Physics of Light and Optics PDF

Physics of Light and Optics
Justin Peatross
Michael Ware
Brigham Young University
2011c Edition
March 2, 2012
Preface
This curriculum was originally developed for a senior-level optics course in the
Department of Physics and Astronomy at Brigham Young University. Topics
are addressed from a physics perspective and include the propagation of light in
matter, reflection and transmission at boundaries, polarization effects, dispersion,
coherence, ray optics and imaging, diffraction, and the quantum nature of light.
Students using this book should be familiar with differentiation, integration, and
standard trigonometric and algebraic manipulation. A brief review of complex
numbers, vector calculus, and Fourier transforms is provided in Chapter 0, but it
is helpful if students already have some experience with these concepts.
While the authors retain the copyright, we have made this book available free
of charge at optics.byu.edu. This is our contribution toward a future world with
free textbooks! The web site also provides a link to purchase bound copies of the
book for the cost of printing. A collection of electronic material related to the
text is available at the same site, including videos of students performing the lab
assignments found in the book.
The development of optics has a rich history. We have included historical
sketches for a selection of the pioneers in the field to help students appreciate
some of this historical context. These sketches are not intended to be authorita-
tive, the information for most individuals having been gleaned primarily from
Wikipedia.
The authors may be contacted at opticsbook@byu.edu. We enjoy hearing
reports from those using the book and welcome constructive feedback. We occa-
sionally revise the text. The title page indicates the date of the last revision.
We would like to thank all those who have helped improve this material. We
especially thank John Colton, Bret Hess, and Harold Stokes for their careful review
and extensive suggestions. This curriculum benefitted from a CCLI grant from
the National Science Foundation Division of Undergraduate Education (DUE-
9952773).
iii
Contents
Preface iii
Table of Contents v
0 Mathematical Tools 1
0.1 Vector Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
0.2 Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
0.3 Linear Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
0.4 Fourier Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Appendix 0.A Table of Integrals and Sums . . . . . . . . . . . . . . . . 20
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1 Electromagnetic Phenomena 27
1.1 Gauss’ Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.2 Gauss’ Law for Magnetic Fields . . . . . . . . . . . . . . . . . . . . 29
1.3 Faraday’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.4 Ampere’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.5 Maxwell’s Adjustment to Ampere’s Law . . . . . . . . . . . . . . . . 33
1.6 Polarization of Materials . . . . . . . . . . . . . . . . . . . . . . . . 36
1.7 The Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2 Plane Waves and Refractive Index 45

2.1 Plane Wave Solutions to the Wave Equation . . . . . . . . . . . . . 45
2.2 Index of Refraction . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.3 The Lorentz Model of Dielectrics . . . . . . . . . . . . . . . . . . . 51
2.4 Index of Refraction of a Conductor . . . . . . . . . . . . . . . . . . 54
2.5 Poynting’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.6 Irradiance of a Plane Wave . . . . . . . . . . . . . . . . . . . . . . . 58
Appendix 2.A Radiometry, Photometry, and Color . . . . . . . . . . . 60
Appendix 2.B Clausius-Mossotti Relation . . . . . . . . . . . . . . . . 63
Appendix 2.C Energy Density of Electric Fields . . . . . . . . . . . . . 66
Appendix 2.D Energy Density of Magnetic Fields . . . . . . . . . . . . 68
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3 Reflection and Refraction 73
v
vi CONTENTS
3.1 Refraction at an Interface . . . . . . . . . . . . . . . . . . . . . . . . 73

3.2 The Fresnel Coefficients . . . . . . . . . . . . . . . . . . . . . . . . 77
3.3 Reflectance and Transmittance . . . . . . . . . . . . . . . . . . . . 78
3.4 Brewster’s Angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.5 Total Internal Reflection . . . . . . . . . . . . . . . . . . . . . . . . 81
3.6 Reflections from Metal . . . . . . . . . . . . . . . . . . . . . . . . . 83
Appendix 3.A Boundary Conditions For Fields at an Interface . . . . 84
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4 Multiple Parallel Interfaces 89

4.1 Double-Interface Problem Solved Using Fresnel Coefficients . . . 90
4.2 Two-Interface Transmittance at Sub Critical Angles . . . . . . . . 93
4.3 Beyond Critical Angle: Tunneling of Evanescent Waves . . . . . . 96
4.4 Fabry-Perot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.5 Setup of a Fabry-Perot Instrument . . . . . . . . . . . . . . . . . . 100
4.6 Distinguishing Nearby Wavelengths in a Fabry-Perot Instrument 101
4.7 Multilayer Coatings . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.8 Repeated Multilayer Stacks . . . . . . . . . . . . . . . . . . . . . . . 109
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5 Propagation in Anisotropic Media 117

