cavity. Fabry–Perot Cavities cavity transmission (arb. units) The same data zoomed in by a horizontal factor of 10 shows a structure of many peaks for each longitudinal mode: . the distances between the peaks are slightly diﬀerent due to the nonlinearity of the piezo stack displacement with voltage.98 Chapter 7.

Hint 1 : The cavity no longer satisﬁes the confocal condition. From the data here. 10¥ zoom) Note that the horizontal units here do not correspond to the plots in (a).2 are the (small) deviations due to mirror roughness. 2 Also assume that they are normally distributed with δd1.4. this new spectrum was taken when the threaded barrel that sets the overall cavity length 1 was rotated by about 1 4 turns. where the angle brackets denote an average over the mirror surfaces. consider a planar resonator of length d0 with perfectly reﬂective mirrors. as a sanity check.7 Exercises 99 cavity transmission (arb. while a typical roughness for a superpolished mirror substrate is 0. Show that if νq = (FSR) q + 1 tan−1 π z2 z0 − tan−1 z1 z0 .5. Assuming that the reﬂective coating of the mirror has the same roughness as the substrate.2 = 0 and δd1. (c) A typical rms surface roughness for a laser-grade mirror substrate is 3 Å. x2 1 exp − 2 f (x.e. The barrel was machined with a nominal pitch of 40 threads/inch. (b) Model the eﬀective cavity length as d = d0 +δd1 +δd2 . where the δd1. estimate the new cavity length by using the cavity line splittings to calculate the new stability parameter. Problem 7. σ is the standard deviation or root-mean-square (rms) deviation. Compute the full width at half maximum δxFWHM of this distribution in terms of σ. the rms roughness is ε). Hint 2 : You should calculate the new cavity length by understanding the above spectrum.7.2 = ε2 (i. One limitation to the ﬁnesse of a Fabry–Perot cavity is the surface roughness of the mirrors. units) Dd (arb. Assume that these are uncorrelated random variables over the mirror surfaces. σ) = √ 2σ 2πσ . . compute the roughness-limited ﬁnesse for each case at λ = 780 nm. Show that the eﬀective cavity ﬁnesse is λ F = √ 8 log 2 ε due to the surface roughness. Here. Note that the spectrum almost repeats after some distance and then really repeats after that.. As a simple estimate of this eﬀect. so don’t treat is as a confocal cavity. (a) Recall the normalized form of the normal (Gaussian) distribution with zero mean. However. units. Problem 7.5 Å.

100 Chapter 7. What is ∆λ? (c) Suppose that the mirrors have very high reﬂectivity. free spectral range 6. survival probability 2. assuming it is longer than the analyzer? . The transmission of the laser light through a Fabry–Perot spectrum analyzer of nominal length d = 5 cm is observed as the analyzer’s length is scanned. Fabry–Perot Cavities is the resonance frequency for a Gaussian mode of a stable. the other laser wavelength is maximally transmitted. (b) Suppose that you change the wavelength by ∆λ before you ﬁnd the next cavity resonance. then the following expression is equivalent: 1 √ νq = (FSR) q + cos−1 g1 g2 . Suppose that the two output wavelengths correspond to two adjacent resonances of the laser cavity. You observe that when the analyzer’s length is changed by ∆d/10. or the same as for a “bad” cavity of the same geometry: 1.7. photon lifetime Problem 7. A laser of nominal wavelength λ = 1 µm emits at two slightly diﬀerent wavelengths. so that the cavity is “good. What is ∆λ? (c) The laser itself consists of a Fabry–Perot cavity ﬁlled with gas (n ≈ 1). Consider a planar cavity of nominal length d = 1 cm. (a) The transmission of the analyzer is observed to be periodic in the analyzer length with period ∆d. The cavity is resonant with light of λ = 1 µm. where the diﬀerence is ∆λ ≪ λ.6. What is the length of the laser cavity. π Problem 7. (a) Compute the ratio Q/F for this cavity at this wavelength.” State whether the following quantities would be larger. round-trip time 5. What is ∆d? (b) Suppose that the analyzer is set so that one wavelength component of the laser is maximally transmitted. spherical-mirror cavity. Q factor 3. smaller. ﬁnesse 4. resonance width (δνFWHM ) 7.

5) (8. E0x = 2|E0x | and E0y = 2|E0y |.1) = E0x x + E0y y . (+) (+) (8. t) = E(+) (z. so we need to work out some details before deﬁning particular polarization states.3) The polarization of the wave is related to the direction of E0 . with the explicit time dependence. (8. x y where Ex = E0x cos(kz − ωt + φx ) Ey = E0y cos(kz − ωt + φy ). = E0 ei(kz−ωt) + c. we can write the real ﬁeld as E(z. (+) (+) E0x = |E0x |eiφx (8.7) + −2 E0x E0y E0x E0y . t) = Ex (z. These equations are the parametric equations for a general ellipse. t)ˆ + Ey (z. is E(z.c.2) The corresponding physical ﬁeld. (+) (+) (8.4) (+) (+) E0y = |E0y |eiφy . t) + c.Chapter 8 Polarization 8. where the vector amplitude has only transverse components: E0 (+) (+) (8. 2 2 Ex Ey Ex Ey cos φ = sin2 φ.c.2 Polarization Ellipse We will now examine the “trajectory” of the electric-ﬁeld vector. ˆ ˆ (+) (+) (8.1 Vector Plane Waves Recall that a plane wave in complex notation has the form E(+) (z) = E0 eikz . With a bit of algebra it is possible to expand the cosines using the sum-angle formula and then eliminate the explicit time dependence to obtain the equation for an ellipse. 8.6) Recall that for consistency among the real and complex notations. this direction is time dependent. Writing out the explicit phases of the components. However. t)ˆ.

1 Simple Cases We will now consider a few of the simplest polarizations. corresponding to special values of φ. In this case. This form isn’t completely transparent. always staying within a single plane: . 1. where E0y (8. Eq. and Ey . 2 2 E0x + E0y (8.12) tan α = E0x E¥ a E≈ Thus.8) In the special case α = 0.2. E0x E0y (8.10) Ex Ey = . In-phase components (φ = 0). the electric-ﬁeld vector remains on a straight line in the Ex -Ey plane (say. where tan 2α = 2E0x E0y cos φ . (8.102 Chapter 8.11) Thus. the electric-ﬁeld vector traces out a sinusoidal waveform. but it turns out that this is the relation for a ellipse rotated by an angle α. Ex . (8. In the full three-dimensional space. we recover the usual form for the unrotated ellipse: Ex E0x 2 + Ey E0y 2 = 1. (8.7) reduces to Ex E0x which is equivalent to 2 + Ey E0y 2 −2 Ex Ey E0x E0y = 0.9) 8. this is the case of linear polarization. Polarization where φ := φx − φy is the relative phase between the components. at z = 0) at an angle α from the Ex -axis.

the wave propagates to the right. 3. Equal-magnitude components in quadrature (φ = π/2. along the +z-direction): E¥ E≈ zoo=ook ^ ^ By convention. E0x = E0y =: E0 ). viewed such that the propagation direction is out of the page (i. and your thumb points along the direction of propagation. This case is the same as for RCP light. (8. In three-dimensional space. the electric-ﬁeld vector traces out a right-handed corkscrew. this polarization is called right-circular polarization (RCP). In this case. In three-dimensional space. but the electric-ﬁeld vector travels counterclockwise: E¥ E≈ zoo=ook ^ ^ Thus.e.7) for the polarization ellipse reduces to 2 2 2 Ex + Ey = E0 . the polarization is linear if φ = 0 or π. the electric-ﬁeld vector traces out a left-handed corkscrew.2 Polarization Ellipse 103 In general.8.13) the equation for a circle of radius E0 . hence the name (curl your ﬁngers in the direction of the electric-ﬁeld trajectory. this is left-circular polarization (LCP). . or if either of E0x or E0y are zero (in which case φ may as well be zero). Equal-magnitude components in opposite quadrature (φ = −π/2. Note that in this picture.. 2. we see that the electric-ﬁeld vector traces out a circle in the Ex -Ey plane in the clockwise direction. Working out the time-dependent behavior. E0x = E0y =: E0 ). Eq. (8.

Often to represent diﬀerent polarizations the Jones vectors are written in normalized form. 8.15) Thus. In the general case.14) A Jones vector is simply this explicit two-component form. (+) E0y (+) as an explicit two- (8. we can write out the electric-ﬁeld vector E0 component vector: (+) E0x (+) E0 = . (8. Polarization One important point to note from this analysis: the only diﬀerence between linear.104 Chapter 8. such that E0x (+) 2 + E0y (+) 2 = 1. linear polarization (angle α): y cos α sin α a x . the electric-ﬁeld vector traces out some ellipse in the Ex -Ey plane. and LCP light is the relative phase between the two component. the some of the basic Jones vectors are as follows: 1. Thus. RCP.3 Polarization States: Jones Vectors To make the notation more explicit. as we noted before. representing an elliptical polarization somewhere between linear and circular. a Jones vector models the polarization state of an electromagnetic wave. linear polarization (along the x-direction): y 1 0 x 2. at least in the proper coordinate system.

Thus. Since we are representing the state of the ﬁeld by a Jones vector.1 Vector Properties Being vectors.4 Polarization Devices: Jones Matrices 105 3. by this deﬁnition. left-circular polarization (LCP): y 1 √ 2 1 i x Notice that the 90◦ phase shift for the circular polarizations are represented by the complex phase diﬀerence between the two components. all of these vectors can have an extra overall phase.3. Jones vectors also have a magnitude given by E0x (+) 2 + E0y (+) 2 . we can model any linear polarization device by a 2 × 2 matrix. Mathematically. We will just reiterate two of the more useful ones here. there is interference. 8. that is. and LCP and RCP are also orthogonal.4 Polarization Devices: Jones Matrices A polarization device is some optical element or system that transforms the polarization state of an optical ﬁeld. (+) (−) (+) (8. which doesn’t amount to anything—it’s like a translation in time but it doesn’t aﬀect the geometry of the way. in which case the phase is relative to some other phase. a Jones vector is normalized if its magnitude is unity. Thus. (+) (+) (+) (+) (8.16) Note that since these are complex-valued vectors.18) where T is the Jones matrix modeling the system. Two Jones vectors are orthogonal if their inner product vanishes: E01 (+) ∗ · E02 = E01 · E02 = 0.8. linear-x and linear-y polarizations are orthogonal. Jones vectors inherit all the usual properties we associate with general vectors. E01 is the input ﬁeld. right-circular polarization (RCP): y 1 √ 2 1 −i x 4. (8. the inner product involves complex conjugation of the ﬁrst vector before the usual dot product. the most general linear transformation of the polarization state is a 2 × 2 matrix. Not unless.17) As we mentioned above. and E02 is the output ﬁeld. . Also. Such a matrix is called a Jones matrix. E02 = TE01 . 8.

The Jones matrix for a wave retarder that retards the phase of the y-direction by ∆φ is T= 1 0 0 ei∆φ . (8.” Let’s look at two special cases more closely.25) 2 1 .24) and thus rotates linearly polarized light by 90◦ . Polarization 8. the Jones matrix for a polarizer that transmits along the y-axis is 0 0 T= . (8. ∆φ = π/2 ∆φ = π −→ “quarter-wave plate” (λ/4-plate) −→ “half-wave plate” (λ/2-plate) These two cases are especially important in the optics laboratory.2 Wave Retarder A wave retarder is a device that adds a phase shift to one component of the ﬁeld. (wave retarder along y) (8. this is a polarizer with its transmission axis along the x-direction. Thus.20) 8. The wave propagation is to the right (along +z).106 Chapter 8. The quarter-wave plate induces the transformation 1 1 −→ 1 i .21) 0 1 −→ (+) E0x 0 . the ﬁnal state is linear-x. The corresponding Jones matrix for a polarizer that blocks the y-component of the ﬁeld is 1 0 T= .1 Linear Polarizer A linear polarizer (usually referred to simply as a “polarizer”) is a device that selectively annihilates the electric ﬁeld along one direction. (8. The initial polarization state is linear at 45◦ . (8.4.22) This is a retardation of phase in time. and the y-direction is horizontal. since our phase convention here is exp(−iωt) for the time dependence.19) 0 0 which induces the transformation (+) E0x (+) E0y Thus. (linear polarizer along x) (8. Obviously. the x-direction is vertical. More generally. Now we can visualize how a wave retarder changes the polarization state of light. (linear polarizer along y) (8. For this retarder. independent of the initial state of the ﬁeld. The half-wave plate induces the transformation 1 1 −→ 1 −1 . the half-wave plate ﬂips the polarization about the fast (x) axis. This is one common laboratory method for producing circular polarization.4. with Jones vector 1 1 √ .23) and thus converts linear polarization (with the right orientation) to LCP. the x-axis is the “fast axis.” and the y-axis is the “slow axis. at least if it has the right initial orientation (of 45◦ from the slow axis).

the polarization changes to LCP (λ/4 retardation) to linear with a 90◦ rotation (λ/2 retardation).components are shown in a lighter color.4 Polarization Devices: Jones Matrices 107 The wave is shown as the dark wave.and y. to RCP (3λ/4 retardation). and ﬁnally back to the original state (λ retardation). and the x. As the y-component is retarded. 0 retardation (∆φ = 0) λ/8 retardation (∆φ = π/4) λ/4 retardation (∆φ = π/2) 3λ/8 retardation (∆φ = 3π/4) λ/2 retardation (∆φ = π) 5λ/8 retardation (∆φ = 5π/4) 3λ/4 retardation (∆φ = 3π/2) 7λ/8 retardation (∆φ = 7π/4) λ retardation (∆φ = 2π) .8.

just like in ray optics (although there are now new matrices).5 Coordinate Transformations Consider a coordinate change from (x. Just as in ray optics.108 Chapter 8. 8. Polarization The same evolution of the polarization is shown here as an animation in the electronic version of this document (currently this only works under Acrobat).27) − sin θ cos θ . the power of this matrix formalism is that this works for cascaded systems via matrix multiplication.4 Cascaded Systems Of course.4. (8. .26) 8. y) coordinates to a new set of coordinates (x′ .3 Polarization Rotator The Jones matrix for a polarization rotator with rotation angle θ is cos θ sin θ This matrix induces the transformation cos θ1 sin θ1 provided θ2 = θ1 + θ. y ′ ) that are rotated by an angle θ with respect to the original coordinates. −→ cos θ2 sin θ2 (8. the order of matrix multiplication is critical: the ﬁrst polarization device that the ﬁeld encounters is the rightmost matrix in the product. 8.4.

the transpose property (AB)T = BT AT . Mathematically.29).31) is the transformation law for rotating the polarization device by θ. we can rewrite the expression (8.33) Rotating both the device and the vectors by an angle θ. we can consider the scalar quantity E02 (+) ∗ · TE01 (+) . We can also derive this transformation more formally. The inner product of two vectors E02 (+) ∗ · E01 (+) (8. (8.29) Notice that the rotation of the coordinates by an angle θ is the same as a rotation of the vector by an angle (+) (+) (−θ). to the Jones matrix for the polarization rotator. the Jones matrix for a polarization rotator of angle θ is R(−θ). (8. Thus. Thus. what is the Jones matrix Tθ for the same device rotated by an angle θ? The intuitive way to handle this is to consider an (+) equivalent picture where the ﬁeld goes through the unrotated device. Eºùoooôæû x¢ q x If E0 is the electric-ﬁeld vector in the original coordinates. we have argued that Tθ E0 (+) = R(−θ)TR(θ)E0 . the input ﬁeld E0 (+) should be rotated by an angle (−θ).28) (8. (8. we have R(−θ)E02 (+) ∗ · Tθ R(−θ)E01 (+) .8. so Tθ = R(−θ)TR(θ) (device rotation by θ) (8. so the two scalars are the same. To compensate. you can see this by comparing the rotation matrix. Using the fact that the dot product is a matrix product with a transposition.32) is a scalar. then the vector E0 given by (+) (+) E0 ′ = R(θ)E0 .34) as E02 (+) ∗ · R(θ)Tθ R(−θ)E01 (+) . What about rotations of polarization devices? That is.35) .27)—that is. so that the input to the rotated device is R(θ)E0 . and the fact that [R(θ)]T = [R(θ)]−1 = R(−θ). (8. and thus is independent of coordinate rotations. (+) (+) ′ in the new coordinates is (8. we should restore things to the original coordinates by undoing the rotation.30) This should be true for any input polarization. After passing through the device. given a Jones matrix T. (8. This is evident from the drawing. since the E0 ′ vector is closer to the x′ -axis than the E0 vector is to the x-axis.5 Coordinate Transformations 109 y¢ y Eºùoooôû. deﬁned by R(θ) := cos θ − sin θ sin θ cos θ . Eq. Eq. (+) (8. Here. a · b = aT b.34) These combined rotations are equivalent to a coordinate rotation by angle (−θ). rotating the output vector by an angle θ. R(θ) is the rotation matrix.

As with the polarization vector.1 Birefringence Now that we have covered the mathematical formalism of polarization. in contrast to the ray matrix case where the Hamiltonian structure ensures a nonvanishing determinant. and ammonium dihydrogen phosphate (ADP).37) (8. Recall that the eigenvectors (eigenpolarizations) are deﬁned by the relation TE0 (+) = λE0 . we must have T = R(θ)Tθ R(−θ). potassium dideuterium phosphate (KD*P). which is the same as Eq.110 Chapter 8.39) where λ (the eigenvalue) is a constant.38) (8. potassium dihydrogen phosphate (KDP).7. for example.λ2 . One important class of optical materials are birefringent materials.33).λ1 + λ2 α2 E0. uniform electric ﬁeld. (coordinate rotation by θ) (8.36) Keeping the signs straight is importnat. Another example of material birefringence is the electro-optic eﬀect.λ2 .. we will discuss some of the materials used to make polarization devices. 8. so there is no general expression for the determinant of a Jones matrix. where the material is anisotropic: the wave sees a diﬀerent index of refraction for diﬀerent polarizations. Examples of electro-optic crystals are lithium tantalate (LTA). the eigenvalues could have a zero value (e. (+) (8. the eﬀect of rotating the coordinates is the opposite: T′ = R(θ)TR(−θ).6 Normal Modes Polarization devices are represented by matrices. due to anisotropic crystal structure or mechanical stress. These eigenvectors are called the normal modes of a system. Polarization For consistency of this expression with (8. The normal modes are easy to propagate through the system. There are at most two such vectors for a 2 × 2 matrix that satisfy this (+) (+) equation. (+) (+) (8. which obviously have eigenvectors and eigenvalues.g. Consider a wave incident on a birefringent plate as shown. A vector can be decomposed into a superposition of the eigenvectors. which we can rewrite as Tθ = R(−θ)TR(θ). since the eﬀect of the Jones matrix is the same as multiplication by a constant factor.λ1 and E0. Polarization devices can be lossy. Materials could be birefringent. but it can be a bit tricky. and we can call them E0. and then easily propagated through the device: TE0 (+) = λ1 α1 E0.31). E¥ n¥ z E≈ d n≈ . respectively. where certain crystals become birefringent when placed in a dc. (8.40) Note that in general. 8.7 Polarization Materials 8. for a polarizer). for the eigenvectors corresponding to λ1 and λ2 .

which sees only the index nz in the material. because it sees an angle-dependent index: n(θe ) = ny cos θe + nx sin θe .46) Thus. The incident ray lies in the x-y plane. (8. the relative phase shift between the two components is 2π(ny − nx)d/λ0 .” the German word for “perpendicular”). we can make a λ/4-plate by choosing λ0 (ny − nx )d = . Then in the notation above.41) Thus. (8.oon¥. (8. λ0 (8. In some cases. the separation of polarizations is useful in . (8. λ0 (8.8. this is a useful eﬀect. Snell’s law gives sin θi = nz sin θo . So for example. whereas the orthogonal polarization. The polarization that lies inside the plane of incidence is P-polarization (for “parallel” to the plane of incidence). For example. with Jones matrix 1 0 0 eiπ/2 = 1 0 0 i . normal to the plane of incidence. which we will call the plane of incidence. n¡o=o1 (air) n≈.7 Polarization Materials 111 Suppose that ny > nx . the y-direction is the “slow axis. where diﬀerent polarizations are refracted diﬀerently. is S-polarization (for “senkrecht. The S-polarized component gives rise to the ordinary wave.” Then the optical path length for x-polarization is kx d = and the optical path length for the y-polarization is ky d = 2πny d . Snell’s law cannot be solved analytically for θe : sin θi = n(θe ) sin θe . Thus. The basic eﬀect is multiple refraction. (8. Birefringent materials also have more complicated behavior for rays at non-normal incidence. this relation leads to a diﬀerent refraction angle for the extraordinary wave.44) as we saw above. which is more complicated. Consider a wave with mixed polarization incident from free space into a birefringent medium.oon (birefringent) extraordinary wave qe qo ordinary wave qi mixed polarization y z x Let’s deﬁne a bit of notation that we will return to when introducing the Fresnel relations.45) The P-polarizated component gives rise to the extraordinary wave.” and the x-direction is the “fast axis. compared to the ordinary wave.43) 4 This gives a π/2 phase retarder along the y-direction.42) 2πnx d .47) First.

48) After passing through an optically active medium of thickness d.in = (+) 1 0 = 1 2 1 i + 1 2 1 −i (8. it is even possible to get separation of the polarizations for normal incidence. we can denote the two indices by n+ and n− for RCP and LCP light. Optical activity causes optical rotation of input linear polarization. The helicity of the medium breaks the symmetry between the two polarizations. the Jones vector becomes E0. Optical activity is characteristic of media with helical molecules. It turns out that for the extraordinary waves. For this reason. Some crystals.out = (+) 1 2 1 i eiφ+ + 1 2 1 −i eiφ− . note that linearly polarized light (say.49) where φ± = k± d = 2πn± d/λ0 . Note that given a material with a particular helicity.out = eiφ0 (+) cos φ sin φ . To see this. Examples of such media include quartz and sugar in the solid phase. quartz can have n+ > n− or n+ < n− . But multiple refraction is even a bit more subtle. the geometrical rays are not in general orthogonal to the wave fronts (hence the name “extraordinary”). For example. Glucose and fructose are two diﬀerent sugars that give rise to opposite senses of optical activity. (8. In mathematical notation. One other important manifestation of optical activity is the Faraday eﬀect. Polarization constructing polarizers (polarizing beam splitters). 8. an optically active material has diﬀerent indices for the two circular polarizations.50) . linear-x) is a linear combination of RCP and LCP: E0. such as YIG (yttrium-indium-garnet). depending on the crystal structure.7. exhibit optical activity when they are placed in a uniform magnetic ﬁeld.112 Chapter 8. We can rewrite this in the form E0. (8. respectively. or sugar solution and turpentine in the liquid phase.2 Optical Activity Whereas a birefringent material has diﬀerent refractive indices for diﬀerent linear optical polarizations. the same material with the opposite helicity exists in principle as well—it would just be the mirror image of the orginal material. Multiple refraction due to a birefringent calcite crystal is shown here in this photograph.

where R(θ) is the rotation matrix. which prevents laser light from being reﬂected back into the laser (which could cause instability or even damage). π/2 from the x-axis. Assume ideal polarizers and note that the rotator produces the same rotation independent of the light’s direction.8 Exercises 113 where φ = (φ− − φ+ )/2 and φ0 = (φ− + φ+ )/2. Compute the intensity transmission coeﬃcient for the system. One important optical polarization system is the optical isolator. An optical isolator typically consists of a polarizer (say. Ax.) (b) A cheaper isolator is a polarizer (say. .2 Ay.8 Exercises Problem 8. 8. and transmits some fraction β of the y polarization. This is a simple realization of the quantum Zeno eﬀect. but that any light traveling backwards through the isolator will be completely extinguished. . where each polarizer acts as a “measurement” of the polarization state—the polarization is “dragged” by the measurements as long as they are suﬃciently frequent.8.1 = A′ C′ B′ D′ Ax. dz (8. (Both α and β should be small for a decent polarizer. A realistic polarizer along the x-direction blocks some fraction α (of the electric ﬁeld) of the x polarization. Write down the Jones matrix for an ideal polarizer that transmits x-polarized light. but will not block arbitrary return polarizations. Problem 8. along x).1 Ay.1 Ay. A beam of light can be described by its position and direction as a geometrical ray as well as by its polarization Jones vector. (Realistic isolators attenuate the reverse beam by about 40 dB. 2π/2N. the last polarizer is oriented along the y-direction. . it is conventional to specify the rotation per unit length as dφ = V B.1.) Write down the Jones matrix for this realistic polarizer. . the Jones matrix for the optic rotated by an angle θ is Tθ = R(−θ) T R(θ). Show that the transmission coeﬃcient approaches unity as N −→ ∞.2. = dz λ0 For the Faraday eﬀect. linear polarization is rotated by the angle φ. Problem 8. π(n− − n+ ) dφ .2 =T Ax. followed by a polarizer at 45◦ .4.3.51) where the constant V is the Verdet constant. Suppose that an input wave is polarized in the x-direction. That is. The polarizers have their transmission axes at angles π/2N. in the order that an input wave sees them. Show that if a laser passes through this isolator. followed by a 45◦ polarization rotator. the isolator will block any light returning from a direct mirror reﬂection. (b) Consider a system of N cascaded polarizers. . (a) Consider a linear polarizer and a wave linearly polarized at an angle θ with respect to the polarizer’s transmission axis.1 . Thus. Note that if T is the Jones matrix for an optic. along x) followed by a quarter-wave retarder oriented at 45◦ .52) (8. Suppose the position/direction vector and polarization vector transform respectively according to y2 θ2 =M y1 θ1 = A C B D y1 θ1 . Show that the intensity of the wave is reduced by cos2 θ after passing through the polarizer (this is called the Law of Malus). Problem 8. The rotation is typically proportional to the magnetic ﬁeld in regimes of interest for laboratory polarization rotators. It is more useful to specify a rotation per unit length. (a) Derive the Jones matrices for forward and backward propagation through the isolator. 3π/2N. Show that x-polarized light passes forward through the system unattenuated.

Assuming the mirrors are perfect reﬂectors. Ay.2 Problem 8.) Problem 8.114 Chapter 8. (Obviously you should not make any low-loss approximations here. Suppose that an ideal (linear) polarizer and a polarization rotator (which rotates the polarization by ∆θ) are placed inside the cavity. ∆φ = π/4. To “Q-spoil” the cavity. and ∆φ = π/2.1 . and the light polarization is perpendicular to the table. what is the ﬁnesse of the cavity? . we can switch on some phase retardation to introduce extra loss. Consider a planar cavity.2 = Ω Ax. Calculate the cavity ﬁnesse for the ring cavity of Problem 1 without the gain medium. You need light that is polarized parallel to the table.7.1 Ay.6. Polarization Write down an expression for the matrix Ω that models both aspects of the optical system. for retardations of ∆φ = 0. the matrix that gives the transformation y1 y2 θ1 θ2 Ax. How can you do this with only two mirrors? (And without turning the laser on its side!) Problem 8. that is. Suppose you have a laser beam parallel to the optical table. the cavity ﬁnesse is unchanged. If we ignore losses other than due to the polarization rotation.5. One method of Q-switching a laser is to include a polarizer and a switchable retarder (Pockels cell) oriented at 45◦ in the cavity.

we considered reﬂection and refraction in ray optics as a consequence of Fermat’s principle. In Chapter 2. We need wave optics to establish the amplitude of the reﬂected and refracted components. We can make whatever assumptions we want.Chapter 9 Fresnel Relations 9. These three waves are not independent: the electromagnetic boundary conditions at the dielectric interface impose constraints among the three waves. The thing to worry about is the generality of the solution. and transmitted waves.1 Optical Waves at a Dielectric Interface One situation where the vector nature of light is crucial is in the reﬂection and refraction at a dielectric interface. so that the three ﬁelds have the form Ei (+) = E0i eiki ·r = E0r eikr ·r = (+) E0t eikt ·r . respectively. n¡ n™ H0r E (+) 0r (+) E0t (+) H(+) 0t qt qr qi E0i (+) y H(+) 0i xoo=oo0 z x We will assume that the waves are monochromatic. and transmitted wave. since we are anticipating the result. The electromagnetic boundary conditions are . and t refer to the incident. r. so long as the solution turns out to be self-consistent—the self-consistency justiﬁes the assumptions. We are assuming a priori that there is an incident. but we will return to this later. (+) (+) Er (+) (9.1) (+) Et where the subscripts i. reﬂected. Note that we are making some pretty speciﬁc assumptions about the ﬁeld. reﬂected. Consider the setup in the diagram. But that only ﬁxed the reﬂection and refraction angles.

ˆ ˆ ˆ (+) (+) (+) (9. 2. (9. (+) (+) (+) (9. since both waves see the same refractive index. where k0 is the vacuum wave number. The geometrical properties of reﬂection and refraction are purely a consequence of the wave geometry. (Snell’s Law) (9. The tangential component is the component in the y-direction. It turns out that the ﬁrst boundary condition is suﬃcient to completely constrain the ﬁelds. This simpliﬁes to give Snell’s Law. (9. we’ll use the fact that E· z (the tangential component) ˆ is continuous across the boundary: ˆ ˆ ˆ (E0i · z )eiki y sin θi + (E0r · z )eikr y sin θr = (E0t · z )eikt y sin θt . we require ki sin θi = kr sin θr . n1 cos θi (+) (+) (9.9) where we have already used the equality of the phases. we recover the Law of Reﬂection: θi = θr . so the continuity condition becomes H0i · y + H0r · y = H0t · y . The tangential components (to the dielectric interface) of E and H are continuous across the boundary. most straightforwardly in integral form.10) = (9. while the right-hand side is the transmitted wave.2) The left-hand side is the sum of the incident and reﬂected waves. We can rewrite this relation as E0i + E0r E0t (+) (+) (+) (+) (+) (+) (9. η0 η0 n2 cos θt . For this to be true for all y.8) which will be more convenient for later manipulations. we have recovered everything that we derived from ray optics.5) (Law of Reﬂection) (9.3) But ki = kr . Thus.7) = 1.11) . The continuity of the electric ﬁeld becomes E0i + E0r = E0t . Using the fact that the magnetic and electric ﬁelds are related by H0 = E0 /η = nE0 /η0 . Equating the incident and transmitted phases. and k = nk0 . the exponents must all be identically equal: the phases of the waves must agree everywhere along the boundary. The continuity of H is only slightly more complicated. These boundary conditions follow directly from the Maxwell equations. Now to consider the amplitudes. We have deﬁned our coordinate system such that the interface is located at x = 0. Fresnel Relations 1.116 Chapter 9.6) (9. The normal components of D and B are continuous across the boundary. First. Let’s now apply the boundary conditions.4) Thus. ki sin θi = kt sin θt . we can rewrite this as E0i − E0r (+) E0t (+) (+) n1 E0i η0 (+) cos θi + n1 E0r n2 E0t cos θr = − cos θt . n1 sin θi = n2 sin θt . where η is the wave impedance and η0 is the vacuum wave impedance. because the electric ﬁelds lie entirely along the z-direction. this boundary condition becomes − Again. For the phases of the incident and reﬂected waves to be equal.

n¡ n™ E0r H0r qr qi E0i (+) (+) (+) E0t (+) (+) H0t qt y H(+) 0i xoo=oo0 z x For the vector convention shown. the Fresnel relations turn out to be rP = n1 cos θt − n2 cos θi n1 cos θt + n2 cos θi 2n1 cos θi tP = . n1 cos θi + n2 cos θt E0t (+) (9. Thus.14) (+) (+) E0i E0i so if we deﬁne the ﬁeld reﬂection coeﬃcient rS as the ratio of the reﬂected to incident ﬁeld amplitude. rS = n1 cos θi − n2 cos θt n1 cos θi + n2 cos θt 2n1 cos θi tS = . n1 cos θt + n2 cos θi (Fresnel relations.1 Optical Waves at a Dielectric Interface 117 Adding Eqs. S-polarization) (9.12) Now we can deﬁne the ﬁeld transmission coeﬃcient tS as the ratio of the transmitted to incident ﬁeld: tS := We can rewrite Eq.13). n1 cos θi n1 cos θi (9. called transverse magnetic (TM) or P-polarization.15) Note that we have assumed a particular input polarization for the light.8) and (9. P-polarization) (9.16) are the Fresnel relations for S-polarization. We can repeat this derivation for the other polarization. = rS := E0r (+) (+) E0i E0t (+) (+) E0i = 2n1 cos θi . (9. The setup is shown here in the following diagram.11). (9.8) as − 1.17) . because the electric-ﬁeld vector is perpendicular to the plane of incidence (the x-y plane in our coordinates here). (9.15) and (9. rewritten here. Eqs. The notation is that the polarization is transverse electric (TE). A more common notation in optics is S-polarization (“S” is for “senkrecht. 2 E0i (+) (+) E0t =1+ n1 cos θi + n2 cos θt n2 cos θt = .” which means “perpendicular” in German). (9.13) E0r (+) = n1 cos θi − n2 cos θt . n1 cos θi + n2 cos θt (Fresnel relations. since the electric ﬁeld is tangential to the interface. n1 cos θi + n2 cos θt (9.9. parallel to the plane of incidence.

from air onto glass). 9. the condition n1 cos θt − n2 cos θi > 0 (9. there is the possibility of a π phase shift of the wave after it encounters the boundary. for either polarization. One the other hand. rS > 0.20) which follows from the general trigonometric identity cos(A + B) sin(A − B) = sin A cos A − sin B cos B. Thus. in turn.1.16) and (9.19) This. For S-polarization. and transmitted ﬁelds at an air-glass interface that matches our setup here (air on the left. Let’s summarize these two cases as follows: 1. so there is no phase shift of the components. and the reﬂected wave is in phase with the incident wave. sin(θt − θi ) cos(θt + θi ) > 0. so the relative phases are in some sense ﬁxed. In this case. This plot is animated in the electronic version.20) is true if the two factors are either both positive or both negative. n2 < n1 : both factors are positive when θt > θi and θt + θi < π/2 . From Eqs. Thus.118 Chapter 9. Eq. glass on the right). except possibly for an overall minus sign. This case is even a bit more complex. Thus. the sign of rP depends on the sign of n1 cos θt −n2 cos θi . Reﬂection: P-polarization. Fresnel Relations These are the Fresnel relations for P-polarization. (9. we know that θt > θi . (9. is equivalent to the condition (9. Reﬂection: S-polarization. (9. and so cos θt < cos θi .1 Phase Changes and the Brewster Angle The phase factors of all the waves must be equal. If n1 < n2 . the reﬂected wave only picks up a π phase shift when the light is incident on a more dense medium (e. Suppose that this quantity is positive.18) is equivalent to sin θt cos θt − sin θi cos θi > 0. Thus. and the reﬂected wave has a π phase shift. By examination of Eqs. The reﬂection and transmission coeﬃcients are also real. then from Snell’s law. Since n1 /n2 = sin θt / sin θi .. rS < 0. (9. reﬂected.17) it is evident that both tS > 0 and tP > 0.g. and so cos θt > cos θi . then θt < θi . If n1 > n2 . Here is a graphical representation of incident. Transmission. the transmitted wave is in phase with the incident wave. The reﬂection coeﬃcient is more subtle.17). the sign of the reﬂection coeﬃcient depends on the sign of (n1 cos θi − n2 cos θt ).

2 (9. the incident and reﬂected ˆ ˆ ˆ waves overlap.27) Now we can use the explicit forms for the energy densities in terms of the ﬁelds: 2 wi = 2n1 ǫ0 |Ei (+) 2 | .24) (9. which means that the reﬂected ray is perpendicular to the refracted ray. To consider the energy balance. the incidence angle for the borderline case is θi.17) vanishes. However. reﬂected. The energy ﬂuxes at the surface due to the the incident. c0 c0 c0 wi cos θi = wt cos θt − wr cos θi . | .2 Reflectance and Transmittance 119 2.23) Energy balance requires that the energy ﬂux towards the interface is the same as the energy ﬂux away: Si · x = St · x + Sr · (−ˆ). 9.4) the Poynting vector can be written in terms of the electromagnetic energy density as ˆ S = c w k.B = tan−1 n2 . we note that (see Problem 4.28) . this is the case where the numerator in the expression for rP in Eq. The borderline case occurs when θi + θt = π/2.25) (9. (9. n2 > n1 : both factors are negative when θt < θi and θt + θi > π/2 There is a π phase change on reﬂection if neither set of conditions holds. = Si · x ˆ n2 wi cos θi 2 (+) wr = 2n1 ǫ0 |Er |2 . There is no analogous behavior for S-polarization. the transmittance as the ratio of transmitted to incident energy ﬂuxes: T = St · x ˆ n1 wt cos θt .B . = R= Si · x ˆ wi wi cos θi n1 Similarly.22) π − n2 sin θi. n1 (Brewster’s angle) (9. so we can go far away from the interface until the incident and reﬂected beams no longer overlap.B = n2 sin θt = n2 sin Thus. n1 n2 n1 We can now deﬁne the reﬂectance as the ratio of reﬂected to incident energy ﬂuxes: c0 wr cos θi wr − Sr · x ˆ n1 = c0 . Thus. it seems that we should consider interference cross terms when calculating the Poynting vectors. and transmitted waves are given in terms of the respective Poynting vectors. Sr · x. (9. so that the reﬂected wave vanishes. Thus. we can treat them separately by realizing that plane waves are an idealization: physical beams have ﬁnite width. Also. while part of the S-polarization and all of the P-polarization transmit. 2 wt = 2n2 ǫ0 |Et (+) 2 (9. ˆ ˆ x Writing this out in terms of the energy densities. and St · x. the this borderline case corresponds to an incidence angle θi.26) (9.9. From Snell’s law.21) which is called Brewster’s angle. Note that the reﬂection coeﬃcient only vanishes for P-polarization.2 Reﬂectance and Transmittance Now we must characterize the energy transport due to the ﬁelds at the dielectric interface.B determined by n1 sin θi. Note that since these are plane waves. (9. so the borderline case corresponds to rP = 0.B = n2 cos θi. Brewster’s angle is also called the polarizing angle because light incident at this angle is polarized on reﬂection from a dielectric interface—only S-polarized light reﬂects. Si · x.

33) (9.c .36) (9. n1 (9. the relative transmitted intensity.c = n2 . In this case. say. the obvious expression in terms of the ﬁeld transmission coeﬃcient (T = |t|2 ). Supposing that θi > θi. 9. we have sin θt = so that cos θt = ± 1 − sin2 θt = ±i sin2 θi − 1. The simple interpretation is that there is no transmitted plane wave. for reﬂection from the surface of a glass window.35) For a bounded solution. Thus.36). The transmitted wave has the form Et (+) = E0t exp {ikt [(cos θt ) x + (sin θt ) y]} n1 sin θi > 1.34) for either polarization. Thus. n2 cos θt 2 |t| . In practice. the transmitted wave becomes 2 n1 (+) (+) sin2 θi − 1 x + (sin θt ) y . As in the geometrical case.30) T = n1 cos θi In terms of the reﬂectance and transmittance. a slab of glass (such as a glass window used as a beam splitter). we must choose the +i branch of the the transmitted wave. Then we see that n1 (sin θi ) y . in which case the transmittance reduces to |t|2 . (9. 2 n1 (+) (+) sin2 θi − 1 Et = E0t exp −kt n2 where the ﬁnal form shows explicitly that the cosine is an imaginary number.3 Internal Reﬂection Recall that Snell’s law reads n1 sin θi = n2 sin θt .120 Chapter 9. the reﬂectance is the relative intensity. n2 n1 n2 2 (+) (9. Rather.38) . In particular. (9. they are measured in the same surrounding medium (say. R = |r|2 . Eq.30) may seem counterintuitively complicated. the transmitted angle becomes a complex number.32) which comes from the condition θt = π/2. (9. (9.25) becomes R + T = 1.29) but the transmittance is not. (9. (9. Fresnel Relations Thus.31) Note that the expression (9. but we will explore this phenomenon in more detail. the incident and transmitted energies are measured outside the dielectric medium. we usually care about reﬂection at a dielectric interface that is at one end of. x exp ikt n2 (9.37) n2 square root in Eq. in general. Et = E0t exp ikt ±i (9. air). the maximum possible refraction angle. Beyond this incident angle. if n1 > n2 there is a critical incident angle given by sin θi. the reﬂected and transmitted intensities (detected outside the window) must add up to the incident intensity.

while the right exponential factor has the form of a traveling plane wave along the y-direction. You can clearly see the propagation of the wave fronts parallel to the interface. δ = kt sin2 θi − 1 . the wave fronts of the evanescent wave propagate along the interface. (9. n¡ n™oo<oon¡ kr exponential damping qi ki transverse wave fronts y z x From the damping factor in Eq.39) is the characteristic distance over which the evanescent wave damps away. (9. so that the skin depth is relatively large. The other. The incoming wave is incident at an angle just slightly past the critical angle.40) . Thus. is that R = 1.3 Internal Reflection 121 which is an exponentially damped harmonic wave: the left exponential factor represents exponential damping along the x-direction. the wave decays as Et where the skin depth δ. This is one way to see that the evanescent wave carries no energy away from the interface.9. (9. Since the (second) propagatingwave factor in Eq. which we will see shortly. Here is a visualization of internal reﬂection at a glass-air interface. as shown. given by n1 n2 2 (+) ∼ e−x/δ .c ) is called total internal reﬂection. −1 (9.38).38) only involves y. This damped ﬁeld is called the evanescent ﬁeld. this situation (θ > θi.

“A simple variable-ratio beam splitter for holography. 9. 602 (1983).41) sin2 θi − 1.36). and R. so we can write sin2 θi − (n2 /n1 )2 cos θi sin2 θi − (n2 /n1 )2 (n2 /n1 )2 cos θi rS = exp −i2 tan−1 rP = exp iπ − i2 tan−1 (9. (9. Polloni. This eﬀect can be used. Hence. M.16) and (9. what I said above about the evanescent wave isn’t quite true. since the reﬂection coeﬃcients themselves are not unity in general. This phenomenon is called frustrated total internal reﬂection and is the analog of quantum tunneling. Bertani.43) .3. But if another glass interface is placed very close to the ﬁrst. (9. so that the air is present only in a thin (∼λ) gap. there is a phase shift on reﬂection. α + iβ (9. for internal reﬂection at a glass-air interface. provided that the evanescent wave is interrupted.46) for the explicit phase shifts of the two polarizations. (internal reﬂection) (9. Fresnel Relations Actually. if both polarization components are present. the wave is damped in the air.122 Chapter 9. for example. Apart from an overall minus sign (π phase shift).17) are n1 cos θi − n2 cos θt n1 cos θi + n2 cos θt n1 cos θt − n2 cos θi rP = .1 Phase Shifts The reﬂection coeﬃcients from Eqs. (9. . n1 cos θt + n2 cos θi rS = From Eq. (up to a minus sign) where tan φ = β/α. the damped wave will begin to propagate again in the second glass region. It can carry away some energy. there is a relative phase shift between them.44) so that the reﬂectance for both polarizations is unity: RS = RP = 1. to make a variable-ratio beam splitter by adjusting the size of a small air gap between two prisms1 . there is a refractive index ratio in the expression for rP that is absent in the expression for rS . (n2 /n1 )2 cos θi − i sin2 θi − (n2 /n1 )2 sin2 θi − (n2 /n1 )2 cos θi + i sin2 θi − (n2 /n1 )2 (n2 /n1 )2 cos θi + i rP = − Notice that both of these reﬂection coeﬃcients are of the form ± α − iβ . For example. The phase angle in each case is exp(−2φ). Cetica.45) However.42) so that we can eliminate θt in the reﬂection coeﬃcients: rS = cos θi − i sin2 θi − (n2 /n1 )2 (9.” Journal of Physics E: Scientiﬁc Instruments 16. we have (upon choosing the positive branch of the square root) cos θt = i n1 n2 2 (9. 1 D.

) The reﬂectance for P-polarization is much smaller because this incidence angle is fairly close to the Brewster angle of θi. cos θi = cos θt = 1.52). let’s work out speciﬁc numbers for an interface between air (n1 = 1) and glass (n2 = 1. One optical element. 9. First. q For glass with n = 1. we discussed how a birefringent material can make a wave retarder (wave plate).7◦ .4 Air-Glass Interface: Sample Numbers 123 This phase shift can be very useful.3%. n1 (9. (These numbers are useful to remember in the laboratory for making simple 10% beam splitters. the correct incidence angle turns out to be θ = 55. At normal incidence. so r = rS = rP = and Thus. around 8%. respectively. the reﬂectances for the two polarizations are quite diﬀerent. but unlike typical wave plates.7◦ .52.5◦ .48) = 9. (9.4 Air-Glass Interface: Sample Numbers To get a feel for the theory presented here. In Section 8. the retardation of a rhomb is relatively insensitive to the optical wavelength used.1.50) = 0. θt = sin−1 n2 Thus. the refraction angle is n1 (9. uses two internal reﬂections to produce a π/2 (λ/4) phase shift. or about 10% and 1%. the Fresnel rhomb. The rhomb has the sometimes inconvenient property of translating the beam.9. a glass window attenuates light due to reﬂection near normal incidence by about twice this amount.B = tan−1 n2 = 56.51) The reﬂectances for the air-glass interface for both polarizations are shown here as a function of the incident angle.47) R = |r|2 = 4. .7% 2 n1 cos θt − n2 cos θi RP = |rP |2 = n1 cos θt + n2 cos θi (9.49) sin θi = 27.7. But the phase shifts due to internal reﬂection obviously also act as wave retarders.21 n1 + n2 (9. At 45◦ incidence. the reﬂectances are RS = |rS |2 = n1 cos θi − n2 cos θt n1 cos θi + n2 cos θt 2 n1 − n2 = −0.94%.

7 90 qi (degrees) On the other hand. 1 Re fle ctance S P 0 0 33.124 Chapter 9.c = sin−1 as is evident from the plot.3 45 90 qi (degrees) The reﬂectances are unity beyond the critical incident angle θi. nair = 41. nglass (9. Fresnel Relations 1 Re fle ctance S P 0 0 45 56. the reﬂectance curves look the same except that they are horizontally compressed. for a glass-air interface.1◦ .52) .

(9. J = σE. we will need to treat wave propagation in a conductive medium. but in reality at optical frequencies it is best modeled by a complex number. due to the lag of the response of the conduction electrons to the electric ﬁeld (see Section 17.56) This means that ρ −→ 0 exponentially with a time constant ǫ/σ. (+) (9. the current density is proportional to the local electric ﬁeld. (9.54) to write ∂ρ σ = −σ(∇ · E) = − ρ.9.5. If the current density at some point is increasing. This is a simple example of an adiabatic approximation. ∇×H= ∂t where ρ is the charge density and J is the current density. (9. We can think of the conductivity as a real constant. so to good approximation we can set ρ = 0.57) so that the only source is the current density. This covers the important case of metallic reﬂectors. t) = E0 ei(kz−ωt) . We can again try the plane-wave ansatz E(+) (r.2.53) becomes ∇ · D = 0. ∂t (9. For our purposes.5 Reﬂection at a Dielectric-Conductor Interface One useful generalization of reﬂection at a dielectric interface is to treat the case where one of the materials is a conductor.59) . We can now repeat the derivation of the wave equationfrom Chapter 4. But to deal with reﬂection from a single conductor. For a good conductor (large σ). Charge and current are related by the continuity equation ∂ρ = −∇ · J. ∂t2 ∂t (9. this means that the ﬁrst Maxwell equation in Eq.2 for a model of the conductivity). We can use the relation D = ǫE and Eq. then there must be a convergence of current density to transport charge to that point.58) with the same result for H.54) where σ is the conductivity. which are some of the most common in everyday experience as well as the laboratory. ∂t ǫ (9. With ﬁeld sources. as this is more subtle. 9.53) (9.55) which is essentially equivalent to the statement that charge is conserved. and this is why we can justify ignoring it. with the result ∇2 E − µ0 ǫ ∂2E ∂E − µ0 σ = 0.1 Propagation in a Conducting Medium In Chapter 4. For an ohmic material. the charge density relaxes to nothing. the damping time is very short. Note that we are ignoring magnetic properties of the conductor by assuming µ = µ0 . (9. we considered only Maxwell’s equations in dielectric media without sources. On longer time scales of optical interest.5 Reflection at a Dielectric-Conductor Interface 125 9. We will not deal explicitly with the case of reﬂection at the interface of two conductors. which we can write in terms of E. the Maxwell equations become ∇·D =ρ ∇·B=0 ∂B ∇×E=− ∂t ∂D + J.