5.1 Constitutive Relation in Crystals . . . . . . . . . . . . . . . . . . . 117
5.2 Plane Wave Propagation in Crystals . . . . . . . . . . . . . . . . . . 119
5.3 Biaxial and Uniaxial Crystals . . . . . . . . . . . . . . . . . . . . . . 123
5.4 Refraction at a Uniaxial Crystal Surface . . . . . . . . . . . . . . . 124
5.5 Poynting Vector in a Uniaxial Crystal . . . . . . . . . . . . . . . . . 125
Appendix 5.A Symmetry of Susceptibility Tensor . . . . . . . . . . . . 128
Appendix 5.B Rotation of Coordinates . . . . . . . . . . . . . . . . . . 129
Appendix 5.C Electric Field in Crystals . . . . . . . . . . . . . . . . . . 131
Appendix 5.D Huygens’ Elliptical Construct for a Uniaxial Crystal . . 134
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
Review, Chapters 1–5 139
6 Polarization of Light 145

6.1 Linear, Circular, and Elliptical Polarization . . . . . . . . . . . . . 146
6.2 Jones Vectors for Representing Polarization . . . . . . . . . . . . . 147
6.3 Elliptically Polarized Light . . . . . . . . . . . . . . . . . . . . . . . 148
6.4 Linear Polarizers and Jones Matrices . . . . . . . . . . . . . . . . . 149
6.5 Jones Matrix for Polarizers at Arbitrary Angles . . . . . . . . . . . 152
6.6 Jones Matrices for Wave Plates . . . . . . . . . . . . . . . . . . . . . 153
6.7 Polarization Effects of Reflection and Transmission . . . . . . . . 156
Appendix 6.A Ellipsometry . . . . . . . . . . . . . . . . . . . . . . . . . 157
Appendix 6.B Partially Polarized Light . . . . . . . . . . . . . . . . . . 159
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
CONTENTS vii
7 Superposition of Quasi-Parallel Plane Waves 171

7.1 Intensity of Superimposed Plane Waves . . . . . . . . . . . . . . . 172
7.2 Group vs. Phase Velocity: Sum of Two Plane Waves . . . . . . . . 174
7.3 Frequency Spectrum of Light . . . . . . . . . . . . . . . . . . . . . 176
7.4 Packet Propagation and Group Delay . . . . . . . . . . . . . . . . . 181
7.5 Quadratic Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . 183
7.6 Generalized Context for Group Delay . . . . . . . . . . . . . . . . . 185
Appendix 7.A Pulse Chirping in a Grating Pair . . . . . . . . . . . . . . 189
Appendix 7.B Causality and Exchange of Energy with the Medium . . 191
Appendix 7.C Kramers-Kronig Relations . . . . . . . . . . . . . . . . . 196
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
8 Coherence Theory 203

8.1 Michelson Interferometer . . . . . . . . . . . . . . . . . . . . . . . 203
8.2 Coherence Time and Fringe Visibility . . . . . . . . . . . . . . . . . 208
8.3 Temporal Coherence of Continuous Sources . . . . . . . . . . . . 209
8.4 Fourier Spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . 210
8.5 Young’s Two-Slit Setup and Spatial Coherence . . . . . . . . . . . 211
Appendix 8.A Spatial Coherence for a Continuous Source . . . . . . . 216
Appendix 8.B Van Cittert-Zernike Theorem . . . . . . . . . . . . . . . 217
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
9 Light as Rays 227

9.1 The Eikonal Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 228
9.2 Fermat’s Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
9.3 Paraxial Rays and ABCD Matrices . . . . . . . . . . . . . . . . . . . 235
9.4 Reflection and Refraction at Curved Surfaces . . . . . . . . . . . . 237
9.5 ABCD Matrices for Combined Optical Elements . . . . . . . . . . 239
9.6 Image Formation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
9.7 Principal Planes for Complex Optical Systems . . . . . . . . . . . 244
9.8 Stability of Laser Cavities . . . . . . . . . . . . . . . . . . . . . . . . 245
Appendix 9.A Aberrations and Ray Tracing . . . . . . . . . . . . . . . . 248
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
10 Diffraction 257
10.1 Huygens’ Principle as Formulated by Fresnel . . . . . . . . . . . . 258
10.2 Scalar Diffraction Theory . . . . . . . . . . . . . . . . . . . . . . . . 260
10.3 Fresnel Approximation . . . . . . . . . . . . . . . . . . . . . . . . . 262
10.4 Fraunhofer Approximation . . . . . . . . . . . . . . . . . . . . . . . 264
10.5 Diffraction with Cylindrical Symmetry . . . . . . . . . . . . . . . . 265
Appendix 10.A Fresnel-Kirchhoff Diffraction Formula . . . . . . . . . . 267
Appendix 10.B Green’s Theorem . . . . . . . . . . . . . . . . . . . . . . . 270
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
viii CONTENTS
11 Diffraction Applications 275