67) ǫ 2ǫ0 1+ σ ǫω 2 +1 (9. so long as you use a complex refractive index. ˜ ω then the dispersion relation (9.60) takes the same form as for an ordinary dielectric. ǫ σ n2 = n2 (1 − κ2 ) + i2n2 κ = ˜ +i . ˜ (9. ǫ0 (9.70) .60) Note that the frequency dependence in this equation is deceptively simple. ˜ ˜ ω 2 µǫ 2 1+ σ ǫω 2 (9. (9.64) and (9.68) −1 . (9. (9.126 Chapter 9. n2 = ǫ . ωǫ0 (9. since the refractive index for a dielectric is the square root of the dielectric constant. ˜ with the replacement ǫ −→ ǫ. ˜ Similarly. we can compute the wave vector in terms of the refractive index as k=ω where 2 k± = √ µ0 ǫ = ω µ0 ǫ0 n = k+ + ik− . k 2 = ω 2 µ0 ǫ. ǫ 2 n2 κ2 = (Im[˜ ]) = n 2ǫ0 σ 1+ ǫω 2 Similarly. ǫ0 (9. Fresnel Relations Substitution shows that the plane wave is a solution provided that k 2 = µ0 ǫω 2 + iµ0 σω = ω 2 µ0 ǫ + i σ . If we deﬁne a complex permittivity σ ǫ=ǫ+i .66) ǫ0 ωǫ0 Matching the real and imaginary parts.61) we can deﬁne the complex refractive index n as the square root of the complex dielectric constant: ˜ n2 = ˜ ǫ ˜ .63) (9. we have the two equations n2 (1 − κ2 ) = These equations have the solutions n2 = (Re[˜ ]) = n 2 ǫ .62) (9. This is the central point of electromagnetic wave propagation in conductors: everything is the same as for an ordinary dielectric. computing n2 from ˜ both Eqs. First. ǫ0 2n2 κ = σ . Let’s compute explicitly the real and imaginary parts of the refractive index. ω (9.65).64) We can also write the complex refractive index itself in terms of real and imaginary parts as n =: n(1 + iκ).65) where κ is the attenuation index or extinction coeﬃcient. as in general σ and ǫ are themselves functions of frequency.69) ±1 .

However. assuming a conductivity of 60 × 106 (m · Ω)−1 . While being undesirable from the point of view of making mirrors for highpower lasers. we can rewrite the plane-wave solution as E(+) (r) = E0 eikz = E0 ek− z eik+ z . the plane wave breaks up into propagating and damping factors—the imaginary part of the refractive index leads to damping of the wave. An induction heater drives a high-voltage.5 km. We can again deﬁne a skin depth δ for the damped wave by e−k− z = e−z/δ . The following photographic sequence shows a quarter being heated and then melted by an inductive heater. k− ω µ0 ǫ0 nκ 2πnκ 2πIm[˜ ] n (9. . and generates heat as it is absorbed.5 Reflection at a Dielectric-Conductor Interface 127 Thus. You can see this in the photo series. Roughly speaking. so that δ ≪ λ0 .15 mm.2 Inductive Heating Since the transmitted wave gets damped. which ideally produces a uniform. causing Joule heating—from our analysis above. This is called the skin eﬀect: for good conductors. so that δ= 1 λ0 λ0 1 = √ = = . the skin depth turns out to be only 0. 9. The corresponding electric ﬁeld is roughly a cylindrical wave collapsing onto the quarter.73) As before. around 1.5.71) As we saw for internal reﬂection. The electromagnetic wave generated by the coil couples into the conductive object. Thus. Of course. The wavelength for the 200 kHz radiation is quite large. with the wave polarization being in the plane of the quarter. oscillating magnetic ﬁeld along the solenoid axis. an electromagnetic wave penetrates the surface by only a small fraction of the vacuum wavelength. you can think of the coil as being a solenoid. because the outside of the quarter is visibly much hotter than the middle. due to the skin eﬀect. the equivalent picture is that the radiation induces currents in the object. radio-frequency signal through a coil that surrounds the conductive object to be heated. nκ is large. Of course.9. Typically. this is the length scale over which the wave damps exponentially away. the copper inside the quarter is an extremely good conductor. since the middle is only being heated by thermal conduction. while the real part represents the propagating aspect of the wave. the radiation is absorbed in a thin shell on the outside of the quarter. this eﬀect can be useful in the machine shop or foundry for heat-treating or melting metals. we saw that it is precisely these induced currents that damp the electromagnetic wave after it penetrates the surface. for a good conductor. (+) (+) (9. this goes into the metal as heat. something must be taking up that dissipated energy.72) (9.

. where you can see the heat ﬂowing in from the outside of the quarter to the inside. Fresnel Relations Here is a movie of the heating process.128 Chapter 9.

n1 is the refractive index of the dielectric (from which the wave is incident). Here is a visualization of wave reﬂection from a dielectric-conductor interface. but now the transmitted wave propagates away from the interface rather than along it.9. we ﬁnd that the skin depth is δ= λ0 λ0 = = 27. This situation is better treated with the matrix formalism for multilayer ﬁlms that we will get to shortly. Putting in numbers.8 nm 2πnκ 28. 82nd ed. P. 2001). David R. though. Note that this is even a bit more complicated than before.0%.3 Fresnel Relations It is worth reiterating: with the conductor present. The speciﬁc indices used are vacuum (n1 = 1) on the left and slightly conductive glass (˜ 2 = 1.” in CRC Handbook of Chemistry and Physics. except that the refractive index is now complex.27 + 4. Boca Raton. This is because the wave penetrates a short distance into the metal. since from Snell’s law. Weaver and H. Lide. “Optical Properties of Selected Elements. the Fresnel reﬂection coeﬃcients still hold with the complex refractive index: n1 cos θi − n2 cos θt ˜ n1 cos θi + n2 cos θt ˜ n1 cos θt − n2 cos θi ˜ rP = . everything is the same as in the dielectric case. H. θt is a complex number too! We won’t go into much detail regarding the wave in the medium beyond what we’ve already done. (9. Frederikse. In particular. but some of ˜ the incident energy is lost in the metal. Ed. 12-133. R. Let’s plug in some numbers for the simple example case of a polished silver mirror at normal incidence.75) n1 + n2 ˜ For silver at 780 nm.47i. and n2 is the complex ˜ refractive index of the conductor.52 + 0.2 n = 0. n1 cos θt + n2 cos θi ˜ rS = (9. since we are usually only interested in this wave if it is then transmitted back into free space where it can be detected. . p. Note n that the wave damping is similar to the case of internal reﬂection.1 (9. 2 J.152i) on the right..74) Here. (CRC Press.5. Thus silver is a good reﬂector. The reﬂectance is 2 n1 − n2 ˜ R= .76) at 780 nm. which give R = 95.5 Reflection at a Dielectric-Conductor Interface 129 9.

Assuming an index of refraction n = 1. We derived the Fresnel relations for S-polarized light giving the reﬂection coeﬃcient rS = and transmission coeﬃcient tS = n1 cos θi − n2 cos θt n1 cos θi + n2 cos θt 2n1 cos θi n1 cos θi + n2 cos θt of the electric ﬁeld at a dielectric interface.2.130 Chapter 9. Problem 9. calculate the round-trip time.5. Model a typical diode laser as a crystal (n = 3.5) of length 300 µm. the output facet is typically antireﬂection coated to couple out more of the light in the cavity. (a) For this resonator. surrounded by air (n = 1).60i) and copper (˜ = 0. For low-power diodes (in the 5-15 mW range). One way of making a cheap but good polarizer (except for wavefront distortion) is to use a stack of microscope slides. (a) Show that in the limit of low frequency. Fresnel Relations 9.52) for both S.1. the change in the band gap is more important.) Problem 9. around 780 nm) are made from GaAlAs crystals. n n Problem 9. Diode lasers in the near infrared (say. Use the setup from the text to derive the Fresnel relations for the reﬂection coeﬃcient.24 + 4. the skin depth for a good (nonmagnetic) conductor can be written 2 . Compute the reﬂectance and skin depth for light at normal incidence from vacuum at wavelength 780 nm for polished surfaces of gold (˜ = 0. the free spectral range. Plot the intensity reﬂection coeﬃcients as a function of incident angle for light incident from air (n = 1) onto crown glass (n = 1.80i). the crystal ends are cleaved to form ﬂat surfaces that act as the two ﬂat reﬂectors of the laser cavity. Problem 9.and P-polarized light. causing much larger average tuning rates as the laser “hops” between longitudinal modes.52 and that the slides are all oriented at Brewster’s angle with respect to a randomly polarized laser beam. (In practice this tuning rate is only valid for small temperature changes. n1 cos θt − n2 cos θi rP = n1 cos θt + n2 cos θi and transmission coeﬃcient tP = for P-polarized light.6.08 + 4. how many slides does it take to attenuate one polarization by a factor of 10−4 in intensity compared to the other? Keep in mind that both sides of each slide are at Brewster’s angle.) (b) Assuming that the diode laser frequency is that of a single longitudinal mode. δ≈ µ0 ωσ 2n1 cos θi n1 cos θt + n2 cos θi . Problem 9. and the frequency width (FWHM) of the modes. Ignore multiple reﬂections in your calculation.4.6 Exercises Problem 9. Repeat for light incident from within the glass. Assume a nominal operating (vacuum) wavelength of 780 nm and a coeﬃcient of thermal expansion (∆L/L) for GaAlAs of 7 × 10−6 /◦ C.3. the photon lifetime. for larger changes. calculate the change in laser frequency per degree Celsius for a temperature change. (Note that for higher power diodes. the ﬁnesse.

Assume that the solenoid length is much larger than all other length scales in this setup.9. just reduce your answer to a reasonably simple fraction or compute the very approximate right answer.) (c) Suppose you have a disc of titanium the size and shape of a quarter. (Use one if you like. but you do not need a calculator for this problem.6 Exercises 131 (b) Evaluate the skin depth at 200 kHz for titanium (σ = 2 × 106 (m · Ω)−1 ). . and it lies in the plane orthogonal to the coil axis. The disc is in the center of a solenoidal coil. What is the distribution of power absorption in the disc? Explain. Recall that µ0 = 4π × 10−7 N/A2 . which is driven at 200 kHz.

.

g. but of course we can also handle conductive ﬁlms simply by using complex refractive indices. rb . For right-to-left incidence.1 Reﬂection-Summation Model We will now consider the multiple reﬂections due to an input wave E0i at a single thin ﬁlm. After the ﬁrst term.Chapter 10 Thin Films Now with reﬂections from single interfaces under our belt. 10. We will start out with a simple treatment of a single-layer ﬁlm in the style of a Fabry–Perot cavity. the coeﬃcients are ra . k1 is the wave vector in medium of index n1 ). and t′ . n1 (the coating). Then the total reﬂected ﬁeld is the sum of all the individual reﬂected waves. ta . a pattern emerges because each successive reﬂection is the same as the previous reﬂection but with one more . We are implicitly considering only a single input polarization (S or P). t′ . rb . we will use the same subscripts to denote waves inside these media (e. We will be implicitly assuming dielectric thin ﬁlms. (+) ni n¡ ns tb ræa rb ta E0i (+) tæa rb ta E0i (+) ræa rb ta E0i (+) (+) tb ta E0i (+) ra E0i rb ta E0i (+) ta E0i (+) E0i (+) a b The convention is that ra . and tb are the reﬂection and transmission coeﬃencients at interfaces a and ′ ′ b for left-to-right incidence.. but the analysis works for either one separately. The interfaces a b a and b divide the media with refractive indices ni (for incident medium). we can tackle multiple interfaces in the form of thin ﬁlms. and then rederive the result in a much more powerful matrix formalism. and ns (the substrate or transmission medium).

3) (10. f¡ qi fº qi q1 h d As shown in the diagram.5) and Thus. 1−x (10. Thin Films round trip inside the thin ﬁlm. (10. φ = 2k1 d cos θ1 .7) ∞ n=0 xn = 1 . Thus. The phase diﬀerence between successive reﬂections is the phase of the ray after one round trip inside the thin ﬁlm as it exits (call this phase φ1 ). (10.8) which is valid provided |x| < 1. The ﬁrst term vanishes by Snell’s law. φ is the propagation phase accumulated between seccussive reﬂections as measured along the same wave front. φ = φ1 − φ0 . the correct form for φ follows from the following argument.9) . (10.2) where φ1 = 2k1 · r1 = 2k1 sin θ1 h + 2k1 cos θ1 d.6) Now we can write out the explicit sum for the reﬂected wave by completing the pattern from Eq. with the result rﬁlm := E0r (+) (+) E0i = ra + t′ rb ta eiφ a . ′ E0r = ra E0i + t′ rb ta E0i eiφ + t′ rb ra rb ta E0i ei2φ + · · · . a a (+) (+) (+) (+) (10. ′ 1 − rb ra eiφ (10.1) Here. φ = φ1 − φ0 = 2h(k1 sin θ1 − ki sin θi ) + 2k1 d cos θ1 . That’s how displaced plane waves interfere: the relative phase of the parts that align along the same wave front (normal to the ray) are what deterimine the interference.4) (10. so we are left with (10. φ0 = k0 · r0 = 2ki sin θi h.1): E0r = E0i We can evaluate this using the formula (+) (+) ra + t′ rb ta eiφ a ∞ n=0 ′ rb ra eiφ n . This is a somewhat subtle point that requires a careful geometric construction. but we need to subtract the propagation of the directly reﬂected ray to the same perpendicular position along the ray (call this second phase φ0 ). (10.134 Chapter 10.

14) Comparing this to the resonance condition for a planar-mirror Fabry-Perot cavity [Eq.1 Example: Single Glass Plate as a Fabry–Perot Etalon To get a better feeling for the meaning behind Eq. let’s look at the special case of ni = ns . (10. or k1 cos θ1 = (10. a glass window can act as a planar Fabry–Perot etalon.10. we ﬁnally come to rﬁlm = which is the main result of this section. kq = πq . d (10. this could be a glass plate surrounded by air. just the same as the incident angle.1 Reflection-Summation Model 135 2 ′ Using the Stokes relations t′ ta = 1 − ra and ra = −ra .15) we see that we have just rederived the Fabry–Perot resonance condition.11). d (10. but now generalized for non-normal incidence.11) 10. (10.17) . ra = −rb . The transmission angle is. we can tune the etalon by changing its orientation with respect to the incident wave.12) (10.1.2)]. with a the result 2 (1 − ra )rb eiφ rﬁlm = ra + . (7. by two applications of Snell’s law. πra 2 1 − ra (10. As we mentioned in Chapter 7.13) Notice that Rﬁlm drops to zero when cos φ = 1. Even though the window thickness is nominally ﬁxed (neglecting thermal expansion and mechanical stress). 4 2 1 + ra − 2ra cos φ πq . we can eliminate the transmission coeﬃcients.10) 1 + ra rb eiφ Rewriting all this as a single fraction. In this case. For example.16) 1 2F π 2 . sin (φ/2) 2 is the ﬁnesse of the etalon. so that rﬁlm = and the reﬂectance becomes Rﬁlm = |rﬁlm |2 = ra (1 − eiφ ) . 2 1 − ra eiφ 2ra (1 − cos φ) . This happens when φ = 2πq for some integer q. 1 + ra rb eiφ (10. ra + rb eiφ . The transmission of the plate is then Tﬁlm = 1 − Rﬁlm = = = 1+ where F = 2 (1 − ra )2 4 − 2r2 cos φ 1 + ra a 2 (1 − ra )2 2 2 (1 − ra )2 + 4ra sin2 (φ/2) (10.

the ﬁnesse depends implicitly on the incident angle via the angle-dependence of ra . The transverse component of the electric ﬁeld (E · z ) must ˆ be continuous across each boundary: E0i + E0r = E0t1 + E0r1 =: Ea (interface a) E0t1 eiδ + E0r e−iδ = E0t =: Eb (interface b) Here.19) This is the same setup as for reﬂection at a single interface. we apply boundary conditions. as in Chapter 9.136 Chapter 10.2 Thin Films: Matrix Formalism Now we will rederive the results from the last section by matching boundary conditions.21) is the phase change traveling directly across the interface. so that the transmitted ﬁeld has zero phase at boundary b. As before. δ := (kt1 · x)d = ˆ 2π n1 d cos θt1 λ0 (10. Thin Films We can think of an arbitrary thin ﬁlm as an etalon. and later modify the results to handle P-polarization. . ni H E (+) 0r (+) 0r n¡ ns H E0r1 E0t1 (+) (+) (+) (+) 0r1 E0t qt (+) H(+) 0t E0i qi qt1 H(+) 0t1 z y x H (+) 0i (xoo=oo0) We are assuming ﬁelds as follows: a (xoo=ood) b Region 0: E0i eiki ·r + E0r eikr ·r Region 1: E0t1 eikt1 ·r + E0r1 eikr1 ·r x Region 2: E0t eikt ·(r−dˆ) . except for the addition of left. We will assume S-polarization. F = 1 − ra rb (10.18) Of course. with ﬁnesse √ π ra rb . The transmitted angle is just given in terms of the incident angle by ni sin θi = ns sin θt . (10. The setup is shown in the diagram.and rightpropagating ﬁelds inside the ﬁlm.20) (10. 10. Notice also the phase oﬀset in the transmitted ﬁeld.

Ea = E0t1 + E0r1 = cos δ Eb − i sin δ Hb α1 (10.22) H0t1 cos θt1 eiδ + H0r cos θt1 e−iδ = H0t cos θt =: Hb (interface b) We can rewrite the magnetic ﬁelds as αi (E0i − E0r ) = α1 (E0t1 − E0r1 ) = Ha (10. (10. The constants for the incident. this works for multiple ﬁlms via matrix multiplication: E0 H0 Here.24) α1 := cos θt1 η0 ns cos θt . α1 Eb + Hb −iδ E0t1 = e 2α1 (10. As usual for matrix formalisms. the transverse component of the magnetic ﬁeld (H · y ) must be continuous across the boundˆ H0i cos θi − H0r cos θi = H0t1 cos θt1 − H0r1 cos θt1 =: Ha (interface a) (10.10. Fj = cos δj −iαj sin δj −i sin δj αj . cos δj (10. ﬁlm.23) α1 (E0t1 eiδ + E0r e−iδ ) = αs E0t = Hb .26) Ha = α1 (E0t1 − E0r1 ) = −iα1 sin δ Eb + cos δ Hb . αs := η0 where as usual. where the media are characterized by the α constants. We can conveniently rewrite this in matrix form: Ea Ha = cos δ −iα1 sin δ −i sin δ α1 cos δ A C B D Eb Hb =: F Eb Hb . E0r1 = 2α1 Using the deﬁnitions of Ea and Ha . Our goal is now to ﬁnd a relation between the ﬁelds at interfaces a and b.2 Thin Films: Matrix Formalism 137 ary: Similarly.27) The matrix F is the thin-ﬁlm transfer matrix.30) = F1 F2 F3 · · · FN EN HN . (10.25) α1 Eb − Hb iδ e . η0 is the vacuum wave impedance. Adding and subtracting the expressions for Eb and Hb .29) . and has the general form F= . and substrate media are ni cos θi αi := η0 n1 (10. (10.28) The matrix “transfers” the ﬁelds from interface a to interface b.

34) This is equivalent to the system of equations 1 + r = At + Bαs t α0 (1 − r) = C + Dα2 t.33) nº n¡ n™ n£ nN ns E0 H0 EN HN Now we can solve for the reﬂected and transmitted ﬁelds. η0 δj = and the transmission angles are given by θtj = sin−1 ni sin θi .31) (10.138 Chapter 10. Thin Films 2π nj dj cos θtj λ0 (dj is the thickness of the jth layer). the geometry coeﬃcients are nj αj = cos θtj . Using the deﬁnitions Ea = E0i + E0r Eb = E0t Ha = αi (E0i − E0r ) Hb = αs E0t in the matrix equation (10.27). we ﬁnd E0i + E0r αi (E0i − E0r ) = A C B D E0t αs E0t .35) (10. we ﬁnd the ﬁlm reﬂection and transmission coeﬃcients αi A + αi αs B − C − αs D rﬁlm = αi A + αi αs B + C + αs D (10.36) where as usual r := E0r /E0i and t := E0t /E0i .32) (10. nj where the phase shifts are (10. Solving these equations.37) 2α0 tﬁlm = α0 A + α0 αs B + C + αs D . (10. (10.

37) are still valid.10.39) . We can then calculate the reﬂectance (intensity reﬂection coeﬃcient) of the ﬁlm as αi = R = |r|2 = 2 2 n1 (ni − ns )2 cos2 φ + (ni ns − n1 )2 sin2 φ . η0 cos θtj (S-polarization/TE polarization) (10. This thickness leads to a phase shift of φ= π 2π n1 d cos θt1 = . (10.43) where it is important to note that the layer is a quarter wavelength for the wavelength inside the medium. The transfer matrix is just the single-layer matrix from Eq. the general idea is simple: compute the αj . For the two polarizations. n1 is the index of the coating and ns is the index of the substrate (say. −iα1 sin φ cos φ Putting the matrix elements into the expression from Eqs. air). α1 = .30) −i sin φ cos φ α1 F= (10. with a slight reinterpretation.38) (P-polarization/TM polarization) Even though the formulae get a bit complicated. we must calculate the geometric coeﬃcients αj diﬀerently. 10.41) η0 η0 η0 and ni is the refractive index of the incidence material (say. = 4 4n1 (10. 2 (n + n )2 cos2 φ + (n n + n 2 )2 sin2 φ n1 i s i s 1 where for normal incidence. (10. but we’ll now make it simpler. The thickness is d= λ λ0 .1 Single-Layer Antireﬂection Coating Consider a single-layer ﬁlm at normal incidence. (10. An important but simple case is the quarter-wave ﬁlm (still at normal incidence). αs = .3 Optical Coating Design 139 in terms of the transfer-matrix elements. we ﬁnd αi αs + iα1 sin φ − αs cos φ αi cos φ − i α1 r= αi αs αi cos φ − i − iα1 sin φ + αs cos φ (10. (10.44) . 10.3. It turns out that for P-polarization. and then compute the reﬂection and transmission coeﬃcients.40) α1 2 n1 (ni − ns ) cos φ − i(ni ns − n1 ) sin φ = . (10.27). (10. λ0 2 (10. compute the transfer matrix for the whole ﬁlm.3 Optical Coating Design Now we’ll consider a few of the simplest optical coating designs in common use.37) for the ﬁlm reﬂection coeﬃcient. the coeﬃcients in Eqs. glass). we have αj = nj cos θtj η0 nj αj = . 2 n1 (ni + ns ) cos φ − i(ni ns + n1 ) sin φ n1 ns ni . The above analysis assumed S-polarization. The only modiﬁcation is that in these relations and in the transfer matrix in Eq.42) This is still a bit messy.

The closest material suitable for making an optical coating (with good optical properties and durability) is MgF2 .3% but not really that close to zero. we ﬁnd R = 1.140 Chapter 10.2 Two-Layer Antireﬂection Coating The basic problem with single-layer antireﬂection coatings is that there is a limited choice of suitable coating materials. then the antireﬂection condition (10.48) The reﬂectance is thus . (10.37) for the ﬁlm reﬂection coeﬃcient. exactly one of the reﬂections inside the ﬁlm will produce a π phase shift. which was for the case ni = ns . For example. which has n = 1. This leads to a reﬂectance of R= 2 ni ns − n1 2 ni ns + n1 2 . The cheapest optical coatings in the visible are single-layer MgF2 . Ideally. 10.50) .52 for optical crown glass. putting the matrix elements into the expression from Eqs. Thus.49) r = αi α2 α1 αs 2 2 n2 ni + n1 ns − − α1 α2 R= 2 2 n2 ni − n1 ns 2n + n 2n n2 i 1 s 2 . single-layer antireﬂection (AR) coating. which is much better than the uncoated reﬂectance of R = 4.45) Notice that the reﬂectance vanishes if the ﬁlm index is chosen such that n1 = √ ni ns . This is diﬀerent from the etalon that we considered early on in this chapter. (10. in the hope that an extra free parameter will make the choice of materials more ﬂexible.3.46) As an aside. which requires a thickness of λ/2 for R to vanish. (10. The reﬂectance vanishes even though there is only λ/2 of propagation over one full round trip. The matrix for a single quarterwave layer is F1 = cos φ −iα1 sin φ −i sin φ 0 α1 = cos φ −iα1 − i α1 . (10. (10. we ﬁnd α1 αs αi α2 + − 2 n 2 ni − n1 ns α1 α2 = 2 . we have ni = 1 and ns = 1. let’s quickly compare this to the Fabry–Perot etalon discussion from the ﬁrst part of this chapter. Let’s assume a stack of two quarter-wave ﬁlms at normal incidence. (10. if ni < n1 .46) implies that n1 < ns (and n1 > ns in the case that ni > n1 ). 0 (10. but now we have to be careful about phase changes on reﬂection. Thin Films which means that cos φ = 0 and sin φ = 1. n1 = 1.38 (for visible wavelengths).23 for an AR coating (where R vanishes). playing the role of an extra λ/2 of propagation distance. Let’s now consider the example of a real.3%. One solution to this problem is to add a second layer. Putting this into the formula (10. Normally. a Fabry–Perot cavity requires an integer multiple of λ of propagation distance per round trip.47) The transfer matrix for the two-layer stack is the product of two of these matrices: i α i 0 − 2 0 − 0 − α1 α2 = α1 F = F1 F2 = α1 −iα1 0 −iα2 0 0 − α2 Again. and so it is diﬃcult to match the optimum index condition very precisely.45). For an air–glass interface.

but we won’t go into that here.52). As an example. a value that is pretty close to the ideal value.65). the idea behind making a high reﬂectance (HR) coating is to essentially reverse the order of the layers in the antireﬂection coating: n2 must have a larger index than n1 for good reﬂection. 10.65)]. the ideal AR ratio is n2 /n1 = 1.3. say. The transfer matrix for one of the double layers at normal incidence is n2 − 0 n1 (10. it is possible to get higher reﬂectances by choosing other materials with a greater index contrast. which gives the AR condition n2 = n1 ns . How can we get a reﬂectance approaching unity? The general idea is to stack many of these double layers to form a quarter-wave stack. the reﬂectance turns out to be R = 18%. But these improvements will be marginal. we note that if the ratio n2 /n1 is very diﬀerent from ns /ni .3 Optical Coating Design 141 2 2 The reﬂectance vanishes if n2 ni = n1 ns .3 High Reﬂector: Quarter-Wave Stack From the analysis of the two-layer AR coating.23.1% for this coating (about as good as you can expect given fabrication tolerances). As an example. a good metallic coating. again for crown glass (ni = 1 and ns = 1. Thus. ni (10. Consider N double layers. In this case. 10 nm or so) of good antireﬂection for a coating in the visible.1) and CeF3 (n1 = 1.1) and CeF3 (n1 = 1. We can choose ZrO2 (n2 = 2.52) F= .10.51) allowing us to pick the ratio of two refractive indices rather than a single absolute index. Re fle ctance lº It is possible to make a more broadband AR coating by adding a third layer. which gives n2 /n1 = 1. but reverse the order of the layers. Of course. although it is not as high as. we ﬁnd R = 0. n1 0 − n2 . let’s take the same materials as for the AR coating [ZrO2 (n2 = 2. then the reﬂectance is relatively large. This kind of coating is called a “vee” coating due to the shape of the reﬂectance plot as a function of wavelength: there is a relatively narrow range (say. Working out the reﬂectance. This is a much higher reﬂectance than the AR coating.27.

The periodic structure of these quarter-wave stacks gives a simple example of a photonic crystal or Bragg reﬂector.) Assume n = 1. +1 (10. We could use fewer layers with a better choice of materials (higher index contrast).52 and n2 /n1 = 1/1.55) (10. Thin Films Thus.54) +1 For the same materials as above. the reﬂectance is FN = n2 n1 N 0 n1 − n2 N . where the design wavelength is λ = 550 nm (in vacuum). Plot the intensity reﬂection coeﬃcients as a function of angle for light incident from air onto crown glass with a single-layer antireﬂection coating of MgF2 . (Make plots for both polarizations. so for N =2: N =3: N =4: N =8: N = 12 : R = 36% R = 53% R = 68% R = 94% R = 99%.142 Chapter 10. ni ns R= ni ns n2 − n1 n2 − n1 2N 2N − 1 . 10.65). although we should note that the intermediate reﬂectances shown here are also useful as beam splitters.38 for MgF2 . Assume a design wavelength of 550 nm (in vacuum) and extend the plot over the visible spectrum (400-700 nm). followed by a λ/4 layer of CeF3 (n = 1.56) Thus. How thick is the coating in nm? Problem 10.1. (10. and a coating thickness of λ/4 (where λ is the wavelength inside the coating!).4 Exercises Problem 10.53) r= − ns − N n1 n2 N n1 − n2 − ns n2 − n1 N = ni ns ni ns 2 − −1 2N n2 − n1 . The thin-ﬁlm stack consists of a λ/4 layer of ZrO2 (n = 2.2. the transfer matrix for the whole stack is n2 − n1 0 N The reﬂection coeﬃcient is ni − ni Thus. How thick are the layers in nm? . ni /ns = 1/1. Plot the intensity reﬂection coeﬃcients as a function of λ for light incident from air onto crown glass with a double layer antireﬂection coating. we see that as the number of double layers increases.1) directly on top of the crown glass. the reﬂectance rapidly converges to unity. n2 n1 2N (10.27.

1) The basis function was exp(−iωt). Roughly speaking. we will soon be dealing with Fourier transforms between space and spatial frequency. So before introducing the convolution. ˜ f (ω) = ∞ −∞ f (t)eiωt dt.3) 11. ˜ f (k)eikx dk. in Chapter 3.4) The interpretation of this integral is as follows: as a function of x. Recall that the formulae for the Fourier transform and the inverse transform are f (t) = 1 2π ∞ −∞ ˜ f (ω)e−iωt dω. However.2) Thus. (11.Chapter 11 Fourier Analysis II: Convolution 11. consider the perfectly localized convolution kernel g(x) = δ(x). 1 2π ∞ −∞ ω −→ k. we can deﬁne the convolution f ∗ g of two functions f and g as the integral (f ∗ g)(x) := ∞ −∞ f (x′ )g(x − x′ ) dx′ . we’ll make a short digression to introduce the diﬀerent convention. The convolution with a function f (x) is (f ∗ δ)(x) = ∞ −∞ f (x′ )δ(x − x′ ) dx′ = f (x). we displace (scan the location) of g(x′ ). ˜ f (k) = f (x)e−ikx dx. So to get the spatial convention from the temporal convention. and this determines the form of the transform integrals. (11.1 Spatial Fourier Transforms Previously.2 Convolution The convolution is a mathematical operation that maps a pair of functions to a function. Notice that this is proportional to the projection of f (x′ ) onto g(x − x′ ). It is most intuitive to consider the (common) case of a localized function g(x′ ). Recall that the right-going plane wave has the form exp[i(kx − ωt)]. which we will refer to as the convolution kernel. As an example. (11. we’ve been dealing with time-frequency Fourier transforms.5) . i −→ −i. (11. we must make the replacements t −→ x. the convolution is an operation that “smears” one function with another. the transform equations for spatial functions are f (x) = ˜ where again f (k) = F [f (x)]. ∞ −∞ (11. Formally. so the basis harmonic function for the spatial part is exp(ikx). Then the value of the convolution at location x is the “amount” of f (x′ ) that “passes through” g(x′ ) when g is displaced by an amount x.

144 Chapter 11. both in an intuitive and a computational sense. continued convolutions will make the distribution look Gaussian. we’ll just compute the explicit form of F [f ∗ g].) We see now the “smoothing” or “blurring” eﬀect of the convolution.1 Example: Convolution of Box Functions As a slightly more complicated example.9) To prove this. both given by f (x) = g(x) = 1. then (f ∗ g)(x) = ∞ −∞ (11.8) The convolution consists of displacing g(x′ ) by x. but the convolution is continuous. |x| > a. in that we’ll just assume the functions are nice enough that all the integrals . The convolution is also wider than the original functions. 11.4). there is no “smearing” due to a perfectly localized function. we will use centered kernels. (11. so the convolution is a triangle function. if g(x) = δ(x − x0 ). f(x) (fo*ofo)(x) -o2a 2a x (Note that f ∗ f in the ﬁgure is the same as f ∗ g for this special case of f = g. This will be very much a physicist’s proof. and integrating. the convolution (product) just turns out to be the area where the two functions overlap. Typically. For this simple case (box functions of unit height). |x| > 2a. Otherwise. The original functions were discontinuous. g(xoo-o x¢) (fo*og)(x) f(x¢) x (-oa) x¢o=oo0 a x¢ When the displacement is large. not a mathematician’s proof.2. |x| ≤ a. 0. the boxes don’t overlap at all. Fourier Analysis II: Convolution The eﬀect of convolution with a δ-function is simply to do nothing. consider the convolution of box functions. multiplying the functions together. For example. the overlap area varies linearly with the displacement.6) f (x′ )δ(x − x0 − x′ ) dx′ = f (x − x0 ). the eﬀect of a displaced kernel is simply to displace the convolution by the same amount. so the convolution is zero. (11. (11. The convolution theorem states that the Fourier transform of the convolution is the product of the Fourier transforms of the individual functions: F [f ∗ g] = F [f ]F [g].7) 11. (11. As we will see.3 Convolution Theorem The convolution theorem gives an easy way to evaluate the convolution integral in Eq.

Thus.11. (11. (11. ˜ (11.16) Recall that the standard (normalized) form of the Gaussian is 1 (x − µ)2 √ exp − 2σ 2 2π σ . let’s use the convolution theorem to convolve two Gaussians.11) = F [f ]F [g]. then compute the inverse Fourier transform. The standard deviation is one measure of the width of a Gaussian function. g(x) has standard deviation β. ′ ∞ −∞ ∞ ∞ f (x′ )g(x − x′ ) dx′ f (x′ )g(x − x′ ) dx′ e−ikx dx f (x′ )e−ikx g(x − x′ ) e−ik(x−x ) dx dx′ . Thus. we invert the Fourier transform to obtain the convolution: (f ∗ g)(x) = F −1 ABαβπe−(α 2 +β 2 )k2 /4 = √ ABαβ π α2 + β 2 exp − x2 α2 + β 2 . multiply them together. but the blurring eﬀect of the convolution makes the convolved Gaussian wider than the original functions. (11. we can write f ∗ g = F −1 {F [f ]F [g]} . and (f ∗ g)(x) has standard deviation α2 + β 2 . just follow this recipe: Fourier transform both functions. ∞ −∞ ′ ′ (11.10) −∞ ∞ −∞ −∞ ∞ −∞ ∞ −∞ ∞ −∞ F [f ∗ g] = = f (x′ ) e−ikx g(x) e−ikx dx dx′ ′ ′ f (x′ ) e−ikx dx′ ∞ −∞ g(x) e−ikx dx (11.13) We computed the Fourier transform of a Gaussian in Chapter 3. the convolution of Gaussians is still Gaussian.12) 11. Note that f (x) has standard deviation α.1 Example: Convolution of Two Gaussians Since it’s easy to compute the Fourier transform of Gaussian distributions. which we’ll come back to shortly. . Let’s write the two functions as f (x) = Ae−x 2 /α2 . Then the product of the Fourier transforms is (F [f ]F [g]) (k) = ABαβπe−(α 2 √ 2 2 F [g](k) = g(k) = Bβ πe−β k /4 . to convolve two functions. so we can write √ 2 2 ˜ F [f ](k) = f (k) = Aα πe−α k /4 . (11.3.14) +β 2 )k2 /4 . g(x) = Be−x 2 /β 2 . Mathematically.15) Finally.17) where the µ is the mean and σ is the standard deviation (σ 2 is the variance). F [f ∗ g] = F = = Letting x −→ x + x . (11.3 Convolution Theorem 145 simply exist.

f+ (x) = ∞ −∞ ∞ −∞ f1 (x′ )f2 (x′′ )δ(x′ + x′′ − x) dx′ dx′′ .146 Chapter 11.22) Var[X] := (X − X ) g(x)f (x) dx. The variance here is deﬁned by for a random variable X. from Eqs. the probability that X1. x′ . when you add the random variables. But then we must sum over all such pairs. The probability that both X1 and X2 will both have particular probabilities is the product of the individual probabilities since the variables are independent. The mathematical problem is as follows: let X1 and X2 be independent random variables with probability density functions f1 (x) and f2 (x). (11. Recalling that the standard deviation is the square root of the variance. the standard deviation adds in quadrature: σX1 +X2 = 2 2 σX1 + σX2 (11.18) We can translate this statement in terms of the probability densities and implement the constraint as a δ-function. (11.2 is between x and x + dx is f1.2 (x) dx. then the uncertainty of the sum is σ1 + σ2 . the variance turns out to be σ 2 .20) (11. σX := Var[X].4 Application: Error Analysis One important application of the convolution is in probability theory. Prob(X1 + X2 between x and x + dx) = Prob(X1 between x′ and x′ + dx′ ) × Prob(X2 between x′′ and x′′ + dx′′ |x = x′ + x′′ ). Fourier Analysis II: Convolution 11. For the Gaussian probability density in the form of Eq. Then we can ask. (11.x′′ (11.24) we can see that for Gaussians.16) and (11. . That is.23) where h(x) is an arbitrary function and f (x) is the probability density function of X. we can note that X1 + X2 = x for any pair of values of X1 and X2 that happen to add up to x.21) (11. (11.19) Evaluating the x′′ integral.25) This is the law of error propagation for addition in error analysis: if two independent measurements have 2 2 uncertainties σ1 and σ2 . what is the probability density of X1 + X2 ? To answer this. Thus. Letting f+ (x) denote the probability density of X1 + X2 . The angle brackets denote the expectation value. (11. we see that the probability density of the sum is the convolution of the individual densities: ∞ f+ (x) = Note that this result is general in that it doesn’t assume any particular form for f1 (x) or f2 (x). which carries over to analysis of statistical error of experimental data in physics.17) we can see that the variances are related by Var[X1 + X2 ] = Var[X1 ] + Var[X2 ]. In the special case where both distributions are Gaussian. expressing what we said in equation form. respectively. h(X) := ∞ −∞ −∞ f1 (x′ )f2 (x − x′ ) dx′ = (f1 ∗ f2 )(x).17). (11.

Körner. Let X1 . scaled statistic SN − N µ √ (11. (11. Deﬁning the sum by N SN := j=1 Xj . XN be independent. So let’s clean this up a bit. with the mean µ = Xj and variance σ 2 = Var[Xj ]. Var[SN ] −→ N σ 2 . identically distributed random variables. so let’s recompute the expansion in Eq.32) is an important tool for the manipulation of probabilities.20) to obtain (11. given by ˜ f (−k) = eikX . what is the probability density fSN (x) of SN ? Evidently. we make many independent measurements of some quantity. Fortunately. This is more cumbersome than necessary. we need to employ the convolution theorem.20). (11. (11.29) ZN := σ N converges to the “standard normal” (Gaussian) distribution 2 1 fZN (x) −→ √ e−x /2 . since “becomes asymptotically Gaussian” is an imprecise statement. 1988). (11. To evaluate the convolutions in Eq. we Taylor-expanded e−ikx and then used the fact that the terms of the expansion were proportional to expectation values X j . Taking the Fourier transform of f (x).11. The central limit theorem states that. However. note that in probability theory the characteristic function of a probability density. This is also the reason that the Gaussian probability distribution is so important. j! 2 (11. (11.31) for the centered. we can iterate Eq. . we might not know the underlying distribution. (11. W. Let f (x) be the probability density function of each of the Xj . starting on p.26) (The mean is an exact result. however. provided that the mean and variance of the Xj exist.31) Here. 2π where the result is the successive convolution of N copies of f (for N − 1 total convolution operations). Fourier Analysis (Cambridge. Let’s prove this now1 .5 Application: Central Limit Theorem 147 11. . 349. the rigorous version is in T. .) This is a rough statement. .28) we will now ask.5 Application: Central Limit Theorem Often. (11.27). the distribution fSN (x) becomes asymptotically Gaussian for large N with SN = N µ. (11. In particular. there is a very important result that often makes it unnecessary to know the underlying distribution. whereas the distribution and variance are valid for large N .30) which is the special Gaussian with mean 0 and unit variance. it turns out that this distribution becomes simple for large enough N .27) fSN (x) = (f ∗ f ∗ · · · ∗ f )(x).33) σ N 1 This is the physicist’s proof. in which case it’s diﬃcult to make use of a result like Eq. . (11. The central limit theorem states that the probability density function fZN (x) of the centered. ˜ f (k) = ∞ −∞ f (x)e −ikx dx = ∞ j=0 ∞ −∞ f (x) (−ikx)j k 2 (σ 2 + µ2 ) dx = 1 − ikµ − + O(k 3 ). scaled variable Xj − µ Zj = √ .

(11. (11. Some one-step distributions. with the result k2 ˜ fZ (k) = 1 − +O 2N k √ N 3 .1 Central Limit Theorem Application: Random Walk The central limit theorem has one (of many) of its important applications in statistical mechanics. (11. independent of the one-step distribution. ∆x ∼ D t1/2 .35) As N becomes large. and then use the formula n− →∞ lim 1+ x n n = ex (11. . (11. 11. The probability distribution thus becomes asymptotically Gaussian with a time-dependent width of σ(t) = σ t . the transform becomes k ˜ fZN (k) = exp − 2 2 . so fZN converges to a standard normal distribution as N −→ ∞.148 Chapter 11. what has the walker done? The central limit theorem says that the probability density of the accumulated displacement N SN = j=1 Xj (11. we can neglect the higher order terms beyond the ﬁrst. and the rescaling changes the factor in front of the variance. the width (standard deviation) is σ N . Suppose a random walker takes a random step of size X with probability density f (x) between periodic intervals of duration ∆t. lead to anomalous diﬀusion where the diﬀusion coeﬃcient diverges.40) This random-walk behavior is characteristic of a diﬀusion process.38) After N steps (N large).39) √ is Gaussian with zero mean and variance N σ 2 . such as the Cauchy distribution. we just compute ˜ fZ (k) to the N th power: ˜ ˜ fZN (k) = fZ (k) N = k2 1− +O 2N k √ N 3 N . (11.5. Var[X] = σ 2 . the ﬁnal distribution is Gaussian.34) The convolution theorem says that to calculate the transform of the N -fold convolution. Fourier Analysis II: Convolution with corresponding probability density fZ (x). Note that within certain restrictions (we’ll explore this more in the homework). The centering eﬀectively zeroes the mean. which is a transport process by which the distribution grows as t1/2 .37) √ But now the inverse Fourier transform of exp(−k 2 /2) is exp(−x2 /2)/ 2π. That is. ∆t (11.41) √ where for the random walker the diﬀusion coeﬃcient is D = σ/ ∆t. Let’s assume that X = 0.36) to see that for large N .

11. (11. √ so µN is Gaussian with mean µ and standard deviation σ/ N . ˙ (11. We can see this by solving Eq. so the oscillator is at rest: x(t < t′ ) = 0. the last two terms in the right-hand integral become negligible (being proportional to ε). . by making many measurements. Since the velocity just before t′ is zero.6 Application: Impulse Response and Green’s Functions 149 11.48) across the impulse. There is no force before the impulse (t < t′ ). t′ +ε t′ +ε x dt = ¨ t′ −ε t′ −ε 2 δ(t − t′ ) − γ x − ω0 x dt. This is why. and the second is the experimental error (statistical ﬂuctuation in the sample mean). (11. the solution for t > t′ is the usual damped harmonic oscillator solution with the initial conditions x(t′ ) = 0 and x(t′ ) = 1. such as a damped harmonic oscillator subject to a forcing function f (t): 2 x + γ x + ω0 x = f (t). Applying the central limit theorem.43) where the ﬁrst term represents the true mean. where σ is the standard deviation of a single √ measurement.45) The solution to this is fairly straightforward. it is possible to increase the accuracy of a measured quantity.6 Application: Impulse Response and Green’s Functions Let’s consider a simple mechanical system. ¨ ˙ (11. Thus. ZN is approximately standard normal for large N . and so we can write (without deriving the solution) ˙ x(t > t′ ) = where ωγ := 2 ω0 − (γ/2)2 1 −(γ/2)(t−t′ ) e sin[ωγ (t − t′ )]. ¨ ˙ (11. Thus. we ﬁnd ∆x = 1 ˙ (11. the velocity just after t′ is 1. Thus. . This makes intuitive sense: an impulsive force changes the velocity.2 Central Limit Theorem Application: Standard Deviation of the Mean Returning again to error analysis.45) and integrating over a short interval around t = t′ .50) is the oscillation frequency in the presence of damping. as ε −→ 0. . let’s assume a particular form for the forcing function: an impulse (δ-function) at time t′ : 2 x + γ x + ω0 x = δ(t − t′ ).46) At t = t′ .42) µN := N j=1 We can rewrite this as µN = σZN SN = µ+ √ . . suppose we make independent measurements X1 . .44) Rather than solving this directly for an arbitrary forcing function.47) In the limit as ε −→ 0. but there isn’t yet time for any position change to take place. the impulse “kicks” the oscillator. 11. (11. XN of some quantity in the laboratory.49) (11. ωγ (11. The sample mean is N 1 Xj . the standard deviation of the mean (also called the standard error) is σ/ N . N N (11.5.

55) so that the general solution is the convolution of f (t) with g(t) := g(t.59) .52). to take the Fourier transform. (11.51). t′ ) satisﬁes Eq. 2 (ω0 − ω 2 − iγω) ′ (11. t ≥ t′ . ¨ ˙ (11. ′ (11. since it is essentially just assuming that the time dependence is of the form e−iωt . (11. Just for fun.52) so we can multiply through Eq.54) Thus.45): (11. (11. the solution to the equation in response to a unit impulsive force at t = t′ : 0. This is easy. we just make the following replacements: ∂ −→ −iω. with the result 2 x + γ x + ω0 x = f (t). Fourier Analysis II: Convolution Let’s rewrite this as the impulse-response solution. t′ ) −→ g (ω. t′ ). (11. let’s look at the Fourier transform of the Green’s function. t′ = 0) = ˜ ˜ 1 . t′ ) dt′ .56) 2 (−ω 2 − iγω + ω0 )˜ = eiωt . t′ ) = ˜ eiωt . ˜ g(t. To employ the convolution theorem. we really just wanted the Fourier transform for the impulse at t′ = 0. This works because of the superposition principle and because any function can be represented as a superposition of δ-functions. g (11. we can write down the solution for any forcing function f (t) in terms of an integral. t′ ) is the Green’s function for the damped harmonic oscillator that is initially undisturbed).58) which we see is a Lorentzian function in frequency. t′ ) = g(t − t′ ). ¨ ˙ Then g(t. (11. γ ωγ 2 g + γ g + ω0 g = δ(t − t′ ). (11. 0). This convolution or impulse-response method is an example of a more general method called the Green’s function solution (g(t. 2 (ω0 − ω 2 − iγω) (11. given the impulse-response function g(t.51) We note also the property of the δ-function f (t) = ∞ −∞ f (t′ )δ(t − t′ ) dt′ .57) g (ω. which is essentially the content of Eq.51) by f (t′ ) and integrate over t′ . (11. This is an example of a convolution method for solving linear system. In fact. we can write x(t) = ∞ −∞ f (t′ )g(t − t′ ) dt′ = (f ∗ g)(t). noting that in this particular case that the time dependence is such that g(t. so g (ω) = g(ω. t < t′ g(t. t′ ) = 1 e−(γ/2)(t−t′ ) sin[ω (t − t′ )]. Thus.53) which is just Eq.44) where x(t) = ∞ −∞ f (t′ )g(t. ∂t The resulting equation is Solving for the Fourier transform g . t′ ).150 Chapter 11. ˜ δ(t − t′ ) −→ ′ ∞ −∞ δ(t − t′ )eiωt dt = eiωt . A quick trick for doing so is to take the Fourier transform of Eq. (11.

we can write the functional dependence of Tcav = Tcav (δν − δνd ) on both the input frequency and cavity length changes.66) . ˜ 11. the kernel g(t) acts as a bandpass ﬁlter (centered on ω0 . ˜ f (ω) = F [f (t)].max 2F π 2 sin 2 πν FSR . 2 (ω0 − ω 2 − iγω) (11. δd is the (small) length change while scanning the cavity.63) Here. so the convolution theorem gives an alternate way to write the general solution: 1 ˜ g x(t) = F −1 f (ω)˜(ω) = 2π ∞ −∞ (11. d0 is the nominal cavity length. so it isn’t so apparent how scanning the cavity length and an input spectrum would ﬁt in. (11. So let’s massage things a little. we can get to an application in optics. and δν is a small frequency deviation from the nominal resonance. however. We only treated the Fabry–Perot cavity in the case of monochromatic light.7 Application: Spectral Transmission Finally. we see that in the frequency domain. As a spectrum analyzer.62) This depends only on a single frequency. Deﬁning φ0 := 2πd0 /λ as the overall oﬀset phase due to the nominal cavity length (essentially a constant).11. = λ λ FSR (11. Thus g (ω) is the transfer function for the damped harmonic oscillator. we can deﬁne a centered line-shape function gc (δν) := g(ν0 + δν). (11. (11. Thus.60) ˜ f (ω)e−iωt dω.7 Application: Spectral Transmission 151 We can also assume we know the Fourier transform of the driving function. we will assume that the line shape is sharply peaked about some center frequency ν0 . we want to analyze the transmission with respect to small length changes. with a width controlled by γ) that passes some of the frequencies of the driving function f (t) to produce the output x(t). so we can write πν FSR = π(νq + δν) FSR = πq + πδν FSR 2πd πδν + = λ FSR 2πδd πδν 2πd0 + + . Suppose we have a Fabry–Perot cavity with transmission function Tcav 1+ Tcav.64) where δνd := − λ/2 δd (11. Thus. we can write πν FSR = π(δν − δνd ) FSR FSR + φ0 . so let’s explore this in more detail.61) In this case. If the input spectrum is the “line-shape function” g(ν).65) is the change in the resonance frequency due to the cavity length change. so that we can view the frequency content of an arbitrary input wave. Recall from Chapter 7 that a Fabry–Perot cavity can act as a spectrum analyzer.

8 Exercises Problem 11. Fourier Analysis II: Convolution where δν = ν − ν0 . . As a spectrum analyzer. identically distributed random variables X1 . weighted by the cavity transmission coeﬃcient at the same frequency. . Problem 11. XN with Cauchy (Lorentzian) probability density functions f (x) = (a) Show that the Fourier transform is given by ˜ f (k) = e−|k| . the cavity line shape (for a given ﬁnesse) represents a resolution limit.total (δd) ≈ ∞ −∞ Tcav (δν − δνd )gc (δν) d(δν). since any input spectrum on transmission through the cavity will have an apparent line width of at least δνFWHM due to the convolution. (11. 11. ¨ ˙ 1 . f (x) = 1. Since the latter is periodic in frequency (due to the free spectral range). Include the corresponding asymptotic distribution on each plot. Problem 11. Consider independent. Thus. the total transmission is still periodic.1. the cavity transmission is the convolution of the input spectrum with the cavity transmission spectrum. which is valid if g(ν) is sharply peaked. . The harmonically forced oscillator has the following equation of motion: 2 x + γ x + ω0 x = f0 sin(ωt). (f ∗ f ∗ f )(x). integrated over all frequencies: Tcav. (f ∗ f )(x). Comment on your results. Let f (x) denote the unit box function.152 Chapter 11. (b) Make plots of f (x). but in (b) you have shown that this is not the case.2. Then we can write the total transmission through the cavity as the amplitude gc (δν) at frequency δν. ˜ Hint: it’s much easier to compute the inverse Fourier transform of f (k) and use the fact that the Fourier transform is invertible for reasonable functions.3. and (f ∗ f ∗ f ∗ f )(x). |x| ≤ 1/2 0 elsewhere (a) Let f ∗N (x) denote the convolution of f (x) with itself N − 1 times. but the individual transmission peaks are broadened by the input line shape due to the convolution. (c) A naive application of the central limit theorem leads to the conclusion that µN has a Gaussian probability distribution for large N . π(1 + x2 ) 1 N N Xj j=1 . Write down the asymptotic form for f ∗N (x) in the limit of large N . Explain why there is no contradiction. .67) The approximation here is the extension of the integration to all frequencies. (b) Show that the probability density of the mean µN := is also f (x).

8 Exercises 153 where Use the Green’s function method in the time domain to show that the motion of the driven. b. 2 ω0 − ω 2 You may ﬁnd the following integral formula useful: ∞ 0 e−ax sin(bx) sin(cx + d) dx = b[2ac cos d + (a2 + b2 − c2 ) sin d] (a2 + b2 )2 + 2(a2 − b2 )c2 + c4 (a. damped oscillator (in steady state) is f0 sin(ωt + δ) x(t) = . a > 0). . d real. 2 (ω0 − ω 2 )2 + ω 2 γ 2 tan δ = ωγ .11. c.

.

3) where the wave vector k = (kx . space takes up three dimensions.Chapter 12 Fourier Optics 12.1 Fingerprints of Propagation The electromagnetic wave equation gives us a formal method for propagating an optical wave forward in space. As opposed to time. ky ) ei(kx x+ky y) dkx dky . Let’s let the optical axis lie along the z-axis. the inverse transform is f (x.2 Wave Propagation 12. (12. That’s where the “Fourier” comes into Fourier optics: we use a Fourier transform to break up an arbitrary ﬁeld into plane-wave components. (12. ky . y)]. . Then the wave vector k has a direction determined by its angles with respect to the z-axis. (12. and any ﬁeld can be regarded as a superposition of plane waves. which is much more complicated) solution to the wave equation: (+) (+) E (+) (r) = E0 eik·r = E0 ei(kx x+ky y+kz z) . ky kx . but two is suﬃcient for our purposes here. Let’s see how this works. 12. this turns out to be simple: just take a one-dimensional Fourier transform along each direction. so we need a Fourier transform in multiple dimensions. Essentially.1 Fourier Transforms in Multiple Dimensions Before introducing Fourier optics. we can do this because we know how plane waves propagate.1) f (x. ky ) = F [f (x.2) ˜ where again f (kx . θy = sin−1 . Fourier transforms can be extended to arbitrarily many dimensions. (12. in two dimensions.2. y) e−i(kx x+ky y) dx dy. For example. we must extend the spatial convention for Fourier transforms that we introduced before.4) θx = sin−1 k k where kx and ky are the transverse spatial frequencies. Recall the scalar plane wave (we won’t do any vector theory. kz ) indicates the propagation direction of the wave. y) = and the Fourier transform is ˜ f (kx . ky ) = ∞ −∞ ∞ −∞ 1 (2π)2 ∞ −∞ ∞ −∞ ˜ f (kx . In Cartesian coordinates.

2 Decomposition A general ﬁeld proﬁle E (+) (x. the electric ﬁeld has harmonic spatial dependence in the x. ky ) = ∞ −∞ ∞ −∞ E (+) (x. Often. y)-plane (say. k θy ≈ ky . y) at z = 0. But for now. so to summarize: 1. (12. We know how individual plane waves travel with z. we wish to use this recipe as a calculational tool for understanding diﬀraction. z = 0) = E0 eikx x eiky y . in the z = 0 plane. y) = (2π)2 −∞ −∞ ei(kx x+ky y) −→ ei(kx x+ky y) eikz z .6) 12. 2.9) So to propagate the full ﬁeld forward in z.156 Chapter 12. (+) (12. k (12. (12.11) . y)e−i(kx x+ky y) dx dy.and y-directions: E (+) (x.5) This is the central point of Fourier optics: harmonic phase variation in the (x. Fourier Optics x qy qx k z y That is.4).2. Start with an initial ﬁeld E (+) (x. ky )ei(kx x+ky y) eikz z dkx dky . where kz = 2 2 k 2 − kx − ky . where these relations are simpler. Compute the Fourier transform of the initial ﬁeld: ˜ E (+) (kx .2. we just multiply each plane wave in the superposition by eikz z : E (+) (x. we can write θx ≈ since kx. the harmonic variation is a “ﬁngerprint” of propagation in a certain direction. kx . y) at z = 0 can thus be written as a superposition of plane waves via the Fourier transform: ∞ ∞ 1 ˜ E (+) (kx . ky )ei(kx x+ky y) dkx dky . 12.y ≪ k.3 Fourier-Transform Recipe The theoretical development thus leads us to a fairly simple (in principle) recipe for propagating the ﬁeld forward in space. (12.7) E (+) (x. we will only want to look at these in the paraxial approximation. y. That is. of the harmonic variation determines the direction of propagation. In the paraxial regime. z) = 1 (2π)2 ∞ −∞ ∞ −∞ ˜ E (+) (kx . the frequency. given the form of the ﬁeld in some plane. ky ). (12. at z = 0) corresponds to a plane wave in a particular direction determined by Eqs. y.8) (12. In this sense.10) It is important to keep in mind here the functional dependence kz = kz (kx . (12. a fact which motivates the ﬁeld of holography. We can see that the information in the entire three-dimensional (monochromatic) ﬁeld is thus encoded in any two-dimensional slice.

18) so that eikz z = e−z/δ(kx . However.17) (12. the free-space transfer function becomes eikz z ≈ eikz e−i(kx +ky )z/2k . kx. Finally. we can write 2 2 2 2 kx + ky 1 kx + ky 2 2 =k− .16) In this form. we can ˜ still keep the limits of integration in Eq. (12. (12.2.2. and the general propagation problem isn’t easy (although it is still straightforward on a computer). In fact.12) Again. we have to live with the square-root form of kz (kx . Note that in the paraxial limit. (12. because the components that carry these details rapidly decay . Multiply the transform E (+) (kx .13) is in general diﬃcult to evaluate because of the functional form of kz in the exponent. ky )eikz z .y ≪ k. (12. because kx and ky are assumed to be small compared to k. Recall that through the Fourier transform.13) This recipe works because the superposition principle holds in linear optics and because it is easy to propagate plane waves.4 Paraxial Propagation The integral in Eq. (12. ky ) has negligible “stuﬀ” at large kx. in the paraxial approximation. (12. However.13) a bit more. y). y) is composed of diﬀerent spatial frequencies.15) where the ﬁrst factor is an overall phase factor corresponding to plane-wave propagation along the optical axis. we can still get some insight by examining the propagation equation (12.13).2 Wave Propagation 157 3. ky )eikz z ei(kx x+ky y) dkx dky . ky ) −→ E (+) (kx . the square root is real and so the wave damps exponentially with a length scale 2 2 δ(kx . which may seem strange in view of the form of Eq. broad features in E (+) (x.5 Nonparaxial Propagation and the Diffraction Limit If the paraxial approximation isn’t valid. ky ). ky ) by the free-space transfer function eikz z : ˜ ˜ E (+) (kx . ky ). these high frequencies correspond to evanescent waves and don’t propagate.14) kz = k 2 − kx − ky ≈ k 1 − 2 k2 2k Thus. y) are represented by small spatial frequencies. Very ﬁne details on spatial scales smaller than λ are represented by spatial frequencies kx. z) = 1 (2π)2 ∞ −∞ ∞ −∞ ˜ E (+) (kx . 4.ky ) for the evanescent-wave solutions. 2 2 (12. we need to apply this to the Fourier transform rather than the initial function because kz = kz (kx . the integral greatly simpliﬁes. (12.12. 12. and the second factor generates the evolution of the spatial proﬁle. 12. invert the Fourier transform to ﬁnd the propagated ﬁeld: E (+) (x.13) to ±∞ because we assume that E (+) (kx . ky ) = kx + ky − k 2 −1/2 .y > k.y . E (+) (x. whereas ﬁne details are represented by large spatial frequencies. y. In particular. (12. This eﬀect gives rise to a diﬀraction limit in imaging: propagation over long distances wipes out ﬁne (sub-λ) details in the initial image E (+) (x. as we can see from the form of the propagation factor: √ 2 2 2 2 2 eikz z = e− kx +ky −k z (if kx + ky > k 2 ). Since.

So when is this approximation valid? There are two conditions to be satisﬁed for the Fraunhofer approximation to hold.y . distant screen q≈ E o(x. In any case. a slit or a pinhole) illuminated by a uniform. and then do the mathematical version in the next section. We’ll go through a heuristic derivation of this result here. Often.19) θx. y). ky ) = E (+) (kθx . z) = k ˜ (+) E 2πz kx ky . ky ) = F E (+) (x. and in terms of angle the relative ﬁeld components in terms of the angles θx. It turns out that the Fraunhofer diﬀraction pattern is simple to calculate: with the proper scaling factors.y are given by ˜ E (+) (kθx .2. this is why it is often said that it is impossible to optically resolve objects that are smaller than the optical wavelength λ.21) (12.3 Fraunhofer Diffraction 12. wiping out yet more of the ﬁne detail. kθy ). 12. Noting that in the paraxial approximation. y. Consider the diﬀraction setup shown here. In the paraxial approximation. As we argued before in Section 12.1. We will speak more generally here. ˜ where E (+) (kx . The ﬁrst is that the size of the diﬀraction pattern must be much larger than the size of the initial aperture pattern E (+) (x. y) . y) in the z = 0 plane. typical imaging systems aperture the light to a much smaller solid angle. The Fraunhofer diﬀraction pattern is clearly an approximation to the “exact” pattern obtained by the propagation recipe. and refer to the diﬀraction pattern of an arbitrary initial ﬁeld E (+) (x. (12. For now it is suﬃcient to realize that Eq. Fourier Optics away. however. the “diﬀraction pattern” refers to the far-ﬁeld pattern of an aperture (say. y) breaks up into its Fourier components. such as an expanded laser beam (plane wave) at normal incidence. y) as ˜ ˜ E (+) (kx .oy) x y zoo=o 0 z z (+) The initial ﬁeld E (+) (x. we can write the Fourier transform of E (+) (x.21) gives the Fraunhofer diﬀraction pattern in terms of the Fourier transform. kθy ) . . z z . we can write E (+) (x.y up to k can propagate. (12. We will establish it more ﬁrmly in the next section. we can write this out as kx. θx ≈ x/z and θy ≈ y/z.1 Far-ﬁeld Propagation Fraunhofer diﬀraction refers to what happens to an optical pattern after propagation over a very large distance z in the paraxial approximation. monochromatic ﬁeld.y ≈ k Thus.3.158 Chapter 12.20) where we have neglected an overall phase. it’s just the Fourier transform of the initial pattern. each Fourier component corresponds to a wave propagating at a particular angle. (12. the 1/z dependence enforces the inverse-square law). Note that this is still true in the ideal limit where all kx. The factor k/2πz is less obvious here. but is what is required to make the total power of the diﬀracted wave equal to the power of the initial wave (in particular. corresponding to propagation up to 180◦ from the optical axis.

(12. . Then each beam at each angle must separate spatially from the others by the time the ﬁeld hits the screen. then we can treat each Fourier component as a separate “spot” on the screen. y) in the x-direction (the same argument will hold for the y-direction).27) δxscreen ≪ λz. we can consider only a region of “radius” δxscreen at the screen.25) that accompanies free-space propagation.12. rather than as a plane wave.3 Fraunhofer Diffraction 159 dx We can view each component at angle θ as a “beam” with a size determined by the size of E (+) (x. it must be that E (+) (x. the uncertainty relation implies a maximum angle of max(θx ) ∼ λ δxsource .3. we can recall from the uncertainty relation (presented in temporal form in Eq. 12. where √ (12. parallel rays propagating at an angle θ with respect to the optical axis get mapped to a transverse position θf in the focal plane of the lens. (7. δxsource (12. or √ (12.22) Using Eq. y). For this phase factor to be negligible. (12. Recall that for a thin lens. In other words.19). (12. y) looks like a point source to an observer on the screen. This is essentially a small-angle restriction. and so this is the far-ﬁeld requirement for the approximation to be valid. The other validity condition comes from the phase factor eikz z −→ exp −i 2 2 kx + ky 2k z (12.26) since we make the identiﬁcation kx −→ kx/z and ky −→ ky/z.24) δxsource ≪ λz.33).23) An angle θx corresponds to a separation over distance z of zθx . so for the size of the diﬀraction pattern to be much larger than the initial pattern. we require λz/δxsource ≫ δx.2 Thin Lens as a Fourier Transform Computer The whole Fourier-transform argument above also works for a thin lens. this phase factor becomes exp −ik x2 + y 2 2z . more restrictive than the usual paraxial restriction. We can write this condition equivalently as z ≫ (δxsource )2 /λ. but equally applicable to space)that the spatially conﬁned ﬁeld implies a spread in spatial frequency of δkx ∼ 1 . At the screen. Letting δxsource denote the size (“radius”) of E (+) (x.

we can model the slits as δ-functions: E (+) (x) = A [δ(x − a/2) + δ(x + a/2)] . we can identify the images at z = ±∞ as being equivalent. 12. E o(x. we can conclude that the image at the focal plane of the lens (i. (12. The slits are uniformly illuminated by a monochromatic plane wave at normal incidence to the mask. but where f is now the propagation length scale: Efocalplane (x.30) = A eikx a/2 + e−ikx a/2 dx = 2A cos kx a 2 .3.oy) (+) F[E (+)o(x.28) Thus. This is the wave-optics version of the parallel-ray argument above.y get mapped to diﬀerent points θx. y) because the image at z = ∞ in the absence of the lens is the Fraunhofer pattern of the original image. y) to z = ∞. thin lenses act as analog Fourier transform computers. where a ﬁeld proﬁle is followed by a thin lens placed one focal length f away..e. f f . the screen) is the Fourier transform of E (+) (x. . Because an image at inﬁnity emits parallel rays.29) e−ikx x [δ(x − a/2) + δ(x + a/2)] dx (12. let’s compute the Fraunhofer diﬀraction pattern due to two narrow slits separated by distance a. Thus. y). The Fourier transform is given by ˜ E (+) (kx ) = A ∞ −∞ (12.3 Example: Diffraction from a Double Slit As an easy example. The lens images E (+) (x. The lens then maps the ﬁeld at −∞ to the focal plane. Plane waves at diﬀerent angles θx. The ﬁeld at z = −∞ is the Fraunhofer diﬀraction pattern of E (+) (x.y f in the focal plane (on the screen). y) = (+) k ˜ (+) E 2πf kx ky . Fourier Optics qf q f With this fact in mind.oy)] f f We can now make a few observations: 1. let’s consider the setup shown here. 3. 2. at least within the paraxial approximation. 4. The lens maps the image at z = −∞ to the focal plane. 5. Assuming the slit widths are much smaller than a.160 Chapter 12. and a screen is another focal length past the lens.

(12. (12.12. which corresponds to angles θx of ±λ/2a. this becomes πE0 ˜ E (+) (kθx ) = k (+) (+) (+) [δ(kx − 2g) + δ(kx + 2g)] .3.31) In terms of angle. Thus. g decreases.33) Using the Fourier-transform relation F eiαx = ∞ −∞ e−ix(kx −α) dx = 2πδ(kx − α). z) = E E k ˜ (+) E (kx/z) = 0 δ(x) + 0 2πz 2 4 (+) (+) [δ(x − 2gz/k) + δ(x + 2gz/k)] (12.1).4 Example: Diffraction from a Sinusoidal Intensity-Mask Grating Another simple example is the diﬀraction of a plane wave by a sinusoidal intensity mask (or intensity grating). cos πz λz (12. the corresponding pattern on a distant screen is E (+) (x. the diﬀraction pattern is (up to an overall factor) ˜ E (+) (kθx ) = 2A cos kaθx 2 = 2A cos πaθx λ . This result is consistent with the result of a simple relative-path-length calculation (Problem 5. It’s important to see the Fourier-transform behavior at work here again—as the period of the grating gets longer. (12.33): πE0 (+) ˜ F E (+) (x) = E (+) (kx ) = πE0 δ(kx ) + 2 Thus.38) when the screen is at distance z. ±3π/2. the far-ﬁeld diﬀraction pattern at distance z is E (+) (x.3 Fraunhofer Diffraction 161 Thus. the angular diﬀraction pattern is πE0 (+) ˜ E (+) (kθx ) = πE0 δ(kθx ) + 2 Using δ(ax) = δ(x)/|a|. (12. then the ﬁeld is E (+) (x) = tmask E0 (+) = E0 (+) cos2 gx = E0 2 (+) + E0 4 (+) cos(2gx) = E0 2 (+) + E0 4 (+) ei2gx + e−i2gx . and so on. Brieﬂy. so this grating is special because it’s sinusoidal: there are no higher-order diﬀraction peaks from this grating (we’ll say more about this in a bit). (12.37) Thus.34) we can easily compute the Fourier transform of Eq. as you’d expect from the uncertainty principle. There are only 3 peaks. and so the diﬀraction pattern narrows. z) = k ˜ (+) E 2πz kx z = kA cos πz kax 2z = kA πax .35) [δ(kθx − 2g) + δ(kθx + 2g)] . 12. the angular separation between fringes is ∆θ = λ/a. (12. and so on.36) δ(θx ) + πE0 2k (+) [δ(θx − 2g/k) + δ(θx + 2g/k)] . . and is illuminated by a normally incident (+) plane wave E0 .32) This function vanishes when the argument is ±π/2. (12. If the intensity mask has a transmission function tmask = cos2 gx. there are diﬀraction peaks at angles θx = 0 and ±2g/k = ±gλ/π. ±3λ/2a.

z) = F −1 F E (+) (x. y) eikz z . Again. not just the far ﬁeld. 12. this becomes E (+) (x. with the diﬀraction peak of order n located at 2gn .46) 2πiz which follows from the Gaussian Fourier transform formula (see Section 3. the transform integral doesn’t actually converge.1).3. this is completely (+) general.4 Fresnel Diffraction 12.44) (12. (12.162 Chapter 12. Fresnel diﬀraction is also used to refer to the case where the wavefronts at z = 0 are not planar. for an arbitrary pattern. as it only modiﬁes the particular form of the initial function E (+) (x.2.42) θn = k and this peak has amplitude cn (or relative intensity |cn |2 ). (12. In the paraxial approximation. y) incident on a phase or amplitude mask (aperture) with transmission taperture . we will be concerned with the propagation of the optical wave to any distance. y. there are many diﬀraction peaks. but this is already accounted for in the Fraunhofer formalism that we presented in the last section.4. But letting z −→ z − iβ. then we have the problem of Fresnel diﬀraction. Thus. z) = F −1 F E (+) (x.41) Thus. y. so this could be an arbitrary wave Ein (x. (12.13)is E (+) (x. y) eikz e−i(kx +ky )z/2k . but we don’t require that the observation screen is in the far ﬁeld.39) where the cn are the Fourier coeﬃcients of the transmission function. but not necessarily in the far ﬁeld.1 Convolution Revisited If we stay in the paraxial approximation.45) k ik(x2 +y2 )/2z e . y) = taperture Ein (x. F −1 e−i(kx +ky )z/2k = 2 2 . Recalling this procedure. so that (+) E (+) (x. we start with a ﬁeld E (+) (x.40) so the angular diﬀraction pattern is 2πE0 ˜ E (+) (kθx ) = k (+) ∞ n=−∞ cn δ(θx − 2gn/k).43) Then the frequency-domain solution (Eq.3. (+) (12. Actually we were already doing this when we introduced the Fourier recipe for wave propagation from Section 12. Since the exponent is purely imaginary.5 Example: Diffraction from an Arbitrary Grating For a general periodic transmission mask of period π/g. y). Now we’ll use the identity 2 2 (12. y). (12. Then the Fourier transform is (+) ˜ E (+) (kx ) = 2πE0 ∞ n=−∞ cn δ(kx − 2gn). Fourier Optics 12. but with a little care. (12. (12.2. Actually. y) at z = 0. we can repeat the above analysis by decomposing the function into its Fourier coeﬃcients: E (+) (x) = ∞ n=−∞ E0 cn ei2gnx .

2 Impulse Response So what exactly is this convolution kernel g(x. E (+) (x.52) becomes E (+) (x. so that the phase factors exp[ik(x2 + y 2 )/2z] and exp[ik(x′2 + y ′2 )/2z] are both approximately unity. y)] = 1.53) Now deﬁning κx := kx/z and κy := ky/z. δ(x. z) = E (+) (x′ . y.46) in Eq. the propagation induces the transformation δ(x. y. we can do the resulting convergent integral and then let β −→ 0 afterwards.48) is the Green’s function solution to the diﬀraction problem (in the paraxial approximation). z) = where ∞ ∞ keikz ik(x2 +y2 )/2z e 2πiz . (12. z) = E (+) (x′ . and Eq.4. y) −→ 2 exp(ikz) exp[−i(kx (12.50) + 2 ky )z/2k] and inverting the keikz ik(x2 +y2 )/2z e = g(x. and let’s propogate it forward. κy ) = E 2πiz 2πiz kx ky .12.3 Far-Field Limit In the far ﬁeld.45). y. including the correct scaling factor (and even an overall phase factor that we dropped in our heuristic treatment). we should be recover the Fraunhofer diﬀraction pattern and thus rigorously derive the results of Section 12. y). y) F By the convolution theorem. y. y − y . y ′ ) E (+) (x. y ′ ) ei(κx x+κy y) dx′ dy ′ . y. So. y ) g(x − x . Then multiplying by the propagation factor exp(ikz z) = transform. z) = F −1 F E (+) (x. g(x. z) := 12. (12. (12. using Eq. y) ∗ g(x. z). Let’s start by writing out the Green’s function solution (12. y) is the impulse-response function or Green’s function for the diﬀraction problem. (12. The Fourier transform is easy: F [δ(x.49) 2πiz iλz This is the convolution or frequency-domain solution to the paraxial propagation or Fresnel diﬀraction problem. so E (+) (x.48) explicitly: keikz ik[(x−x′ )2 +(y−y′ )2 ]/2z ′ ′ e dx dy .3.47) (12. (12. y. (12.55) Thus. (12. z) dx dy . so Eq. (12.52) 2πiz −∞ −∞ √ √ We can assume the Fraunhofer approximations δxsource ≪ λz (far ﬁeld) and δxscreen ≪ λz (narrow angle range).4 Fresnel Diffraction 163 where β > 0. (12. z) = keikz 2πiz ∞ −∞ ∞ −∞ ∞ ∞ E (+) (x′ . (12. g(x. y. y)? Suppose we have an optical source that is a unit point. 12.4. y ′ ) exp −i keikz 2πiz ∞ −∞ ∞ −∞ kx z x′ exp −i ky z y ′ dx′ dy ′ .48) ′ ′ ′ ′ E (+) −∞ −∞ (x . y. we recover the Fourier transform form for the Fraunhofer diﬀraction pattern. z) = E (+) (x.54) keikz ˜ (+) keikz ˜ (+) E (κx . we can write E (+) (x. z z . z) = This is just the Fourier transform. y. ′ ′ keikz ik(x2 +y2 )/2z eikz ik(x2 +y2 )/2z e = e . (12. we can write E (+) (x.51) 2πiz Thus. .

(12.4. z) = E0 (+) −ieikz k 4πz C k a x+ πz 2 +iS −C k a x− πz 2 − iS a k x− πz 2 .60) a k x+ πz 2 The Fresnel integrals can be evaluated graphically using a tool called the “Cornu spiral” (basically a plot of S(x) vs. (12.8 -10 -1.8 0.57) 2πiz x−a/2 Now letting x′ −→ πz/k x′ . z) = (+) E0 −ieikz √ k/πz (x+a/2) ′2 k eiπx /2 dx′ . Notice the gradual transition from something rectangular to the familiar far-ﬁeld sinc pattern as the distance increases.8 0. (12. E (+) (x. In this case. but they are just as easily evaluated on a computer.56) where we assume the y ′ -integration has already been completed.58) Now deﬁning the Fresnel integrals x C(x) := 0 cos πt2 2 x dt S(x) := 0 sin πt2 2 dt. Eq. C(x)). which ultimately gives rise to the fringes in the diﬀraction patterns. (12. z) = E0 eikx /2z dx′ . (12. √ 4πz k/πz (x−a/2) (12.48) becomes E (+) (x.5 -0. The slit edges are marked by dashed lines.41 0 1.5 z = 0.01(a2 /2λ) 0. let’s consider the Fresnel diﬀraction pattern from a single slit of width a. and we’ve also suppressed the y-dependence.4 Example: Fresnel Diffraction from a Slit As one of the simplest examples. x+a/2 ikz ′2 (+) ke E (+) (x. and that they asymptotically settle to ±1/2.5 -0. The Fresnel integrals are plotted here. z) = E0 (+) keikz 2πiz a/2 −a/2 eik(x−x ) ′ 2 /2z dx′ . 0.59) we can write the diﬀraction pattern in standard form as E (+) (x.164 Chapter 12.5 z = 0.41 10 x x Some diﬀraction patterns for the single slit are shown here.1(a2 /2λ) C (x) 0 S (x) -1 0 1 10 0 -0. Fourier Optics 12. .8 -10 -0. Letting x′ −→ x′ + x. Note their oscillatory nature.

.5 Spatial Filters The fact that a thin lens computes the Fourier transform of an optical ﬁeld proﬁle opens up a lot of intriguing possibilities—namely.01(a2 /2λ) z = 0. For simplicity we’ll assume that the two lenses have the same focal length f . Then the propagation is over a distance 4f as shown here. The basic setup for doing this is a system of two lenses.12. that of optical information processing.1(a2 /2λ) z = 0.5(a2 /2λ) z = 1(a2 /2λ) z = 2(a2 /2λ) z = 20(a2 /2λ) 12.5 Spatial Filters 165 z = 0.

61) The spatial locations in the Fourier plane correspond to diﬀerent spatial frequencies and thus to details in the images of diﬀerent length scales.) The point of this setup is that we can put in ﬁlters to modify the Fourier transform of the image. blurring the image. which is equivalent to the inverse transform with x −→ −x.63) ka . the output image is the convolution of the original image with the Fourier transform of the aperture. we can perform many useful enhancements or modiﬁcations to the image. f f . Low-pass ﬁlter. 1.166 Chapter 12. The aperture in the Fourier plane deﬁnes the transfer function for the system. Fourier Optics (+) F[E in o(x. ky ) by Efocalplane (x. k¥ fine details (y) fine details (x) k≈ fine details (x) broad features fine details (y) The low frequencies (and thus broad image features) are near the center of the transform plane. This observation motivates some of the simplest spatial ﬁlters. causing the blurring. so the ﬁnal image is inverted.62) are removed from the image. and so the image is “smeared” by the Airy disk. in either the amplitude or phase. . The Fourier transform of the original ﬁeld proﬁle appears in the plane midway between the two lenses. 2f (12. the second lens performs a second Fourier transform.oy) f f f f (+) E out (x. From the convolution theorem. the ﬁeld is given in ˜ (+) terms of the Fourier transform Ein (kx . then in the Fourier-transform plane. the cutoﬀ frequency is given by the condition kr = kr/f . We’ll ignore this in the discussions here. with the dc component being right in the center. structures smaller than the length scale of 2π 4πf 2λf = = kcutoﬀ ka a (12. The ﬁne details are encoded in the high frequencies away from the center. as we’ll come back to later. If the ﬁlter is an aperture of diameter a. The second lens undoes the Fourier transform and reconstructs the original image on the output plane. A circular aperture in the Fourier plane will block the high spatial frequencies and pass the low ones.oy)] E in o(x. (Technically. The aperture’s Fourier transform is the Airy disk. y). In doing so. (12. giving a cutoﬀ frequency of kcutoﬀ = That is. The eﬀect is to remove the ﬁne details.oy) o (+) For this reason this is called the 4-f setup for spatially ﬁltering an image. y) = (+) k ˜ (+) E 2πf in kx ky . (+) Recall that in the original wave is Ein (x.

A circular. The original photo (of Oregon moss on trees) has structures on many diﬀerent scales. 3. notice that only the branches perpendicular to the slit survive the ﬁlter. Thus. The high. This ﬁlter is also called an edge detector because sharp edges generate high spatial frequencies and thus make it through the ﬁlter.12. the cutoﬀ is the same as for the low-pass ﬁlter. Directional ﬁlter. For a spot of diameter a. while the moss detail is preserved. Examples of these ﬁlters are shown here in simulations. original photo (intensity) low-pass ﬁlter high-pass ﬁlter directional mask . The directional mask in the ﬁnal photo is oriented at 45◦ .5 Spatial Filters 167 2. leaving only the branches. which get wiped out. the image ﬁlter keeps only those structures smaller than 2λf /a. a vertical slit preferentially passes vertical structures while blocking horizontal structures.and low-pass ﬁlters can also work only in one direction. The low-pass ﬁlter wipes out the ﬁne details of the moss. with the obvious modiﬁcation of the aperture. That’s because branches parallel to the slit represent high spatial frequencies perpendicular to the slit. The high-pass ﬁlter does just the opposite: the structure of the branches is mostly wiped out. For example. blocking the low frequencies and passing the high. High-pass ﬁlter. while the parallel branches are mostly erased. absorbing spot in the Fourier does the opposite.

168

Chapter 12. Fourier Optics

**12.5.1 Spatial Filtering of a Gaussian Beam
**

One important application of spatial ﬁltering in real optical setups is the cleaning of a Gaussian beam. When propagating through real optical elements, Gaussian beams get contaminated with spatial intensity noise due to dust, imperfections, diﬀraction at the edges of objects, etalon fringes (when, say, passing through an uncoated window), aberrations, and so on. For example, an acousto-optic modulator acting as a switch tends to distort the beam proﬁle, and so it is often desirable to restore it. The idea is to use a low-pass ﬁlter, here in the form of a pinhole. The Gaussian beam is focused through the pinhole, and under the right conditions a nearly ideal Gaussian proﬁle appears on the other side. Let’s simplify things from the basic 4-f setup. We’ll use a lens of focal length f to focus the beam through the pinhole, and let’s assume that the waist of the unfocused beam occurs at the lens. The pinhole is much smaller than the unfocused beam, so the pinhole is a distance f from the pinhole. After the pinhole, a second lens of focal length fout recollimates the beam. 2wº

input (dirty) beam recollimated beam

f pinhole

fout

In the absence of the pinhole, this is simply a telescope beam expander, and so the diameter of the beam is magniﬁed by a factor fout /f . The most important part of the setup is the pinhole, so let’s focus on that. Let the incident beam have radius w0 . Then we can use the results in Section 6.5 from the ABCD Law, which gives the beam radius at the pinhole: w0,pinhole = λf . πw0 (12.64)

This assumes, of course, a clean Gaussian beam. The commonly recommended pinhole diameter is one that passes 99% of the incident power of a clean beam, which is Dpinhole = 2 · 1.5 · w0,pinhole . (12.65)

In this case, nearly all of the Gaussian beam passes through the pinhole. But noise, especially high-frequency noise, has its power distributed away from the pinhole in the Fourier-transform plane. So the pinhole blocks the power associated with spatial intensity noise on the beam. Thus, the recollimated beam is a very clean Gaussian to good approximation. There are two caveats here. One is that a small amount of intensity noise is left, corresponding to the same length scales as the beam diameter. This is usually not noticeable, and corresponds to shifts or slight distortion of the proﬁle. In cases where there is signiﬁcant low-frequency noise (such as the highly distorted output of a single-mode diode laser), it may be necessary to use a smaller pinhole and lose much more power. The other caveat is that the output isn’t quite Gaussian. From the convolution theorem, the output is the convolution of a Gaussian and the Airy disk, so there are generally faint rings around a spatially ﬁltered beam. These can be clipped oﬀ if necessary with an aperture after the recollimating lens. All this information is best visualized in the simulated example shown below. The ﬁrst plot shows the intensity proﬁle of a clean Gaussian beam. The second is the same proﬁle, but with intensity noise superposed. Now the task is to clean this up as much as possible. Focusing the beam produces a very narrow Gaussian proﬁle, shown in the next plot. The Gaussian itself is rather clean, but what is only apparent with a large vertical magniﬁcation (fourth plot) is that the baseline is contaminated with broadly distributed noise. This is what we expect from the Fourier transform argument above. We clip this noise away with pinhole (ﬁfth and sixth plots, edges of pinhole marked by vertical lines). The very tips of the Gaussian proﬁle are clipped as well, but essentially all of the noise is gone. The beam on the output is a nice-looking Gaussian (seventh plot), but notice that the amplitude is reduced compared to the ﬁrst plot:

12.5 Spatial Filters

169

power is lost in the cleaning process, since we are essentially projecting the beam into the Gaussian mode. The faint Airy rings are not visible on this plot, but they are on a logarithmic plot (last plot). The side lobes in this plot do not occur on the original beam. These are visible to the human eye, which itself has a logarithmic response to intensity. original Gaussian beam beam contaminated with noise

I(x)

x

I(x)

x

focused beam before pinhole

focused beam (vertical zoom)

I(x)

x

I(x)

x

after pinhole (vertical zoom)

after pinhole (horizontal zoom)

I(x)

x

I(x)

x

cleaned beam

cleaned beam (log plot)

x

log[I(x)]

I(x)

x

170

Chapter 12. Fourier Optics

**12.5.2 Visualization of Phase Objects
**

The “obvious” spatial ﬁlters in the beginning of this section (low-pass, high-pass, directional) are intuitive examples of what you can do with optical spatial ﬁlters. But these days, it seems far easier just to digitize the image and do the same processing on a computer. So why do optical information processing? Sometimes it is convenient or useful to do so optically, as in cleaning a Gaussian beam. But sometimes it is impossible to digitize or even record optical information before optical processing. An important class of examples is in imaging phase objects, or objects that only inﬂuence the phase of probe light. We can model a phase object by a transmission coeﬃcient of the form t(x, y) ∼ eiφ(x,y) . If this object is illuminated by a plane wave E0 E0 The intensity is I(x, y) ∼ 2|E0 |2 , η

(+) (+) (+) (+)

(12.66)

**(in the z = 0 plane), the resulting ﬁeld becomes
**

(+)

−→ E0 t(x, y) ∼ E0 eiφ(x,y) .

(12.67)

(12.68)

which is a constant and thus has no information about the phase φ(x, y). The point is that the transmission function has unit amplitude everywhere, and so doesn’t modify the intensity of the transmitted light. So we can’t just record the light on ﬁlm and post-process the information. We could get at this information via interferometry or holographic techniques. Instead, we will look at how spatial ﬁlters can reveal this information. 12.5.2.1 Zernike Phase-Contrast Imaging One widely used phase-imaging method was developed by Frits Zernike, who won the 1953 Nobel Prize in Physics for this work. This method is commonly used, for example in biological microscopy to view transparent objects such as bacteria (that would otherwise require staining). Let’s consider a phase object with a small phase shift of the form t(x, y) = eiφ(x,y) ≈ 1 + iφ(x, y) + O φ2 . As we saw before, when uniformly illuminated, the intensity transmitted is I ∼ |1 + iφ(x, y)| = 1 + O φ2 ,

2

(12.69)

(12.70)

so that there is no intensity change (we are discarding the higher order terms in φ to be consistent with the initial expansion). Essentially, this is because the linear cross-terms cancel when squaring the ﬁeld. The idea behind the Zernike method starts with the observation that the unity part of Eq. (12.69) is the dc component of the signal, while φ(x, y) represents spatial structure, and thus has a wideband spectrum. The vanishing cross-terms are thus a product of the dc component and the rest of the spectrum. The idea is, what if we modify one of these to prevent the cancellation? Speciﬁcally, let’s try a π/2 phase shift of the dc component: I ∼ eiπ/2 + iφ(x, y)

2

= |i [1 + φ(x, y)]|2 = 1 + 2φ(x, y) + O φ2 .

(12.71)

Now the linear term stays, so the intensity reﬂects the phase information. Actually, since the intensity change is linear in the phase (for small phase shifts), this method gives an image that is simple to interpret: other than an overall background, the intensity is a direct image of the phase. The other imaging methods we will discuss here produce more complicated functions of the phase.

12.5 Spatial Filters

171

Since the intensity increases with increasing phase shift, this method is called positive phase contrast imaging. We can also try a 3π/2 phase shift of the dc component, with the result I ∼ ei3π/2 + iφ(x, y)

2

= |(−i) [1 − φ(x, y)]| = 1 − 2φ(x, y) + O φ2 .

2

(12.72)

Now the intensity decreases with increasing phase, so this version is called negative phase contrast imaging. The requisite phase shift is accomplished here by a ﬁlter in the Fourier plane of the 4-f system that could be, for example, a glass plate with a transparent dielectric thin-ﬁlm dot at the center that causes a relative phase shift of ±π/2 of the small portion at the center of the image. 12.5.2.2 Central Dark-Ground Method An alternative to the Zernike method is the central dark-ground method. The idea is very similar, but instead of a mask with a phase-shifting dot, the mask has an opaque dot at the center to block the dc component. The details are left as an exercise, but we can summarize the results. The advantage is that the background (dc component) is suppressed, giving a dark background. The disadvantage is that the intensity varies nonlinearly with phase (as φ2 for small φ). 12.5.2.3 Schlieren Method The last method we will discuss here is important in the visualization of ﬂuid ﬂows, but requires considerably more mathematical sophistication to handle than the last two cases. The upshot of all this is that the schlieren method shows gradients in the phase of the object in a particular direction. The name comes from the German word schlieren, which means “streaks.” One application of this method was to test glass lenses for impurities introduced during the fabrication process that would cause inhomogeneities in the refractive index. These would show up as streaks in the schlieren test, and hence the name. The setup is the same 4-f ﬁltering setup. But rather than some sort of spot in the Fourier plane to modify the dc component, the schlieren method uses a knife edge in the Fourier plane to block half of the frequency spectrum. So we can represent the transmission function of the Fourier-plane mask as the Heaviside step function: 1, x > 0 tmask (x, y) = UH (x) := (12.73) 0, x < 0. Since the y-direction is trivial to handle, we’ll only do a one-dimensional analysis. Following the ﬁeld through the 4-f setup, 1. Input ﬁeld: Ein (x) ˜ (+) 2. Transform plane: Ein (kx ) ˜ (+) 3. After knife edge: Ein (kx )UH (kx )

(+) ˜ (+) ˜ 4. Image plane: F Ein (kx )UH (kx ) = Ein ∗ UH (x) (+)

˜ The last expression follows from the convolution theorem, and UH (x) is the Fourier transform of the Heaviside function, which we must now evaluate. So let’s now compute F [UH (x)]. Unfortunately, this is a little tricky, so we’ll spend some time working this transform out. It turns out that the answer is 1 ˜ F [UH (x)] ≡ UH (kx ) = πδ(k) + . (12.74) ik Rather than derive this, we’ll simply verify that the inverse transform recovers the Heaviside function. The inverse transform is 1 ˜ F −1 UH (kx ) = 2π

∞ −∞

πδ(k)eikx dk +

∞ 1 – 2π −∞

1 ik

eikx dk.

(12.75)

172

Chapter 12. Fourier Optics

Note the special notation in the integral in the last term. This is because this integral is singular due to the 1/k dependence. To handle the singularity, we take the Cauchy principal value by splitting the integral about the singularity, and then approaching the singularity symmetrically: –

∞

−∞

· · · ≡ lim

−δ −∞

δ→0

··· +

∞ δ

···

.

(12.76)

In this way, the singular part cancels; this works because the 1/k changes sign on either side of k = 0. Thus, – even though

∞

−∞

dx = 0, x

∞

(12.77)

diverges. Of course, if the singularity occurs elsewhere, the splitting is understood to be there. Now back to evaluating Eq. (12.75). 1 1 ˜ lim F −1 UH (kx ) = + 2 2πi δ→0 Flipping the integration limits in the last term, 1 1 ˜ F −1 UH (kx ) = + 2 π Here we used the result

0 ∞ ∞ 0 ∞ δ

−∞

dx x

(12.78)

eikx dk + k

−δ −∞

eikx dk . k

(12.79)

1 1 sin(kx) dk = + sgn(x) = UH (x). k 2 2

(12.80)

π sin(kx) dk = ± , k 2

(12.81)

where the result is π/2 for x > 0 and −π/2 for x < 0 (proving this requires a contour integral, which we won’t get into here). So now we’ve veriﬁed that we have the correct Fourier transform of UH (x). But that’s not quite what ˜ we wanted. What we really need is the inverse Fourier transform of UH (kx ), to get UH (x). So to handle the overall scaling and sign conventions, we just need to divide our result by 2π and let i −→ −i to obtain 1 i ˜ UH (x) = δ(x) + . 2 2πx (12.82)

Now, as we concluded above in tracing the ﬁeld through the ﬁltering system, the output ﬁeld is the ˜ convolution of the input ﬁeld with UH (x):

(+) (+) ˜ Eout = Ein ∗ UH (x).

(12.83)

Recalling the form of the convolution integral from Eq. (11.4), we can thus write the output ﬁeld for the schlieren setup as (+) ∞ i E (+) (x′ ) ′ E (x) (+) dx . (12.84) + – Eout = in 2 2π −∞ x − x′ If Ein (x) is produced by a uniformly illuminated phase object, then for small phase shifts Ein (x) = E0 eiφ(x) ≈ E0 (1 + iφ(x)). Then the output intensity is

∞ Iout 1 iφ(x) 1 φ(x′ ) = dx′ + O φ2 + − – Iin 2 2 2π −∞ x − x′ 2 (+) (+) (+) (+)

(12.85)

,

(12.86)

Theory and Applications of Holography (Addison-Wesley.92) to write φ(x) = − Diﬀerentiating. we ﬁnd ˆ 1 ∞ φ(x′ ) – dx′ . up to a minus sign: H −1 = −H .88) more compactly as ˆ 1 φ(x) Iout = + . In any case. DeVelis and George O. This statement can be proved by using the fact that the Hilbert transform is linear.93) ˆ Here is the ﬁrst problem: the integral with 1/x2 doesn’t converge for generic functions φ. (12.5 Spatial Filters 173 where we have already used – ∞ Multiplying out the intensity expression and keeping only ﬁrst-order terms in φ.11). f (x) := π −∞ x′ − x As for the Fourier transform. Now something even 1 from ˆ 1ˆ 1 1 ∞ φ(x′ ) dφ dx′ = − φ(x) ∗ 2 . and from Eq. we can also use an operator notation ˆ f ≡ H [f ] (12. π −∞ x′ − x (12. x − x′ (12. So.88) is of special signiﬁcance. So we can write Eq.12. For a function f (x). the Hilbert transform is the convolution of a function f (x) with 1/x. while the derivative is deﬁned by information arbitrarily close a single point. (12. Reynolds. in the limit of small phase shifts. (12.90) (12. the result is ∞ 1 Iout 1 φ(x′ ) = − dx′ . . the derivative is a reasonable intuitive guide. (12. When you multiply f (x) by 1/x and integrate. and it is called the ˆ Hilbert transform.84) and (12. 1967). The complication is that 1/x falls oﬀ so slowly that the Hilbert transform picks up long-range information about the function. That is. the value will be large when f (x) has a large slope at x = 0. you thus expect that it will take on a large value only when something upsets the cancellation at x = 0.91) Iin 4 2 But what does this mean intuitively? Well. In this sense. (12.84) that the output ﬁeld of the schlieren system is (up to an oﬀset and scaling factor) the Hilbert transform of the input ﬁeld. (12. and right where it diverges it suddenly changes sign. there isn’t even really a limit (like “slowly varying functions”) for arbitrary functions in which the Hilbert transform closely approximates the derivative. But let’s go with it.89) to simplify some algebra. – Iin 4 2π −∞ x − x′ −∞ dx′ = 0. (12. and computing the Hilbert transform of the harmonic functions (see Problem 12. since it gives a good intuitive result. We’ll start oﬀ with the property that the Hilbert transform is its own inverse. we now see how the phase information is reﬂected in the intensity. But 1/x diverges at x = 0. we see from Eq.88) we see that the output intensity is the Hilbert transform of the input phase. we can use Eq. the Hilbert transform f of f is given by 1 ∞ f (x′ ) ˆ – dx′ . Let’s go through a sloppy argument1 that “shows” that the schlieren image is the derivative of the phase.94) John B.87) (12. even using the Cauchy principle value.92) or H 2 = −1. But as far as visualizing the schlieren image is concerned. =− – ′ − x)2 dx π −∞ (x π x (12. so that the two sides of 1/x are unequally weighted. decomposing a test function into harmonic components (via the Fourier transform). The integral that appears in Eqs.88) Thus. So the Hilbert transform is “something like” a derivative.