11.1 Fraunhofer Diffraction Through a Lens . . . . . . . . . . . . . . . 275
11.2 Resolution of a Telescope . . . . . . . . . . . . . . . . . . . . . . . . 279
11.3 The Array Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
11.4 Diffraction Grating . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
11.5 Spectrometers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
11.6 Diffraction of a Gaussian Field Profile . . . . . . . . . . . . . . . . 287
11.7 Gaussian Laser Beams . . . . . . . . . . . . . . . . . . . . . . . . . 289
Appendix 11.A ABCD Law for Gaussian Beams . . . . . . . . . . . . . . 291
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
12 Interferograms and Holography 301

12.1 Interferograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
12.2 Testing Optical Components . . . . . . . . . . . . . . . . . . . . . . 302
12.3 Generating Holograms . . . . . . . . . . . . . . . . . . . . . . . . . 303
12.4 Holographic Wavefront Reconstruction . . . . . . . . . . . . . . . 304
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
13 Blackbody Radiation 315

13.1 Stefan-Boltzmann Law . . . . . . . . . . . . . . . . . . . . . . . . . 316
13.2 Failure of the Equipartition Principle . . . . . . . . . . . . . . . . . 317
13.3 Planck’s Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
13.4 Einstein’s A and B Coefficients . . . . . . . . . . . . . . . . . . . . . 322
Appendix 13.A Thermodynamic Derivation of the Stefan-Boltzmann
Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
Appendix 13.B Boltzmann Factor . . . . . . . . . . . . . . . . . . . . . . 326
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
Index 331
Physical Constants 337

Chapter 0
Mathematical Tools
Our study of optics begins with Maxwell’s equations in Chapter 1. Before we start,
look over this chapter to make sure you are comfortable with the mathematical
tools we’ll be using. The vector calculus material in section 0.1 will be used
beginning in Chapter 1, so you should review it now. In Section 0.2 we review
complex numbers. You have probably had some exposure to complex numbers,
but if you are like many students, you haven’t yet fully appreciated their usefulness.
Please be warned that your life will be much easier if you understand the material
in section 0.2 by heart. Complex notation is pervasive throughout the book,
beginning in chapter 2.
You may safely procrastinate reviewing Sections 0.3 and 0.4 until they come
up in the book. The linear algebra refresher in Section 0.3 is useful for Chapter 4,
where we analyze multilayer coatings, and again in Chapter 6, where we discuss
polarization. Section 0.4 provides an introduction to Fourier theory. Fourier trans-
forms are used extensively in optics, and you should study Section 0.4 carefully René Descartes (1596-1650, French)
was born in in La Haye en Touraine
before tackling Chapter 7. (now Descartes), France. His mother
died when he was an infant. His father
was a member of parliament who en-
couraged Descartes to become a lawyer.
0.1 Vector Calculus Descartes graduated with a degree in
law from the University of Poitiers
Each position in space corresponds to a unique vector r ≡ x x̂ + y ŷ + z ẑ, where in 1616. In 1619, he had a series of
dreams that led him to believe that he
x̂, ŷ, and ẑ are unit vectors with length one, pointing along their respective axes. should instead pursue science. Descartes
Boldface type distinguishes a variable as a vector quantity, and the use of x̂, ŷ, became one of the greatest mathemati-
cians, physicists, and philosophers of
and ẑ denotes a Cartesian coordinate system. Electric and magnetic fields are
all time. He is credited with inventing
vectors whose magnitude and direction can depend on position, as denoted by the cartesian coordinate system, which
E (r) or B (r). An example of such a field is E (r) = q (r − r0 ) 4π²0 |r − r0 |3 , which