But. The second set of images is in the large phase-shift regime. so all our linearized conclusions above hold.174 Chapter 12. The ﬁrst set of images is in the small-phase regime (the largest phase shift is 0. the conclusion is that the schlieren image is approximately an intensity background plus the derivative of the phase. The images are visually more interesting but harder to interpret.88) becomes 1 π dφ Iout − ∼ . But if we’re concerned about actually being correct. (12. and 1/x2 has no length scale. The derivative nature of the schlieren image is apparent in the last image. Eq. Iin 4 2 dx (12. Again. while the integral of the δ-function is not. except that the “wrapping” of the large phase shifts is apparent in the image.2π). The other three images show intensity patterns at the output of the 4-f imaging system (neglecting inversion and magniﬁcation). but the Gaussian spots are somewhat distorted. dx π So. Fourier Optics ˆ sloppier: if φ varies slowly. so it isn’t clear what we mean by slowly varying. The darkground image has a dark background. where the largest phase shift is 3π. 12.4 Numerical Examples Now we’ll look at simulated images of a phase object to get a better idea of how the diﬀerent phasevisualization methods compare. but has an intensity background.5. . giving a “three-dimensional” sort of look. The original image is a plot of the phase shift. where the phase shifts are concentrated in randomly placed Gaussian blobs. Similar comments to the ﬁrst set of images hold here.96) (12.2. The Zernike image almost perfectly reproduces the original phase. then the convolution is with the δ-function and so ˆ φ(x) dφ ∼ . it’s better to stick with the Hilbert transform. then 1/x2 is something like a δ-function. if we make the replacement.95) So. the integral of 1/x2 is divergent.

bright: 0.5 Spatial Filters 175 phase plot (dark: 0.2π) Zernike phase-contrast image dark-ground image schlieren image .12.

We can exploit this idea that the three-dimensional ﬁeld can be represented in two dimensions. say E (+) (x. 0). y) ∝ E (+) (x.6 Holography Recall from our earlier discussion of Fourier-optics propagation that an arbitrary monochromatic ﬁeld E (+) (x. y. z) is completely determined by the ﬁeld in a single plane. z = 0).97) . y. (12. This is because we can simply apply the propagation factor eikz z to E (+) (x. y. z = 0) (in frequency space) to obtain the ﬁeld at any location z. bright: 3π) Zernike phase-contrast image dark-ground image schlieren image 12. y.176 Chapter 12. Fourier Optics phase plot (dark: 0. A hologram (at least ideally) is a thin optical element with a complex transmission coeﬃcient t(x.

y) = ei(k sin θ)x . Illuminating this (+) with an on-axis plane wave E (+) (x.6 Holography 177 When uniformly illuminated by a monochromatic ﬁeld. But now getting the information back out is a bit subtle because of the encoding. y)E (+) (x. Then 2 (−) (+) (+) (+) (+) (+) E (+) = tEr ∝ Er Ir + Er Io + Ir Eo + Eo Er η 2 . which is a plane wave with wave vector k = k (ˆ sin θ + z cos θ) .2 Film Holograms But now the question is. given by (+) (+) Eo = |Eo |eiφo (x. this z hologram acts like a prism. Irecord = + Er |2 η 2 (+) (−) 2 (−) (+) = Io + Ir + Eo Er + Eo Er η η 2|Eo (+) (+) (12.0 . Both the phase and amplitude information are encoded in the intensity and hence the photograph (like in an interferometer).y) . A hologram needs both intensity and phase information. y) = Er. where φo is the object phase.100) The original direction was k = kˆ. so we need to employ some tricks. z).. The photograph records the intensity of the superposition. a prism). the holographic mask reconstructs E (+) (x.e. y. (12. z) = E0 ei(k sin θ)x ei(k cos θ)z . y) = E0 ei(k sin θ)x . Thus the hologram reconstructs the entire three-dimensional ﬁeld E (+) (x. In other words.99) (12. 12. (12. How do we recover the information recorded on the ﬁlm? The idea is to illuminate the holographic √ (+) mask with the same reference light Er ∝ Ir . how can we make a hologram? Reconstructing the phase information in the last example requires a glass wedge of varying thickness (i. Then propagating this ﬁeld forward to arbitrary z. y) (+) (+) (the “object ﬁeld” that we want to record) with a reference plane wave Er (x. up to an overall constant. literally is a prism. (12.101) (+) (+) (12. 0).98) so that it modulates the signal with a single spatial frequency k sin θ in the x-direction. y.1 Example: Single-Frequency Hologram Suppose that the hologram has a transmission of t(x. y)E (+) (x. y) = E0 . we ﬁnd t(x. In fact. where d = x sin θ. y. x ˆ (12.12. y)]. not to the phase of the light. But photographic detectors such as ﬁlm and ccd cameras respond to intensity. at least in the z > 0 region past the holographic mask. (+) The key point in holography is this: we can use interference to make a hologram.104) . mixing Eo (x.102) = Io + Ir + 2 Io Ir cos[φo (x. so this hologram deﬂects a plane wave by an angle θ.103) The hologram is the developed ﬁlm that is exposed to this superposition.6. the transmitted (reconstructed) ﬁeld is t(x.6. 12. a phase delay of exp[i(k sin θ)x] = exp(ikd).

106) Eo √ ∼ Ir + Io + Ir Ir Io ei(k sin θ)x + Ir Io e−i(k sin θ)x . The original ﬁeld is recovered in the last two terms. while the fourth term is the extra conjugate wave. . ∝ Ir + Io (x. Then the reconstructed wave is (+) (+) Eo ∼ Io eik(sin θ)x . Fourier Optics Dividing through by Er (+) . (12. The third term is proportional to the original ﬁeld.6. This hologram turns out to be just the sinusoidal diﬀraction grating from Section 12.105) The ﬁrst term is just a constant oﬀset. (+) called oﬀ-axis holography. the reconstruction is not perfect because the ﬁlm is fundamentally a real-valued transmitter.3.1.107) Again. hence the conjugate wave. which is what we’re trying to get. this example motivates a method to get at the reconstructed object ﬁeld along.178 Chapter 12. Then the forward wave Ir + Io and conjugate wave Eo are all spatially separated from (+) the reconstructed wave Eo . y) + 2Er η (−) (+) Eo (x.4. The fourth term is the conjugate wave. film developed film reconstructed Eo o(+) reference wave Er o(+) reconstruct reference wave object wave Er o(+) (but get E o too) o(-) Eo o(+) 12. However. (12. the complex conjugate of the original wave. y) + E (+) (+) Er 2Er η (+) (−) Eo (x. film developed film Eo o(+) reference Er o(+) reconstruct Er o(+) Iorooo+ooI o Eo o(+) Eo o(-) Once again. This term is a consequence of the fact that the ﬁlm hologram is fundamentally a real-valued transmission mask.3 Hologram of a Plane Wave and Off-Axis Holography A simple example. similar to the example of Section 12. and the second is the intensity proﬁle of the object. y). The idea is to ensure that Eo is at a suﬃciently large angle (larger than its (−) angular spread). (12. The conjugate wave turns out to be a plane wave at the same angle but reﬂected about the z-axis. the third term is the reconstructed plane wave. is the hologram of a single plane wave.6.

so it is important to minimize the vibrations of the optics.7 Exercises 2 Problem 12. with half illuminating the object to form the object wave. laser hologram (film) virtual image observer Another interesting thing to note is that it is possible to make synthetic holograms of objects that never really existed. the idea is to illuminate the hologram with the same reference wave (although the exact details of the reference wave aren’t too important). for example on a computer. (+) . A laser beam (monochromatic light source) is split. Holographic plates must also have high resolution and are consequently rather expensive. Also.7 Exercises 179 12.1. The appropriate pattern can be computed and printed out on a transparency. laser beam splitter reference wave hologram (film) object wave To reconstruct the hologram. and half going directly to the screen to form the reference wave. y. Consider the Gaussian ﬁeld proﬁle E (+) (x. 12.4 Setup: Off-Axis Reﬂection Hologram Then the optical setup for an oﬀ-axis reﬂection hologram is as shown here. and thus show that your results are consistent with the Gaussian beam solution to the paraxial wave equation. while recording. y. Then the wave after the hologram forms a virtual image of the original object in the same location behind the ﬁlm.12. Also.6. Use the (+) Fourier-transform formalism for propagating the ﬁeld to derive the form of E (x. the relative phase of the two waves must be held very constant. generally holographic ﬁlm takes the form of glass plates for structural rigidity during exposure. z = 0) = E0 exp[−(x2 + y 2 )/w0 ]. although usually the pattern will still need to be photographically reduced to achieve the required spatial resolution. Notice that both beams must be expanded appropriately to cover the object and the ﬁlm. z) (in the paraxial approximation).

circular aperture of radius a is E0 (2πa2 ) (+) (+) J1 (2πar/λz) .) Assume the slits are uniformly illuminated. intermediate. Suppose you have a cruddy Gaussian beam of radius w0 = 1 mm and wavelength λ = 780 nm that you want to clean up and expand. and far ﬁelds. (a) Show that the far-ﬁeld (Fraunhofer) diﬀraction pattern due to a uniformly illuminated (by a normally incident plane wave). Problem 12.2. We’ll ignore the y-direction for simplicity.4. (a) Compute the far-ﬁeld (Fraunhofer) diﬀraction pattern on a screen at location z due to a rectangular aperture of width b at z = 0. (b) Make an intensity plot of this diﬀraction pattern. regularly spaced array of narrow slits.7. . (This is a simple model of a diﬀraction grating. an aperture diﬀracts into an intensity pattern I(x. Problem 12. r is the radial distance from the center of the screen.3. You may restrict your calculation to one dimension. (a) Compute the far-ﬁeld (Fraunhofer) diﬀraction pattern due to an inﬁnite. where the distance between adjacent slits is a. Problem 12. and J1 (x) is an ordinary Bessel function. What is the eﬀect on the intensity pattern if the incident wave makes a small angle θ with respect to the x-axis? The answer should be reasonably obvious but you should prove your answer mathematically. Problem 12. z) on a screen a long distance z away. (a) Write down an expression for the Fresnel diﬀraction pattern for a knife edge illuminated by a plane wave at normal incidence. and it may help you to know that they satisfy the recurrence relation ′ Jn−1 (x) − Jn+1 (x) = 2Jn (x) and Jn (x) = (−1)n J−n (x).6. Assume the aperture is uniformly illuminated by a plane wave at normal incidence to the aperture. 2πar/λz where E0 is the amplitude of the illuminating wave. Fourier Optics Problem 12. (b) Compute the far-ﬁeld intensity pattern due to N regularly spaced slits with spacing a. This is a well-known pattern called the Airy disk.5. Write your answer in terms of the Fresnel integrals C(x) and S(x).180 Chapter 12. (c) Use the convolution theorem to write down the far-ﬁeld diﬀraction of an aperture consisting of two parallel slits of width b with centers separated by distance a. Suppose that when illuminated by a plane wave at normal incidence. The Bessel functions are deﬁned by the integral Jn (x) = 1 π π 0 cos(nφ − x sin φ) dφ. (b) Write down the far-ﬁeld diﬀraction pattern if the aperture is instead a pair of narrow slits separated by a distance a. (b) Make a few representative plots showing the diﬀraction pattern in the near. Problem 12.

the Hilbert transform is proportional to the derivative. ﬁlling in the missing steps where appropriate. One method to prove this is as follows. π −∞ x′ − x Show that the Hilbert transform is its own inverse. Assume that the system input is from a phase object. provided you throw in an extra minus sign. of the form exp[iφ(x)]. Problem 12. Compute the convolution of fG (x. (b) If you want to expand the beam to a ﬁnal radius of w0 = 10 mm. Hint: use the convolution theorem to evaluate the transform integral. that is. Go through the derivation in the notes of the output intensity of a schlieren imaging system. Do this in detail.10. σ) be a Gaussian probability density function with variance σ 2 . which is a 4-f system with a ﬁlter consisting of an absorbing dot that blocks the dc component in the Fourier plane. and suppose that there is nothing between the two lenses of focal length f . without any silly approximations.oy) o (+) . Problem 12. and ﬁnally do the same for H 2 [f (x)]. Problem 12. Let fG (x. Model the blurring of a sinusoidal image as follows.11. Consider the 4-f imaging system shown here.4 mm to focus the beam. Problem 12. where the input is a phase object. Interpret your result. Also go through the argument that the Hilbert transform is “something like” a derivative.7 Exercises 181 (a) Using a lens of focal length f = 25. Following the treatment for the Zernike phase-contrast imaging system.9.12. what is the best pinhole diameter according to the “rule of thumb” to spatially ﬁlter the beam? You may assume that the lens is at the focus of the original beam. and exp(ix). σ) with cos(x). Compute the Hilbert transforms of the following functions: sin(x). Recall that the Hilbert transform is deﬁned by 1 ∞ f (x′ ) ˆ H [f (x)] ≡ f (x) := – dx′ . (+) F[E in o(x. cos(x). Then compute what the function H [f (x)] looks like in frequency space.12.oy)] E in o(x. so the schlieren image is something like the derivative of the phase shift. what focal length lens should you use to recollimate the beam after the pinhole? Problem 12.8. explain the following statement: for narrowband signals. From this you can show that H [eikx ] = sgn(k)ieikx .oy) f f f f (+) E out (x. show that H −1 = −H . work out the output intensity of a central dark-ground imaging system. Recall from Problem 10 that H [e±ix ] = ±ie±ix . From your answer.13. Problem 12.

this is only true in the paraxial approximation. What is the new diﬀraction limit for this situation? For simplicity. Then recall the argument for why the Fourier transform of the input image appears in the focal plane of the ﬁrst lens. using the Fourier-transform propagation recipe. by computing the Fourier transform (not the inverse transform) of the focal-plane ﬁeld. Fourier Optics Recall that the ﬁeld in the focal plane is given by Efocalplane (x. f f . you may assume that the image object represented (+) by Ein (x. and suppose again that there is nothing between the two lenses of focal length f . (b) Consider the 4-f imaging system from Problem 3. y) is much smaller than D. This is the image inversion of the 4-f system that we ignored. y) = Ein (−x. that the ﬁeld in the output plane is (+) (+) Eout (x. Hint: Start by sketching some rays from the object through the system.14. How is the ﬁeld in the transform plane aﬀected by the ﬁnite lens diameter? . Strictly speaking. Show. Problem 12. −y). (a) What is the diﬀraction limit that occurs in optical-wave propagation? Describe the eﬀect mathematically.182 Chapter 12. Suppose also that the lenses are of diameter D. y) = (+) k ˜ (+) E 2πf in kx ky . and you can also ignore the eﬀect of the right-hand lens. to see how the image forms in the output plane.

and the coils in the modulator form an impedance-matching circuit to maximize the coupling eﬃciency into the piezo. the other end of the crystal is cut at an angle to avoid a backreﬂected acoustic wave and thus a standing wave. Notice the red piezo transducer on the right-hand side of the crystal. The crystal is surrounded by aluminum to help it dissipate heat from the acoustic wave.Chapter 13 Acousto-Optic Diffraction One important application of diﬀraction occurs when you can make a dynamical diﬀraction grating. this device is useful for modulating the amplitude of a wave (since the “amplitude” of the diﬀraction grating can be controlled electronically). For a monochromatic acoustic traveling wave. TeO2 modulator designed for 780 nm light. This is an 80 MHz. The crystal is the chunk of transparent material on the left-hand side of the modulator. . Typically. The acousto-optic eﬀect is one way to accomplish a dynamical grating. (13. input light q sound wave diffracted beams x z piezo ~ rf drive The sound wave leads to local compressions in the crystal and thus local density variations. δn ≪ n0 .1) where n0 is the natural refractive index of the crystal and δn is the modulation amplitude induced by the acoustic strain. The idea is that a piezoelectric transducer driven by a radiofrequency (rf) wave (typically in the 10 MHz to 1 GHz range) causes an acoustic wave to propagate through an optical crystal. As implied by the name. The rf power enters via the coaxial cable on the right. Generally. A commercial acousto-optic modulator is shown below. This in turn leads to a spatial variation in the refractive index due to the acoustic strain. and notice the angle-cut of the left-hand side of the crystal to suppress the reﬂected acoustic wave. and it turns out that this device can also change the frequency of a laser beam. the resulting refractive index variation is n(x) = n0 + δn cos(krf x − ωrf t).

as we will discuss below). We will ﬁrst treat the simpler Raman–Nath regime. the beam is diﬀracted into several spots.1 Raman–Nath Regime There are distinct regimes of acousto-optic diﬀraction. In the far-ﬁeld. 13. but in this regime and alignment multiple diﬀracted orders are visible. Acousto-Optic Diffraction This photo shows the same modulator passing a He-Ne laser beam. we don’t have to worry about the fact that the beam . where the acoustic wave is conﬁned to a small width ∆z. What this means is that we can treat the eﬀect of the modulator as simply leaving a phase imprint on the beam.184 Chapter 13. That is. essentially depending on the optical path length of the crystal. This modulator is designed to maximize the power in a single diﬀraction order (operating in the Bragg regime. as if it were an aperture.

13.1 Raman–Nath Regime

185

diﬀracts a bit from the front of the crystal before getting to the back of the crystal. Thus, this problem is essentially that of computing the Fraunhofer diﬀraction pattern due to a sinusoidal phase grating, although we will be a bit more careful and go beyond the paraxial approximation. The phase imprint due to the acoustic wave is just eik·∆r , where ∆r is the path length through the acoustic wave and the dependence on the refractive index is implicit in the longitudinal spatial frequency kz . We can write this out explicitly: exp(ik · ∆r) ≈ exp ik∆z [n0 + δn cos(krf x − ωrf t)] , n0 cos θ (13.2)

where θ is the incident angle (with respect to the z-axis), and the factor of cos θ accounts for the fact that the acoustic beam “looks” wider to the optical wave when it propagates through at angle θ from the normal (in the past, we ignored this in the paraxial approximation). Again, this is an approximation assuming that ∆z is small, so that the diﬀraction grating acts as a thin optic. It is also important to note the convention here: we are doing all the diﬀraction analysis inside the crystal, so k and λ refer to the spatial frequency and wavelength inside the crystal, so that they are changed by a factor of n0 and 1/n0 from their respective vacuum values. Also, θ refers to the incident angle inside the crystal. Generally it is most useful to analyze the incident and diﬀracted beams outside the crystal, which requires a simple application of Snell’s law. We can now write Eq. (13.2) in two parts as exp(ik · ∆r) ≈ exp ik∆z δn ik∆z exp cos(krf x − ωrf t) . cos θ n0 cos θ (13.3)

The ﬁrst factor is just an overall phase, which we can discard. We can deﬁne K := as the amplitude of the phase imprint, so that exp(ik · ∆r) ≈ exp [iK cos(krf x − ωrf t)] . Now we need to break this phase imprint into simpler components. One important identity is the generating function for the ordinary Bessel functions: x exp 2 1 t− t =

∞ j=−∞

k∆z δn n0 cos θ

(13.4)

(13.5)

Jj (x)tj .

(13.6)

This function generates the Bessel functions Jn (x) in the sense that Jn (x) is the coeﬃcient of tn in the power-series (Laurent) expansion. Letting t −→ ieix and x −→ K, we obtain a useful identity: eiK cos x =

∞ j=−∞

ij Jj (K)eijx .

(13.7)

Essentially, we just computed the Fourier series of eiK cos x . (+) If the input is the plane wave Ein exp[i(k·r−ωt)], then after the phase imprint, we can use Eqs. (13.5) and (13.7) to write the eﬀect of the diﬀraction as Ein exp[−(k · r − ωt)] −→ Ein exp[i(k · r − ωt)]

(+) (+) ∞ j=−∞

ij Jj (K) exp[i(krf x − ωrf t)].

(13.8)

Again, we have technically just computed the Fraunhofer pattern, taking into account a diﬀerent incident angle, but we will improve on this in a moment.

186

Chapter 13. Acousto-Optic Diffraction

**13.1.1 Diffraction Amplitudes: Bessel Functions
**

Each component in the sum corresponds to a diﬀerent diﬀracted wave. The jth wave, or the jth order diﬀracted wave, has an amplitude of Jj (K), ignoring an overall phase. Each wave has a diﬀerent wave vector and thus propagates in a diﬀerent direction. We will analyze this fact more carefully below, but for now we will simply note that this means that it makes sense to compute the intensities of each diﬀracted wave separately. We are treating the waves as plane waves, but really they are beams with ﬁnite diameter, and we will generally be interested in the intensities of the waves in the far ﬁeld when they are spatially separated. The intensity of the jth diﬀracted wave is thus given by |Ej |2 Ij = (+) = Jj2 (K). Iin |Ein |2 To get a feeling for what this means, we need to explore the Bessel functions a bit more. For our purposes here, we can deﬁne the Bessel function of order ν via the integral (x/2)ν Jν (x) = √ πΓ(ν + 1/2)

π (+)

(13.9)

**cos(x cos θ) sin2ν θ dθ,
**

0

(13.10)

for ν > −1/2, and where Γ(ν) is the Gamma function (which satisﬁes Γ(n + 1) = n! for integer arguments). In Eq. (13.9), we only need the Bessel functions of integer order n, where the integral simpliﬁes to Jn (x) = 1 π

π 0

cos(x sin θ − nθ) dθ.

(13.11)

The Bessel functions satisfy the recurrence relations Jν+1 (x) = 2ν Jν (x) − Jν−1 (x); x ν ′ Jν (x) = Jν−1 (x) − Jν (x). x (13.12)

For integer orders, it is handy to know that Jn (x) = (−1)n J−n (x), and for real x, we can also write Jn (−x) = (−1)n Jn (x). For small arguments, the Bessel functions have the form Jn (x) ≈ xn 2n n! (13.13)

for n ≥ 0. In particular, J0 (0) = 1 and Jn=0 (0) = 0. As x increases from zero, the higher-order Bessel functions “turn on later” compared to their low-order counterparts. The Bessel functions oscillate, and as x becomes large, they decay as x−1/2 . The ﬁrst few ordinary Bessel functions are plotted here.

13.1 Raman–Nath Regime

187

1 noo=oo0 noo=oo1 0.5 noo=oo2 noo=oo3 noo=oo4

Jño(x)

0 -0.5 0

5

10

x

15

20

**For ﬁxed argument x and large order ν, the Bessel functions are approximately 1 Jν (x) ≈ √ 2πν ex 2ν
**

ν

.

(13.14)

In particular, notice that for large order ν, the amplitude decays exponentially with ν. We can conclude that as a function of n, Jn (x) has possibly large amplitude (oscillating with n) until n exceeds x. Then, the amplitude of Jn (x) drops quickly oﬀ to zero. Here is a plot of Jax (a) for several diﬀerent values of a. Whenever x exceeds 1, the value of the Bessel function drops rapidly oﬀ, as we expect.

0.5

**aoo=oo5 aoo=oo10 aoo=oo20 aoo=oo100 aoo=oo500 Jå≈o(a)
**

0

-0.5

0

x

1

2

188

Chapter 13. Acousto-Optic Diffraction

The point of all this Bessel stuﬀ is this: Practically, for Raman-Nath diﬀraction, the maximum diﬀraction order that contains substantial power is K. So we expect to see about 2K + 1 spots in the diﬀraction pattern of the device.

**13.1.2 Frequency Shifts
**

The jth order diﬀracted wave has a time dependence of the form exp(−iωj t), where the frequency is ωj = ω + jωrf . (13.15)

Thus, each diﬀracted wave also experiences a frequency shift. There are two intuitive interpretations of this shift. The ﬁrst is the view that the diﬀraction is a nonlinear scattering process of photons from a stream of phonons. A phonon is the sound-wave analog of a photon, or in other words a quantum of the acoustic ﬁeld. Diﬀraction into the +|j| order corresponds to the absorption of j phonons by the photon, hence raising the photon energy. Diﬀraction into the −|j| order, on the other hand, corresponds to the stimulated emission of j phonons by the photon, hence lowering the photon energy. Thus, we can view the frequency shift as a conservation of energy principle, where part of the energy is carried by the phonon. The other interpretation is purely classical: The diﬀraction grating is moving, and thus the deﬂected (diﬀracted) beams are frequency-shifted due to the Doppler eﬀect.

**13.1.3 Momentum Conservation
**

Now we need to examine the spatial dependence of the diﬀracted waves. From Eq. (13.8), the wave vector of the jth diﬀracted wave is kj = k + jkrf , (13.16) where krf = krf x. Since the momentum of a photon is proportional to its wave vector, we can regard this ˆ relation as conservation of momentum. This is a vector relation, though so it is a bit more complicated than conservation of energy.

k q k∆ jokrf

Dq∆

x z

Note that in this diagram, jkrf points opposite krf , so this corresponds to some negative diﬀraction order. Optical and acoustic frequencies are very diﬀerent; optical frequencies are around 1014 Hz, while acoustic (rf/microwave) frequencies are in practical limited to the few GHz range, and are much more commonly around 100 MHz. Thus, ωrf ≪ ω, (13.17)

so while relative frequency shifts are observable in very sensitive setups, for our purposes here we can write for absolute frequencies ωj ≈ ω, (13.18)

which implies that the spatial frequency of the diﬀracted wave is equal to that of the incident wave to very good approximation: kj ≈ k. (13.19) Thus, the momentum-vector triangle must be isosceles, so that k and kj are symmetrical about the z-axis.

13.2 Bragg Regime

189

Dq∆

k q k∆ jokrf

x z

From the diagram, we can see that sin or

∆θj krf = , 2 2k λ krf , = 2 sin−1 2k 2λrf

(13.20) (13.21)

∆θj = 2 sin−1

where λrf = 2π/krf is the wavelength of the rf acoustic wave. Now, it’s important to realize that from the argument we just presented, the statement krf = krf x ˆ (13.22)

can’t be quite true, even though the acoustic wave propagates in the x-direction. If we assume that kj ≈ k to conserve energy, the only way to satisfy momentum conservation (i.e., to get the triangle to close) is to have the isosceles triangle shown above. But if krf = krf x, then for any given input angle θ, there can be at ˆ most one diﬀraction order that closes the momentum triangle. In particular, this can only be true for the jth order if ∆θj θ=− , (13.23) 2 so that ∆θj . (13.24) θj = θ + ∆θj = 2 This seems to imply that there is only one diﬀracted beam. However, in the Raman–Nath regime, many diﬀraction orders are observed. How is this possible? Well, we started oﬀ with the assumption that the acoustic wave was conﬁned to a very narrow beam. The acoustic wave is subject to diﬀraction eﬀects, so the narrow spatial proﬁle implies that it has a large spread in the direction of krf , just as it would if it were an optical wave—diﬀraction patterns from smaller apertures are larger in the far ﬁeld for precisely this reason, and this is another manifestation of the uncertainty relation δx δkx ∼ 1 that we saw in the discussion of Fraunhofer diﬀraction. Since there are many diﬀerent available orientations of krf available, conservation of energy and momentum “pick out” the right phonon that closes the triangle. It is the thinness of the acoustic beam that makes the proper phonons available for many diﬀraction orders.

Dq∆

k jokrf

x z

k∆

The length of krf is the same; it’s only the orientation that changes, so for “nonresonant” diﬀraction orders, the momentum triangle has the same shape, but the entire thing is rotated appropriately.

**13.2 Bragg Regime
**

The Bragg regime of acousto-optic diﬀraction is the opposite extreme from Raman–Nath diﬀraction. Instead of a narrowly conﬁned acoustic wave, in the Bragg regime, we assume that the acoustic wave is very broad but relatively weak. In this case, the acoustic wave is approximately a plane wave, so that krf

dz 2 = 0. Expanding the left-hand side. There is an interpretation to this. ∂z (13. It is thus desirable to have as much modulation range as possible as well as to not waste too much of the incident beam.. we can write E (+) (z + dz) = E (+) (z) + ∂E (+) dz. we can similarly write ei dφ E (+) (z) = E (+) (z)[1 + i dφ]. x z ~ We will thus consider only ﬁrst-order diﬀraction. θ = θB := λ ∆θ1 = sin−1 2 2λrf (13.28) . (13. which is as follows: each of the “bumps” in the refractive-index proﬁle reﬂects a little bit of the light in the opposite direction. when the transverse optical wavelength exactly matches the acoustic wavelength): these larger angles correspond to higher diﬀraction orders. The arguments of the last section hold here. since nearly all of the incident wave can be diﬀracted into a single order. since the diﬀracted beam is the intensity-modulated beam. Recall that forward propagation over a distance z is equivalent to multiplication by exp(ikz z). whereby if the transverse optical wavelength matches the grating wavelength. that is. Expanding the exponential in the right-hand side. Thus. If the small reﬂections from successive bumps interfere constructively.190 Chapter 13. eﬃcient diﬀraction occurs. We will want to be a little more general than this. Simplifying the setup from the Raman–Nath discussion above. Of course this argument works for larger angles.27) (13. This turns out to be a useful fact. Acousto-Optic Diffraction has a well-deﬁned direction. when the transverse optical wavelength is smaller (e. k qb k¡ qb qb kº k rf x z The diﬀracted wave makes an angle 2θB with respect to the incident wave. so we can already conclude that diﬀraction can only occur into a single order. Notice that the Bragg condition is equivalent to the following condition: the variations in the optical wave in the direction of the acoustic wave must have twice the wavelength of the acoustic wave. we see that the incident angle must satisfy the Bragg condition.25) Again. Diﬀraction to other orders in this geometry is suppressed.g. For this reason most acousto-optic modulators in laboratory use are Bragg-regime devices. the array of bumps acts as a sort of generalized Fabry–Perot cavity.26) since for inﬁnitesimals. then a lot of the light can be reﬂected in the opposite direction. the ﬁeld picks up some phase change dφ: E (+) (z + dz) = ei dφ E (+) (z). diﬀraction in this conﬁguration corresponds to absorbing a single phonon. so let’s say that when you propagate the wave forward in the z-direction from z to z + dz. The condition for this is that one round-trip between two bumps is 2π in phase—one round trip in the direction of the acoustic wave.

since they don’t satisfy the Bragg condition and thus they shouldn’t be excited. ∂z ∂z (13.13.1 ei(krf x−ωrf t) + e−i(krf x−ωrf t) . (13. but we will consider this more general setup to handle angular misalignment.1 n0 cos θ0.34) Now let’s consider the evolution of the zeroth and ﬁrst diﬀraction orders.1 2n0 cos θ0. This is ﬁne as long as there isn’t anything like a dielectric interface that would generate a reverse wave. the phase shift becomes ∂φ0. We will also deﬁne the Rabi frequency Ω := k δn . We’ll denote the zeroth-order wave by . the Bragg condition is |θ0 | = |θ1 | = θB for eﬃcient diﬀraction. we can write the phase change as dφ0. we can write Eq.x k ≈k+ . but this equation is ﬁrst-order: it implicitly considers only forward-propagating waves (kz > 0). Thus. k qº k¡ q¡ qº kº k rf x z From our discussion above. Note that we will want to keep track of the phase evolution along each of the diﬀracted waves.26) in diﬀerential form as ∂φ (+) ∂E (+) =i E . Let’s simplify this by deﬁning the phase-rotation rate κj := k = cos θj 2 kj.1 = n n k dz k dz k dr0.1 . n0 cos θj (13.1 . However. any diﬀerence in the Rabi frequencies is a much smaller eﬀect and thus ignorable. this evolution equation is the same as the wave equation. Now let’s calculate the phase change dφ as the wave propagates in the acousto-optic medium.x /k)2 (13.1 is dz/ cos θ0. Recalling that k = k0 ≈ k1 . The two angles will still be subject to the constraint |θ0 | + |θ1 | = 2θB . the path length along k0.30) where dr0. Then k δn k ∂φ0.32) where the last expression holds in the paraxial approximation. As z −→ z + dz. 2k 1 − (kj.1 Ω i(krf x−ωrf t) e + e−i(krf x−ωrf t) .1 are the angles made by k0.1 from the z-axis. whereas the full wave equation handles propagation in any direction. We will see that even small diﬀerences in the two angles have a substantial impact on the diﬀraction via the κj .33) In principle the Rabi frequency is diﬀerent for each wave. and the third term represents coupling of diﬀraction order j −→ j − 1.1 corresponding to a displacement dz along z. but since δn/n0 is small. so we should keep track of two Rabi frequencies Ω0.1 = = [n0 + δn cos(krf x − ωrf t)] . where θ0.1 + ∂z 2 (13.29) In principle. + = ∂z cos θ0.31) The ﬁrst term here represents free-space evolution in the absence of the acoustic wave. the second term represents coupling of diﬀraction order j −→ j + 1. = κ0. We’ll ignore the others.1 (13. With these deﬁnitions.1 (13. we will assume that |θ0 | is very close to |θ1 |.1 is the path length along k0.2 Bragg Regime 191 since dφ2 = 0. n0 n0 cos θ0. so we need to compute the phase evolution along the wave vectors k0 and k1 .

36) where ωj = ω0 + jωrf . reducing the wave equation to a pair of ordinary diﬀerential equations. z) = iκ0 E (+) (k0 . Diﬀerentiating the ∂z c0 equation. 2 (13. As long as we keep this in mind. z) e−iω0 t + iκ1 E (+) (k1 . by multiplying through by exp(iω0 t) and integrating over a time interval of length 2π/ω0 ) and discarding the common time dependence. (13. so all terms in Eq. ∂z 2 Similarly. the factors of exp(±ikrf x) account for the momentum change on absorbing or emitting a phonon.39) where ∂z ≡ ∂/∂z. z) + E (k0 .192 Chapter 13. 2 Using the other equation for ∂z c1 . (13. iΩ (+) ∂ (+) E (k1 .29) becomes ∂ E (+) (k0 .38) have the common dependence of exp(ik1 · r). Ω Ω ∂z2 c0 = iκ0 ∂z c0 + i (iκ1 c1 ) + i 2 2 2 c0 .g. z) eikrf x e−iω2 t + 2 iΩ (+) E (k0 . we’ve made things much simpler. z) e−iω0 t + E (+) (k1 . z) e−ikrf x . Deﬁning the normalized electric ﬁelds. Notice that E (+) (kj .41) ∂z2 c0 = iκ0 ∂z c0 + i ∂z c1 . Thus.38) become Ω ∂z c0 = iκ0 c0 + i c1 2 Ω ∂z c1 = iκ1 c1 + i c0 . z) eikrf x e−iω1 t + 2 iΩ + E (+) (k1 .. 2 (13. z) e−ikrf x e−iω−1 t 2 iΩ (+) E (k1 . z) ∼ exp(ikj · r) and kj = k0 + jkrf x. we obtain the equation of motion ∂ (+) iΩ (+) E (k0 . z) eikrf x . z) Ein (+) . and all terms in Eq. z) and the ﬁrst by E (+) (k1 . (13. We can solve these coupled equations by decoupling them. (13. where the wave vectors are given by kj = k0 + jkrf = k0 + jkrf x.40) E (+) (kj . z) e−iω1 t .42) . t) = E (+) (k0 .37) have the ˆ common dependence of exp(ik0 · r). z) e−iω0 t + E (+) (k1 . ˆ The total ﬁeld is the sum of the two waves. Acousto-Optic Diffraction E (+) (k0 . the evolution equation (13. z) e−ikrf x e−iω0 t . and with the explicit time dependence we can write E (+) (z.37) In these two equations. z) e−iω1 t = ∂z iκ0 E (+) (k0 .35) Note that we’ve already included the exp(ikj · r) in the z-dependence of E (+) (kj . cj (z) := Eqs.37) and (13. (13. z) = iκ1 E (+) (k1 . Picking out the frequency component of this equation at ω0 (e. z) + E (k1 . z). (13. ∂z 2 (13.38) (13. so we haven’t written it out explicitly here. Then recalling that the phase evolution φ depends on the wave vector. we don’t have to write out the explicit phase dependence. since what we’re really after is the relative amplitudes of the two waves. Ω (13. z) e−iω1 t + iΩ (+) E (k0 . z). for the component at frequency ω1 .

13.51) Since we’re only interested in relative amplitudes. We can factor Eq. (13.47) c1 (z) = A+ eiΩ+ z/2 + A− eiΩ− z/2 .48) where A± are constants to be determined by the initial conditions.46) (13. 2 while diﬀerentiating Eq. we can equivalently write ˜ ˜ Ωz Ωz + A2 sin . (13.49) gives ∂z c1 (z = 0) = ˜ so that A2 = iΩ/Ω. from the ∂z c1 equation in Eq.2 Bragg Regime 193 and using the ∂z c0 equation to eliminate c1 . 2 2 (13. the solution is c1 (z) = iei(κ0 +κ1 )/2 Ω sin ˜ Ω ˜ Ωz 2 . (13. Thus. we can obtain |c1 | from the conservation constraint |c0 |2 + |c1 |2 = 1.53) where of course I1 + I0 = Iin . (13. (13.52) ˜ ΩA2 . we can see that A1 = 0. ˜ Ω := (κ0 − κ1 )2 + Ω2 . we ﬁnd ∂z2 c0 − i(κ0 + κ1 )∂z c0 − κ0 κ1 − Ω2 4 c0 = 0.45) ˜ Ω± = (κ0 + κ1 ) ± Ω. (13.44) Let’s concentrate on obtaining an explicit solution for c1 (z). Thus. The intensity of the diﬀracted beam is given by I1 = |c1 (z)|2 = Iin Ω ˜ Ω 2 sin2 ˜ Ωz 2 . (13. (13. so c0 (z = 0) = 1 and c1 (z = 0) = 0. Note that given the form of Ω± . upon entering the sound ﬁeld.49) c1 (z) = ei(κ0 +κ1 )/2 A1 cos 2 2 where A1.2 are (other) constants to be determined by the initial conditions. (13. Initially.50) (13. Now we are in a position to make a few observations about acousto-optic Bragg diﬀraction . all of the incident light is undiﬀracted.45) is Ω+ 2 ∂z − i Ω− 2 c1 = 0 (13.54) ˜ In this form we can see that the intensity oscillates sinusoidally with frequency Ω.40) gives Ω ∂z c1 (z = 0) = i . Also. 2 (13. (13. (13.43) Repeating this procedure yields the same equation of motion for c1 : ∂z2 c1 − i(κ0 + κ1 )∂z c1 − κ0 κ1 − Ω2 4 c1 = 0.44) and write it as ∂z − i where the quadratic roots are given by where The general solution to Eq. We can also write this in the form I1 = |c1 (z)|2 = Iin Ω ˜ Ω 2 1 1 ˜ − cos Ωz .

x |.194 Chapter 13. kº qb k qb qb k–¡ 5.55) = 2 ˜ Ω + (κ0 − κ1 )2 Ω2 Since this amplitude ≤ 1. The intermediate waves are not resonantly coupled.x |.53) is Ω2 Ω2 . but the analysis works just as well for diﬀraction into the −1 order. Resonant selection. but in particular this is equivalent to a two-level atom (the two optical modes) driven by a classical laser ﬁeld (the sound ˜ wave. so the resonance condition occurs when − k0. (13. The (peak-to-peak) amplitude of the intensity oscillations from Eq. we can see that for best diﬀraction eﬃciency there is an optimum rf drive power for a given crystal length.x | = |kj. and Ω is the generalized Rabi frequency. 6. But ∆kx = krf for ﬁrst-order diﬀraction. The intensity oscillates between the incident and diﬀracted wave at rate Ω. 4. there are incomplete Rabi oscillations unless the resonance condition κ0 = κ1 is satisﬁed. Higher-order scattering. Acousto-Optic Diffraction ˜ 1.. Resonance condition. which in turn requires that the zeroth and jth order diﬀracted waves to subtend equal angles about the normal to krf .58) (13. Then the diﬀraction eﬃciency becomes I1 = sin2 Iin Ωz 2 = sin2 k δn z . but higher-order Bragg diﬀraction is also possible by absorbing or emitting multiple phonons. so this process proceeds exponentially more slowly (in terms of interaction length for ﬁxed δn) than ﬁrst-order diﬀraction. Thus. Rabi ﬂopping. We did the analysis for the +1 diﬀraction order. With this analogy. Expanding this further. practical devices usually stick to ﬁrst-order diﬀraction.56) Since the index modulation amplitude δn is related to the rf drive power.x = k1. in this case).57) (13. This behavior (oscillation of energy between two coupled oscillators) occurs often in physics. this condition is equivalent to |k0.x | = |k1. . the diﬀerences being the orientation and that the diﬀracted beam is downshifted in frequency instead of upshifted like the +1 order. so we need a particular input angle θ0 . On resonance (κ0 = κ1 ). We only considered ﬁrst-order diﬀraction. Only one order can be on resonance (i.x = krf /2. of course.e. (13. the intensity oscillations are Rabi oscillations. satisfy the Bragg condition) at a time. Mirror image. Diﬀraction to the jth order requires that |k0. Putting in the deﬁnitions of the κj . ˜ 3. Resonant Rabi ﬂopping. Again krf is well-deﬁned.59) (13. the generalized Rabi frequency Ω reduces to the ordinary Rabi frequency Ω. 2n0 cos θB (13. 2. just leads to the Bragg condition θ0 = θB = sin−1 (λ/2λrf ).

13. (13. The crystal dimensions are z = 15 mm thickness and h = 1 mm height.63) Iacoustic = zh where Pacoustic is the total power in the acoustic wave. (13.8 W. the ﬁgure of merit is M = 34.62) to write an expression for the optimal acoustic intensity. For TeO2 . so after exiting the crystal.64) θB ≈ 4πvrf The deﬂection angle of the diﬀracted beam is 2θB = 1. we’ll look at a commercial TeO2 (“tellurium dioxide”) modulator designed to work with a 150 MHz rf drive at 780 nm (vacuum) optical wavelength. which is of the same order of magnitude. the index modulation amplitude is given by λ δn ≈ = 1.60) = sin z = sin Iin 2n0 cos θB n0 λ cos θB n0 λ 2 cos θB Now we can translate the index modulation δn into properties of the crystal and acoustic wave. (13. We will use the paraxial approximation throughout.1 W. We can write 2(δn)2 M Iacoustic := (13. (13.2 Example: TeO2 Modulator (Bragg Regime) To get a feeling for the numbers. we can use Eq. (13. optimum = λ2 .2 Bragg Regime 195 13.6◦ . Maximum diﬀraction eﬃciency occurs when I1 = sin2 (π/2).2. Thus.35.59): √ I1 πz k δn π δn z δn 2 2 2 2 √ = sin .66) (13. we can write the acoustic intensity as Pacoustic . (13.79◦. Recall that all geometrical quantities refer to their values inside the crystal. (13. (13. Pacoustic.7◦ .1 Efﬁciency We can manipulate the form of the (optimally aligned) diﬀraction eﬃciency in the Bragg regime from Eq. Iin Since cos θB ≈ 1.1 × 10−5 .62) Noting that the dimensions of the acoustic wave are the interaction length z and the crystal height h. but implies that much power is wasted in the piezo transducer in converting electrical to acoustic energy. (13. the deﬂection angle is 3. optimum = or equivalently for the optimal power λ2 h = 0.2.67) 2zM The actual power applied to the piezo for the best diﬀraction eﬃciency is actually 0. 13.60) to see that at maximum diﬀraction eﬃciency.5 × 10−15 s3 /kg. and the refractive index is n0 = 2. but its eﬀect builds up coherently over the optical path length of the modulator.68) n0 2z This is a very small index change. The Bragg angle is λωrf = 14 mrad = 0. the acoustic-wave velocity is vrf = 4. Iacoustic.65) . n0 where Iacoustic is the intensity of the acoustic wave and M is the ﬁgure of merit that characterizes the acousto-optic crystal. We can also use Eq. we can write the eﬃciency as I1 πz = sin2 √ Iin λ 2 cos θB M Iacoustic . 2z 2 M (13.26 km/s.61) 2 .

whose refractive index is modulated by an applied ac signal with frequency ωmod . which is some intermediate regime where neither treatment is really appropriate. by running the wave through an electro-optic crystal. Bragg (small angular spread of phonons): Q ≫ 1 The borderline is usually taken to be from 1 to 2π. for example. 2πvrf (13. we can deﬁne the instantaneous frequency as ωinst := dφ .4 Exercises Problem 13. Raman–Nath (large angular spread of phonons): Q ≪ 1 2. "Uniﬁed approach to Ultrasonic Light Diﬀraction. so this device is clearly in the Bragg regime. . (a) For a wave with time dependence exp[−iφ(t)]. 1 W. with the general form ∞ j=−∞ cj ei(kx−ωj t) . dt (13. Q = 1 is a practical dividing line between the two regimes. (13. Klein and B. For the TeO2 modulator in the previous example. But how do we distinguish which regime applies to a particular device? Recall that what distinguishes the two regimes is whether or not the direction of krf is well-deﬁned. That is. which says whether or not diﬀraction to many orders is allowed. Cook. Then we can distinguish the two regimes as follows: 1.69) where we used the paraxial form for θB and again vrf is the phase velocity of the acoustic wave in the medium.72) Hint: start by writing down a Bessel series for the function exp(iK sin x). Acousto-Optic Diffraction 13." IEEE Transaction on Sonics and Ultrasonics.1. phase modulation is in some sense the same as frequency modulation. One approach is to examine the Klein–Cook parameter1 Q := = ≈ = = (angular spread scale of diﬀracted beams) (angular spread of rf phonons) (2θB ) (1/krf z) (krf /k) 1/krf z 2 krf z k 2 λωrf z 2 . 123 (1967). (+) (13.71) Compute the instantaneous frequency of the phase-modulated wave and thus show that the frequency oscillates about ω. However. 13. SU-14.196 Chapter 13. t) = E0 ei[kx−ωt+δφ sin(ωmod t)] . Q = 39.3 Borderline We have discussed the operation of acousto-optic modulators in the Raman–Nath and Bragg regimes. A phase-modulated optical wave has the form E (+) (x. (b) Write the phase-modulated wave as a sum of plane waves.70) where ωmod is the modulation frequency. Such a wave could result.

5. note that the best diﬀraction eﬃciency is given by Ω2 /Ω2 . Problem 13. This drops to 1/2 when (κ0 − κ1 ) = Ω2 . argue that the intensity spectrum (as viewed through a Fabry–Perot spectrum analyzer) consists of a series of peaks with relative intensity Jj2 (δφ). assuming that the piezo transducer is suddenly switched on? Assume that the only contribution to the rise time is due to the ﬁnite propagation velocity of the acoustic wave front. This phasemodulation technique is commonly used in the laboratory to shift the frequency of a laser or to generate multiple laser frequencies. Now draw out the diﬀraction geometry for the misaligned case. what would the index modulation δn/n0 need to be to achieve this maximum intensity? (c) What would the index modulation δn/n0 need to be to completely deplete the input beam (zeroth order)? For parts (b) and (c) you will need to consult a table of Bessel functions or use some mathematical software. what is the new diﬀraction angle? (b) What is the rf power required for most eﬃcient diﬀraction at the new wavelength? (c) The coupled-wave theory of Bragg diﬀraction predicts that the best diﬀraction eﬃciency is 100%. Assume the modulator is made from TeO2 . Consider a Gaussian beam focused to a waist w0 in a Bragg-mode acousto-optic modulator. (b) What is the maximum achievable relative intensity in any diﬀracted order (other than the zeroth)? This is why Bragg modulators are potentially much more eﬃcient at diﬀracting light into a single order.1 mm. You may assume the response of the Fabry–Perot analyzer is slow compared to the modulation frequency. However.8 W of input rf power. (a) If the modulator is 80 MHz. Use these to eliminate κ0. what is the maximum thickness of the crystal in the direction of beam propagation? Assume the beam is normal to the acoustic wave. noting that the total diﬀracted angle is still 2θB .4. What is the turn-on time of the diﬀracted beam. Give an example of something we didn’t take into account that could explain the lower practical eﬃciencies of acousto-optic modulators.2. in practice the best eﬃciencies are more like 90%. (a) If 850 nm light is used instead. and this is the part of the eﬃciency that 2 is sensitive to angle.4 Exercises 197 (c) From your answer to (b). Suppose you observe that for a particular acousto-optic modulator in the Bragg regime. Problem 13.x /k| ≈ |θB − δθ| and |k0. The most eﬃcient diﬀraction is observed for 0. with an angle of 2◦ between the undiﬀracted and ﬁrst-order beams. In the Bragg regime.3. so obviously we made some assumptions that aren’t quite true. and ignore any distortion of the wave front due to dispersive . Assuming z = 0. what is δθ? causes the diﬀraction eﬃciency to drop by a factor of 2.x /k| ≈ |θB + δθ|. Problem 13.13. Problem 13. (a) Show that for small θB .1 and obtain the result above. Consider an acousto-optic modulator operating in the Raman-Nath regime at λ0 = 780 nm (vacuum) wavelength. You can use the following general outline: ˜ First. 780 nm light (wavelength in air) diﬀracts eﬃciently into the ﬁrst order. the optimum diﬀraction eﬃciency occurs when the incident angle is θB from the normal to the acoustic-wave propagation direction. an angular misalignment of δθ = δn 2n0 θB (b) For the 150 MHz TeO2 modulator that we discussed as an example in class. so that |k1.

198 Chapter 13. (You need not compute any particular parameter.7. Describe how you can use an acousto-optic modulator as a spectrum analyzer for an rf signal. A glass acousto-optic modulator (n0 ≈ 1.5) diﬀracts He–Ne laser light (vacuum λ ≈ 200π nm).6. Problem 13.) (c) Suppose that you observe that I0 ≈ J02 (17). just go through the reasoning for what distinguishes the Raman–Nath and Bragg regimes. However. Assume that the modulator operates in the regime where there are many diﬀracted peaks. you are measuring this angle outside the modulator). . Estimate the number of diﬀracted beams you would expect. Problem 13. Acousto-Optic Diffraction propagation in the crystal. How did you get yourself into this situation. Also describe how set up the analyzer to maximize the frequency resolution. anyway? (a) What is the Bragg angle θB ? (b) Under what conditions would you expect there to be many diﬀracted beams? Give a rough range of optical path lengths of the crystal that would lead to many diﬀracted beams. You may assume that any diﬀraction angles are small. Iin where I0 is the the intensity of the zeroth diﬀraction order and Iin is the input intensity. and that the acoustic-wave velocity is independent of frequency. deﬁne the rise time as the time it takes the diﬀracted beam to go from 10% to 90% of its maximum intensity. Suppose that you observe that the angle between the zeroth and ﬁrst diﬀracted order is 3◦ (obviously. some diabolical ﬁend has arranged it such that you cannot tell if any other diﬀracted orders are present. In order to save the world. you must ﬁgure out whether or not the modulator diﬀracts substantial power into other orders. For concreteness.

3) The optical terms of the form exp(±i2ωt) vanish in the time average. . Then φ = −ωτ . and so Isum = I1 + I2 + 1 (−) (+) −iωτ E E e + c. Suppose that the phase diﬀerence represents a diﬀerence in optical path length. so we obtain Isum = 2 (−) (+) 2 (−) (+) 1 (−) (+) iφ 1 (−) (+) −iφ E E + E20 E20 + E10 E20 e + E20 E10 e η 10 10 η η η 1 (−) (+) iφ E E e + c. η 10 20 (14. we’ll assume that the two waves have equal amplitude and come from a common source. we made the comment that assuming a monochromatic wave with time dependence of the form exp(−iωt) is implicitly computing a Fourier transform.c. . Isum = 1 η E10 ei(kz−ωt) + E20 ei(kz−ωt) eiφ + E10 e−i(kz−ωt) + E20 e−i(kz−ωt) e−iφ (+) (+) (−) (−) 2 1 2 (Esum ) .c. To simplify things. Recall that we are just writing the positivefrequency components. in the form of a time delay τ . Now that we have the proper mathematical machinery in place. η (14. The intensity of the superposition is Isum = |Esum × Hsum | = Writing this out explicitly in terms of the component ﬁelds. (+) (+) (14.c. the interference is in the terms with the relative phase φ. t) = E10 ei(kz−ωt) + E20 ei(kz−ωt) eiφ . we can write the superposition of the two waves as (+) Esum (r. so that the physical ﬁelds also must include the negative-frequency parts. . Recalling the interference of two monochromatic plane waves.6) . we will explore this idea in more detail.Chapter 14 Coherence When considering interference of monochromatic waves in Chapter 5.5) Now let’s handle the case of multiple frequencies. η 0 (14. = I1 + I2 + η 10 20 (14. Then the intensity density at frequency ω is Isum (ω) = 2I(ω) + 1 (+) |E (ω)|2 e−iωτ + c. Let’s now consider an interference experiment that involves a range of frequencies. φ is a relative phase diﬀerence between the two waves. (14.4) Again.1) Here.2) .

(14. it compares f to itself but shifted by an amount x by computing an overlap integral. (14. and so |F |2 is the energy spectral density of f . There is one subtle point to these deﬁnitions: for some signals. Essentially.9) becomes ∞ −∞ ∞ −∞ f ∗ (−x)e−ikx dx = ∞ −∞ f ∗ (x)eikx dx = (F [f (x)])∗ . the energy spectrum is the square of the usual spectrum. which is common when working with intensities and powers. ∞ −∞ f ∗ (x′ )f (x + x′ ) dx′ = |F [f ]|2 . Now we can sum over all frequencies to ﬁnd the total intensity: Itotal = 0 ∞ Isum (ω) dω 1 Isum (ω) dω + η ∞ 0 (+) |E0 (ω)|2 e−iωτ =2 0 ∞ (14.c. (14. when we don’t know the general form of the electric ﬁeld? 14.7) dω + c. (14. (14.9) If we make the particular choice g(x) = f ∗ (−x). the Wiener–Khinchin theorem states that the Fourier transform of the autocorrelation function gives the energy spectrum. Essentially. Thus the intensity spectrum Isum (ω) is a one-sided spectrum. the frequency-dependent quantities I(ω) and E (+) (ω) don’t have the respective dimensions of intensity and electric ﬁeld.8) f (x′ )g(x − x′ ) dx′ = F −1 [F [f ]F [g]] . (14.14) . with the phase information removed. We can now recognize the (+) second integral in the last expression as a Fourier transform of |E0 (ω)|2 . ∞ −∞ (14.1 Wiener–Khinchin Theorem Recall the convolution theorem for functions f and g from Section 11. rather. Note that the frequency integral ranges only over positive frequencies.13) is the autocorrelation function of f (x).12) f ∗ (x′ )f (x′ + x) dx′ .11) Inverting the transform and letting x′ −→ x′ + x. such as steady optical signals. the correlation integral diverges: ∞ −∞ f ∗ (t)f (t + τ ) dt −→ ∞. Writing out the convolution integral explicitly. Coherence Note that the notation here is a little funny.200 Chapter 14.10) f (x′ )f ∗ (x′ − x) dx′ = F −1 |F [f ]| 2 . But what is this.3: F [f ∗ g] = F [f ]F [g]. I(ω) dω and E (+) (ω) dω are the intensity and electric ﬁeld in the frequency interval between ω and ω + dω. Eq. . we obtain the Wiener–Khinchin theorem: F The function on the left-hand side. we’ve already accounted for the negative frequencies by including the complex conjugate terms. then F [g(x)] = F [f ∗ (−x)] = Thus. We can understand the right-hand side by noting that F [f ] is the spectrum of f . This is consistent with the notion of energy going as the square of a signal amplitude. Thus. ∞ −∞ (14.

we should consider a time average instead of the normal integral.. Denoting this long-time averaging limit as f ∗ (t)f (t + τ ) = f ∗ (t)f (t + τ ) we can thus write the Wiener–Khinchin theorem as 1 F [ f (t)f (t + τ ) ] = lim T →∞ T ∗ T /2 2 T →∞ . long enough to resolve the frequency spectrum). and in this form it is more clear that this is the energy density per unit time.18) f (t)e −T /2 iωt dt .16) Here we have deﬁned a ﬁnite-time Fourier transform by FT [f ] := 1 T T /2 −T /2 f (t)eiωt dt. Then we can extend the averaging time T −→ ∞.19) Again. which gives the total intensity: ∞ 0 I(ω) dω = 2 2 (−) E (t)E (+) (t) = |E (+) (t)|2 . it is conventional to deﬁne a normalized correlation function g (1) (τ ) := so that E (−) (t)E (+) (t + τ ) . (14. (14. (14. or the energy spectral density per unit time. η η (14. Note that from the inverse Fourier transform. this integral is guaranteed to converge. T should be a suitably long observation time (e. (14. f ∗ (t)f (t + τ ) T := 1 T T /2 −T /2 f ∗ (t)f (t + τ ) dt.19) becomes ∞ 0 I(ω)e−iωτ dω = 2 (−) E (t)E (+) (t + τ ) .20) This is because the intensity density I(ω) = |S| is the time-averaged power spectral density of the optical ﬁeld. but it avoids problems with singular spectra. E (−) (t)E (+) (t) ∞ 0 (14. To be physically sensible.2 Optical Wiener–Khinchin Theorem In terms of optical ﬁelds.21) In optics.14. For such signals. as in statistics and other ﬁelds. Now the Wiener–Khinchin theorem says that the Fourier transform of the (time-averaged) correlation function is the power spectral density. (14. the right-hand side is the power spectral density. For an averaging time of T . We can see this from the boundary condition at τ = 0. (14. but it is missing here because it is already implicitly included in I(ω). η (14. the correlation function is independent of t (generally for a suﬃciently long averaging time T ).23) .2 Optical Wiener–Khinchin Theorem 201 In this case. we can write the Wiener–Khinchin theorem as F [ f ∗ (t)f (t + τ ) T] = F ∗ [f ] FT [f ]. the Wiener–Khinchin theorem (14. For a stationary process.15) For bounded signals.22) 2 (−) E (t)E (+) (t + τ ) = η I(ω) dω g (1) (τ ).17) Deﬁning this seems a bit funny. 14. there is conventionally a factor of 1/2π.g.

7). (14. Returning to the interference result of Eq. The idea in FTIR spectroscopy is to use a Michelson interferometer to monitor a stationary input signal E (+) (t).27) (+) 2 (+) (14. (14. say. the correlation function is subject to the constraint g (1) (−τ ) = [g (1) (τ )]∗ . a correlation value of unity indicates perfect correlation.c.c.1 Application: FTIR Spectroscopy One example of where the optical Wiener–Khinchin theorem is very useful is in Fourier-transform infrared (FTIR) spectroscopy. a recording the same spectrum with the same detector on a diﬀraction-grating spectrometer. The idea is then to digitize the output of the photodetector as a function of the mirror displacement. (14. (14.26) . (14. and thus measures g (1) (τ ).2. eﬀectively recording Re[g (1) (τ )].2. and the path-length diﬀerence τ varies with the displacement of one of the mirrors. up to an overall constant. because it turns out that obtaining a spectrum this way gives better rejection of thermal detector noise compared to.c.202 Chapter 14. 14.2 Example: Monochromatic Light As a simple example. this function becomes g (1) (τ ) = e−iω0 τ . That is. vary t E oo(t) o(+) detector (+) ·Eo(-) oo(t)oE o (too+oot)Ò + c. (−) = =2 2+ g (τ ) + c. The correlation function is E (−) (t)E (+) (t + τ ) = E0 In normalized form. let’s compute the correlation function for monochromatic light and verify that the Wiener–Khinchin theorem makes sense. Let’s take a monochromatic wave of the form E (+) (t) = E0 e−iω0 t .) This technique is common in molecular spectroscopy. ∞ 0 1 + Re g (1) (τ ) 14. Computing the Fourier transform of the signal on the computer gives the spectrum I(ω). and so the imaginary parts of g (1) (τ ) don’t contribute to the spectrum. Notice also that because the spectrum I(ω) is real. The interferometer splits and interferes the beam with itself.25) e−iω0 τ . we ﬁnd Itotal = 2 0 ∞ ∞ 0 Isum (ω) dω + I(ω) dω I(ω) dω 1 η ∞ 0 (1) |E0 (ω)|2 e−iωτ dω + c. The photodetector at the output measures the time average of the product of the ﬁelds. Coherence The normalization is such that g (1) (τ = 0) = 1.24) . (Note that I(ω) is real and even. just as in Eq.24).

36) I1 + I2 The important thing to note is that for interference of monochromatic light. (14. a wave with harmonic time dependence leads to a harmonic correlation function. that is. (14.3 Visibility 203 Thus. (14. there is also a matching δ(ω + ω0 ) component. the harmonic time dependence produces a single frequency in the power spectrum. (14.35) For monochromatic light. using Eq. corresponding to the complex conjugate of the correlation function. so the visibility is just the magnitude of the correlation function: V = |g (1) (τ )|. The power spectrum is easy to calculate via the Wiener–Khinchin theorem: F E (−) (t)E (+) (t + τ ) = E0 (+) 2 F e−iω0 τ = E0 (+) 2 δ(ω − ω0 ). the spectrum is F ∗ E (+) (t) FT E (+) (t) = E0 δ(ω − ω0 )E0 sinc [(ω − ω0 )T /2] = E0 (−) (+) (+) δ(ω − ω0 ). The ﬁnite-time transform is FT E (+) (t) = 1 T T /2 −T /2 (+) (+) (+) (14. . For example. For no interference.16).3 Visibility The above example of the correlation function for a monochromatic wave is special in the sense that the correlation function does not decay with τ . the intensity does not vary with phase and so V = 0.28). The normal spectrum is F E (+) (t) = E0 δ(ω − ω0 ). Let’s now compute the power spectrum directly. V = 1. and so Eq. Of course. with our convention of positive and negative frequencies. Partial coherence is represented by intermediate values of the visibility. To quantify this better. the visibility becomes √ 2 I1 I2 V = .24) becomes Itotal = 2 0 g (1) (τ ) = |g (1) (τ )|eiφ(τ ) .31) This result is consistent with Eq. we would run into something bad involving the square of a δ-function. ∞ (14. 14.33) I(ω) dω 1 + |g (1) (τ )| cos φ(τ ) .34) The cosine varies from −1 to 1. and so V = 1.32) where Imax and Imin are the maximum and minimum intensities achieved for phase variations on the order of several π. Imax + Imin (14. (14. For two monochromatic waves of unequal amplitude. (14. because the correlation function is the magnitude of the interference terms. (14. This is important. This is true independent of the phase of the input ﬁeld. the correlation function does not reﬂect any extra phase in (+) E0 . Writing out the explicit phase of the correlation function. complete interference results in intensity variation from 0 to some maximum. Note that without being careful with ﬁnite-time transforms.14. 2 (14. (14.28) Again.29) E0 ei(ω−ω0 )t dt = E0 sinc [(ω − ω0 )T /2] . Thus. this is only true if the amplitudes of the input waves are equal.30) Note that the value of the sinc function is 1 at ω = ω0 . we can deﬁne the fringe visibility as V := Imax − Imin . Actually. the visibility is independent of τ.

.23) gives g (1) (τ ) = 0 ∞ s(ω)e−iωτ dω (14. (Cambridge. (14. (14. Thus.38) (recall that in the ω-t convention. Intuitively. Coherence Length. we can say that the equality is more or less satisﬁed to δωrms δτrms ∼ 1 . these uncertainties can diverge (e. since g (1) (τ ) −→ 0 as τ −→ ∞..37) I(ω) dω form a Fourier-transform pair.39) For many distributions. the monochromatic components tend to dephase. but as a practical relation for the temporal and frequency widths. Let’s deﬁne the average value of the normalized spectrum as s= ¯ 0 1 as ∞ s2 (ω) dω. 7th (expanded) ed.42) in Max Born and Emil Wolf. That is. this is because as light waves with diﬀerent wavelengths propagate.40) and thus see that the “widths” of g (1) (τ ) and s(ω) are inversely related.4 Coherence Time. we can express this as an uncertainty relationship. In the generic case. 2πs(ω) is the second half of the pair. 2 (14. (14. 1999). and Uncertainty Measures However. this is a peculiarity of monochromatic light. The uncertainty inequality is always satisﬁed. as in quantum mechanics: δωrms δτrms ≥ 1 . The problem is that for some useful functions in optics. 2 (14. Coherence 14.20) with Eq. this is less useful. the visibility drops as the phase diﬀerence increases. δτ is the width of a unit-height box signal with the same power as g (1) (τ = 0) = 1. |g(1)o(t)|2 o same area 1 t dt We’ll take a slightly diﬀerent convention for the width of the frequency spectrum. since the inverse transform of Eq. there is normally a factor of 1/2π in front of the inverse-transform integral). Actually. Thus we will adopt other uncertainty conventions in time and frequency1 We can deﬁne the coherence time as the power-equivalent width of the correlation function: δτ := ∞ −∞ g (1) (τ ) dτ. we can write down an uncertainty relation. if we deﬁne the root-mean-square (rms) widths of these two functions. with the same area as |g (1) (τ )|2 . Principles of Optics. Mathematically.204 Chapter 14. Note that the correlation function g (1) (τ ) and the normalized spectral density (lineshape function) s(ω) := 0 I(ω) ∞ (14. for Lorentzian functions) or vary by orders of magnitude.g. of course. where light is composed of a range of frequencies.41) The idea here is to ﬁrst note that g (1) (τ = 0) = 1. 2 (14. Thus the width δτ is the width of a box of unit height.

46) This “uncertainty relation” is a strict equality. because it remains coherent on the scale of about a foot. but the argument carries through for the one-sided case as well. the laser line width is δν = δω/2π ∼ 1 GHz. Coherence Length. we’re regarding s(ω) as a probability distribution (because it has the appropriate normalization).44) 2 δτ δω = −∞ ∞ 0 We can use Parseval’s theorem to evaluate this ratio. but it’s very diﬃcult because the path lengths must be matched to µm accuracy.43) Thus. and then we’re calculating the expectation value of s(ω) with respect to itself. This corresponds to a coherence time of δτ ∼ 1 ns. On the other hand. The deﬁnitions for δτ and δω look like inverses.14. we can use this to write δτ δω = 2π. (14. (14. eh? The point is.4 Coherence Time. the line width is δν ∼ c δλ ν δλ = 2 = 300 THz. it’s much easier to observe interference or record a hologram with light from a He-Ne laser.48) This gives a coherence time δτ ∼ 3 fs and a coherence length ℓc ∼ 1 µm. which states that the signal power is equivalently measured in either the time or frequency basis: ∞ −∞ |f (t)|2 dt = 1 2π ∞ −∞ ˜ f (ω) 2 dω. valid for any functions as long as the measures exist and are ﬁnite. . ¯ s(w) unit area s dw w Note that we have constructed the diagram as if s(ω) is two-sided.47) which is called the coherence length. the fringe visibility is mostly gone. The big advantage of these deﬁnitions is that they are related in a simple way. but they’re really quite diﬀerent because of the diﬀerent normalizations of the two functions. On the other hand for a light bulb that spans the visible wavelength range of 400-700 nm. λ λ (14. For a He-Ne laser. We can write ∞ g (1) (τ ) dτ .45) Noting that g (1) (τ ) and 2πs(ω) form a Fourier-transform pair. for time (optical path length) delays larger than the coherence time δτ . or a coherence length ℓc ∼ 30 cm. s2 (ω) dω (14. (14. Here are some examples. This coherence time corresponds to a physical path length diﬀerence ℓc := c δτ. Then we’ll deﬁne the eﬀective frequency width by the reciprocal of the average value: δω := (¯) s −1 = 0 ∞ s2 (ω) dω −1 . δω is the width of the box function of unit area and height s. and Uncertainty Measures 205 That is. So in fact it is possible to see interference of white light in a Michelson. (14. Neat.

where ∆τ is the full width at half maximum.206 Chapter 14. Assume that the splitting is small.49) = I1 + I2 + 2 (1) I1 I2 Re g12 (τ ) . we can now look at the interference pattern between two partially coherent sources. Prove Parseval’s Theorem: if f (x) and f (k) form a Fourier transform pair. |∆λ| ≪ λ.c. then ∞ −∞ |f (x)|2 dx = 1 2π ∞ −∞ ˜ |f (k)|2 dk Problem 14. Coherence 14. the visibility is √ 2 I1 I2 (1) g (τ ) .51) This tells us that very diﬀerent beams. (14.5 Interference Between Two Partially Coherent Sources In general.2. Problem 14. where ∆τ is the 1/e time. 14. resulting in little correlation. Consider a Michelson interferometer.50) Again. Show that the visibility of the output fringes varies periodically as one mirror is displaced. with period ∆d = in terms of the mirror displacement. don’t interfere very well. Problem 14.3. (b) Gaussian: g (1) (τ ) = exp[−τ 2 /(∆τ )2 ]. Then the intensity of the superposition of these waves is I= 2 η 2 = η E1 (t) + E2 (t + τ ) E1 (t) (+) (+) (+) + 2 η E2 (t + τ ) (1) (+) + 2 η E1 (t)E2 (t + τ ) (−) (+) + c. represented (+) (+) by the two ﬁelds E1 (t) and E2 (t). where ∆τ is the 1/e time.1. where ∆λ := λ2 − λ1 and λ := (λ1 + λ2 )/2.4. (c) Exponential: g (1) (τ ) = exp[−|τ /∆τ |]. Find the coherence times δτ according to the power-equivalent width deﬁnition for the following coherence functions: (a) Lorentzian: g (1) (τ ) = 1/[1 + (τ /∆τ )2 ]. where the input is bichromatic with air wavelengths λ1 and λ2 .6 Exercises ˜ Problem 14. (14. The second ﬁeld has an adjustable time delay of τ . Go through the derivation in the notes and ﬁll in the details of the “uncertainty relation” δτ δω = 2π. where the coherence time δτ is the power-equivalent width of g (1) (τ ) and δω is the eﬀective spectral width. where g12 (τ ) is the normalized cross-correlation function: (1) g12 (τ ) := E1 (t)E2 (t + τ ) E1 (t)E2 (t) (−) (+) (−) (+) . V = I1 + I2 12 (14. λ2 2∆λ .

since an optical resonator is a frequency-selective ﬁlter. Invariably the gain medium involves quantum systems. There are three key components to any laser: 1. The name “laser” is an acronym for “light ampliﬁcation by the stimulated emission of radiation. these are the ingredients for an active optical oscillator. Directionality. A laser tends to have a relatively narrow frequency spectrum. 2. Also. Examples of pumps that directly involve electricity are gas discharges and injection currents in semiconductor lasers. The optical resonator. . The pump is the energy source that is converted to laser light. the energy comes from electricity. recycles the light through the gain medium to achieve better ampliﬁcation. by amplifying light passing through it. Less direct are optical pumps. High power. for the “pop-up x-ray laser” defense system. where the transition energies determine the output wavelength.Chapter 15 Laser Physics 15. The output of a laser has small beam divergence. Laser light has a number of potentially desireable properties. but many of the most intense man-made optical sources are lasers. Now let’s examine the main components of a laser in more detail. although sometimes indirectly. 15.1. Not all lasers have high output powers. amplifying device. in the form of ﬂashlamps or even another laser. 15. The gain medium performs the actual conversion of energy to coherent light. for “transform-limited” laser pulses). typically diﬀraction-limited or nearly so. The output of a laser is commonly a Gaussian beam.1 Overview Now we’ll explore in a fair amount of detail the operation of a laser. 3. A more bizarre example of a pump is a nuclear device.” emphasizing that the laser is an active. Monochromaticity. 3.1 Laser Pumps The laser pump is the energy reservoir for the laser. either in the absolute sense or in the sense of being limited by the time-frequency uncertainty principle (i.1. Among its key features are: 1.2 Gain Media The gain medium converts the pump energy into the desired light.. Often. in contrast to the passive optical elements that we have studied so far. it narrows the frequency spectrum of the output and thus increases the coherence of the laser light. Put together to form a laser. which is very important as a source of intense and/or highly coherent light. which we have already studied in detail.e. 2.

The output mirror is to the left of the area shown. 15. as in the manifold organic laser dyes. Important examples are Nd:YAG.2. Molecules can also be contained in liquid solvent. and Yb:ﬁber.1.2.6 µm for CO2 ) to the ultraviolet/visible/near-infrared. Brewster etalons.6 µm). although pulsed gas lasers are possible (e. A gas-phase molecular laser is CO2 . in the range of 1-10%. and can operate as either CW or pulsed lasers. Output powers are in the range of 100 mW (CW lasers) to PW peak pulse powers for some truly huge lasers (P = peta = 1015 ).05% for an Ar+ laser. 15. nitrogen and copper-vapor lasers).3 Molecules Molecular lasers come in two main ﬂavors. for example. and this is a particularly important laser in industrial cutting and welding.208 Chapter 15. The eﬃciencies of these solid-state lasers are much higher than gas lasers. The blue-green pump light (from an Ar+ laser) is clearly visible as it enters through the input mirror (with a special coating to transmit blue-green light but reﬂect the 850 nm laser light).g. Commonly these lasers operate in CW (continuous-wave) mode.” These lasers operate mostly in the near-infrared (in the range of 0.7-1. a laser-pumped Ti:sapphire laser can have an eﬃciency approaching 10%. . Laser Physics 15. The eﬃciency varies widely depending on the details of the pump. ICA) all help force the laser to oscillate in a single cavity mode. the notation is “dop-ant:medium. The wavelengths of these lasers range from the infrared (10. in the neighborhood of 0. in the neighborhood of 1%. These gain media typically lase in the visible to ultraviolet range. and they are pumped by electrical discharge.1. Examples include the He–Ne laser and the argon-ion laser. Output powers vary from 1 mW (He–Ne) to 20 W (Ar+ ). Er:ﬁber. Birefringent filter Input lens ICA Brewster tuning plates Ti:sapphire crystal Input mirror The input lens aligns and matches the pump-beam proﬁle to the laser cavity mode. The eﬃciencies of gas lasers tend to be appalling.2. The eﬃciencies of molecular lasers are also relatively high.. Here is a closeup view of part of a Ti:sapphire ring-cavity laser. ruby (Cr:sapphire). The Ti:sapphire crystal is Brewster-cut to suppress reﬂection loss at the surfaces and glows bright red due to ampliﬁed spontaneous emission. The other components (birefringent ﬁlter. Output powers range from around 100 mW (the low end of dye lasers) to 10 kW for CO2 lasers. depending on the details of the pump.1 Gas-Phase Atoms One important class of gain media is atoms in the gas phase.1.2 Atoms Embedded in Transparent Solids Atoms can also be suspended in (doped into) transparent solids. Ti:sapphire (sapphire is crystalline Al2 O3 ). the wide range covered by the various available laser dyes. Nd:glass. Here.

1.3) If τrt = 2d/c is the cavity round-trip time (ignoring any contribution due to the thin gain medium). the intensity after a round trip is Iafter = GR1 R2 Ibefore . and output powers are in the range of 1 mW to 50 W (for diode arrays).1 Overview 209 One of the most amusing (and nearly edible) lasers falls in this category: the gelatin laser. Pernier. although InGaN/GaN/AlGaN lases in the 400-420 nm range. composed of a two-mirror resonator.. 1 T. .to near-infrared.4 Semiconductor Lasers Diode lasers have become enormously important in recent years. The reﬂectances R1. Then we can expand to lowest order in τrt : dI(t) 2 I(t) + τrt + O(τrt ) = GR1 R2 I(t).2 Ibefore . representing loss. Typically. The laser contains a thin gain √ medium. Eﬃciencies of diode lasers are as high as it gets (up to 65%). we can write I(t + τrt ) = GR1 R2 I(t). (15. GaAlAs. “Laser Action of Dyes in Gelatin. The “normal slight tremor” when holding the gain medium by hand was observed to enhance the operation of the medium. Schawlow. due to scattering loss). thin gain medium R¡ √ G R™ d The eﬀect of the mirrors after one round trip on the intracavity intensity is Iafter = R1. with single-pass intensity gain G.4 A Simple Model of Laser Oscillation: Threshold Behavior We can get a great deal of insight about laser operation just by considering a very simple model. Consider a laser. Most diodes lase in the red.1. (15. 15.2 < 1. and InGaN/GaN/AlGaN.5) dt W.2) (15. the laser resonator is stable for CW operation. L.” IEEE Journal of Quantum Electronics QE-7.1 15.2 .4) Let’s assume that the intensity varies slowly on the time scale of τrt . The eﬀect of the gain medium after a single pass is Iafter = √ GIbefore . M. 15. 45 (1971).1. where G > 1 for net gain and G < 1 for net loss (e. Hänsch. Diode lasers are pumped via injection current ﬂowing across the p-n junction. but sometimes unstable resonators are used in pulsed-laser designs. (15. (15. with mirror reﬂectances R1.2. Materials include GaAs. Thus.g.15.3 Optical Resonator The laser cavity or resonator is often a spherical-mirror resonator.1) Note that G is the round-trip gain. where the molecules included sodium ﬂuorescein (marginally edible) and rhodamine 6G (much less edible). A.

210

Chapter 15. Laser Physics

Rearranging gives the diﬀerential equation (GR1 R2 − 1) dI(t) I(t). = dt τrt (15.6)

Note that this is a very simple model: we are ignoring the fact that the gain G should depend on the pump, and we are assuming that the laser light is resonant with the cavity. But with the assumption of constant G, we can write the explicit solution to Eq. (15.6) as I(t) = I0 exp (GR1 R2 − 1) t . τrt (15.7)

Note that there is a threshold behavior represented in this solution, depending on the value of the gain: G> 1 R1 R2 1 G< R1 R2 =⇒ I(t) −→ ∞ =⇒ I(t) −→ 0.

(15.8)

That is, the gain must be larger than the threshold gain of 1/R1 R2 (i.e., the gain must exceed the loss) in order for the intensity to start growing. But an exponentially growing (divergent) solution is unphysical! We’ve arrived at this funny solution because the linear model of constant G is too simple.

**15.1.5 A Less-Simple Model of Laser Oscillation: Steady-State Oscillation
**

A better (but still quite simpliﬁed) model for the gain is G(I) = 1 + g0 1+ I Isat , (15.9)

where g0 /ℓg is the small-signal gain coeﬃcient (with ℓg the length of the thin gain medium), and Isat is the saturation intensity. For small intensities (I ≪ Isat ), G(I) ≈ 1 + g0 =: G0 , (15.10) and so the small-signal gain G0 is a constant, and is also the maximum possible value of G. For large intensities (I ≫ Isat ), the gain becomes g0 Isat G(I) ≈ 1 + , (15.11) I and thus the gain decreases as I increases, with G −→ 1 as I −→ ∞. That is, there is no gain in the limit of large input intensity. This nonlinear eﬀect is called gain saturation, and is a universal feature of physical ampliﬁers, which have ﬁnite energy reservoirs. The saturation intensity Isat marks the crossover between these two regimes. If we write the gain as G(I) = 1 + g(I), where the integrated gain coeﬃcient g(I) is related to the small-signal gain coeﬃcient by g(I) = g0 1+ I Isat . (15.13) (15.12)

Then the saturation intensity is deﬁned such that when I = Isat , the gain drops to half its small-signal value: g0 g(Isat ) = . (15.14) 2 Now let’s look at the solution to Eq. (15.6) when we include gain saturation, assuming we start with some small, positive intensity.

15.2 Light–Atom Interactions

211

I(t)

settles to steady state gain saturation kicks in

exponential growth (I < oooI sat ) <

t

Assuming we are above threshold, we only see the exponential rise in intensity while I ≪ Isat . Once I becomes comparable to Isat , gain saturation begins to kick in, tempering the growth of the intensity. The gain continues to decrease due to gain saturation until it just balances the losses, and the intensity settles into a steady state. We can calculate the steady-state intensity Iss by setting dI/dt = 0. This happens when G(Iss )R1 R2 = 1. Putting in the gain from Eq. (15.9) and solving for Iss , Iss = (g0 + 1)R1 R2 − 1 Isat . 1 − R1 R2 (15.16) (15.15)

Again, this is only true if the small-signal gain G0 is above threshold (G0 R1 R2 > 1). If the small-signal gain is below threshold (G0 R1 R2 < 1), then any initial intensity decays away, and so Iss = 0. Roughly speaking, Isat is a ﬁxed property of a given gain medium, and g0 is proportional to the pump power (at least for small g0 ). Noting that the steady-state laser output is Iout = (1 − R2 )Iss , (15.17)

where the second mirror is the output coupler, we can see the more physical threshold behavior of a laser. Iout(t) o

(steady state) linear increase after threshold

I ss oo=oo0

threshold (gºoooth ,o ooooooo)

(~ pump power)

gº

For a given pump, and thus gain g0 , there is an optimum output coupler reﬂectance R2 that maximizes the output power, because this slope and the threshold gain are coupled together. However, we will defer this calculation until we have treated the laser in more detail.

Above the threshold value g0,th of the small-signal integrated gain coeﬃcient, the steady-state output intensity rises linearly with slope R1 R2 (1 − R2 ). (15.18) 1 − R1 R2

**15.2 Light–Atom Interactions
**

15.2.1 Quantization

Here we are going to adopt a fairly simplistic view of how the electromagnetic ﬁeld interacts with atoms. In particular, we will be using the language of photons, but we will stick to a strictly semiclassical treatment, not really treating the atoms or the ﬁeld quantum mechanically.

212

Chapter 15. Laser Physics

But we will start with the observation that the energy in an electromagnetic ﬁeld is quantized. This means that a monochromatic ﬁeld of frequency ω (typically restricted to some “quantization volume” such as an optical cavity) has possible energies given by E= n+ 1 2 ω, (15.19)

where n is a nonnegative integer, representing the number of photons in the ﬁeld. This may be familiar as the energy-level structure of the quantum harmonic oscillator. The photon number is always deﬁned with respect to a particular mode (ﬁxing the direction, polarization, and frequency characteristics). The energies for atoms and molecules are also quantized, although the exact energy-level structure depends on the speciﬁc atom or molecule. If we denote the quantized energies by En , then the diﬀerences in energy levels correspond to frequencies via ∆Emn := En − Em = ωmn . (15.20)

The idea is that atoms with an energy diﬀerence ∆Emn prefer to interact with resonant ﬁelds of frequency ωmn . In this case, the energy of a single photon matches the atomic energy diﬀerence, and energy is conserved. There are diﬀerent types of transitions, generally corresponding to diﬀerent types of radiation. Electronic transitions in atoms are the most energetic of the type we will consider, and they correspond to visible optical frequencies. Vibrational transitions in a molecule correspond to diﬀerent amplitudes and types of motion internal to the molecule, and generally correspond to radiation in the infrared. Rotational transitions in molecules have yet lower energy, and they correspond to microwave radiation (which enables the maser, the microwave predecessor to the laser).

**15.2.2 Fundamental Light–Atom Interactions
**

There are three fundamental interactions between light and atoms. In all cases we will consider only a twolevel atom with ground-state energy E1 and excited-state energy E2 . We will also assume resonant light, ω = (E2 − E1 )/ . 1. Absorption (stimulated). In the absorption process, a photon is destroyed and the atom is promoted to the excited state. More generally, if there are n photons to start with in some resonant mode, then there are n − 1 photons after the absorption process.

E™ E™

photon

ﬁ

E¡ E¡

(no photon)

2. Stimulated Emission. This process involves the atom initially being in the excited state, in the presence of n photons in some resonant mode. After the stimulated-emission event, the atom is demoted to the ground state and the ﬁeld is left with n + 1 photons. In some sense, this process is the opposite of stimulated absorption, although absorption ending with 0 photons is possible while stimulated emission beginning with 0 photons is not.

E™ n photons E™

ﬁ

E¡ E¡

noo+o 1 photons

3. Spontaneous Emission. This process is much like stimulated emission, but when the atom is demoted, a photon is created in some mode that is initially unpopulated. Thus, a photon can go into a wide range of possible modes by spontaneous emission. It is possible to view spontaneous emission as stimulated emission due to quantum vacuum ﬂuctuations in addition to classical radiation reaction (the bound electron radiates because it is accelerating).

15.2 Light–Atom Interactions

213

E™ (no photons) E¡

E™

ﬁ

E¡

1 photon

Generally, we can associate stimulated absorption and emission with the laser mode, while spontaneous emission is additionally associated with all other modes. Absorption and stimulated emission are the most important once the laser gets going—absorption attenuates the laser mode, while stimulated emission ampliﬁes the laser mode. ‘

**15.2.3 Einstein Rate Equations
**

New let’s consider an ensemble of two-level atoms interacting with light. Let N1,2 denote the number density of atoms with energy E1,2 . Then the Einstein rate equation for the excited state is dN2 = −A21 N2 − B21 ρ(ω)N2 + B12 ρ(ω)N1 . dt (15.21)

Here, ρ(ω) is the energy density of the electromagnetic ﬁeld (the energy density in the frequency interval ω to ω + dω). The ﬁrst term corresponds to spontaneous emission, and we can see that it reduces the excited-state population, even in the absence of any ﬁeld. The second and third terms are proportional to ρ(ω), and correspond to stimulated emission and absorption, respectively, as we can see from their overall signs. By convention, the constant A21 is called the Einstein A coeﬃcient, while B21 and B12 are called the Einstein B coeﬃcients. To be consistent, N1 + N2 must add up to some constant, assuming that we really have two-level atoms and that the atoms stay in place (something that even works fairly well for gas lasers as long as we modify A21 appropriately). Thus, dN2 /dt = −dN1 /dt, and so it is easy to write down the rate equation dN1 = A21 N2 + B21 ρ(ω)N2 − B12 ρ(ω)N1 . dt (15.22)

for the ground-state population N1 . It is most convenient to look at the steady-state behavior to see if we can have laser operation. Steady state occurs when dN2 /dt = 0, whence it follows from Eq. (15.21) that B12 ρ(ω) N2 = . N1 A21 + B21 ρ(ω) (15.23)

If the energy levels are not degenerate, it turns out that B12 = B21 , as we will see in the next section. That is, stimulated emission and absorption are exactly symmetric from the rate-equation point of view. Then we can rewrite the steady-state solution as N2 = N1 1 . A21 +1 B21 ρ(ω) (15.24)

We can see from this that N2 < N1 in steady state. Thus, there is no steady-state population inversion in a two-level system. The spontaneous emission breaks the symmetry of the stimulated emission and absorption processes and guarantees that there will be always more atoms in the ground state than in the excited state. In the limit of large intensity, ρ(ω) −→ ∞, the best we can do is to equalize the population. Thus, we can’t use a two-level system for a laser gain medium. We need a population inversion for stimulated emission (ampliﬁcation) to win out over absorption (attenuation).

214

Chapter 15. Laser Physics

**15.2.4 Relations Between the Einstein Coefﬁcients
**

Now we brieﬂy outline Einstein’s derivation of the relation between the A and B coeﬃcients. If the energy levels are degenerate, we can deﬁne the degeneracy factors g1,2 as the number of ways of having energy E1,2 . For example g1,2 = 2J1,2 + 1 for atomic angular-momentum states. Then the steady-state population ratio from Eq. (15.22) can be written also via Boltzmann statistics as g2 N2 = e− N1 g1 Solving for ρ(ω), ρ(ω) = A21 B21 1 B12 g1 e B21 g2 8π λ3 e

ω/kB T ω/kB T

=

B12 ρ(ω) . A21 + B21 ρ(ω) . −1

(15.25)

(15.26)

This is equivalent to the Planck blackbody distribution2 ρ(ω) = if we make the identiﬁcations g2 B21 = g1 B12 . A21 8π = 3 . B21 λ Recall that λ here is the wavelength within the gain medium. and 1

ω/kB T

−1

(15.27) (15.28) (15.29)

**15.2.5 Line Shape and Spectral Distributions
**

So far, we’ve considered only monochromatic light and two-level atoms with sharply deﬁned energy levels. Now it’s time to improve our model of the two-level atom and its interaction with light. We will ﬁrst introduce a line-shape function s(ω) to model the fact that the energy levels have some width. The line shape is deﬁned such that s(ω) dω is the probability that a spontaneously emitted photon will have frequency between ω and ω + dω. We can also interpret this as the relative probability of stimulated emission or absorption of a photon with frequency between ω and ω + dω. Since s(ω) represents a probability density, it is appropriately normalized:

∞ 0

s(ω) dω = 1.

(15.30)

Note that as in our discussion of coherence in Chapter 14, we are using a “one-sided spectrum” that ranges only over positive frequencies. In terms of the “two-sided spectrum” s± (ω) with both positive and negative frequencies, the one-sided spectrum satisﬁes s(ω) := s± (ω) + s± (−ω) for ω ≥ 0 and s(ω) = 0 for ω < 0. When we apply the line shape and sum over all frequencies, the rate equation becomes dN2 = −A21 N2 − B21 N2 dt

∞ 0

ρ(ω)s(ω) dω + B12 N1

0

∞

ρ(ω)s(ω) dω.

(15.31)

Qualitatively, we can picture the line shape function as a relatively sharply peaked distribution centered around the resonant optical frequency ω0 . s(w)

7 dw/2p ~ 10oo -1012 Hz o w

**wº/2p ~ 10oo14 -1015 Hz o
**

2 P.

W. Milonni and M.-L. Shih, “Zero-point energy in early quantum theory,” American Journal of Physics 59, 684 (1991).

Note that in the absence of radiation. Both are important in understanding laser operation.1 Broadband Light Light is broadband (relative to the transition) if ρ(ω) is much broader than s(ω). (15.34) The cross section has the dimensions of area. we can evaluate the integral with the same slowly varying approximation as for the broadband case: ∞ 0 ρ(ω)s(ω) dω ≈ s(ωﬁeld ) ∞ 0 ρ(ω) dω.2 Nearly Monochromatic Light For nearly monochromatic light. their combined eﬀect can be modeled by the convolution of the individual line shapes.2.35) (15. which has an exponentially damping solution. the ﬁeld spectrum is narrow. and local crystal structure eﬀects on the dopant atoms. ∆ω . so g2 Iω dN2 N2 − N1 . with a half-width at half maximum of A21 . Then we can evaluate the integral in Eq. Now we will consider two limiting cases for the light spectrum.38) . we recover the previous rate equations.15. Collisions are often modeled by a spontaneousemission-like term.5. with sharp energy levels. summed over all frequencies.31) by noting that ρ(ω0 ) varies slowly over the width of s(ω). a Gaussian. dt The energy density is related to the intensity by ρω = Iω /c (see Problem 4. (15. so the line shape for spontaneous emission (the “natural line shape”) is Lorentzian. ∆ω 2π σ(ω0 ) = (15.37) s(ω) = 2π [(ω0 − ω)2 + (∆ω/2)2 ] the resonant cross section σ(ω0 ) is given by A21 λ2 .2 Light–Atom Interactions 215 Often. independent broadening eﬀects contribute. = −A21 N2 − σ(ω) dt ω g1 Here. (15. collisions.4). s(ω) turns out to be a Lorentzian.32) 0 Thus. If multiple. we have deﬁned the laser cross-section σ(ω) = A21 λ2 s(ω).21). 15.36) (15. The line-shape function models transition width due to spontaneous emission. Doppler shifts in gas lasers. or a convolution of the two (a Voigt proﬁle). and is deﬁned such that σ(ω)Iω is the power absorbed by a single atom when irradiated by intensity Iω (in the weak-excitation limit). Then the rate equation becomes dN2 = −A21 N2 − B21 N2 s(ω)ρω + B12 N1 s(ω)ρω . but the second is the more useful case for understanding laser gain. Doppler shifts lead to Gaussian line shapes because the Maxwell–Boltzmann velocity distribution is Gaussian. The Fourier transform of an exponential is a Lorentzian. Note that for a Lorentzian line shape s(ω). the rate equation is dN2 /dt = −A21 N2 . 15. and thus also lead to Lorentzian line shapes. Thus.5. so s(ω) is much broader than ρ(ω). (15.2.33) The integral on the right-hand side is the total ﬁeld energy density. 4 (15. Let’s denote this by ρωﬁeld . (15. so that we can pull it out of the integral: ∞ 0 ρ(ω)s(ω) dω ≈ ρ(ω0 ) ∞ s(ω) dω = ρ(ω0 ). corresponding to Eq.

however.27) assumes isotropic radiation. Stimulated-emission events occur at the rate of B21 Iω /c per atom. . In fact. according to the general template of Eq. multiplied by the energy per photon to give the rate at which energy is absorbed or emitted.41).41) per unit weight atom photon area due to the gain medium of thickness dz.3. To do so. the gain medium can in principle radiate into many diﬀerent modes. 2π (15.1 Stimulated Emission The intensity change for stimulated emission.1 Gain Coefﬁcient Now we can compute the change in intensity through the gain medium and then through the ﬁltering and detection system above. becomes dIω (stimulated emission) = B21 Iω c + ω s(ω) N2 dz . For atomic dipole moments aligned with the ﬁeld polarization. so that the natural cross section is λ2 (15. N2 dz.216 Chapter 15.39) σ(ω0 ) = 0 . (15. the emitted photons are in the same mode as the incoming photons. Note that we will assume the input light to be in a single mode. (15. 15. and then we ﬁnally account for the atomic number density of the gain medium. and so we count every stimulated-emission event as contributing to the ampliﬁcation of the input mode. we will consider the ampliﬁcation of a monochromatic input intensity Iω by a thin gain medium of thickness dz. We have simply the rate at which photons are absorbed/emitted by a single atom. Then we multiply by the spectral line shape to account for the frequency dependence (resonance behavior) of the atom. We will treat the contribution of each of the three fundamental processes separately. 15. and every emission event adds the photon energy ω to the atom. since the blackbody distribution of Eq. (15. ∆ω is the natural line width given by ∆ω = A21 . 15.3. and we will need to physically distinguish input photons from ampliﬁcation photons or spontaneously emitted photons. 2π This answer is consistent with a fully quantum-mechanical calculation. This must be weighted by the relative emission probability s(ω) at the incident optical frequency. Laser Physics For homogenous broadening. We then multiply this rate per atom by the number of excited-state atoms per unit area.3 Light Ampliﬁcation Now we would like to establish the value of the gain coeﬃcient in terms of atomic parameters. we will consider an inﬁnitesimal intensity change of the form # of photon energy line-shape atoms dIω = rate per per (15.1. In each case. the resonant cross section is σ(ω0 ) = 2 3λ0 . Due to the nature of stimulate emission. this answer assumes an average over all possible atomic orientations.40) since the coupling that would normally be “distributed” among three orthogonal directions is concentrated into one.42) Let’s go through these factors again in a bit more detail.