±
is named after him. For the rst time,
geometric shapes could be expressed as
is the static electric field surrounding a point charge located at position r0 . The algebraic equations. (Wikipedia)
absolute-value brackets indicate the magnitude (or length) of the vector given by
¯ ¡ ¢ ¯
|r − r0 | = ¯(x − x 0 ) x̂ + y − y 0 ŷ + (z − z 0 ) ẑ¯
q
¢2 (0.1)
= (x − x 0 )2 + y − y 0 + (z − z 0 )2
¡
1
2 Chapter 0 Mathematical Tools
Example 0.1
¡ ¢
Compute
¡ the electric
¢ field at r = 2x̂ + 2ŷ + 2ẑ Å due to an charge q positioned at
r0 = 1x̂ + 1ŷ + 2ẑ Å.
Solution: As mentioned above, the field is given by E (r) = q (r − r0 ) 4π²0 |r − r0 |3 .

±
We have ¡ ¢ ¡ ¢
r − r0 = (2 − 1)x̂ + (2 − 1)ŷ + (2 − 2)ẑ Å = 1x̂ + 1ŷ Å
and p p
|r − r0 | = (1)2 + (1)2 Å = 2Å
The electric field is then ¡ ¢
q 1x̂ + 1ŷ Å
E=− ¡p ¢3
4π²0 2 Å
Figure 0.1 The electric field vec-

tors around a point charge. In addition to position, the electric and magnetic fields almost always depend
on time in optics problems. For example, a common time-dependent field is
E(r, t ) = E0 cos(k·r−ωt ). The dot product k·r is an example of vector multiplication,
and signifies the following operation:
¡ ¢ ¡ ¢
k · r = k x x̂ + k y ŷ + k z ẑ · x x̂ + y ŷ + z ẑ
= kx x + k y y + kz z (0.2)
= |k||r| cos φ
where φ is the angle between the vectors k and r.
Proof of the final line of (0.2)
Consider the plane that contains the two vectors k and r. Call it the x 0 y 0 -plane. In
this coordinate system, the two vectors can be written as k = k cos θx̂0 +k sin θŷ0 and
r = r cos αx̂0 +r sin αŷ0 , where θ and α are the respective angles that the two vectors
make with the x 0 -axis. The dot product gives k · r = kr (cos θ cos α + sin θ sin α).
This simplifies to k · r = kr cos φ (see (0.13)), where φ ≡ θ − α is the angle between
the vectors. Thus, the dot product between two vectors is the product of the
magnitudes of each vector times the cosine of the angle between them.
Another type of vector multiplication is the cross product , which is accom-

plished in the following manner:1
¯ ¯
¯ x̂ ŷ ẑ ¯¯
¯
E × B = ¯¯ E x E y E z ¯¯
¯ B (0.3)
x B y Bz
¯
¡ ¢ ¡ ¢
= E y B z − E z B y x̂ − (E x B z − E z B x ) ŷ + E x B y − E y B x ẑ
1 The use of the determinant to generate the cross product is merely a fortuitous device for
remembering its form.

0.1 Vector Calculus 3
Note that the cross product results in a vector, whereas the dot product mentioned
above results in a scalar (i.e. a number with appropriate units). The resultant
vector is always perpendicular to the two vectors that are cross multiplied. If the
fingers on your right hand curl from the first vector towards the second, your
thumb will point in the direction of the result. The magnitude of the result equals
the product of the magnitudes of the constituent vectors times the sine of the
angle between them.
Proof of cross-product properties
We label the plane containing E and B the x 0 y 0 -plane. In this coordinate system, the
two vectors can be written as E = E cos θx̂0 + E sin θŷ0 and B = B cos αx̂0 + B sin αŷ0 ,
where θ and α are the respective angles that the two vectors make with the x 0 -axis.
The cross product, according to (0.3), gives E × B = E B (cos θ sin α − sin θ cos α)ẑ0 .
This simplifies to E × B = E B sin φẑ0 (see (0.14)), where φ ≡ θ − α is the angle be-
tween the vectors. The vectors E and B, which both lie in the x 0 y 0 -plane, are both
perpendicular to z 0 . If 0 < θ − α < π, the result E × B points in the positive z 0
direction, which is consistent with the right-hand rule.
We will use several multidimensional derivatives in our study of optics, namely

the gradient, the divergence, and the curl.2 In Cartesian coordinates, the gradient
of a scalar function is given by
¡ ¢ ∂f ∂f ∂f
∇ f x, y, z = x̂ + ŷ + ẑ (0.4)
∂x ∂y ∂z
the divergence, which applies to vector functions, is given by
∂E x ∂E y ∂E z
∇·E = + + (0.5)
∂x ∂y ∂z
and the curl, which also applies to vector functions, is given by