46) Thus. (15.42) and (15.2 Threshold Behavior and Single-Mode Operation Again. Generally. Thus. This is the fundamental root of the gain saturation that we explored earlier. but the overall rate per atom is now given in terms of the Einstein A coeﬃcient. because spontaneous emission can occur into any mode.28) between the Einstein coeﬃcients and the deﬁnition of the laser cross section. (15. (15.1. dIω (absorption) = B12 Iω c − ω s(ω) N1 dz . the gain coeﬃcient γ(ω) ≈ γ0 (ω). dz where we have deﬁned the gain coeﬃcient (gain per unit length) γ(ω) := σ(ω) N2 − g2 N1 . for small signals. so we include the factor dΩ/4π. the light does not change the populations. Finally.43) with the only two changes being the sign of the photon energy (and hence the overall sign). N2 − (g2 /g1 )N1 −→ 0 for strong intensities. and so we can write the total (small-signal) round-trip power gain as G0 = exp[2γ0 (ω)ℓg ]. the gain is related only to the cross section and the population inversion. since spontaneous emission can occur into either of 2 possible polarizations. since each added photon reduces the inversion. 15. we have a more complicated expression. 15. and only one will match the input mode. since it always provides a source of power.2 Absorption Absorption leads to a similar expression for the intensity change. we can ignore it for the purposes of computing gain once the laser is going.45) (15. since absorption attenuates the wave. (15. We can use the relations of Eqs. since only the ground-state atoms are relevant here. (15. and is thus constant. the population relaxes to some steady state depending on other pumping ﬁelds and decay mechanisms. even for an initially empty cavity.47) . but it is still important to get the laser going. 15. we include a factor of 1/2. In the limit of small intensity. 2 (15. Thus. we see that in terms of the atomic parameters. and the gain coeﬃcient reduces to zero.1.3 Light Amplification 217 15.28) and (15.3. the angle and frequency factors are the most restrictive. however. but we want to count only those in the input mode: dIω (spontaneous emission) = + ω A21 s(ω) N2 dz dΩ 4π ∆ω ∆ωgain 1 . The fraction of photons with the proper frequency in the width ∆ω of the input mode is ∆ω/∆ωgain . (15. The other diﬀerences lie in the fraction of spontaneously emitted photons that are emitted into the input mode.45) is therefore just an exponential.36).3. Strong light.44) The emission rate is similar to the stimulated-emission version. How does gain saturation ﬁt into this? Without any optical intensity.1. Eq. and hence spontaneous emission accounts for only a small fraction of the detected intensity. while neglecting spontaneous emission. Only a small fraction of the photons have the right direction to match the input mode. which we assume subtends the solid angle dΩ.3. g1 (15.15.43). and of course the energy density of the input mode does not enter here.3 Spontaneous Emission For spontaneous emission. and so the gain coeﬃcient keeps the value of the small-signal gain coeﬃcient γ0 (ω). The solution to Eq. where ∆ωgain is the width of s(ω). and the number of atoms per unit area is now N1 dz. to write dIω = γ(ω)Iω .4 Combined Effects Now we can combine the contributions of Eqs.3. depletes the population inversion.

consider this graphical example. As gain saturation kicks in. This occurs if s(ω) is the same for all the atoms in the medium (where the width of s(ω) is said to be due to “homogenous broadening”). For oscillation in a two-mirror linear cavity. for many lasers. we can gain great insight into the frequency dependence of laser operation. and are thus extinguished. even when there are several candidate modes. gº(w) a w 3 modes could oscillate Three modes are above threshold and can potentially oscillate. Laser Physics where ℓg is the length of the gain medium. The factor of two here arises because we are assuming a linear laser cavity. and all the atoms interact with the cavity modes in the same way. the gain proﬁle γ(ω) is reduced in magnitude. we should have G0 R1 R2 ≥ 1.218 Chapter 15.49) (15. the threshold condition is γ0 (ω) ≥ a for oscillation to occur. distributed along the length of the gain medium. Any cavity modes that fall within this range are candidates for laser operation. Under these conditions. γ0 (ω) ≥ 1 log 2ℓg 1 R1 R2 =: a. In equilibrium. In terms of this coeﬃcient. Thus.48) where a is the eﬀective “loss per unit length” per round trip. we get single-mode operation: only the mode with the largest gain will oscillate. they all start to oscillate. the laser operates in only a single cavity mode. To see how this works. (15. the gain proﬁle will continue to decrease in magnitude until the equilibrium condition is satisﬁed at the laser frequency. so that a round trip entails two passes through the medium. there will be a range of frequencies that will be above threshold. Initially. gº(w) (gain/unit length) (loss/unit length) a wº w range of possible laser operation The small-signal gain proﬁle γ0 (ω) reﬂects the line shape s(ω). and so with a suﬃcient pump rate. or solving for γ0 . Recall that the threshold condition is that the gain matches the loss. a w The other modes with smaller gain end up below threshold (γ < a). the gain must balance the loss. so the equilibrium condition is that γ(ω) = a. We can rewrite this as exp[2γ0 (ω)ℓg ]R1 R2 ≥ 1. { g(w) . However. Depicting this graphically.

without a pump. The level scheme is shown here. The ruby laser is one example of a three-level laser. and thus eﬀectively have diﬀerent resonant frequencies. where “Doppler broadening” is a major component to the width of s(ω). 2 B¡™ A™¡ 1 15. (15.21) with B21 = 0] dN2 = −A21 N2 = −N2 /τ21 . This is so named because this rate equation has the exponential solution (15.50) dt where τ21 = 1/A21 is the lifetime of the excited state.51) N2 (t) = N2 (0)e−A21 t = N2 (0)e−t/τ21 .1 Three-Level Laser The simplest change that we can make to achieve a population inversion is to add a third level. (15. Thus.21) as σ(ω)Iω dN2 = −A21 N2 − (N2 − N1 ). (15. the gain proﬁle is reduced only locally around each of the laser frequencies. In multi-mode operation. we need to develop some details of speciﬁc gain media to compute the inversion. He–Ne).52) dt ω where ω is the pump frequency. Doppler broadening arises because atoms with diﬀerent velocities experience diﬀerent Doppler shifts of the laser light. Without inversion. Since spontaneous decay is asymmetric. so there isn’t any loss of generality in doing this. 3 A£™ (fast) pump R¡£ 2 A™¡ (slow) 1 } laser transition . The basic point that we made earlier is that because the pump enters the rate equation with a term of the form (N2 − N1 ). g(w) a w In fact. there is no possibility for population inversion in the two-level atom.) Recall that for a two-level atom. Let’s simplify things by assuming equal ′ degeneracies. g1 = g2 . the pump tries to equalize N1 and N2 .g. we can write the rate equation (15. diﬀerent atomic velocity classes can contribute to diﬀerent modes. and thus no laser oscillation.15. there is no gain. When we include the pump. 15. it is possible for many modes to oscillate simultaneously.4 Pumping Schemes Now that we have established how the laser gain coeﬃcient is related to the atomic inversion.4 Pumping Schemes 219 This argument does not hold for certain lasers such as gas lasers (e.4. (This is equivalent to setting N1 = (g2 /g1 )N1 . and so the gain proﬁle is not necessarily reduced uniformly by gain saturation.. has the rate equation [from Eq. and thus τ21 is the time scale for exponential decay from the excited level.

≈ 0 =⇒ N3 ≈ dt A32 + R13 A32 (15.32) to write R13 = Iω0 ω0 ∞ 0 σ(ω) dω = A21 λ2 Iω0 . A21 (15. we canuse Eq. which gives the equilibrium population ratio N2 R13 = . we see that we have a population inversion if the ratio exceeds unity and thus if the pump rate exceeds the spontaneous emission rate from level 2. we canuse Eq. note that N2 − N1 = N2 − 1 N1 = N1 R13 − 1 N1 . R13 /A32 ≪ 1. In steady state. Then we can write the rate equations as dN3 = −R13 (N3 − N1 ) − A32 N3 dt dN2 = A32 N3 − A21 N2 dt dN1 = A21 N2 + R13 (N3 − N1 ).220 Chapter 15.. A32 .55) where of course one of the equations is redundant since N1 + N2 + N3 must be a constant of the motion. and so we can assume that it is always in quasiequilibrium. Thus.58) Again.54) where ω0 = (E3 − E1 )/ is the pump transition frequency. Also. dt These are the rate equations for a two-level atom. for a monochromatic pump (e. so level 3 is essentially unpopulated.53) R13 = ω while if the pump is broadband (e.57) dN1 = A21 N2 − R13 N1 .35) to write σ(ω)Iω . That is. but with an asymmetric pump. like a friction term) for N3 . we will assume A21 ≪ R13 . while the laser (2 −→ 1) transition decays slowly. dN2 /dt = dN1 /dt = 0. 4 ω0 (15. (15. N1 A21 (15. and it is now a redundant quantity since it is completely determined by N1 .55) for the remaining two levels as dN2 = R13 N1 − A21 N2 dt (15. Laser Physics The new level (with highest energy) decays quickly. so N3 ≪ N1 . a ﬂash lamp).e.59) . dt (15..56) The population N3 is said to adiabatically follow the N1 population.g. 2 R¡£ A™¡ 1 The presence of the third level breaks the symmetry of the pump. Then we can use N3 − N1 ≈ −N1 and write eﬀective rate equations (15. In the limit of large A32 (A32 ≫ R13 ). R13 > A21 . Furthermore. the equilibrium condition for N3 gives R13 R13 N1 dN3 ≈ . the pump is another laser). Now we will make the adiabatic approximation to eliminate energy level 3. The idea is that the A32 term in the dN3 /dt equation is a damping term (i.. (15. the adiabatic approximation says that N3 damps to equilibrium very quickly.g. (15. and now we can get a population inversion if the pump rate exceeds the spontaneous decay rate. To compute the inversion.

62) dN1 = A21 N2 − A21 N2 dt dN0 = A20 N2 A10 N1 + R03 (N3 − N0 ). 3 2 A™¡ (slow) 1 A¡º (fast) 0 A£™ (fast) pump Rº£ A™º (slow) } laser transition Other decay paths besides the ones we have included. as we will see.61) where R′ := R13 /A21 .59). Four-level media are particularly important since.60) to eliminate N1 in the right-hand side of Eq. but we assume them to be slow.2 Four-Level Laser Another important level scheme is the four-level laser. Again. we ﬁnd an inversion of N2 + N1 = N = N2 − N1 = R13 − A21 R13 + A21 N= R′ − 1 R′ + 1 N. we can neglect the small populations in levels 1 and 3.4. the eﬀective rate equations for the remaining two levels are dN2 = R03 N0 − (A21 + A20 )N2 dt dN0 = (A21 + A20 )N2 − R03 N0 . (15.15. Using Eq. A10 ≫ R03 . we can see that a pump rate R′ > 1 corresponds to a population inversion. so we can adiabatically eliminate both of them. and we are neglecting the population N3 .4 Pumping Schemes 221 Similarly. and in particular N3 − N0 ≈ −N0 . we will assume that all decay rates except those from level 2 are fast: A32 . A20 . (15. with the laser transition (2 −→ 1) sandwiched between the two auxiliary levels. Now both levels 1 and 3 have fast damping terms. A21 . (15.63) N3 ≈ A32 while setting dN1 /dt ≈ 0 leads to A32 N1 ≈ N0 ≪ N0 . The Nd:YAG laser is one example of a four-level laser.64) A21 Again. Thus. in which case they do not aﬀect our basic results. The level scheme is shown here. 15. (15. where we add yet another extra level.60) A21 where N is the total atomic number density.65) . The rate equations are dN3 = −R03 (N3 − N0 ) − A32 N3 dt dN2 = A32 N3 − (A21 + A20 )N2 dt (15. dt (15. (15. (15. The new level 0 is now the ground state. they are more eﬃcient in general than three-level media. R13 + 1 N1 . dt Again. Setting dN3 /dt ≈ 0 leads to R03 N0 ≪ N0 .

Again. this corresponds to a two-level atom with an asymmetric pump. The ground level of the laser transition is level 1. (15. The last term accounts for the interaction of the atom with the laser ﬁeld. so that a certain minimum pump was required.66) N0 A21 + A20 Using N0 + N2 ≈ N and a procedure similar to the three-level calculation. (15. However. so that σ. we will consider it to be pumped by a monochromatic laser at a rate σpump Ipump R2 = . I. for the four-level laser we can neglect N1 compared to N2 . and this is why it is easier to develop an inversion in the four-level laser than in the three-level version.68) ωpump and we can regard this to be a unidirectional pump since we assume level 3 will decay quickly to level 2. whereas we saw that in the three-level laser. we can eliminate N0 to get an expression for N2 . The complication is that now we will have to consider the interaction of the gain medium with two optical ﬁelds: the pump ﬁeld and the laser ﬁeld. Going back to the four-level laser scheme. but similar results hold for the three-level laser. dt ω (15. except that they interact with diﬀerent levels. the diﬀerence is that the ground level 0 of this eﬀective atom is not involved in the laser transition. For simplicity.69) where A2 := A21 + A20 . Other auxiliary levels might make the pumping process easier.222 Chapter 15. and they have diﬀerent eﬀective signs after the adiabatic elimination of levels 1 and 3. .67) where R′ := R03 /(A21 + A20 ). Since N1 ≈ 0. with the result dN2 σI ≈ R2 N0 − A2 N2 − N2 .65) that the eﬀective rate equation is dN2 σI = R2 N0 − A2 N2 − (N2 − N1 ).70) dt ω Of course. we can proceed to explore the gain coeﬃcient in more detail. (15. Again. and ω all refer to the laser ﬁeld. the inversion was of the form N (R′ − 1)/(R′ + 1). 15. this is the population inversion: N2 − N1 ≈ N2 ≈ R′ 1 + R′ N. the stimulated absorption/emission terms for the pump and laser light have the same form.5 Gain Coefﬁcient With the speciﬁcs of the atomic media in hand. (15. which is almost completely depleted. we will only do this for the four-level laser. 3 fast decay effective pump R™ A™¡ (slow) fast decay 0 2 h—w 1 } laser transition Recall from Eq. the whole point of this is: auxiliary levels break the symmetry of the pump and allow for population inversion. (15. In steady state. R03 N2 ≈ . Note that there is an inversion for any pump R′ > 0. Laser Physics Again.

when we have an equation of the form N2 = (stuﬀ)N0 with N0 + N2 ≈ N .72) where (15.78) γ(I = Isat ) = . Gain saturation. the gain coeﬃcient drops to half its small-signal value: γ0 (15. In the limit of a very strong pump.1. Consider the limit of a weak pump (R2 τ2 ≪ 1).74) From this expression. the saturation intensity is deﬁned such that at the saturation intensity.77) I 1+ Isat Again.72) becomes σ(ω)R2 τ2 N . γ(I −→ ∞) = 0. we can see that there are two types of saturation in the gain medium.71) Following the algebraic steps of the last section. (15.79) This limit has a nice physical interpretation: there is only a ﬁnite number N of atoms. (15. Thus. R2 τ2 −→ ∞. 1. and each can only add an energy ω of energy as soon as it is stimulated by an incoming laser photon. which gives the steady-state population N2 = R2 N0 . . = R2 σI I 1+ A2 + 1 + R2 τ2 + + R2 σI ω Isat A2 + ω A2 + Isat = ω . σI A2 + ω (15. Pump saturation. I 1 + R2 τ2 + Isat (15. Then the gain coeﬃcient (15. so that eventually the gain coeﬃcient does not continue to increase with an increasing pump intensity.15. γ(ω) = 2.75) γ(ω) = I 1+ Isat In the small-signal regime (I ≪ Isat ). (15. the gain coeﬃcient takes on the small-signal value γ0 (ω) = σ(ω)R2 τ2 N. the gain coeﬃcient saturates at a value γ(ω) −→ σ(ω)N. the solution is (N2 /N ) = (stuﬀ)/(1 + stuﬀ). Thus.5 Gain Coefficient 223 Steady state occurs when dN2 /dt = 0.73) −1 and τ2 = A2 . so that (15. the gain can also be limited by the ﬁnite number of atoms in the gain medium.76) γ0 (ω) .5. Thus. 2 In the limit of large intensity. the gain coeﬃcient for the four-level laser is γ(ω) = σ(ω)[N2 − N1 ] = σ(ω)R2 τ2 N . στ2 (15. This is precisely the gain-saturation eﬀect that we discussed before in the context of the simple laser model of Section 15. R2 N2 = N σI R2 R2 τ2 ω = .

and the saturation intensity Isat is given by Eq. (15. we solve for dz: 1 1 + Iω Isat Integrating over the length of the gain medium. (15. and compute the gain of the medium in terms of the gain coeﬃcient. (15. the solution is more complicated than a simple growing exponential. but rather we can get a transcendental equation that relates them.84) G := we can rewrite Eq. (15.224 Chapter 15.80) (15. we recover the small-signal result: G0 = exp[γ0 (ω)ℓg ]. Laser Physics 15. In this case.83) and thus log Deﬁning the single-pass gain as + I2 − I1 = γ0 (ω)ℓg . (15.84) as log G + (15. To ﬁnd the solution. I1 (15.86) This is the central result of this section.82) 1 1 + Iω Isat I2 I1 ℓg dIω = 0 γ0 (ω) dz.5. (15.86). then we can ignore the log G term in Eq. In the small-signal regime (I1 .73). We cannot get a general expression for G in terms of γ0 . dz so we can write this explicitly in terms of the small-signal gain coeﬃcient as 1 dIω γ0 (ω) . = I Iω dz 1+ Isat (15. 3 fast decay effective pump R™ A™¡ (slow) fast decay 0 2 h—w 1 } laser transition The gain coeﬃcient is deﬁned such that dIω = γ(ω)Iω . (15. The medium has length ℓg . Since the right-hand side of this diﬀerential equation involves the intensity. with the result I2 − I1 = [γ0 (ω)Isat (ω)]ℓg .1 Gain in a Medium of Finite Length Now let’s revisit the gain medium.88) . then we can ignore the (G − 1) term in Eq. Isat I2 .86). (15. I2 I1 dIω = γ0 (ω) dz.85) I1 (G − 1) = γ0 (ω)ℓg .76) in the weak-pump regime. Isat (15. respectively.81) Recall that the small-signal gain coeﬃcient γ0 (ω) is deﬁned by Eq. (15. and the input and output intensities are I1 and I2 . I2 ≪ Isat ).87) In the high-intensity limit (I1 ≫ Isat ).

so that our above analysis of the single-pass gain remains valid (in a linear cavity. Thus. (15. the gain equals the loss.6 Laser Output: CW 225 The interpretation of this relation is that the gain medium adds an amount γ0 Isat ℓg to the input intensity.86) becomes log Solving for I2 . σ(ω)τ2 (15.1 Optimum Output How can we optimize the output of the laser? In other words. That is. Isat (15.6 Laser Output: CW Now we’ll examine in more detail the output characteristics of a continuous-wave (CW) laser. Let’s consider a ring-cavity laser as shown. In steady state. given a certain gain and loss in the laser. there is no loss in the output coupler (mirror).94) . Writing out the gain coeﬃcient and the saturation intensity. 15. multiplied by the pumping rate per atom. and in particular look at its eﬃciency and how to optimize it. (15. we will assume that there is an “optical diode” in the cavity to force the laser to operate in only one direction (typically.90) This expression can easily be generalized to include other cavity losses (e..92) (15. multiplied by the photon energy. the gain equation (15. R£ R™ I¡ R¢ gain medium I™ R¡ output In addition. γ0 Isat ℓg = [σ(ω)R2 τ2 N ] ω ℓg = (N ℓg )( ω)R2 .15. We can conclude from this that we can only extract energy from the medium at the rate that energy is pumped into it.91) Isat [γ0 ℓg − log(1/Ps )] .g. so that T1 = 1 − R1 . which leaves the polarization unaﬀected in only one direction). what reﬂectance should we choose for the output coupler to maximize the output power? Let’s work in the low-loss approximation where (1 − Ps ) is small. the gain medium adds an intensity given by the number of atoms per unit of transverse area. an intense input signal extract all possible energy from the gain medium. Then the output intensity is simply I2 T1 . If all the losses are due to partial mirror transmission. and so GPs = 1. making things more complicated). (15. the two counterpropagating waves saturate each other. we ﬁnd 1 Ps I2 = + I2 (1 − Ps ) = γ0 (ω)ℓg . due to reﬂections at the gain medium surfaces).6. Then log 1 Ps = − log[1 − (1 − Ps )] ≈ 1 − Ps .93) 1 − Ps Ideally. an optical diode consists of a Faraday rotator and a birefringent plate. then the survival probability is Ps = R1 R2 R3 R4 .89) Thus. 15.

on the other hand.8 nm wavelength corresponding to ∼ 2 eV. .226 Chapter 15.101) (15. and represents the fundamental limit to eﬃciency for any given laser gain medium. (15. Then we can write ηq as the ratio of the output power to the input power: ηq = ω ω · (total pump rate) . (15. Then from Eq. The laser wavelength is λlaser = 1064 nm. The laser is on a Ne transition.1. (15.97) (15. giving a relatively high quantum eﬃciency of ηq ∼ 0. where we have deﬁned the loss probability Pl := 1 − R2 R3 R4 . ∆Epump is the energy of the pumping transition. 1 − Ps = 1 − (1 − T1 )R2 R3 R4 = (1 − R2 R3 R4 ) + T1 R2 R3 R4 = Pl + T1 (1 − Pl ) ≈ L + T1 . Pl + T1 (15. Thus the output power is Pout = ω · (total pump rate). Iout = T1 Isat [γ0 ℓg − log(Pl + T1 )] = T1 Isat Pl + T1 γ0 ℓg −1 . For example. Other limits to eﬃciency include cavity losses (extraction eﬃciency) and the fact that not all pump photons lead to stimulated emission (pumping eﬃciency). This ideal eﬃciency is called the quantum eﬃciency of the laser. and thus N ℓg A is the total number of atoms in the gain medium. which gives an optimum transmission T1 (opt) = The corresponding optimal output intensity is Iout (opt) = Isat γ0 ℓg − Pl 2 γ0 ℓg Pl − Pl .98) . with a 632.2 Quantum Efﬁciency In the ideal case. Laser Physics Also. and most of this excitation energy is not extracted by the laser ﬁeld.93). (15.95) This is an optimum when ∂Iout /∂T1 = 0. but it isn’t quite: the pump photon is more energetic than the laser photon because they correspond to diﬀerent energy transitions. This sounds like perfect eﬃciency. Suppose that ηq is the laser eﬃciency in this ideal case. This is one reason why gas-discharge lasers tend to be less eﬃcient than optically pumped lasers: the discharge leads to very high excitation. 15. the output power consists of one output photon energy per input photon.102) Here. Then the optimal output intensity is Iout (opt) = γ0 ℓg Isat = ωR2 ℓg N.8. then the corresponding output power is Pout (opt) = AIout (opt) = ωR2 N (ℓg A). around ηq ∼ 0. operates by electric discharge to a state of ∼20 eV in He.100) where ℓg A is the volume of the gain medium.99) Sometimes. The energy is transferred via collision to a state of ∼20 eV in Ne. = ∆Epump · (total pump rate) ∆Epump (15. a Nd:YAG laser has a pump wavelength of λpump = 810 nm when pumped from a diode laser (these lasers can also be ﬂashlamp-pumped in the blue with correspondingly less eﬃciency). This can enhance the stability of the laser at the expense of some output power. in real lasers it is necessary to choose an output coupler that has lower reﬂectance than this theoretical minimum. A He–Ne laser. If A is the area of the gain medium. The quantum eﬃciency in this case is much lower. there is no loss (Pl = 0).96) (15. In other words. and we have made the low-loss approximation 1 − Pl ≈ 1.6.

. This happens on a fast time scale. so the process repeats. especially for a weak or slowly activated pump where the laser intensity adiabatically follows the inversion until steady state is reached. as a result.7 Laser Output: Pulsed Many lasers don’t work in CW mode. The cavity intensity becomes so large that the gain medium drops below threshold. before the cavity intensity can respond. Also. we might expect a CW laser to turn on smoothly. The intensity. 15.7 Laser Output: Pulsed 227 15. it might not be possible to sustain the pump or to cool the laser system eﬃciently enough. There are many reasons why pulsed lasers are preferred in some applications. Below is the measured intensity from a diode-laser pumped Yb:ﬁber laser operating at 1085 nm that shows laser spiking.1. From the simple laser model of Section 15. quickly depleting the gain medium as the laser oscillates.7. there is another possibility. This process causes the intensity oscillations or laser spikes. 1. However. For example. especially with high-gain lasers such as ﬂashlamp-pumped Nd:YAG lasers. 2. when the laser is used to probe or visualize a fast process. the initial transient can look like a periodic sequence of short intensity pulses. 211). The laser spikes are also called relaxation oscillations. builds up rapidly.15. We are more or less at the initial condition again. called laser spiking. With high powers or certain gain media. To treat pulses. although this term often refers to the low-amplitude oscillations after the gain spikes settle down to near equilibrium (or when the laser is only slightly perturbed). This process occurs due to the following process. but instead work in pulsed mode. I(t) N™o-oN¡ Nth t t As in the simulation shown here. though we must treat the laser dynamics.1 Laser Spiking In the treatment of CW lasers above. 3. A fast pump rapidly excites the gain medium far above the threshold inversion. pulsed lasers are capable of much higher peak powers than CW lasers (with state-of-the-art lasers reaching the petawatt range). where the dynamics of the inversion N2 − N1 and the intensity I conspire to produce intensity oscillations.5 (see the plot on p. the oscillations damp away and the laser eventually relaxes to steady state. Since both the laser output and spontaneous emission act as damping (friction-like) processes. we focused on the steady-state behavior of lasers. short pulses may be needed to strobe and eﬀectively freeze the dynamics during the probe. One kind of pulsed laser output occurs in the initial transient behavior of a CW laser.

so that the cavity Q is only high at the moment when the mirror reaches the right angle. The saturable absorber will spoil the Q until the gain reaches the point where an ampliﬁed spontaneous wave exceeds the saturation threshold. big pulse. When the voltage is applied. for a ﬂashlamp-pumped laser. This is a medium that absorbs light. the intensity is extracted in one short.e. the output pulse can be on the order of the cavity round-trip time (10 ns pulses are typical for a 1 m linear Nd:YAG laser). which is essentially a nonlinear crystal that becomes birefringent when high voltage is applied across it. but due to the same saturation eﬀect as in gain medium.. so that the laser ﬁnds itself way above threshold. the cavity is restored to its former high-Q condition. One common method is to use a Pockels cell (electro-optic crystal) in the cavity. the laser jumps to well above the steady-state power. the birefringence is suﬃcient to spoil the Q. the polarization is rotated going through the cell. Near the end of the ﬂashlamp pulse.) Another method is to use a rotating cavity mirror. an “automatic” Q-switch can be made from a saturable absorber. During the ﬁrst spike. and a polarizer in the cavity strongly attenuates the mode. For example. (Actually it turns out that a polarizer isn’t necessary. .228 Chapter 15. 15. Finally. Laser Physics 30 output power (W) 0 0 500 t (µs) The pump diodes are turned on quickly.2 Q-Switching The idea behind Q-switching is to avoid the laser spiking by compressing all the relaxation oscillations in to a single. the cavity Q is “switched low”) during the ﬂashlamp pulse to allow the inversion to build up way beyond the high-Q threshold value. and the resulting laser spikes are clearly visible before the laser power settles down to the steady-state value of about 10 W. high intensities pass through the medium nearly unaﬀected (this saturation eﬀect is also called “bleaching”). Several techniques are available for switching the cavity Q. the cavity is made lossy (i. enormous pulse. as shown in the simulation below. In this case. For a high-gain laser (such as Nd:YAG with an unstable resonator). at which point the absorber bleaches and the cavity Q is switched high.7.

since it is not itself a laser.15.7. The idea is to take a ring cavity and inject CW laser light into one port.3 Cavity Dumper Another simple source of pulse laser light is the cavity dumper (see Problem 7.7 Laser Output: Pulsed 229 flashlamp N™o-oN¡ Nth t I(t) t I(t) t t 15. the idea is to suddenly turn on an optical switch (such as an acousto-optic modulator) inside the cavity to couple out the high-intensity light.2). the pulse duration is equal to the round-trip time of the cavity. optical switch . For eﬃcient out-coupling. so a high cavity ﬁnesse is important in attaining large pulse intensities. this is more of a device to convert CW light into pulses. The intensity in the cavity builds up to ∼(F /2π)Iin in steady state. Once the light has built up. Actually.

This is also the method that provides some of the shortest possible pulses. Let’s consider a slightly less simple model of mode locking. giving a noisy output that ﬂuctuates about the mean intensity I. The idea is that a gain medium with inhomogeneous broadening can support simultaneous lasing on many modes. and zero phase oﬀset. as we discussed in Section 15. sin2 [(δω)t/2] (15.7. period τrt . we can expect the output to be a series of short. (15. intense pulses. with duration ∼1/∆ω. we have a sum of the form N 2 I0 e n=1 iω0 t i(n−1)(deltaω)t e . sin2 (φ/2) (15.2.103) We can interpret the left-hand side as the sum of many waves of equal amplitude.105) which leads to an intensity I = I0 sin2 [N (δω)t/2] . frequency ωn = 2πn.106) ¯ This is a sequence of short pulses. if we can lock the relative phases of many modes.3. dwoo=oo2p(fsr)oo=oo2p/trt Dw w Usually. Interpreting φ −→ (δω)t. I (CW) I 0 trt 2trt 3trt t But what if it is possible to give them all the same relative phase? Recall the Poisson sum rule (Problem 3.5). (15. The laser output is thus the interference of ¯ all the beams. In terms of the laser. and thus do not compete for pump energy. . The output for 10 equalamplitude. This is because diﬀerent modes interact with diﬀerent atoms.104) where I0 is the intensity of a single wave. compared to the random-phase version shown as a dashed line. The right-hand side is a periodic series of arbitrarily short pulses. ∞ n=−∞ cos(2πnt) = ∞ n=−∞ δ(t − n).8 that the interference of N equal-amplitude waves of the form exp[i(n − 1)φ] is I = I0 sin2 (N φ/2) . Suppose that the gain medium can support modes in a range of order ∆ω.230 Chapter 15. where δω is much smaller than the central optical frequency ω0 . the modes will have no particular phase relationship. Recall from Section 5. Laser Physics 15. phase-locked modes is shown here.4 Mode Locking Mode locking is another method for converting a CW laser into a pulsed laser. The modes are spaced by the usual cavity spacing δω = 2π/τrt . and height N I. but with random phases.

. which takes advantage of the nonlinear optical behavior of the gain medium. = dt τrt where the gain G saturates according to G(I) = 1 + g0 1+ I Isat .7 fs. which gives ∆ω/2π ∼ 9. a Ti:sapphire laser is popular for mode-locking because of its very wide emission range. where the modulation frequency matches the cavity free-spectral range. the sidebands lead to coupling of the modes. the refractive index of the gain medium increases with the applied intensity. A similar eﬀect can be achieved by modulating the amplitude of the intracavity light. to derive an expression for the cavity photon lifetime τp . (b) Expand the above expression to ﬁrst order in τrt . so the pulse width is δt ∼ 1/∆ω ∼ 1. and tend to synchronize (you can also think of the sidebands seeding the other modes with light of the proper phase). just as in the Qswitched laser. a saturable absorber can select the mode-locked behavior. For example.) (a) Write down the cavity intensity at time t + τrt in terms of the cavity intensity at time t. by periodically driving an acousto-optic modulator. separated by the modulation frequency (see Problem 13. thus selecting again the highintensity pulses. Real lasers have much longer pulse lengths due to a number of eﬀects.15. A similar eﬀect is achieved by the Kerr lens. The mirrors have intensity reﬂection coeﬃcients R1 and R2 . The modes then act as coupled nonlinear oscillators.1).8 Exercises 231 NI I trt 1 ~ N Dw ~ (CW) I 0 trt 2trt 3trt t If there are many modes available. spanning 600-1050 nm. dI (G(I) R1 R2 − 1) I. There several ways to mode-lock a laser.8 Exercises Problem 15. but fs-scale pulses are now routine through this method. since the pulses will bleach the absorber. deﬁned as the time required for the cavity intensity to decay to 1/e times its initial value. (All this is in the text. an electro-optic phase modulator in the cavity can put sidebands on each mode. where τrt is the round-trip time of the cavity. 15. Thus.1. What are the conditions under which this approximation is justiﬁed? (c) Apply the result of part (b) to a cavity without a gain medium. Consider a two-mirror cavity with a gain medium. e. Lasers can also be mode-locked actively. a high-intensity Gaussian beam eﬀectively creates a lens for itself in the gain medium. Problem 15. For example. while the CW light will be blocked. The full width at half maximum of the emission range is more like ∆λ ∼ 200 nm (from 700-900 nm). An appropriately placed aperture (or eﬀective aperture) can block the cavity light except when the Kerr eﬀect focuses the mode through it. For example.g. and hence obtain a diﬀerential equation for I(t). At high intensities. The gain medium is thin and has a round-trip intensity gain coeﬃcient G. If the modulation frequency matches the free spectral range of the laser cavity (or possible some multiple of it). (a) Obtain an analytic solution to the laser cavity intensity equation.2..5 THz. the point is to go through it and ﬁll in the details. One way is to “encourage” only the large intensities of the mode-locked operation and “discourage” the random-phase CW operation. the pulses can be extremely intense and short.

take g0 = 2. and the far-ﬁeld divergence angle of your laser. You then ﬁll the tube with the appropriate mixture of helium and neon.232 Chapter 15. Solve the diﬀerential equation numerically to make the plot. 10. assuming it lases. Then modify these in two ways. (a) Write down the rate equations for the three-level atom. Suppose you build your own He-Ne laser. write the equations in terms of the scaled time parameter t′ := A21 t. Isat show that in the limit of a thin gain medium. and good (say. Assume that the laser operates on the red line at 632. R1 R2 = 1 (for this part only). and so on. Also assume that the output mirror has a reﬂectance of Rout = 99%. (d) Calculate the ﬁnesse and Q factor of the resonator. (Hint: we did something very similar to this in the text. and plot the solution as a function of t/τrt . You use a symmetric cavity (geometrically speaking) by sealing two concave mirrors with the same radius of curvature R = 100 cm onto the ends of a glass tube of length L. Include the plot of the solution of the adiabatic rate equations for comparison. install proper electrodes. Treat the case of a perfect cavity. You will not be able to obtain an explicit solution.99.) (b) Find the steady-state intensity for a laser modeled by the above equations. Make plots of n2 (t′ ) for diﬀerent values of A32 .8 nm. and then use conservation of atom number.4. (e) Calculate the frequency width of the resonator modes. hook up the power supply. measure time in units of 1/A21 . This is a very crude estimate of the frequency width of the laser output. This is only a good estimate if the laser operates close to threshold. Isat = 1 W/cm2 .3. . In other words. Examine the validity of the adiabatic approximation in the three-level laser model as follows. Problem 15. log G + I1 (G − 1) = γ0 (ω)ℓg . Second. For what range of length L might you expect a working laser? (b) Suppose you settle on a length L = 75 cm. (15.9): G(I) = 1 + g0 1+ What is g0 in terms of the gain parameters? Problem 15. the gain coeﬃcient takes the form that we considered in the simple laser model of Eq. and 100). Calculate the beam waist parameter w0 .5. Write the equations in terms of the normalized populations nj := Nj /N . (b) Assume a pumping rate of R13 = 2A21 . R1 R2 = 0. and so the light must pass through the gain medium many times before exiting the cavity to sustain oscillation. I Isat . Laser Physics and where g0 is a constant. Problem 15. I0 = 1 µW/cm2 . A32 /R13 = 1. In all cases start with the atoms entirely in the ground state. corresponding to where the adiabatic approximation is bad. marginal. both with and without the adiabatic elimination of the third level. as it assumes that the gain medium plays negligible role in determining the frequency spectrum. (c) Calculate the free spectral range of the resonator. while the other (back-end) mirror is perfectly reﬂecting. For the purposes of the plot. Starting with the gain relation for a medium of length ℓg . and the order q of the lasing mode. but you can get an implicit solution t(I). (c) Plot the solution I(t) to the above equations. I(t). the spot size w at the mirrors. First. (a) Gas lasers have relatively small single-pass gain. and that the refractive index of the helium/neon mixture is n = 1. eliminate N1 by letting N be the total number density of the atoms.

In this problem you will examine numerically the dynamics of a CW Nd:YAG laser. 1991) p.0 and donut modes compete so that only the mode with the least loss lases at any given time. and as it moves around slightly the laser jumps between the Gaussian and donut modes.6. What is the Einstein A coeﬃcient for the transition? What is the corresponding resonant laser cross section σ(ω0 )? The saturation intensity Isat on resonance? (Check your answers: Isat = 262 W/cm2 .3 GHz due to Doppler broadening at room temperature. For the same longitudinal mode. (Part of this exercise is to locate a reputable source for optical material properties. (h) Suppose you change the temperature of the laser resonator by 1 ◦ C. What is the threshold population inversion density N2 − N1 required for the laser to oscillate? (Assume g2 = g1 for this laser.0 or the TEM1.0 ). (g) Suppose that the laser can oscillate in either the TEM0.8 Exercises 233 (f) The He-Ne gain medium can support laser operation within a frequency range of about 1. Calculate the optical power of the light circulating in the resonator (assuming that all the power is either reﬂected from or transmitted through the output mirror) and the peak intensity inside the resonator. 3 Bahaa E.83 at the laser wavelength (be careful to note that 1.064 µm is the wavelength in vacuum!). we can use the rate equation dN2 = Rpump N0 − A21 N2 dt to model the level population densities. (2) in addition to the transmission through the output mirror. ∆ω s(ω) = . and N ≈ N2 + N0 . 478 . (4) the TEM0.) (c) Assume a cavity with a perfectly reﬂecting back mirror and a 95% reﬂecting output mirror (5% transmission). (a) What does “CW Nd:YAG” mean? (b) Assume the following parameters for the gain medium:3 the lasing transition has a lifetime τ21 = 1.0 mode.) Find an expression for the steady-state inversion (N2 − N1 )/N . N1 ≈ 0. Assume that the laser is well modeled as a four-level laser producing output at a wavelength of 1. (Right now we are still excluding the eﬀect of laser light. (j) Suppose that the output power of your laser is 5 mW. calculate the diﬀerence in the lasing frequencies of these two modes. You observe that the bug is resting near the center of the output mirror. By how much does the frequency of the output light change? Assume that you constructed your laser using a glass tube made out of BK7 borosilicate crown glass. and is otherwise impervious to the laser radiation. and (5) the gains of the two modes in the medium are roughly the same. with all other losses negligible. Use this form to write down an expression for s(ω0 ). Saleh and Malvin Carl Teich.) (i) Calculate the minimum round-trip intensity gain G required for your laser to oscillate (lase). with N3 ≈ 0. and that the laser is not too much hotter than room temperature.2 ms and a linewidth ∆ω/2π = 120 GHz. (3) the bug is much smaller than the beam diameter.064 µm. you ﬁnd that you managed to trap a tiny bug inside the cavity while you were ﬁlling it.15. such as the thermal expansion coeﬃcient of a popular optical glass. Do you expect your laser to operate in a single longitudinal mode? Assume that the laser operates in a single transverse mode (TEM0. and that the refractive index of YAG is n = 1. Fundamentals of Photonics (Wiley. Estimate the diameter of the bug. Also assume a 10 cm long gain medium. 2π [(ω0 − ω)2 + (∆ω/2)2 ] where ω0 is the resonance frequency. Problem 15. A.) (d) Assuming an ideal 4-level laser. Upon closer inspection. using these assumptions and observations: (1) the bug is a perfectly absorbing disc. say by cooling it oﬀ with air current from a fan. the only sources of loss are absorption by the bug and loss around the edges of the 2 mm diameter mirrors. Assume that you can model the lineshape function as a Lorentzian. (k) You notice that your laser is behaving strangely.

and nth := (1 − R1 R2 )/(2ℓg σ(ω0 )N ) is the threshold inversion density. The output coupler has an intensity reﬂection coeﬃcient of R1 = 80%.2 /N . Solve these simultaneous equations and make plots of the laser intensity and the population inversion N2 /N . which are diﬀerent from the last problem: 1. n0. and the gain medium sits in one leg of the cavity. We will assume the cavity intensity equation dI ηSE ℓg ω0 A21 N2 (R1 R2 − 1) ℓg σ(ω0 )N2 I+ . e. Everything else should be the same as last time: . note that we are not properly treating eﬀects due to ampliﬁed spontaneous emission). τp := τ /(1 − R1 R2 ) is the photon lifetime.5%. frequency. These equations are valid when the laser intensity changes slowly. Any light not reﬂected is transmitted through the mirror.7. This is a follow-up to Problem 15. n2 := N2 /N .234 Chapter 15.6 that models the CW Nd:YAG laser. The laser will be pumped at a rate Rpump that produces an inversion density N2 of 10 times the threshold inversion Nth (in the absence of laser oscillation). Therefore you should switch to dimensionless variables. The other three mirrors have reﬂectances of R2 = R3 = R4 = 99.. and generate solutions over a few photon lifetimes. comprising four mirrors at each of the corners of the square beam path. Assume a cavity length of 1 m (make sure to account for the laser rod in calculating the round-trip time τrt !). = I+ dt τ τrt τrt where the second term now accounts for the gain (saturation is built into the rate equation for N2 ). ′ dt nth nth τp τ21 2I ′ n2 ′ Here. numerical integrators don’t like really big or really small numbers. by calculating expressions for steady-state values to compare with your data. Assume also a small initial intensity (due to spontaneous emission) to get things started.g.. and assume ηSE = 10−12 . 3.8 s−1 . Use the parameters you calculated or were given above to set up the parameters for these scaled equations. The laser uses a ring cavity. Rpump := Rpump τp . which is equivalent to the set of unscaled equations above: dn2 τp ′ n2 − = Rpump (1 − n2 ) − ′ dt τ21 ′ n2 n2 ′ dI = −I ′ + I + ηSE .) Use initial conditions n2 (0) = 0 and I(0) = 0. t′ := t/τp . Laser Physics (e) At what rate Rpump must you pump the medium to achieve an inversion density of 5 times the threshold value you found in (c)? Assume a dopant density N = 1019 cm−3 . Each leg of the beam path is 50 cm long. and the third term is a small contribution to the intensity due to spontaneous emission.) (f) Now including the laser light. Solve this set of diﬀerential equations. etc.2 = N0. nor units for that matter. = Rpump N0 − A21 N2 − dt hω0 where I is the laser intensity. (Use scaled variables in the plots as well. 2. the rate equation is 2σ(ω0 )I dN2 N2 . polarization. Problem 15. Here you will examine numerically the dynamics of a pulsed Nd:YAG laser. Appropriate measures have been taken to ensure the laser oscillates in only one direction. I ′ := I/Isat . Here we have deﬁned ηSE as the fraction of spontaneously emitted photons that end up in the cavity mode (with proper direction. (Check your answers: Rpump = 1. and the extra factor of 2 in the last term accounts for the fact that the laser light counterpropagates through the medium. Use these assumptions. For solving ODEs. How does changing the ratio τp /τ21 aﬀect the laser dynamics? Perform sanity checks on your results.

4. τp := τrt /(1 − Ps ). Isat := ω0 /σ(ω0 )τ21 . Ignore degeneracy of the laser transition levels. dt′ nth nth τp τ21 I ′ n2 (d) Assume that the gain medium is pumped by a ﬂashlamp modeled by a square pulse of duration τpulse = 1 ms (i. and −1 A21 = τ21 .. with the initial conditions n2 (0) = I ′ (0) = 0 (assume ηSE = 10−12 ). then make the replacement nth −→ nth fQ (t′ ) in the above equations (in addition to the modiﬁcation you made in (d)). Plot the solution to the scaled rate equations for these conditions. Let fQ (t′ ) have the value 12 during the ﬂashlamp pulse (to model a Q-spoiled cavity) and then have the value unity immediately afterwards. dt τ τ τ where Ps := R1 R2 R3 R4 is the round-trip survival probability. These parameters produce a resonant laser cross section 2 σ(ω0 ) = 5. = Rpump N0 − A21 N2 − dt ω0 (Ps − 1) ℓg σ(ω0 )N2 ηSE ℓg ω0 A21 N2 dI = I+ I+ . The laser is well modeled as a four-level laser producing output at a wavelength of 1. (a) Recalculate the threshold inversion density N2 required for this laser to oscillate.) (b) The output mirror we chose is better for creating short pulses. n2 := N2 /N .064 µm. Compare your results to what you saw for the “long-pulse mode” of (d). The refractive index of YAG is n = 1. note that we are not properly treating eﬀects due to ampliﬁed spontaneous emission). ′ nth := Nth /N . The lasing transition has a lifetime τ21 = 1. which can be written in terms of the Heaviside step function as fpulse (t′ ) = UH (t′ ) − UH (t′ − τpulse /τp ). and ηSE is the fraction of spontaneously emitted photons that end up in the cavity mode (with proper direction. We will model Q-switching the cavity by allowing nth to be eﬀectively time dependent. The Nd dopant density is N = 1019 cm−3 . I ′ := I/Isat .9 × 10−19 cm2 and a saturation intensity Isat = 262 W/cm .. polarization. Nth := (1 − Ps )/ℓg σ(ω0 ). since it quickly couples energy out the resonator.8 Exercises 235 1. These equations are really only valid for a high-ﬁnesse resonator. 3. τrt is the cavity round-trip time. deﬁne a pulse function fpulse (t′ ) which is unity between t = 0 and τpulse . but with spontaneous emission into the cavity mode explicitly accounted for. so they will be marginally accurate but good enough for our purposes. (e) Now plot the laser output for a Q-switched pulse. Using the scaled and otherwise deﬁned variables t′ := t/τp .e.15. Rpump := Rpump τp . What is the corresponding pump rate that produces an inversion of 10 times the threshold value (if the laser is not oscillating)? (Check your answer: Rpump = 35 s−1 . You must generalize the formula you used last week slightly.2 ms and a line width ∆ω/2π = 120 GHz (with a Lorentzian line-shape function s(ω)). What is the duration of the Q-switched pulse (in ns)? . and ′ ′ then replace Rpump −→ Rpumpfpulse (t′ ) in the above equations). These are the same equations as before.83 at the laser wavelength. What reﬂectance of the output coupler would maximize the output intensity for CW operation? What is the corresponding output intensity? (c) We will model the laser with the pair of equations σ(ω0 )I dN2 N2 . etc. frequency. show that the above rate equations are equivalent to the following equations: τp dn2 ′ n2 − = Rpump(1 − n2 ) − dt′ τ21 ′ dI n2 n2 ′ = −I ′ + I + ηSE . do not use the formula given below for Nth . so the laser wavelength inside the gain medium is 580 nm. The gain medium has length ℓg = 10 cm. 2. I is the intensity circulating in the cavity.

show that this model is equivalent to the model in Problem 15. t′ := t/τp . Isat := ω0 /σ(ω0 )τ21 .236 Chapter 15. You can do this via the following outline. Rpump := Rpump τp .7) dn2 τp ′ n2 − = Rpump(1 − n2 ) − ′ dt τ21 ′ n2 n2 ′ dI = −I ′ + I + ηSE . and hence recover the equation for the second model above. Then use this to argue that the ﬁrst model can be rewritten (approximately) as (R1 R2 − 1) (G − 1) dI = I+ I.th ′ ′ and Rpump. I ′ := I/Isat . Under the conditions where this equation is valid. Show that the intensity oscillation frequency and damping rate are given by ωγ and γ/2. dt τ τ which is speciﬁc to the four-level laser (note that we have dropped the spontaneous emission term).6 dI (R1 R2 − 1) 2ℓg σ(ν0 )N2 = I+ I. Be careful to make sure your results are valid for a round trip. where ωγ = ω = ω 2 − (γ/2)2 r−1 τp τ21 r γ = τ21 (1 − nth ) ′ Rpump . Do this in the following steps: (a) Argue that both G and R1 R2 must be close to 1. Laser Physics Problem 15. nth := −1 ′ Nth /N . and A21 = τ21 . dt τ where the gain G saturates according to G=1+ g 1+ I Isat . Problem 15. n2 := N2 /N . you will compute the frequency and damping rates of the relaxation oscillations for small disturbances about the equilibrium (these are the same oscillations as in the laser spikes when they have nearly damped away). r := ′ Rpump. I is the intensity circulating in the cavity. τrt is the cavity round-trip time. For a ring laser modeled by the gain equations (as in Problem 15. ′ dt nth nth τp τ21 I ′ n2 where Ps is the round-trip survival probability. . dt τ τ (b) Derive an expression for the round-trip gain G in the regime where G is close to 1. not just a single pass through the gain medium. (c) Complete the argument by using the results from the four-level laser to write an expression for g. Recall the simple laser model dI (GR1 R2 − 1) = I.8. τp := τrt /(1 − Ps ).th is the scaled threshold pumping rate (the value of Rpump above which the laser oscillates).9. Nth := (1 − Ps )/ℓg σ(ω0 ). respectively.

the laser operates only in a single mode.10. Problem 15. so you must undo this to get the above answers. What is the underlying eﬀect that causes single-mode operation? Explain how this eﬀect leads to single-mode operation. Remember that the above rate equations measure time in units of τp .15. Show that the harmonic oscillator equation can be rewritten as the pair of ﬁrst-order equations x= ˙ p m p = −mω 2 x − γp. and write down expressions for the oscillation frequency and damping rate. Assume a low-loss ring cavity (nth ≪ 1. and discarding terms beyond ﬁrst order in δn2 and δI ′ . (d) What is mode-locking? Give an example of how to achieve mode locking in a multimode laser.2 ms). the solution with p(0) = 0 is x(t) = x(0)e−(γ/2)t cos(ωγ t). (This is a simple review question to help you tie together some of the concepts. Now draw a formal analogy between the laser gain equations and the damped harmonic ˙ oscillator. Roughly how many oscillations do you expect to see before they damp away? Problem 15. ˙ where p = mx. Your answers to the questions below should be qualitative and need not involve calculations or equations. (a) What is the condition for the laser to oscillate? (b) Suppose that many cavity modes satisfy the condition in (a). τp ∼ 10τrt ) just above threshold (r = 1. 2 A™¡ pump Rº™ 1 A¡º 0 } laser transition . where ωγ is deﬁned above. Explain how the medium is diﬀerent from the medium in part (b). (c) A diﬀerent gain medium in the same cavity causes the laser to oscillate in many modes simultaneously.th? (b) Ignore spontaneous emission and linearize the above pair of rate equations about the equilibrium ′ solution by letting n2 = nth + δn2 and I ′ = Iss + δI ′ .11.8 Exercises 237 ′ (a) Neglect spontaneous emission and show that in steady state. (c) Recall that the damped harmonic oscillator is governed by the equation m¨ + mγ x + mω 2 x = 0. and a Nd:YAG gain medium (τ21 = 1. (d) To get a sense of scale. What is ′ Rpump. put in some numbers to see a typical oscillation frequency. but in steady state. Consider the inverted three-level laser scheme shown here.1) with a ∼1 m cavity path length (τrt ≈ 3 ns).) Consider a laser whose gain medium is described by a frequency-dependent gain coeﬃcient γ(ω). x ˙ Recall that in the underdamped regime. putting these into the rate equations. You should now have a pair of rate equations for the small disturbances δn2 and δI ′ . n2 = nth and Iss = r − 1.

respectively. Argue qualitatively that in the best possible cases (optimal pumping and decay rates). (e) Give a qualitative argument for why the saturation intensity for the inverted three-level scheme will be twice that of the saturation intensity for the usual three-level scheme.4. laser cross section. A21 . Recall that the gain coeﬃcient is given by γ(ω) = σ(ω)[Ne − Ng ]. Assume the laser cross sections are the same in both cases. Hint: you don’t need to do any calculations.238 Chapter 15. Laser Physics (a) Write down the rate equations for this laser. the small-signal gain coeﬃcient for the usual scheme is twice the small-signal gain coeﬃcient for the inverted scheme shown here. where “e” and “g” refer to the excited and ground levels of the laser transition. (d) The usual three-level laser that we covered in Section 15. just reason out what the steady-state populations would be in the optimal cases. and A10 that ensure the best laser operation? (c) In the regime you described in part (b). (b) What are the conditions on R02 . for the same pumping rate. and atomic number density. carry out the adiabatic elimination of the appropriate state to obtain a set of rate equations for an eﬀective two-level atom. .1 has the lower two levels deﬁning the laser transition.

from the convolution. but this time in .4) we can compute the Fourier transform of Eq. the frequency-space multiplication in Eq. we will look at the eﬀects of dispersion—dependence of dielectric material properties on the optical frequency—on light propagation. P = ǫ0 χE.5. causality requires that P(t) can’t depend on E(t′ ) for future times t′ > t. By the convolution theorem.3 that i ˜ F [UH (t)] ≡ UH (ω) = πδ(ω) + . (16. we can write causality as the mathematical statement gχ (t) = gχ (t)UH (t). (16.1 Causality and the Kramers–Kronig Relations Recall from Section 4. we can put substantial constraints on its form. (16. For dispersive media.5) (16. using the convolution theorem again. Equivalently. Recalling from Section 12. We can also think of χ(ω) as the (complex-valued) frequency-space transfer function for the medium. It turns out that just by knowing that χ(ω) is a response function corresponding to a physical system. ˜ ˜ P(ω) = ǫ0 χ(ω)E(ω).Chapter 16 Dispersion and Wave Propagation Now that we have discussed the generation of laser light and the propagation of laser light through a gain medium.1) where χ is the susceptibility.4).2) ˜ ˜ where E(ω) and P(ω) is the magnitude of the component exp(−iωt).3) Since P(t) is the material response to E(t). (16. (16. we can see that causality implies gχ (t) = 0 for t < 0. the susceptibility is frequency-dependent (χ = χ(ω)). In particular.2) is equivalent to a convolution in the time domain of the electric ﬁeld with the inverse fourier transform gχ (t) of χ(ω): P(t) = ǫ0 ∞ −∞ gχ (t′ )E(t − t′ ) dt′ . Thus. ω (16.3 that the polarization density of a dielectric is the material response to the electric ﬁeld.2. and thus describing a causal response. where UH (t) is the Heaviside step function. 16. we will take a deeper look at propagation in media.

6) dω ′ The factor of 1/2π in applying the convolution theorem in the above derivation comes again from applying the convolution theorem in the reverse sense: the usual convolution theorem reads F [f (t) ∗ g(t)] = ˜ ˜ F [f (t)]F [g(t)]. (16. from Eq. Dispersion and Wave Propagation the opposite direction than usual: χ(ω) = F [gχ (t)] = F [gχ (t)UH (t)] = F F −1 [χ(ω)]F −1 [UH (t)] 1 ˜ χ(ω) ∗ UH (ω) 2π ∞ 1 i = χ(ω ′ ) πδ(ω − ω ′ ) + 2π −∞ ω − ω′ ∞ χ(ω) i χ(ω ′ ) = dω ′ + – 2 2π −∞ ω − ω ′ χ(ω) i = − H [χ(ω)] . Writing out the real and imaginary parts of this equation. deﬁned by H [f (x)] := 1 ∞ f (x′ ) – dx′ . we arrive at the standard form of the Kramers–Kronig relations: 1 ∞ Im[χ(ω ′ )] ′ – dω Re[χ(ω)] = π −∞ ω ′ − ω (16.9) (16.7) (16. (16.8) Writing out the Hilbert transform explicitly.240 Chapter 16. due to ˜ g the factor of 2π that appears in the inverse Fourier transform equation (3. Thus. 2 2 = Here again. The Kramers–Kronig relations for the permittivity then read ˜ ˜ ǫ 1 ∞ Im[˜(ω ′ )] ′ – dω π −∞ ω ′ − ω 1 ∞ Re[˜(ω ′ )] − ǫ0 ǫ Im[˜(ω)] = − – ǫ dω ′ . H denotes the Hilbert transform. π −∞ ω ′ − ω Notice that this is a strong constraint on the form of χ(ω): the real part of χ determines the imaginary part via the Hilbert transform and vice versa.10) 1 ∞ Re[χ(ω ′ )] ′ Im[χ(ω)] = − – dω . Re [χ(ω)] = H {Im [χ(ω)]} Im [χ(ω)] = −H {Re [χ(ω)]} . π −∞ x′ − x (16.6) we have that χ(ω) = −iH [χ(ω)] .3 that the electric displacement is given in terms ǫ ˜ of the permittivity by D(ω) = ǫ(ω)E(ω). π −∞ ω′ − ω Re[˜(ω)] = ǫ0 + ǫ Again. where we recall from Section 4. but in the opposite direction we have F −1 [f (ω) ∗ g (ω)] = 2πF −1 [f (ω)]F −1 [˜(ω)].21). ǫ (16. The Kramers–Kronig relations are often written in terms of the complex permittivity (dielectric constant) ˜(ω) = ǫ0 [1 + χ(ω)].11) . the real part of ˜ determines the imaginary part via the Hilbert transform and vice versa.

the material electrons fail to respond to the electric ﬁeld. (16. ω→±∞ lim ǫ(ω) = ǫ0 . n 1 ∞ Im[˜ (ω ′ )] ′ – dω π −∞ ω ′ − ω n 1 ∞ Re[˜ (ω ′ )] − 1 ′ dω .13) 16.15) ˜ ˜ This is another input-output relation.1 DC Component But one should then ask. in(w) response(w) out(w) For example. it most relevant to note that the Kramers–Kronig relations also apply to the refractive index. ˜ dz (16. the response function gχ (t) picks up the additional term χ∞ δ(t). which carries over ˜ into the dispersive case: (+) n n ˜ ˜ (+) E (+) (z) = E0 ei˜ k0 z −→ E (+) (ω. ǫ ˜ from D(ω) = ǫ(ω)E(ω). any causal response function with its asymptotic value subtracted away satisﬁes the Kramers–Kronig ǫ relations. with input E (+) .1. z) ˜ = ik0 n(ω)E (+) (ω.12) Thus. and transfer function proportional to n(ω). z)[1 + i˜ (ω)k0 dz]. Im[˜ (ω)] = − – n π −∞ ω′ − ω Re[˜ (ω)] = 1 + n (16. Eqs. why do the Kramers–Kronig relations for the susceptibility [Eqs. the refractive index (˜ (ω) − 1.17) . Thus. Recall the propagation of a plane wave in a medium with a complex refractive index n.0. we ﬁnd that if χ had an asymptotic dc component of χ∞ := χ(ω → ∞). ǫ looks like an input-output response function just like χ.1. If we repeat the derivation of Eqs.0. so that χ(ω) − χ∞ = −iH [χ(ω) − χ∞ ] . It works out if UH (t) passes “half” of δ(t). after subtracting the asymptotic dc part as in the discussion ˜ n of Section 16. ˜ n or in ineﬃcient form.1. since n tends to unity at high frequencies) obeys the Kramers–Kronig relations ˜ n(ω) − 1 = −iH [˜ (ω) − 1] .1.11) follow from the fact that [˜(ω) − ǫ0 ] is the appropriate response function for the Kramers–Kronig relations. (16. ˜ (16. where ǫ0 needs to be subtracted from Re[˜(ω)]? After all. z) = E0 (ω)ei˜ (ω)k0 z .16) (16. they apply to any frequency-space transfer function that corresponds to a physical. (16.18) (16. which is absent in χ.11)] have diﬀerent forms. This is because at very large frequencies. (16. causal response.16. Rather. z). (16.1 Causality and the Kramers–Kronig Relations 241 16.1 Refractive Index Notice that the Kramers–Kronig relations are not speciﬁc to electromagnetism.14) The corresponding inﬁnitesimal evolution is ˜ ˜ E (+) (ω.6). Thus. The diﬀerence lies in the ˜ ˜ ˜ dc component of ǫ(ω). n which is equivalent to the diﬀerential equation ˜ dE (+) (ω. they could apply to the response function of an electronic ampliﬁer or feedback loop. output dE (+) /dz. z + dz) = E (+) (ω.10)] and the permittivity [Eqs. and thus ω→±∞ lim χ(ω) = 0. (16. For our purposes.

2π[(ω0 − ω)2 + (∆ω/2)2 ] Im[˜ (ω)] ≈ n so that Re[˜ (ω)] − 1 ≈ n a0 π∆ω δ(ω − ω0 ). variation in the absorption coeﬃcient implies regions of strong (real) refractive index. n Since Re[˜ (ω)] controls the phase shift of propagating light and Im[˜ (ω)] controls the absorption (gain). and a0 = a(ω0 ). The second expon nential factor represents exponential decay (or gain if Im[˜ (ω)] < 0).c. where we can write the solution as n n ˜ ˜ (+) E (+) (ω. = 2 0 i + c.16). It isn’t too diﬃcult to compute the Hilbert transform without approximations. = 2(1 − ix) x . (16.c. z) = E0 (ω)eiRe[˜ (ω)]k0 z e−Im[˜ (ω)]k0 z .20) The convention here is that a(ω) is the intensity absorption coeﬃcient. for example. =− 1 + x2 (16. while the absorption coeﬃcient scales as (ω0 − ω)2 . approximate method. Similarly.3).11): H 1 = −F −1 [iπ sgn(k) exp(−|k|)] 1 + x2 −i ∞ π exp(−k)eikx dk + c.25) . n n we see why the Kramers–Kronig relations are so signiﬁcant.c. Since the Hilbert transformation is “something like” a derivative (see Section 12. 16. Let’s ﬁrst do this with a simple. We can do this using the convolution theorem. This is evident from Eq. (16. Then we can approximate the Lorentzian by a δ-function. Dispersion and Wave Propagation Recall that the complex refractive index corresponds to both phase delay and absorption (or gain).19) The ﬁrst exponential factor is a propagating plane wave with wave number Re[˜ (ω)]k0 .5.1 Example: Lorentzian Absorption One particularly important absorption line shape is the Lorentzian [see Eq. and the fact that the Fourier transform of 1/x is i sgn(k) (see Problem 12.37)].242 Chapter 16. Let’s compute the Hilbert transform of the scaled Lorentzian 1/(1 + x2 ). which we can write as Im[˜ (ω)] = n (a0 /2k0 ) (∆ω/2) a(ω) = . (15.22) In this case. the fact that the Fourier transform of 1/(1 + x2 ) is π exp(−|k|) (see Problem 11.2). ˜ and vice versa.1. Note that a0 > 0 corresponds to absorption.21) to ﬁnd the phase behavior.24) ′−ω 4k0 ω 4k0 (ω0 − ω) −∞ Notice that the refractive index scales as (ω0 − ω) for large |ω0 − ω|. while a0 < 0 corresponds to gain. Now we need to evaluate the integral Re[˜ (ω)] − 1 = n 1 ∞ Im[˜ (ω ′ )] ′ n – dω ′−ω π −∞ ω (16. (16. Thus. variation with frequency of the (real) refractive index implies regions of absorption. derivatives in the real part of n(ω) induce large imaginary parts. Assume that a(ω) is sharply peaked around ω0 (∆ω ≪ ω0 ).2.23) ∞ ∆ω a0 δ(ω ′ − ω0 )] ′ a0 ∆ω – dω = . so that I(z) = I0 exp[−a(ω)z]. 2k0 2 (16. This turns out to be an important result in atom optics.1. by considering the normalized form of the line shape: ∆ω ≈ δ(ω − ω0 ). (16. 2k0 (ω0 − ω)2 + (∆ω/2)2 2 (16. = 2π 0 −i ∞ −(1−ix)k e dk + c.

and then we will generalize to the case of pulses with a spectrum of ﬁnite width.2 Group Velocity But what happens with pulses. Then the phase is the same before and after the transformation.28) so that k dx = ω dt. (16. Recall the plane-wave solution to the electromagnetic wave equation: E (+) (x. This deﬁnes the phase velocity vph as vph := dx ω = . vph = ω c0 =c= . We can deﬁne the phase velocity as the velocity of the points of constant phase.30) √ where n is the refractive index.2 Pulse Propagation and Group Velocity 16.1. for a dispersive dielectric medium. the refractive index depends on the optical wavelength and thus on the wave number.2 Pulse Propagation and Group Velocity 243 Then with x = 2(ω − ω0 )/∆ω.2. To derive an expression for the phase velocity. we know that the dispersive refractive index must also be complex. so when we write n here.16.27) The phase of the wave is thus φ = kx − ωt. ˜ 16.29) Recall that for a plane wave to be a solution of the wave equation. φ = k(x + dx) − ω(t + dt) = kx − ωt. k n (16.26) We will derive this formula again when we examine models of atomic media. the phase velocity vph = vph (k) is likewise wavelength-dependent. which contain a continuum of frequency components? In a dispersive medium. We will start out by considering the monochromatic case. dt k (16. t) = E0 ei(kx−ωt) . (16. while φ is held constant. each component sees a diﬀerent phase velocity: E (+) (x.1 Phase Velocity Now we will consider the velocity of light propagation in a dispersive medium. n ∆ω (16. consider the inﬁnitesimal transformation x −→ x + dx and t −→ t + dt. Im[n~(w)] w Re[n~(w)]o-o1 16. (+) (16. we can write Re[˜ (ω)] = 1 + n 2(ω0 − ω) Im[˜ (ω)]. t) = 1 2π ∞ 0 ˜ E (+) (k)ei(kx−ωt) dk. Thus.31) .2. In general. and c0 = 1/ ǫ0 µ0 is what we deﬁned to be the speed of light in vacuum in Section 4. By the Kramers–Kronig relations. we are referring only to the real part of n. n = n(k).

n= vph so we can deﬁne an analogous group index for the group velocity by ng := dk c0 dn = c0 =n+ω .31) is of the form x− ω0 t = (x − vph t) . and all the spacetime dependence is of the form E (+) (x. (16. the pulse propagates at the group velocity vg = dω/dk. so the pulse propagates at the phase velocity vph = ω0 /k. all the components have the same phase velocity. k0 ω (16. dω ω(k) = ω0 + (k − kc ) = ω0 + vg (k − kc ).39) . For ω = ω0 . k (16.35) Thus.31) as E (+) (x. E (+) (x. t) = E (+) (x − vg t).32) Thus. but let’s assume that the dispersion relation is linear. kc (16. vg Now the phases outside the integral are uninteresting overall phase factors. t) = E (+) (x − vph t). In either case. which controls the phase velocity. More generally. where ω0 is a constant. Note that there are multiple relevant propagation velocities here: while the phase fronts of the individual waves still propagate at the appropriate phase velocities.38) since k = nω/c0 . (16. Dispersion and Wave Propagation Here we account for dispersion by regarding ω = ω(k) and use k as the independent variable. we can expand an arbitrary dispersion relation to ﬁrst order about a center wave number kc . the vacuum wavelength is λ0 = 2π 2πc0 = . (16.34) whose signiﬁcance will become apparent presently. but the ideas here are generally applicable.244 Chapter 16.36) vph Recall that the usual refractive index. (16. so all the time dependence in Eq. This works for small deviations of the refractive index from unity. which we can do via ω(k) = kvph (k) = c0 k/˜ (k).37) (16. and we have deﬁned the group velocity vg := dω dk . such n that k(ω) is an invertible function. Now suppose we have a dispersive medium. Also.33) dk kc where ω0 = ω(kc ). Now we can write the pulse function (16. t) = ∞ 1 ˜ E (+) (k)ei[kx−ω(k)t] dk 2π 0 ∞ 1 ˜ E (+) (k)ei(kx−ω0 t) e−ivg (k−kc )t dk = 2π 0 1 ik0 vg t −iω0 t ∞ ˜ (+) e e = E (k)eik(x−vg t) dk 2π 0 (16. The expansion of Eq.33) is valid if the pulse spectrum is narrow enough for the higher-order terms to be neglected. can be written c0 . the overall envelope of the pulse propagates at the group velocity. vg dω dω (16.

1. (dn/dλ0 > 0) Here the pulse envelope travels more quickly than the phase fronts. Normal dispersion. 1. =n+ω = n − λ0 dω dλ0 dω dλ0 dλ (16. since the normal refractive index n is commonly speciﬁed in terms of wavelength. there is a region of width ∆ω (usually very narrow) of anomalous dispersion.2. 2. For the Lorentzian line-shape function (Section 16.40) where the last equality follows from assuming that n(λ) is a slowly varying function. For example. a more careful analysis of the pulse shape shows that the signal can’t be sent at a higher velocity than c0 . 16. Im[n~(w)] normal dispersion w anomalous dispersion Re[n~(w)]o-o1 normal dispersion In the anomalous case. Then dλ0 λ0 2πc0 =− 2 =− . Here. 3. dω ω ω and so we can write the group index (16. broadly speaking three cases to consider with respect to the group index.1.43) 1 d2 ω 2 dk 2 . this happens when the index increases with wavelength. since it seems possible to send a signal pulse faster than the speed of light.38) as ng = n + ω dn dn dn dλ0 dn ≈n−λ . when the group and refractive indices are equivalent: ng = n.3 Pulse Spreading Higher order terms in the dispersion relation lead to pulse spreading (second order) or distortion (third and higher orders). up to second order. where the slope of the index curve changes sign. t) = 1 ik0 vg t −iω0 t e e 2π ∞ 0 2 ˜ E(k)(+) eik(x−vg t) eiω2 (k−kc ) t dk (16.2 Pulse Propagation and Group Velocity 245 where k0 = k/n is the vacuum wave number.44) .16. and are thus deﬂected more by the prism. (dn/dλ0 < 0) Here the pulse envelope travels more slowly than the phase fronts. This case is called “normal” because this is the observed behavior when white light is dispersed by a glass prism: blue wavelengths see a larger refractive index than red wavelengths.41) (16. and implies that ng < n.31) becomes E (+) (x. There are. This “superluminal” propagation seems to violate (relativistic) causality. normal dispersion occurs everywhere except near the line center. kc (16.1). and implies that ng > n. where ω2 := Then the pulse function (16. No dispersion. However. it is even possible to have vg > c0 . (dn/dλ0 = 0) This is the nondispersive case that we already discussed.42) (16. Anomalous dispersion. this happens when the index decreases with wavelength. This form for the group index is convenient. ω(k) = ω0 + vg (k − kc ) + ω2 (k − kc )2 .

t) = E0 (+) ∆x q exp ik(x − vg t)2 ikc x e . (16. 71). (6. 2 (16. we can deﬁne a q parameter for this problem by q := τ − iq0 .5 (p. while the ones with negative x-components move towards the other. ω2 is related to the group velocity dispersion.13)]. w(z) = w0 1+ z z0 2 . (16. (16. To see this. E (+) (r) = E0 (+) q0 ikr2 exp(ikz). Since the Gaussian beam satisﬁes the expansion law [Eq.45) (16. the plane waves with wave vectors with positive x-components will move towards one edge of the beam proﬁle. so that the propagation replacement becomes (∆x)2 −→ (∆x)2 + iω2 t = iq .46) Thus. where q0 = 2(∆x)2 and τ = 2ω2 t. exp q(z) 2q(z) (16. 2q (16.54) R(z) = z 1 + z The Gaussian beam acquires curvature because it is a superposition of plane waves with slightly diﬀerent directions in their wave vector k.51) This has the same form as the Gaussian beam in terms of the q parameter [Eq. and as the beam propagates.47) which disperses the pulse in the same way as the complex q parameter of the Gaussian beam of Section 6. which controls this dispersion eﬀect. Dispersion and Wave Propagation For a Gaussian pulse. The Fourier transform after propagation through time t is 2 (+) √ ˜ E (+) (k. The other aspect of the complex q parameter is that it controls the radius of curvature of the Gaussian beam [Eq.53) the analogue in pulse propagation in a dispersive medium (by comparing the two forms of the q parameter) is that the pulse width ∆x increases with a similar functional form with the time parameter τ . (16. the propagation to time t leads to a larger eﬀective width (∆x)2 −→ (∆x)2 + iω2 t. the inverse transform is E (+) (x.52) 2 where q = z − iz0 = z − iπw0 /λ.40)].246 Chapter 16. (6. (6. t = 0) = E0 e−x so that the Fourier transform is approximately 2 2 (+) √ ˜ E(k) ≈ E0 2 π(∆x)e−(∆x) (k−kc ) . these begin to separate. given by dvg /dω. Here.48) Now we can take the inverse Fourier transform. t) = E0 2 π(∆x)eiq(k−kc ) e−ikvg t . E (+) (x. z0 2 .12)]. and the red/blue wavelength components begin to move to the front/back of the pulse (or the other way around. (16.49) (16.50) and so up to an overall phase. depending on the sign of ω2 ). The analogue in the pulse propagation is that the Gaussian pulse is a superposition of plane waves of diﬀerent wavelength. (16. . For propagation in the +z-direction.44). (+) 2 /4(∆x)2 ik0 x e . thus evaluating the integral of Eq.

1. c0 (16. Wolf.max = n + ω dn dω =1+ 1+ max ω δnmax . see Robert W. how is it possible to make vg very small or very large? Answering this question is fundamentally interesting. but the line width is quite narrow (∆ω/2π ∼ 6 MHz).62) Thus. Boyd and Daniel J. ed. 2∆ω (16. 2k0 (ω0 − ω)2 + (∆ω/2)2 2 (16. vol. we can write the maximum group index as ng.59) Let’s put in some typical numbers to see the magnitude of the eﬀect. we need to be more crafty when setting up a medium with an extreme group index. The maximum value √ of the derivative occurs at (ω − ω0 ) = ± 3∆ω/2. giving dn dω = max a0 δnmax = .38)].1. We roughly follow some of their arguments here.56) ∆ω The maximum value of |n − 1| occurs for (ω − ω0 ) = ±∆ω/2. “‘Slow’ and ‘Fast’ Light.” in Progress in Optics. (16. giving a maximum deviation of n from unity of n = Re[˜ (ω)] = 1 + n δnmax = a0 . 497.3 Slow and Fast Light A recent “hot topic” in the quantum optics community has been the slow and fast propagation of light in atomic media. n (16. 43.60) an absolutely enormous index (so that pulses should propagate at ∼ 100 m/s.58) Thus. For a vapor at room temperature.3 Slow and Fast Light 247 16. p.1 The general question is. 1 For a good review.61) However.max ∼ 2 × 104 . we need to consider the derivative dn/dω.1. 4k0 (16. but still quite large: ng. (Elsevier. and can also be technologically interesting: for example. E. it is practically impossible to observe these eﬀects in ordinary vapors. Again. The corresponding group index is smaller. the line width is more like ∆ω/2π ∼ 1 GHz due to Doppler broadening (in fact the line-shape function should be Gaussian rather than Lorentzian.16. For alkali vapors. (16. . From Section 16. Gauthier. Let’s consider a medium with Lorentzian absorption (appropriate for stationary gas-phase atoms). Putting in these numbers gives ng. the resonant frequency is in the optical (ω0 /2π ∼ 400 THz).1).57) (a0 /2k0 ) (∆ω/2) a(ω) = . 8k0 ∆ω 2∆ω (16. Amsterdam.max ∼ 3 × 106 . this is valid for stationary atoms. the absorptive part of the refractive index is thus given by Im[˜ (ω)] = n while the phase index is 2(ω0 − ω) Im[˜ (ω)]. 2002). but we’ll ignore this here).55) To evaluate the group index [Eq. because a large group index is accompanied by strong absorption: a ∼ (δn)4k0 = 4ω(δn) ∼ 3 × 104 cm−1 . (16. a “slow” medium can act as a compact delay line or even memory storage for optical information. the refractive index is not too large (n ∼ 1.

which we will not get into here. A pulse propagating through the medium is thus delayed compared to pulse propagation without the medium. Im[n~(w)] w Re[n~(w)]o-o1 The physical signiﬁcance of the dip is that the medium is transparent to the probe ﬁeld. For small intensities. where two ﬁelds (the pump and probe) couple the two ground states to the excited state. even though the absorption is small. Rather. this eﬀect comes about in a medium of three-level atoms in the “Λ” conﬁguration.3. this is a quantum-mechanical eﬀect. the width of the dip can be small. we need a fully quantum-mechanical description of the atom. the group index is correspondingly large. Again. on the order of the decay rate between the ground states (which can be as low as Hz to kHz). the pump saturates the probe transition. Due to quantum interference. Since dn/dω is large.63) ωprobe − ωeg2 = ωpump − ωeg1 . so we can’t describe this via the rate-equation approach that we used to model lasers.248 Chapter 16.1 Quantum Coherence: Slow Light The solution to the problem of acheiving a high group index in a transparent medium is to use a quantum interference eﬀect called electromagnetically induced transparency (EIT). Dispersion and Wave Propagation 16. when it is properly tuned. |eÒ probe pump |g™Ò |g¡Ò This creates a narrow “dip” in the usual Lorentzian absorption spectrum for the probe transition. Roughly speaking. . The dip also induces a steep slope in the refractive index in the same frequency range. but only if the two detunings match precisely: (16.

(16. “Gain-assisted superluminal light propagation.” Nature 406. 3 L. M. Dogariu.2 Fast Light The alternative to slow light is “fast” light. as we can see from Eq.3 is to use two pump beams to induce two gain peaks in the refractive-index proﬁle. so that the group velocity was vg = −c0 /310. Rochester.” 2 D. Given the continuity conditions at the boundaries. Thus.16.” Physical Review Letters 83. “Nonlinear Magneto-optics and Reduced Group Velocity of Light in Atomic Vapor with Slow Ground State Relaxation. and V.3.2 16. These appear as negative peaks in the absorption curve. . Kuzmich and A. F. Kimball. V. In the experiment of Wang et al.3 Slow and Fast Light 249 n nogo o>o n o n t A number of slow-light experiments have achieved impressively small group velocities. and the propagation appears to be “faster than c0 . down to 8 m/s for the slowest achieved velocities. A. 277 (2000). where the backwards-propagating pulse meets the incoming pulse and annihilates it. 1767 (1999). this happens if a “pulse pair” is created at the far boundary. The magnitude of the derivative can be large enough that the group index can become negative. Budker. J. Yashchuk.38). One approach that was realized experimentally by Wang et al. so that the sign of dn/dω becomes negative. S. Wang. The outgoing pulse is thus advanced with respect to the “normal” propagation of the incoming pulse. Re[n~(w)]o-o1 w Im[n~(w)] This arrangement ﬂips the slope of the phase index. D.. a pulse entering the medium must propagate backwards through it. the group index was ng = −310.

Lasers (Wiley. for example. and thus ˜(−ω) = ǫ∗ (ω). 4 See. Dispersion and Wave Propagation nogo o<o 0 o n t Again. Use the following outline. there can be no violation of causality.250 n Chapter 16. Eberly. – π 0 ω ′2 − ω 2 Re[˜(ω)] = ǫ0 + ǫ To do this. you will need to argue that the time-domain polarization response gχ (t) applies to real ﬁelds and polarizations.3. c0 dλ0 where ∆λ0 is the wavelength range of the pulse. Show that for a rareﬁed medium (|χ| ≪ 1).2. and d is the propagation distance. Show that the pulse spread due to propagation in a ﬁber composed of a dispersive material can be written4 λ0 d2 n ∆tpulse ≈ 2 (∆λ0 ) d. and in fact this behavior is completely consistent with the Kramers–Kronig relations that enforce causality.1. 16. the refractive index n ≡ Re[˜ ] and intensity n absorption coeﬃcient a can be written in terms of the susceptibility as n≈1+ Re[χ] 2 a ≈ k0 Im[χ]. . Problem 16. ǫ ˜ Problem 16.4 Exercises Problem 16. Peter W. 1988). Show that the Kramers–Kronig relations for the complex permittivity ǫ 1 ∞ Im[˜(ω ′ )] ′ – dω π −∞ ω ′ − ω 1 ∞ Re[˜(ω ′ )] − ǫ0 ǫ Im[˜(ω)] = − – ǫ dω ′ π −∞ ω′ − ω Re[˜(ω)] = ǫ0 + ǫ can be written in the equivalent form ǫ 2 ∞ ω ′ Im[˜(ω ′ )] ′ dω – ′2 − ω 2 π 0 ω 2ω ∞ Re[˜(ω ′ )] − ǫ0 ǫ Im[˜(ω)] = − ǫ dω ′ . Milonni and Joseph H.

d2 n/dλ0 = −0.68374049400 λ0 0. (a) Show that the optical phase of the wave propagates at velocity ω/k. fusedsilica ﬁber where the material makes the dominant contribution to dispersion. (b) What are the conditions for the envelope to propagate backwards? 5 “HPFS Fused Silica KrF Grade.0296 µm−2 at 850 nm. Consider a bichromatic optical wave of frequencies ω1 and ω2 propagating through a dispersive medium.00112 µm−2 at 2 2 −2 1310 nm. k := (k1 + k2 )/2. 2003. Problem 16. Give your answers in units of ps/(nm · km). where ω := (ω1 + ω2 )/2. Assume that |∆ω| ≪ ω and that the two frequency components have equal amplitudes. ∆k := k2 − k1 .” Corning Incorporated data sheet.00472 µm at 1550 nm. 1310 nm. and n1 and n2 are the refractive indices at frequencies ω1 and ω2 .49327320000 . we can use the Sellmeier formula for fused silica5 n2 (λ0 ) − 1 = (c) Compute the pulse spread per unit length per unit of wavelength spread for a single-mode.58502748000 λ0 2 − 0. .00460352869 + λ 2 − 0. n n dλ0 (b) Then use the fact that the pulse propagation time is tpulse = d/vg and ∆vg ≈ (dvg /dλ0 )∆λ0 to 2 obtain the above expression. ∆ω := ω2 − ω1 . To get the refractive-index derivatives. ω1 = c0 k1 /n1 . λ0 0 0 2 2 where λ0 is in µm. ω2 = c0 k2 /n2 . Calculate the dispersion at the popular diode-laser wavelengths of 850 nm. and d n/dλ0 = −0. This gives d2 n/dλ0 = 0.16.4. while the envelope of the interference pattern propagates at velocity ∆ω/∆k. respectively.01339688560 + λ 2 − 64. and 1550 nm. in the regime where (dn/dλ0 )2 ≪ |d2 n/dλ0 |.4 Exercises 251 (a) Show that the group velocity may be written vg ≈ λ0 dn c0 1+ . 2 2 2 0.42032361300 λ0 0.

.

when an atom is a manifestly quantum-mechanical object? It turns out that the classical calculation gets many phenomena correct. Essentially.4) where ε is the unit polarization vector.Chapter 17 Classical Light–Atom Interactions We will now model the interaction between light and atoms. We will model the atom as a classical harmonic oscillator. Thus. x (17. Generally me ≪ mn . In writing down this expression. it requires a bit of patching to make it quantitatively correct. using a classical model of the atom. (17. Thus. so m ≈ me 1 − me mn . and mn is the nuclear mass. The above equation is also in centerof-mass coordinates. the classical calculation is good for small excitations. the electron is not localized. even in the limit of small intensity. In particular. the lowest-order approximation to an arbitrary potential. x represents the average position of the electron. m is the reduced mass of the electron.5) . The force on the electron due to the ﬁeld is F(+) = −eE(+) . (17.3) and generally. ˆ (+) (17. and ω0 is the resonant frequency of the harmonic potential.1 Polarizability We will now consider the interaction of the atom with a monochromatic ﬁeld of the form E(+) (t) = εE0 e−iωt . we are making the dipole approxˆ imation: we are assuming that the size of the atom is much smaller than the optical wavelength. (17. Why use a classical calculation. 17. is an accurate model. we need not consider the spatial dependence or propagation direction of the ﬁeld. when the harmonic potential. so that the electron only sees the ﬁeld at the nuclear position. an electron bound to the nucleus by a harmonic force (linear spring): 2 m¨ + mω0 x = 0. since quantum-mechanically. and as we will see. and these results can be justiﬁed by a quantum calculation.1) Here. so that we can ignore the motion of the nucleus.2) m= me + mn where me is the electron mass. it is a good approximation to use m ≈ me . given by me mn . This will allow us to treat a variety of phenomena from the refractive index of atomic vapors to the conductivity of metals to laser cooling and trapping of atoms. the classical model does not predict any saturation eﬀects.

17. We now see if we can assume that the solution has the same time dependence as the ﬁeld: x(+) (t) = εx0 e−iωt . where the interactions between the atoms are negligible. the magnitude of the electron charge (so that the electron charge is −e). (17. ˆ With this solution. .254 Chapter 17. this is the frequency-space response function. 2 ω0 − ω 2 (17. The dipole moment of the atom is d(+) = −ex(+) .10) where x = εx. 2E ω 2 − ω0 0 (17.11) From Eqs.9) Again.8) to obtain the solution x0 (+) = eE0 /m 2. We can thus write the susceptibility for the vapor as N e2 /mǫ0 χ(ω) = 2 (17. Of course. (17.9) and (17.12) The polarizability completely characterizes the response of the atom to the applied ﬁeld.3) is the dipole moment per unit volume. Thus. In dense media. Eq. we will ignore any motion except that induced by the ﬁeld. (17.13) This expression is valid for a rareﬁed medium. we can write the polarizability as α(ω) = e2 /m . correlations between dipoles cause deviations from these results. Then the equation of motion for the electron becomes 2 m¨ + mω0 x = −ˆeE0 e−iωt . ω 2 − ω0 (+) (17. Since the dipole moment is induced by the ﬁeld (the electron displacement is zero in ˆ equilibrium). Classical Light–Atom Interactions where e is the fundamental charge.10). as we will justify when considering the damped version of the harmonic oscillator.1 Connection to Dielectric Media Recall that the polarization density P (from Section 4.6) becomes − mω 2 x0 which we can solve for x0 (+) (+) 2 + mω0 x0 (+) (+) (17.1. P(+) = N d(+) = N α(ω)E(+) = ε ˆ N e2 /m (+) −iωt e . we are breaking the electron displacement into its positive and negative components x(t) = x(+) (t) + x(−) (t).14) 2 . x ε (+) (17. ω − ω0 in view of the deﬁning relation P = ǫ0 χE.7) = −eE0 e−iωt . which we have obtained via an implicit Fourier transform of the applied ﬁeld. we can deﬁne the polarizability α to describe how easily the ﬁeld induces the dipole moment by d(+) = α(ω)E(+) . for an atomic vapor of number density N . (17. (+) (17.6) We need only worry about the electron motion in the direction of the electric ﬁeld.

this model is valid for a rareﬁed plasma. n is purely imaginary.2 Damping: Lorentz Model 255 17. the damping rate is the same as the Einstein A coeﬃcient (spontaneous emission rate): γ = A21 . (17. (17.17. reﬂecting a phase lag of the displacement behind the ﬁeld. 2.20) ω − ω0 + iγω Now the displacement is complex.1. In this case.17) n(ω) = 1 + χ(ω) = 1 − ˜ ω The behavior is quite diﬀerent in the limiting cases of low and high frequency. Here n −→ 1. (17. and the equation of motion is 2 m¨ + mγ x + mω0 x = −ˆeE0 e−iωt . Then for this case. x ˙ ε (+) (17. Following the method above. (+) Again.16) ωp N e2 /mǫ0 =− ω2 ω 2 . Again. so the reﬂection coeﬃcient (Section ˜ 9.5) is of the form n1 − in2 . the polarizability becomes e2 /m . (17. (17. the plasma frequency occurs in the ultraviolet.2 Conducting Media: Plasma Model In conducting media. the electrons are not tightly bound to the nucleus. with phase angle γω .15) is the plasma frequency. modeling the conductor as a plasma.5) is given in terms of the susceptibility by ωp 2 . we assume a solution of the form x(+) (t) = εx0 e−iωt . so it is only qualitatively correct for metals.22) α(ω) = 2 ω0 − ω 2 − iγω . High frequency (ω ≫ ωp ).19) The damping (“friction”) term models radiation reaction due to the charge acceleration (the classical analogue of spontaneous emission) and collisions with other atoms.18) r= n1 + in2 where n1 and n2 are real. we ﬁnd perfect reﬂection from a metal surface for frequencies below the plasma frequency. 17. The complex refractive index (see Section 9. Then the susceptibility becomes χ(ω) = − where ωp := N e2 mǫ0 (17. Thus.2 Damping: Lorentz Model A better model of the atom is a damped harmonic oscillator. We can treat this by letting ω0 −→ 0.21) δ = tan−1 2 ω0 − ω 2 The phase lag approaches zero for ω ≪ ω0 and π for ω ≫ ω0 (δ = π/2 exactly on resonance). ˜ For typical metals. This improved model is known as the Lorentz model of the atom. 1. so that metals are transparent to ultraviolet and shorter wavelengths. A quantum-mechanical calculation shows that for an isolated atom. the ˆ solution is (+) eE /m (+) x0 = 2 0 2 . so that |r| = 1. so that the metal becomes transparent at high frequencies. (17. where density-induced correlations are more important. Low frequency (ω < ωp ).

Classical Light–Atom Interactions The susceptibility likewise becomes χ(ω) = It is worth reiterating here that α and χ are complex quantities deﬁned for the positive-rotating ﬁelds via d(+) = α(ω)E(+) and P(+) = ǫ0 χE(+) . n(ω) = ˜ 1 + χ(ω) ≈ 1 + 2 χ(ω) (ω0 − ω 2 ) N e2 γω N e2 +i =1+ 2 − ω 2 )2 + γ 2 ω 2 2 − ω 2 )2 + γ 2 ω 2 . . With this approximation.25) This is eﬀectively equivalent to the rotating-wave approximation in quantum optics.1. 2 mǫ0 c0 (ω0 − ω 2 )2 + γ 2 ω 2 (17.28) Hence.26) 2 N e2 (ω0 − ω 2 ) .27) (17. The oscillator strength can only be obtained from a quantum calculation. we will explore it in more detail. f0j is the absorption oscillator strength. and is necessary to make the classical calculation quantitatively correct. If χ is small (as for a dilute vapor). In general.24) Then we can read oﬀ the phase index as the real part.29) as required by the Kramers–Kronig relations [Eq. we see that in the same regime. (17. |ω − ω0 | ≪ ω0 . mǫ0 c0 γ (ω0 − ω)2 + (γ/2)2 The region where signiﬁcant absorption occurs for small detunings of the ﬁeld from the atomic resonance.256 Chapter 17.5 and 16. Also. Because of the quantitative importance of this factor. the phase index and absorption become n(ω) ≈ 1 + (ω0 − ω)/2ω N e2 2mǫ0 (ω0 − ω)2 + (γ/2)2 (γ/2)2 N e2 a(ω) ≈ . Then the polarizability and susceptibility become f0j e2 α(ω) = 2 m ω0j − ω 2 − iγj ω j (17. The quantum-mechanical (and correct) interpretation of these expressions is that each term in the sum represents a transition from the ground level 0 to excited level j. n−1= 2(ω0 − ω) N e2 ω (γ/2)2 2(ω0 − ω) = Im[˜ (ω)]. 2 2mǫ0 (ω0 − ω 2 )2 + γ 2 ω 2 (17.30) f0j N e2 χ(ω) = . 2 2mǫ0 (ω0 2mǫ0 (ω0 N e2 /mǫ0 .2. n γ 2mǫ0 γ (ω0 − ω)2 + (γ/2)2 γ (17. we can have atoms with multiple electrons that we need to sum over.1 (hence the name) with full width at half maximum ∆ω = γ and resonant absorption coeﬃcient a0 = N e2 /mǫ0 c0 γ. which acts as a weighting factor for each electron. n(ω) ≈ 1 + while the absorption coeﬃcient becomes a(ω) = 2k0 Im[˜ (ω)] ≈ n N e2 ω 2 γ . 2 ω0 − ω 2 − iγω (17. Then 2 ω0 − ω 2 = (ω0 − ω)(ω0 + ω) ≈ 2ω(ω0 − ω). 2 mǫ0 ω0j − ω 2 − iγj ω j Here. we recover the Lorentzian absorption proﬁle from Sections 15.1.26)].23) (17. and therefore must be treated appropriately. (16.

17.2 Damping: Lorentz Model

257

**17.2.1 Oscillator Strength
**

Since the absorption coeﬃcient scales as the susceptibility and thus the oscillator strength (for a dilute gas), the oscillator strength also scales with the cross section [see Eq. (15.36) for the laser cross section that we deﬁned earlier]. On resonance, the cross section for absorption is deﬁned by a(ω0 ) = σ(ω0 )N = σ0 N. Thus, using Eq. (17.26), we can write the classical absorption cross section as σclassical (ω0 ) = γ e2 ω 2 2 mǫ0 c0 (ω0 − ω 2 )2 + γ 2 ω 2 =

ω=ω0

(17.31)

e2 . mǫ0 c0 γ

(17.32)

This cross section is not quantitatively correct, as the correct cross section from Eq. (15.39) for the transition to level j is 2 2 λ0j πc0 (17.33) = 2 . σ0j = σj (ω0 ) = 2π ω0j Note that this cross section assumes an orientational average (and thus a factor of 1/3) that is generally appropriate for our purposes. We can then deﬁne the absorption oscillator strength to be the “fudge factor” to ﬁx the classical cross section: 3 σ0j 2πǫ0 mc0 γj f0j := = . (17.34) 2 σ0, classical e2 ω0j We can also write the cross section as σ0j = f0j e2 , mǫ0 c0 γj (17.35)

which will be useful later. More commonly, the absorption oscillator strength is deﬁned to include the degeneracy of the level structure,1 3 2πǫ0 mc0 γj gj f0j = , (17.36) 2ω 2 e 0j g0 with a separate expression deﬁned for the emission oscillator strength f0j (which just ﬂips the degeneracy ratio). Also, in the limit of large frequency, the susceptibility of Eqs. (17.30) becomes χ(ω) −→ − N e2 mǫ0 ω 2 f0j .

j

(17.37)

In this limit, the induced electron displacements are small, and thus the damping and harmonic-potential forces are not important. We thus expect to recover the behavior of the free-electron plasma of Eq. (17.15) in the high-frequency limit (i.e., the conductor without damping): χ(ω) = − ωp ω

2

=−

N e2 . mǫ0 ω 2

(17.38)

Comparing these two expressions, we ﬁnd the Thomas–Reiche–Kuhn sum rule for the oscillator strength: f0j = 1.

j

(17.39)

Since f0j > 0, the sum rule tells us that f0j < 1. The interpretation is that the classical cross section represents the maximum possible cross section, which turns out to be distributed over all the possible transitions from the ground level. Note that transitions to unbound (ionized) states are also included in this sum, making it diﬃcult to verify this with atomic transition data.2

1 Alan 2 See

Corney, Atomic and Laser Spectroscopy (Oxford, 1987). Peter W. Milonni and Joseph H. Eberly, Lasers (Wiley, 1988), p. 239.

258

Chapter 17. Classical Light–Atom Interactions

**17.2.2 Conductor with Damping: Drude Model
**

We can again use the Lorentz model to derive an improved model of conductors or plasmas. This model, the Drude model, accounts for damping, and accounts for electron collisions in the metal or plasma. We again take the Lorentz-model results and let ω0 −→ 0 so that the electrons are not bound. In this limit, Eqs. (17.30) give 2 −ωp −N e2 /mǫ0 = 2 . (17.40) χ(ω) = ω 2 + iγω ω + iγω The complex refractive index becomes n(ω) = ˜ In the dc limit (ω −→ 0), n(ω) −→ ˜ 1+i 1 + χ(ω) = 1−

2 ωp . ω 2 + iγω

(17.41)

2 ωp ≈ γω

i

2 ωp . γω

(17.42)

Thus, the refractive index is not purely imaginary at any nonzero frequency, even though the imaginary part is much larger than the real part. This leads to a reﬂectance that is slightly less than unity, which is a distinct improvement over the plasma-model result of perfect reﬂectance. In the high-frequency limit, the damping part is not important, and the Drude model reduces to the undamped plasma model. The imaginary (damping) part of the conductivity here becomes signiﬁcant at infrared and higher frequencies, gradually rolling oﬀ the material reﬂectance below the plasma frequency. Another useful quantity for a conductor is the induced current density, which we can write as J = −N ex. ˙ Since x = −iωx, we can use the solution for x0 ˙

(+)

(17.43)

from Eq. (17.20) to write N e2 /m E γ − iω (17.44)

J=

where

in the unbound limit. Recalling that the conductivity σ is deﬁned by J = σE, we can write the Drude-model conductivity as N e2 /m σ0 σ= = , (17.45) γ − iω 1 − iω/γ σ0 := N e2 mγ (17.46)

is the dc conductivity. We thus have a model for the frequency dependence of the metal conductivity that we ignored in Section 9.5. It is worth reiterating that this is only a simple model and valid for rareﬁed plasmas.

**17.3 Atom Optics: Mechanical Effects of Light on Atoms
**

Now we will have a brief look at the ﬁeld of atom optics, or optics with matter (de Broglie) waves. We will only be looking here at how to trap and cool atoms with laser light using the classical Lorentz model of the atom, so in a sense we will be doing “geometrical atom optics.” Broadly speaking, there are two types of mechanical forces that light can have on atoms. The ﬁrst, the dipole force, is related to the potential energy of the induced dipole in the electric ﬁeld, and is thus related to the real part of α(ω) [see Eq. (17.53)]. The second is radiation pressure due to absorption and rescattering of the incident light, which is thus related to the imaginary part of α(ω) [see Eq. (17.89)].

17.3 Atom Optics: Mechanical Effects of Light on Atoms

259

**17.3.1 Dipole Force
**

The dipole moment of the atom is induced by the external ﬁeld, so we can write the potential energy of the induced dipole as dE d·E =− . (17.47) Vdipole = − 2 2 The extra factor of 1/2 compared to the usual dipole energy is because the dipole is induced, and thus

E E

Vdipole = −

0

(d) dE = −

0

1 αE dE = − αE 2 . 2

(17.48)

Since we found the solution for the positive-frequency component of the ﬁeld, we should write out the potential in terms of the same components: Vdipole = − Noting that we can see that the terms of the form d(±) ∼ e∓iωt , E(±) ∼ e∓iωt , (17.50) (17.51) 1 d(+) + d(−) · E(+) + E(−) . 2 (17.49)

rotate at twice the optical frequency, which is too fast for the atoms to respond mechanically. So we will drop these terms in the time average (the same average that leads to the intensity). The terms of the form d(±) · E(∓) ∼ 1 are dc, so we can keep these. Thus, 1 1 Vdipole = − d(+) · E(−) − d(−) · E(+) 2 2 1 1 α(ω)E(−) · E(+) = − α(ω)E(+) · E(−) − 2 2 = −Re[α(ω)] E =−

(+) 2

d(±) · E(±) ∼ e∓i2ωt

(17.52)

(17.53)

η0 Re[α(ω)]I(r). 2

Here, η0 is the vacuum wave impedance from Eq. (4.60), and we are regarding the electric-ﬁeld envelope E (+) (r) to be a slowly varying function of position. Putting in the explicit form for the polarizability, we can write the dipole potential as Vdipole =

2 ω0 − ω 2 −e2 I(r). 2 2mǫ0 c0 (ω0 − ω 2 )2 + γ 2 ω 2

(17.54)

Thus, the atom sees a spatial potential proportional to I(r) and to (n − 1). This potential-shift eﬀect (also known as the ac Stark shift) is the dual of the phase shift (due to n − 1) of a beam propagating through a vapor. Both eﬀects follow from the coupling of the ﬁeld to the atom. The corresponding force is given by the potential gradient Fdipole = −∇Vdipole ∝ ∇I(r). (17.55)

Thus, the dipole force responds to intensity gradients. If the dipole is viewed as two slightly separated, opposite charges, there is only a net force if the two charges see a diﬀerent electric ﬁeld, which is only possible in the ideal dipole limit if the ﬁeld has a gradient.

260

Chapter 17. Classical Light–Atom Interactions

The sign of the dipole potential is set solely by the detuning of the ﬁeld from the atomic resonance. Deﬁning the detuning ∆ := ω − ω0 , we can write the dipole potential as Vdipole = e2 (ω0 + ω)∆ I(r). 2mǫ0 c0 [(ω0 + ω)∆]2 + γ 2 ω 2 (17.56)

Everything in this expression is positive except for the factor of ∆ in the numerator. Thus, for positive ∆ (ω > ω0 , or blue detuning), Vdipole > 0, while for negative ∆ (ω < ω0 , or red detuning), Vdipole < 0. That is, a bright spot in space (e.g., due to a tightly focused Gaussian beam) will repel an atom for blue detunings, forming a potential barrier, while for red detunings, the spot attracts atoms and forms a potential well. The sign dependence of Vdipole makes sense in terms of the phase lag (17.21). Recall that for small frequencies (∆ < 0), the phase lag of the dipole behind the ﬁeld is smaller than π/2, while for large frequencies (∆ > 0), the phase lag is between π/2 and π. Since Vdipole ∝ −d · E, the phase lag is important because then d and E are mostly aligned or mostly opposed for ∆ < 0 and ∆ > 0, respectively. Thus, Vdipole ≷ 0 for ∆ ≷ 0. 17.3.1.1 Dipole Potential: Standard Form By writing the dipole potential in a more standard form, we can see that it matches the result of a quantum calculation, at least in the limit of low intensity. To do this, we ﬁrst need to patch the classical result of Eq. (17.54) as before by including the oscillator strength and summing over all transitions: Vdipole = −

2 ω0j − ω 2 e2 f0j I(r). 2 2mǫ0 c0 (ω0j − ω 2 )2 + γj2 ω 2

(17.57)

j

Now to put this in more standard form, we need to deﬁne the saturation intensity for the atom. We deﬁned the saturation intensity before in Eq. (15.73) for a four-level laser, but we need to redo this for the two-level atom. Recall that a nondegenerate, two-level atom has the steady state solution when driven on resonance from Eq. (15.23) N2 B12 ρ(ω0 ) = . (17.58) N1 A21 + B21 ρ(ω0 ) Using N1 + N2 = N and B12 = B21 , we can solve this to ﬁnd N2 I/2Isat B21 ρ(ω) σ0 I/ ω0 = , = = N A21 + 2B21 ρ(ω) A21 + 2σ0 I/ ω0 1 + I/Isat where σ0 is the resonant cross section and we have deﬁned the saturation intensity as Isat := ω0 γ , 2σ0 (17.60) (17.59)

and we have identiﬁed γ = A21 . Note that this deﬁnition diﬀers by a factor of 2 from the four-level laser deﬁnition (15.73), since here the ground state is not naturally depleted. For the maximum possible resonant cross section of σ0 = 3λ2 /2π (where there is no average over the dipole orientation), the saturation intensity is Isat = 1.10 mW/cm2 for 133 Cs on the D2 transition (852 nm), while for 87 Rb on the same transition (780 nm), the saturation intensity is Isat = 1.67 mW/cm2 . We can also write the saturation intensity in terms of the oscillator strength by using Eq. (17.35), with the result Isat,j = ω0j mǫ0 c0 γj2 . 2e2 f0j (17.61)

Even though the above numerical values are often quoted for the saturation intensity, this is actually a context-dependent quantity (as we can see from the diﬀerence of its value in the two- and four-level cases). A safe but cumbersome approach is to use the quantum-mechanical formalism for angular momentum to directly calculate the cross section and thus saturation intensity.3

3 Daniel

A. Steck, “Cesium D Line Data,” 2003. Available online at http://steck.us/alkalidata.

17.3 Atom Optics: Mechanical Effects of Light on Atoms

261

**Using Eq. (17.61), we can write the dipole potential (17.57) as Vdipole = −
**

2 ω0j γj2 ω0j − ω 2 I(r) . 2 − ω 2 )2 + γ 2 ω 2 I 4 (ω0j sat,j j

(17.62)

j

This is the general expression for any frequency, so long as the intensity is small. To simplify this, we can look at the functional form far away from all resonances (|ω0j − ω| ≫ γj for all j) so that Vdipole =

j

=

j

γj2 8

ω0j γj2 1 I(r) 2 −ω2 4 ω 0j Isat,j 1 1 − ω − ω0j ω + ω0j I(r) . Isat,j

(17.63)

The ﬁrst term in the parentheses is the inverse of the detuning, and represents the Stark shift due to the atomic resonances. The second term can be interpreted as the weak, additional Stark shift due to resonances at the corresponding negative frequencies. This secondary shift is always negative (like a red detuning), and corresponds to the Bloch–Siegert shift. Note that this expression also recovers the dc Stark shift (or equivalently, the dc polarizability up to some universal factor) when ω = 0, when both terms contribute equal, negative energy shifts. If one resonance is dominant (that is, the laser is tuned far away from resonance, but much closer to one than all the others), then we can make the rotating-wave approximation and neglect the second term in the parentheses of Eq. (17.63) to obtain Vdipole = γj2 I(r) , 8∆ Isat (17.64)

where again ∆ = ω − ω0 is the detuning from resonance. Note that for a far-detuned, linearly polarized laser 2 creating this potential, it turns out that σ0 = λ0 /2π is the appropriate resonant cross section, so the above saturation intensity values should be multiplied by 3 before being used in this formula. Typically, a focused, red-detuned, Gaussian laser beam is used to make a dipole trap or far-oﬀ resonance trap (FORT)4 for atoms via the dipole force.5 Below is an example image of about 105 87 Rb atoms conﬁned in a dipole trap formed by a 10 W, 1090 nm Gaussian laser beam (far below the 780 and 794 nm main resonances) focused to a 31 µm beam waist (1/e2 radius), implying a Rayleigh length (depth of focus along the beam direction) of 2.8 mm.

The dipole trap clearly runs from left to right with a slight downward angle; the dimensions of the image are 270 × 29 CCD pixels (6.59 × 0.71 mm). This is an absorption image, where the shadow cast by the atoms in a brief, resonant laser probe is imaged and recorded on a CCD camera. (The image greyscale is inverted so the atoms appear bright rather than dark.) But now the important question to address is under what conditions the trap is stable, since as the atom scatters photons, it heats up until it boils out of the trap. So we will need to take a closer look at the radiation of the Lorentz atom.

**17.3.2 Radiation Pressure
**

Now we will examine the forces due to absorption and reemission of the incident light. We will begin by examining the radiation process in detail.

4 Steven L. Rolston, Christoph Gerz, Kristian Helmerson, P. S. Jessen, Paul D. Lett, William D. Phillips, R. J. Spreeuw, and C. I. Westbrook, “Trapping atoms with optical potentials,” Proceedings of SPIE 1726, 205 (1992) (doi: 10.1117/12.130392). 5 The ﬁrst observation of atoms trapped in a dipole-force potential was Steven Chu, J. E. Bjorkholm, A. Ashkin, and A. Cable, “Experimental Observation of Optically Trapped Atoms,” Physical Review Letters 57, 314 (1986) (doi: 10.1103/PhysRevLett.57.314).

. Milonni and Joseph H. √ 2 2. + (ˆ × r ) ε ˆ H(+) (r. Only the 1/r terms actually transport energy to inﬁnity (i. t) ≈ 4πc0 r E(+) (r. t) = (17. Eberly. and ε is the polarization unit vector of the applied ﬁeld (and thus ˆ the dipole orientation vector).3. 1988). There are two main possibilities for the polarization vector: the incident light can be linearly or circularly polarized.65) where tr = t − r/c0 is the retarded time.68) for the angular dependence. Classical Light–Atom Interactions 17. which we can write as S = E(+) × H(−) + c. This is the usual “doughnut-shaped” radiation ε ˆ r ˆ pattern for an oscillating dipole.c. 1. Jackson. while φ is the angle around the azimuth.c. ε ˆr ˆ ε ˆ r2 ¨ |d(+) |2 r ˆ 2 1 − |ˆ · ε| + c.e.66) (17. Note that any arbitrary polarization can be represented as a superposition of these three basis vectors.c. Here. D.2.. Lasers (Wiley. 2nd ed. they correspond to radiation). The (intensity/power) radiation patterns for the linear and circular dipole cases are shown here. p. 2 . t) = 2 2 4π c0 r c0 r E(+) (r. p.262 Chapter 17. t) ≈ The energy transport is governed by the Poynting vector. so we can drop the rest to obtain ¨ 1 d(+) (tr ) ε ˆr ˆ 2 [(ˆ · r )ˆ − ε] 4πǫ0 c0 r ¨(+) (tr ) d 1 (ˆ × r ) ε ˆ . r ˆ = 2ǫ c 3 16π 0 0 r2 = 1 16π 2 ǫ 3 0 c0 (17. 6 See J. ¨ |d(+) |2 [(ˆ · r)ˆ − ε] × (ˆ∗ × r ) + c. Linear polarization (ˆ = z ): 1 − |ˆ · ε| = sin2 θ.67) where we have used [(ˆ · r )ˆ − ε] × (ˆ∗ × r) = 1 − |ˆ · ε|2 r ε ˆr ˆ ε ˆ r ˆ ˆ (17. H(+) (r. Circular polarization (ˆ = ε± := ∓(ˆ ± iˆ)/ 2): 1 − |ˆ · ε| = (1 + cos2 θ)/2. 1975). 395 or Peter W. This is a “peanutε ˆ x y r ˆ shaped” radiation pattern for a rotating dipole.1 Dipole Radiation The electric and magnetic ﬁelds for an oscillating dipole are6 ˙ ¨ 1 d(+) (tr ) d(+) (tr ) d(+) (tr ) 1 + [3(ˆ · r )ˆ − ε] ε ˆr ˆ + [(ˆ · r )ˆ − ε] ε ˆr ˆ 2 3 2 4πǫ0 r c0 r 4πǫ0 c0 r ¨ ˙ c0 d(+) (tr ) d(+) (tr ) . θ is the angle from the z-axis. (Wiley. 44. Classical Electrodynamics.

3 Atom Optics: Mechanical Effects of Light on Atoms 263 z circular dipole linear dipole x The three-dimensional distributions are generated by sweeping these patterns around the z-axis.69) For linear and circular polarization. φ) := ˆ 3 2 1 − |ˆ · ε| .70) This function has the nice property that it is normalized. this takes the form 3 sin2 (θ) 8π 3 1 + cos2 (θ) . f± (θ. while for circular polarization (ˆ = ε± = ∓(ˆ ± iˆ)/ 2).71) The power radiated per unit solid angle is then ¨ |d(+) |2 dPrad = r2 S · r = ˆ ˆ 3 fε (θ. this factor turns out ε ˆr ˆ ε ˆ √ ˆ to be sin θ θ. we can see that the polarization vector is proportional to (ˆ· r)ˆ− ε.66). the polarization vector is proportional ε ˆ x y √ ˆ ˆ to (cos θ θ ∓ iφ)e∓iφ / 2. ˆ 3 3πǫ0 c0 r2 (17. r ˆ 8π (17. (17. For linear polarization (ˆ = z ). Now let’s deﬁne the angular-distribution function via fε (θ. φ). φ). From Eq. ˆ Here. and thus represents a probability distribution for photon emission in quantum mechanics: fε (θ. φ) dΩ = 1. dΩ = sin θ dθ dφ is the usual solid-angle element. Now we can write the Poynting vector in terms of the angular-distribution function as S = ¨ r ˆ |d(+) |2 fε (θ. φ) = 16π fz (θ. The corresponding electric ﬁelds for the dipole radiation are polarized. dΩ 3πǫ0 c0 (17. φ) = ˆ (17.17.72) (17.73) .

2. the incident intensity is contained implicitly in the electron acceleration x. dΩ 3πǫ0 c0 3πǫ0 c0 (17.7 Eq. there are some subtleties here regarding the cross sections and orientational averages that are better handled by angular-momentum algebra. no orientational average for the dipole). Atomic and Laser Spectroscopy (Oxford.79) then we recover the damping term in the harmonic oscillator: F(+) = −mγ z (+) . since we used the complex representation. 6πmǫ0 c0 (17. F(+) = rr If we deﬁne γ= 2 e2 e2 ω0 (+) ˙˙˙ x(+) ≈ − ˙ .6). (17. we can make the replacement |¨(+) |2 −→ x x2 ¨ . which isn’t quite correct. 7 This argument follows Alan Corney. Then the average work done by radiation reaction must balance the energy emitted into the ﬁeld: x x0 Frr · dx′ = t t0 Frr x(t′ ) dt′ = − ˙ t e2 (¨)2 dt′ x 3 6πǫ0 c0 t0 t t e2 x¨ − ˙x x˙˙˙ dt′ .77) Going back to the complex representation and noting that the displacement is a harmonic function.80) This is the classical result for the spontaneous emission rate.76) Here Frr refers to the radiation-reaction force. Then the radiation-reaction force is Frr = e2 ˙˙˙ 3 x.e. (17. ˙ rr (17. (17. 230.264 Chapter 17.81) 2 This is consistent with Eq. Classical Light–Atom Interactions and the total radiated power is Prad = dΩ ¨ e2 |¨(+) |2 x |d(+) |2 dPrad = = 3 3 .34) if we take σ0j = 3λ0 /2π (i. p. In terms of the real displacement.3. 3 3x 6πǫ0 c0 6πǫ0 c0 2 e2 ω0 3.75) where the angle brackets denote the time average.2 Damping Coefﬁcient Now we can connect the radiated power to the damping term in the Lorentz model. we can patch this with the substitution e2 /m −→ (e2 /m)f0j . the boundary term vanishes (it is also negligible for large t − t0 ). If we pick t− t0 to be an integer multiple of the optical period..78) (17. Again. 2 (17. with the result γj = 2 e2 ω0j f0j 3. 6πǫ0 c0 (17. 1987).74) Of course.74) is the time-averaged power. Note that the radiated power in Eq. 6πmǫ0 c0 (17. Again. ˙x =− 3 6πǫ0 c0 t0 t0 (17. . ¨ 17.

while red wavelengths are preferentially transmitted.84) Using Eq. 6πǫ0 c0 The overall scaling is as ω 4 (neglecting the small modiﬁcation due to the polarizability).j . 2 2 Prad = 6πc0 j γj ω2 2 ω 2 − ω 2 − iγ ω ω0j 0j j 2 I 2 = 2 j ω √ 2 ω0j ω0j γj − ω 2 − iγj ω 3/2 (17. As a brief aside.87) j Again. We can write the total radiated power from Eq. we ﬁnd Rsc = η0 ω 2 2 Im[α]I(r). given the deﬁning relation Prad = σI: ω 4 |α(ω)|2 (17. We can continue by writing out explicitly the polarizability in Eq.3 Atom Optics: Mechanical Effects of Light on Atoms 265 17. (17. using Eq.30) restricted to a single resonance. This is the usual explanation for why the sky is blue and sunsets are red: blue wavelengths are preferentially scattered by the atmosphere.82) where we used d(+) = α(ω)E0 e−iωt .86) The photon scattering rate Rsc is the radiated power divided by the photon energy ω: Prad = Rsc = ω γj ω 3/2 2 − ω 2 − iγ ω 2ω0j ω0j j 3/2 2 I Isat.17. I Isat.j 2 where we used σ0j = 3λ0 /2π to write the saturation intensity as 3 ω0j γj ω0j γj = 2 .82). (17.88) ω3 I γ3 = 2 2ω0 (ω0 − ω 2 )2 + γ 2 ω 2 Isat Using Eq.89) . (17. 6πǫ0 c0 (17.j = (17. 2σ0j 4πc0 Isat. (17. this expression simpliﬁes greatly for certain detunings.83) σRayleigh = 2 4 .85) . (17. ω0 (17. (17. we can ignore the contribution of the others: ω3 I γ3 Rsc ≈ 2 − ω 2 − iγω|2 I 2ω0 |ω0 sat (17.81) to eliminate the oscillator strengths. (17.2. we can write down the scattering cross section from the total radiated power.3 Photon Scattering Rate Now we can compute the rate of photon scattering as a way to get to the rate of momentum transfer.74) in terms of the polarizability as Prad = (+) ω 4 |α(ω)|2 2 4 I.3.30): 2 Prad e2 ω 4 = 2 2 6πm2 ǫ0 c0 j 2 ω0j f0j − ω 2 − iγj ω I. though. Near one resonance.

as we will discuss.92) ˆ where k is the unit vector along the direction of the wave vector (propagation direction) k.5 mm/s. 17.e. Thus. .4 Scattering Force Each photon carries a momentum of k0 . which will be convenient when discussing the heating rate.. and for 87 Rb at 780 nm. whether or not the hyperﬁne structure is resolved). since the classical ﬁeld momentum is related to the absorbed beam power: F = dp/dt = Pabs /c0 = σI/c0 .94) Again. so the recoil velocity is orders of magnitude smaller than typical room-temperature velocities. To get a sense of scale of the momenta involved we can compute the recoil velocity vr . (17.90) Rsc ≈ 2 I 8∆ sat ∆ where in the last formula we have used Eq.3 Laser Cooling: Optical Molasses Now let’s explore how we can use the radiation-pressure force to cool atoms. but still close enough for the resonance to still dominate.88) is Rsc = so that the radiation pressure force becomes Frad = k0 (γ/2)3 I ˆ k. (17. so some caution is necesary in applying these formulae. the appropriate value of the saturation intensity might be very diﬀerent. 2 + (γ/2)2 I ∆ sat (17.266 Chapter 17. Classical Light–Atom Interactions which shows the connection of the scattering rate (and hence the radiation pressure force below) to the absorptive part of the polarizability. deﬁned as the velocity corresponding to one photon recoil momentum k0 : vr = k0 . Note that even though we have invoked the concept of the photon here. and compensate by increasing the intensity. dipole traps with small heating rates and hence long lifetimes (up to minutes for dipole traps created by CO2 laser light) can be created in this way. The photon scattering rate represents heating of the atoms.3. we ﬁnd that γ γ3 I = Vdipole . m (17. vr = 3. as discussed above. far-detuned dipole trap where these scalings are valid.9 mm/s. (17.3. exposed to identical but counterpropagating laser ﬁelds along the velocity direction. Vdipole ∝ . because of the random nature of photon emission. But the scattering rate and dipole potential scale as I I Rsc ∝ 2 . The simplest setup we can consider is an atom moving with velocity v. 2 + (γ/2)2 I ∆ sat (17. depending on the polarization and exactly how close the detuning is (i. the scattering (heating) rate can be made very small by making the detuning ∆ large. Far away from the dominant resonance (|∆| ≫ γ). Close to a single resonance. Thus. vr = 5.64) for the dipole potential.2.93) For 133 Cs at 852 nm.91) ∆ ∆ so that for a given desired potential depth. the scattering rate from Eq. This result is of prime importance in the design of an optical dipole trap. the same saturation intensities are to be used to calculate the dipole force and scattering rate. the photon scattering rate implies a rate of momentum transfer and thus a radiation pressure force of ˆ Frad = k0 Rsc k. 17. Note that for a linearly polarized.95) I (γ/2)3 . (17. everything here is really classical. (17.

Cable.1 × 106 s−1 . although it turns out that the optimal detuning is actually something else for reasons we will soon discuss. and the two oﬀset Lorentzians are shown as dashed lines. only fairly slowly moving atoms can be cooled Chu. Isat (17. this expression is the diﬀerence of two Lorentzians. but the idea behind Doppler cooling is to tune the lasers below the atomic resonance. A. γ)/k0 ]. typically leading to heavily overdamped motion for trapped atoms. √ this light conﬁguration is called optical molasses. The velocity capture range is the range in velocity for which the force is appreciable. Thus. and A. (17. With the pictured setup. (17. the capture range is about ±2 m/s. E. each displaced by |∆|/k0 from zero velocity.99) Frad = 0 2 + (γ/2)2 ]2 I 2 [∆ sat Because this force is damping for ∆ < 0.96) where ∆1. the frequency of laser 1 is Doppler shifted (red shifted) by −k0 v.98) Regarded as a function of velocity.2 − ω0 .97) ∆2 = ∆ + k0 v. “Experimental Observation of Optically Trapped Atoms. we can expand to lowest order in v to obtain the viscous damping (“friction”) force: ∆ I k 2γ3 v.” Physical Review Letters 57. λ0 = 852 nm) and 87 Rb (γ = 38. J. Ashkin.3 Atom Optics: Mechanical Effects of Light on Atoms 267 laser 1 v laser 2 The radiation-pressure force on the atom due to the two ﬁelds from Eq. Bjorkholm.8 The maximum damping rate occurs for ∆ = −γ/2 3.2 = ω1. (17.100) ± k 2k 4π assuming ∆ ∼ −γ/2. Thus. thus tending to stop the atom. so that the beam that opposes the atomic velocity is Doppler-shifted into resonance. λ0 = 780 nm). the capture range is on the order of γ γλ |∆| ∼± =± . Frad v ±|D|/kº For small velocity [v ≪ max(|∆|. This force is plotted for ∆ = −γ/2. we can write ∆1 = ∆ − k0 v (17.17. 8 Steven . Then the force is Frad = k0 (γ/2)3 1 1 − 2 + (γ/2)2 (∆ − k0 v) (∆ + k0 v)2 + (γ/2)2 I Isat . For both 133 Cs (γ = 32.2 are the eﬀective detunings of the two lasers. where ∆ = ω − ω0 is the detuning in the laboratory frame. 314 (1986).8 × 106 s−1 . The detunings of the two lasers are the same in the laboratory frame.95) is Frad = k0 (γ/2)3 1 1 − 2 2 ∆1 + (γ/2)2 ∆2 + (γ/2)2 I . while the frequency of laser 2 is Doppler shifted (blue shifted) by +k0 v. Since the detunings are given by ∆1. (17.

a scattering event is eﬀectively equivalent to two steps in a random walk in velocity space. atoms were slowed by other methods from hot atomic beams to below the capture velocity and then trapped. v = 0 dt ma [∆2 + (γ/2)2 ]2 Isat (17.268 Chapter 17. (17. ma (17. There are also ﬂuctuations of the cooling force that lead to a temperature limit. However. and the direction is random but along one of the six beams. Our treatment so far makes it appear as though the atomic velocity may be damped completely away. The emission is also in a random direction (not in a dipole-radiation pattern if we assume all polarizations to be equally present). and the angle brackets denote an ensemble average.99) for the average cooling force.3. the velocity damps to zero for ∆ < 0. Let’s look at the variance of the velocity distribution: dv d 2 v =2 v· dt dt = 2 ma v· dp dt = 2 v · Frad . With the small-velocity expression (17.1 Doppler Cooling Limit For laser cooling in three dimensions. However. this equation of motion becomes d 2 I ∆ k 2γ3 v2 . In the course of scattering a photon from one of the laser beams. one along each axis. Now we will include the force ﬂuctuations heuristically. the Doppler limit. according to this diﬀerential equation. Frad = 2 k0 γ 3 ∆ I v.103) Again. These scattering events happen at the scattering rate 6I (γ/2)3 . 2 [∆2 + (γ/2)2 ]2 Isat (17.102) Here. for the cooling mechanism presented here. we have only considered the average cooling force. Thus. there is a photon absorption and a photon emission. where the step size is k0 /ma . it is suﬃcient to simply combine three of the above one-dimensional setups. leading to a second kick of magnitude k0 in a random direction.3. 17. We will now derive this temperature limit.101) where I is still the intensity of a single beam. .9 Then we can write the force vector for small velocities as. Classical Light–Atom Interactions at all with this method. Traditionally to load atomic traps.104) Rsc = 2 ∆ + (γ/2)2 Isat 9 Graphics by Windell Oskay. since the ﬂuctuations are quantum–mechanical in origin (although there is a more general connection between damping and ﬂuctuations known as the ﬂuctuation–dissipation theorem). Each absorption leads to a momentum “kick” of magnitude k0 . it is possible to load a trap from room-temperature vapor with this method by capturing only the small fraction of the atoms with small enough velocity. ma is the atomic mass.

3 Atom Optics: Mechanical Effects of Light on Atoms 269 since there are six beams present. E. Phillips. R.17. It turns out that for alkali vapors. which is the temperature corresponding to atoms with an average momentum of one photon recoil k0 (i. 133 Cs can be laser cooled with the same general setup described above to about 2. Dalibard and C.e. which we can convert to a temperature via 1 3 ma v 2 = kB T. Since the (one-dimensional) rms velocity is vrms = TD Tr k0 ma . L. The probability distribution is also inversion symmetric in either case. “Laser cooling below the Doppler limit by polarization gradients: simple theoretical models.105) so that v 2 increases at the rate 2 2Rsc Including the heating rate in Eq.11 10 Sub-Doppler temperature were observed in some of the ﬁrst laser cooling experiments. Clairon. These temperatures are extremely low.110) γ . typical laser-cooled samples exhibit temperatures well below the Doppler limit. Cohen-Tannoudji. The ﬁrst reported observation is P. given by the variance after one step starting from the origin.5 µK.113) which is 8. each step increases v 2 by a ﬁxed amount. 11 C. 2084 (1989). A classic treatment of sub-Doppler cooling mechanisms is J..107) (17. N. For 133 Cs at 852 nm. and for 87 Rb at 780 nm. The three-dimensional probability distribution for a single scattering event is conﬁned to a shell of radius k0 /ma in velocity space for either the absorption or emission event.112) 133 The temperature is minimized for the detuning ∆ = −γ/2.5. TD = 125 µK. D. Such “sub-Doppler” cooling is due to the degenerate level structure of alkali atoms. A.” Europhysics Letters 12.8 cm/s for 133 Cs and 12 cm/s for 87 Rb.111) 2 This temperature is the best expected for Doppler cooling. 4ma (−2∆/γ) k0 ma 2 . and for 87 Rb. 2023 (1989). Dalibard. We can compare these temperatures to the recoil temperature Tr . TD = 146 µK.” Journal of the Optical Society of America B 6. C. Rolston. D. (17.1 that for a random walk. Recall from Section 11. 683 (1990). Cs and (17. and S.108) This is an expression for the equilibrium kinetic energy.” Journal of the Optical Society of America B 6. we ﬁnd . and C. W. we can write 2 vinitial = 0 −→ k0 ma 2 vﬁnal = k0 ma 2 . Salomon. Tanner. Lett. S. (17. I. then after one step in the random walk.109) 2 2 where kB is the Boltzmann constant.10 For example.106) 1 I I ∆ 3γ 3 k 2γ3 d 2 v2 + v = 0 2 + (γ/2)2 ]2 I dt ma [∆ 2 ∆2 + (γ/2)2 Isat sat In steady state. we can set the right-hand side to zero. This gives kB T = γ 1 + (2∆/γ)2 . (17. Phillips. with the result v2 = 3 γ 1 + (2∆/γ)2 . Guellati. giving the Doppler temperature TD : For 133 Cs. D. W. (17. (17. (17. Westbrook. ma (17. Watts. “Laser Cooling of Cesium Atoms below 3 µK.103). Thus if the atom is initially at rest. so the Doppler limit is TD = 631 Tr for TD = 403 Tr for 87 Rb. 4 (−2∆/γ) (17. J. Tr = 198 nK. “Optical molasses. a one-dimensional rms momentum of one photon recoil): kB TD = kB Tr = ( k0 )2 . These velocites are about three orders of magnitude slower than room-temperature rms velocities. Tr = 362 nK. .

Steven Chu. The additional complication is that the laser beams must all be correctly (circularly) polarized to address magnetic substates in the degenerate excited level.4 Exercises Problem 17. Pritchard.2 Magneto-Optical Trap Optical molasses tends to stop atoms.3. Classical Light–Atom Interactions 17. Dalibard proposed the idea for the magneto-optical trap. 13 Jean 12 Graphics . the appropriate beam comes into resonance and pushes it towards the trap center. This arrangment is called the magnetooptical trap (MOT). “Trapping of Neutral Sodium Atoms with Radiation Pressure. Alex Cable. What are the two types of optical forces on atoms? How are they related to the atomic polarizability? by Windell Oskay.3.” Physical Review Letters 59. thus deﬁning a point for atoms to accumulate. such that if the atom is away from the center of the trap. 2631 (1987). L.13 17. making them “stuck. M. but essentially the idea is very similar to laser cooling. and the MOT was ﬁrst demonstrated by E. and D. The magnetic ﬁeld gives a position-dependent “Zeeman” shift of the transition frequency. E. Prentiss. We will not go into the operation of the trap in detail. Raab.” but it does not conﬁne atoms to a particular place.270 Chapter 17.1.12 The magnetic ﬁeld vanishes at the center point of the trap. A slight modiﬁcation to the three-dimensional optical molasses is to impose the magnetic ﬁeld due to two opposed current loops in the “anti-Helmholtz” conﬁguration.

203–205 time. 66 beam splitter. 67 birefringence. 184–189. 11 coherence. 144 of two Gaussians. 74 ABCD matrix. 18. 196 Raman–Nath regime. 14 Cauchy principal value. 257–258 Fresnel relations for. 87 Q switch. 258–270 dipole force. 183–198 Bragg regime. 125–129 confocal parameter. 58. 228–229 Abbé v-constant. 66 Bessel function. 258 conductors Drude model. 73 Q factor. 114 Q-switching. 228 Boltzmann statistics. 142 Brewster’s angle. 139–140. 206 complex dielectric constant. 115–116 Bragg reﬂector. 214 boundary conditions. 126 conductivity. 266–269 radiation pressure. 181 central limit theorem. 71 factorization of general. 143 of box functions. 198 action functional. 27 antireﬂection (AR) coating single layer. 189–198 Klein–Cook parameter. 172 Cauchy probability distribution. 143–153 deﬁnition. 171. 71–75 cascading rule. 142 two layer. 195–196 acousto-optic modulator as spectrum analyzer. 31 aberration theory.Index ABCD law. 60 beam waist parameter. 232 air-glass interface cool movie. 254–255 wave propagation in. 203–205 coherence time. 197–198 TeO2 modulator. 208 anamorphic prism pair. 70 confocal resonator. 146–149 characteristic function. 180 Bessel functions. 126 complex notation. 185 generating function for. 121 Airy disk. 268–269 magneto-optical trap (MOT). 181 atom optics. 17 absorption. 73 deeper meaning. 256–257 acousto-optic diﬀraction. 147 characteristic polynomial. 119 bug in laser. 196–198 rise time. 258–261 Doppler temperature. 48–50 complex refractive index. 145 271 . 185 big mess. 140–142 approximations silly. 141 central dark-ground imaging. 31 bleaching. 152 cavity dumper. 30 adiabatic approximation. 216–217 absorption image. 229 CeF3 . 96. 261 absorption oscillator strength. 26 convolution. 180 ampliﬁed spontaneous emission. 126 beam radius. 199–206 length. 233 calculus of variations. 261–269 attenuation index. 125 Drude model. 110–112 BK7 borosilicate crown glass. 128–129 plasma model. 212. 269–270 optical molasses.

136 ﬁsh underwater. 245 fast light. 22–26 roughness limit to ﬁnesse. on resonance. diabolical. 158–163. 85 free spectral range. 37 Fourier transform. 93 resonance condition. 133–139 matrix formalism. 48 complex. 49 computed by lens. 30 dipole force. 39 multiple dimensions. 148 diode laser. 48 electro-optic eﬀect. 180 corner-cube reﬂector. 133. 37–40. 180 diﬀraction limit. 27 ﬂuctuation–dissipation theorem. 240 electric ﬂux density. 47. 157 recipe. 248 energy density. 161–162. 157–158 paraxial regime. 156 diﬀraction limit. 180 . 155–179 central point. 83 Fabry–Perot interferometer. 87 confocal. 38–39 spatial convention. 180 circular aperture. 83–100. for optical ﬁber. 126 dielectric medium simple. 268–269 Drude model. 145–146 Euler-Lagrange equation. 257 laser. 239–251 anomalous. 86 photon lifetime. 40–42 dielectric constant. 167 dispersion. 165–176 Fourier series. 258–261 directional ﬁlter. 246–247. 84 Gaussian modes. 268 Fourier optics. 95–99 convolution and. 13–16 Fermat’s principle. 215. 136–139 reﬂection-summation model. 157–158 holography. 179–180 Airy disk. 143 thin lens as analog computer. 54 electromagnetically induced transparency. 213–216 EIT. 121 extinction coeﬃcient. 30 evanescent ﬁeld.272 Index with δ-function. 47–48 diﬀraction acousto-optic. 219. 249–250 material. 179–180 Fresnel. 113 Fermat’s Principle. 215. 180 diﬀraction grating. 83 far-oﬀ resonance trap (FORT). 162–165. 133–136 ﬁnesse. 110 electromagnetic ﬁeld energy density. 158–163. 233 Doppler temperature. 155 of Gaussian. 126 extraordinary wave. 29 ﬁend. 31. 130 diopter. 99 survival probability. 135–136 Hermite–Gaussian modes. 27 cross section absorption. 257–258 Einstein A and B coeﬃcients. 215 natural. 248 electric displacement. 151–152 ﬁnesse. 148 diﬀusion process. 43 of rectiﬁed sine wave. 216 delta function. 92–95 glass plate etalon. 180 diﬀraction grating. 198 ﬁlm reﬂection. 54 error analysis. 112. 151–152 Q factor. 159–160 Fraunhofer diﬀraction. 86 Fabry–Perot etalon. 176–179 nonparaxial regime. 85. 87 physical mode condition. 95 line width. 135. 156 decomposition into plane waves. 31 Fabry–Perot cavity. 213–214 Einstein rate equations. 113 Verdet constant. 161–162. 144–145. 261 Faraday eﬀect. 143 convolution theorem. 157–158 diﬀusion coeﬃcient. 245 slow light. 156–157 spatial ﬁlters. 183–198 Fraunhofer. 250–251 normal. 47 electric permittivity. 111 F2 ﬂint glass. 35–37 accuracy of. 246–249 Doppler broadening. 83 resonators in ray optics.

199–206 constructive. 167–170. 118 interference between partially coherent sources. 226 Nd:YAG. 130 He–Ne. 231 Klein–Cook parameter. 227–228 lasers. 76 Hermite–Gaussian beam. 60–61. 196 Kramers–Kronig relations. 64. 141–142 high-pass ﬁlter. 243–245 group velocity dispersion. 30 laser Q switch. 113 rotation of. 149–151 induction heater. 42 gain coeﬃcient. 180 knife edge. 104–105 normalization. 233–235 Ti:sapphire. 121 Jones matrix. 105 orthogonality. 121 skin depth. 58 destructive. 129 interface. 108 linear polarizer. 160–161. 31 free spectral range. 161–162. 70 vector. interferometric. 66–75 ABCD law. 52–53 of vacuum. 226. 57–58 two tilted plane waves. 202 function. 179–180 thin lens. 178–179 homogenous medium. 250 Lagrangian. 218 Gaussian beam. 108 realistic polarizer. 29 harmonic oscillator. 149–151 group index. 122 FTIR spectroscopy. 77 Gaussian beams. 181. 13–26 Gibbs phenomenon. 58 multiple waves. 14 impedance. 225–226 . 205–206 coherence. 203 interferometer Mach-Zehnder. 110 polarization rotator. 176–179 oﬀ-axis. air-conductor cool movie. 105–106 normal modes. 127 intensity. 58–60. 74 minimum spot size for focusing lens. 120–123 evanescent ﬁeld. 50 Hermite polynomials. 71 grating. air-glass cool movie. 105–110 cascaded system. 231 laser spiking.Index 273 double slit. 253–258 damped. 105 Kerr lens. 180 gravimeter. diﬀraction. 115–130 Fresnel rhomb. 202 internal reﬂection. 75–77 high reﬂection (HR) coating. 106 Hamilton’s principle. 246 half-wave plate. 123 frustrated total internal reﬂection. 159–160 validity conditions. 163–165 Fresnel relations. 229 continuous-wave. 106–108 Jones vector. 239–243. 62–63 visibility. 179–181 knife-edge measurement (10-90 rule). 180 single slit. 244–245 group velocity. 173. 108–110 wave retarder. 64 Michelson. 207–238 Q-switching. 74–75 parameter speciﬁcation of. 255–258 Heaviside step function. 210–211. 166–167 Hilbert transform. 30 Hamiltonian mechanics. 149–153. 70–71 geometrical optics. 9 functions orthogonal. 68–69. 162–165. 216–219 gain saturation. 84 Fresnel diﬀraction. 61 Green’s function. 53 impulse-response function. 158–159 Fraunhofer lines. 46 interface. 171–172 Fourier transform. 43 Gouy phase. 71–75 focusing by thin lens. 228–229 cavity dumper. 114 diode. 63 two beams. 240 holography. 179–180 rectangular aperture. 171–172 Helmholtz equation.

210 key components of. 9–11 linear transformation. 50–53 . 10 sum. 112–113 optical axis. 30 Lensmaker’s formula. 54 in dielectric media. 47 magneto-optical trap (MOT). 80 Lorentz model. 227–228 light–atom interactions. 215 loss probability. 216 optical activity. 26 Planck blackbody distribution. 16 mode locking. 46–49 in vacuum. 80 planar resonator. 209 single-mode operation. 12 product. 226 low-pass ﬁlter. 266–269 optical path length. 140 Michelson interferometer. 111 orthogonal functions. 16. 28 plane. 61 line shape. 209. 128 MgF2 . 214 natural. 211–216 LIGO. 232 threshold behavior. 207 pulsed. 58–60. 207–209 gain saturation. 211–216 mode locking. 226–231 pump saturation. 30 optical spectrum analyzer. 211 properties of. 255–258 damping coeﬃcient. 223 gain. 54 plane waves. 205 permeability. 207 laser spiking. 219–221. 11 identity. 83. 47 magnetization density. 64 magnetic ﬂux density. 142 pinhole camera. 10 inverse. 264 Lorentzian absorption. 207 resonator. 202 Michelson-Morley experiment. 218. 222–225 gain media. 226 optimum output coupler. 9 determinant. 242–243. 210–211. 13 optical power (of lens). 216–219. 90–92 confocal. 11–12 eigenvalues. 247. 17 Liouville’s theorem. 256–257 paraxial approximation. 243 photon lifetime. 113 optical molasses. 45 permittivity. 14 spherical. 11–12 eigenvectors. 111 natural line width. 269–270 Malus. 15. 17 paraxial wave equation. 12.274 Index four-level. 229–231 monochromatic waves. 10 linear transformation as. 65–66 Parseval’s Theorem. 60–61. 218–219 optimum output. 113 Legendre transformation. 11–12 deﬁnition. 206 Parseval’s theorem. 10. 211 three-level. 10 trace. 264–266 photonic crystal. 12 transpose. 16 parabolic. 221–222 gain. 223 pumping schemes. 240 phase velocity. 223–225 gain coeﬃcient. 10 Maxwell equations. 166 Mach-Zehnder interferometer. 215 linear algebra. 95–96 ordinary wave. 218 steady-state behavior. 214 plane wave. 219–222 pumps. small-signal. 113 matrix characteristic polynomial. 17 optical isolator. 61 mirror elliptical. 231 photon scattering rate. 217–218 Law of Malus. 22 light–atom interactions. Law of. 42 oscillator strength. 64. 209–211. 256 line shape. 229–231 multi-mode operation. 45. 45–46 melting coin cool movie. 30 Lord Rayleigh. 48–49 multiple refraction.

254 telescope. 22 periodic rays. 78 spontaneous emission. 171–172 Fourier transform. 110–112 circular. 18 Rayleigh length. 250–251 quantum eﬃciency. 166 phase-object visualization. 180–181 directional. 261–264 angular distribution. 15 spatial ﬁlter. 9 cartesian product. 47 polarization density. 26 resonators. 269 recoil velocity. 108 electro-optic eﬀect. 105–110 Jones vector. 171–174. 263–264 random walk. 106 quarter-wave stack. 268 rate equations Einstein’s. 228. 181 Schrödinger equation. 241–243 fused silica. 110 ellipse. dipole. 18 ray optics. 71 TEMl. 13. 127. 166–167 low-pass. 121. 76 quantum Zeno eﬀect. 130 skin eﬀect. 13–26 ray-transfer matrix. 23–26 rotation matrix. 166. 167 high-pass. 145. 170–176 schlieren imaging. 261–269 radiation reaction. 112. 47. 174. 109 sample mean.Index 275 plasma frequency. 254 polarizer (linear). 239. 171–172 stimulated emission. 266 reﬂectance. 167–170. 119 power-equivalent width. 168. 126. 213–216 ray matrix. 101–114 birefringence. 148. 148 of mean. 136–139 . 225 susceptibility. 148–149 step function. 66 Sellmeier formula for fused silica. 171–174. 31 TEM (transverse electromagnetic) wave. 149 saturable absorber.m mode. 141–142 radiation pressure. 26 confocal–planar. 217 standard deviation. 119 refractive index complex. 181 cleaning Gaussian beam. 117 polarization (of dielectric). 133–139 matrix formalism. 239. 76 thin-ﬁlm reﬂection. 212. 54. 251 set. 113 quarter-wave plate. 26 planar. 204. 210 Snell’s Law. 146. 44. 181–182 central dark-ground imaging. 59–60 applied to beam splitter. 251 refractive interface. 254–255 Pockels cell. 171. 206 Poynting vector. 230 polarizability. 24–25 stability condition. 212–213. 26 spherical. 66 recoil temperature. 260 scattering rate. 104–105 linear. 60 survival probability. 113 Jones matrix. 170–171 speed of light. 216 Stokes relations. 22–26 equivalent waveguide. 104 Faraday eﬀect. 223 for two-level atom. 181 Zernike phase-contrast imaging. 103–104 cool movie. 165–176 4-f setup. 231 saturation intensity. 46. 9 skin depth. 254 polarization. 255 plasma model. 264–266 schlieren imaging. 112–113 S and P. 210 for four-level laser. 226 quantum harmonic oscillator. 55 pulse spread in optical ﬁber. 171. 102–103 optical activity. photon. 101–102 elliptical. 114 Poisson sum rule. 15 resonator confocal. 264 radiation. 46 spherical wave. 127 small-signal gain coeﬃcient. 106 polarizing angle.

54 paraxial. 203 Voigt proﬁle. 63. 122 transmittance. 201–203 Wigner transform. 29 Thomas–Reiche–Kuhn sum rule. 170–171 ZrO2 . 215 wave equation. 9–10 Verdet constant. 160–161 Zernike phase-contrast imaging. 120–123 frustrated. 204–206 variationial principle.276 Index reﬂection-summation model. 10 vector space. 133–136 thin-lens law. 65–66 Wiener–Khinchin theorem. 200–203 optical. 141 . 119 uncertainty relation. 46. 257 total internal reﬂection. 74 Young double slit. 113 visibility. 14 vector.

- Jenkins White - Fundamentals of Optics - Libro
- Classical Mechanics - Physics - Week 1-3 Summary Formulas
- Exercises in Physics 1000023899
- ΑΝΤΙΒΑΡΥΤΗΤΑ
- The Nazi UFO Mythos
- Fundamental Optics
- Chem- Redox Formula Sheet (Never Completely Finished), Electrolytic Cells, Voltaic Cells, Electric Potentials
- Physics- Relationships between the equations Linear and Angular Motion. torque, momentum, angular velocity etc
- Future Energy Evolution - Thomas Valone, 2010
- Chem Mastery MathCard Sample
- 0200 Sets and Classes
- Roger J Anderton- On the Misconceptions about Einstein
- the Physics of Waves
- Table of the simple relationships between the basic values within an electrolytic and galvaic cells in respect to the redox half equations
- Class VII FOUNDATION & OLYMPIAD
- Analytical Chemistry
- Scalar Stress Wave - Tesla Wave (Tom Bearden)
- James P. Sethna-Statistical Mechanics Entropy Order Parameters and Complexity [Solutions to Exercises] (2011)
- Quantum & Atomic Physics (Eg Photoelectric Affect) Formula sheet & Study tool Physics A
- Waves
- 2.2.1 a Formula for the Powers of the Transition Matrix
- Magnetic Current Ed Leedskalnin Coral Castle
- llinalg9
- Waves Formula Sheet
- Gerry Vassilatos - Lost Science (Complete Edition)
- Matrix Algebra[1]
- Thomas F. Valone (2005) Electrogravitics II. Validating Reports on a New Propulsion Methodology (ISBN 0964107090)
- Future Science:Life Energies and the Physics of Paranormal Phenomena - John White, Stanley Krippner [OCR]
- Optics Notes
- Mat Lab Eigenvalue

Sign up to vote on this title

UsefulNot usefulRead Free for 30 Days

Cancel anytime.

Close Dialog## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Loading