¯ ¯
¯ x̂ ŷ ẑ ¯¯
∇ × E = ¯¯ ∂/∂x ∂/∂y ∂/∂z ¯¯
¯
¯ E Ey Ez ¯
x (0.6)
∂E z ∂E y ∂E z ∂E x ∂E y ∂E x
µ ¶ µ ¶ µ ¶
= − x̂ − − ŷ + − ẑ
∂y ∂z ∂x ∂z ∂x ∂y
Example 0.2
Derive the gradient (0.4) in cylindrical coordinates defined by the transformations
x = ρ cos φ and y = ρ sin φ. (The coordinate z remains unchanged.)
2 See M. R. Spiegel, Schaum’s Outline of Advanced Mathematics for Engineers and Scientists, pp.
126-127 (New York: McGraw-Hill 1971).

Solution: By inspection of Fig. 0.2, the cartesian unit vectors may be expressed as
x̂ = cos φρ̂ − sin φφ̂ and ŷ = sin φρ̂ + cos φφ̂
In accordance with the rules of calculus, the needed partial derivatives expressed
in terms of the new variables are
∂ ∂ρ ∂ ∂φ ∂ ∂ ∂ρ ∂ ∂φ ∂
µ ¶ µ ¶ µ ¶ µ ¶
= + and = +
∂x ∂x ∂ρ ∂x ∂φ ∂y ∂y ∂ρ ∂y ∂φ
Meanwhile, the inverted form of the coordinate transformation is

q
ρ = x 2 + y 2 and φ = tan−1 y/x
from which we obtain the following derivatives:

∂ρ x ∂φ y sin φ
Figure 0.2 The unit vectors x̂ and =p = cos φ =− 2 =−
∂x x + y2
2 ∂x x + y2 ρ
ŷ may be expressed in terms of
components along φ̂ and ρ̂ in ∂ρ y ∂φ x cos φ
=p = sin φ = 2 =
cylindrical coordinates. ∂y x + y2
2 ∂y x +y 2 ρ
Putting this all together, we arrive at

∂f ∂f ∂f
∇f = x̂ + ŷ + ẑ
∂x ∂y ∂z
∂ f sin φ ∂ f ¡
µ ¶
= cos φ cos φρ̂ − sin φφ̂
¢
−
∂ρ ρ ∂φ
∂ f cos φ ∂ f ¡ ¢ ∂f
µ ¶
+ sin φ + sin φρ̂ + cos φφ̂ + ẑ
∂ρ ρ ∂φ ∂z
∂f 1 ∂f ∂f
= ρ̂ + φ̂ + ẑ
∂ρ ρ ∂φ ∂z
where we have used cos2 φ + sin2 φ = 1 (see Ex. 0.4).
We will sometimes need a multidimensional second derivative called the

Laplacian. When applied to a scalar function, it is defined as the divergence of a
gradient:
∇2 f x, y, z ≡ ∇ · ∇ f x, y, z
¡ ¢ £ ¡ ¢¤
(0.7)
In cartesian coordinates, this reduces to
Pierre-Simon Laplace (1749-1827, ¢ ∂2 f ∂2 f ∂2 f
∇2 f x, y, z = 2 + 2 + 2
¡
French) was born in Normandy, France (0.8)
to a farm laborer. Some wealthy neigh-
∂x ∂y ∂z
bors noticed his unusual abilities and
took an interest in his education.
Since the Laplacian applied to a scalar gives a result that is also a scalar, in Carte-
Laplace is sometimes revered as the sian coordinates we deal with vector functions by applying the Laplacian to the
Newton of France with contributions
scalar function attached to each unit vector:
to mathematics and astronomy. The
∂2 E y ∂2 E y ∂2 E y
Ã !
Laplacian dierential operator as well as µ 2
∂ E x ∂2 E x ∂2 E x
¶
Laplace transforms are used widely in 2
∇ E= + + x̂ + + + ŷ
applied mathematics. (Wikipedia) ∂x 2 ∂y 2 ∂z 2 ∂x 2 ∂y 2 ∂z 2
µ 2 (0.9)
∂ E z ∂2 E z ∂2 E z
¶
+ + + ẑ
∂x 2 ∂y 2 ∂z 2
0.1 Vector Calculus 5
This is possible because each unit vector is a constant in Cartesian coordinates.

The various multidimensional derivatives take on more complicated forms
in non-cartesian coordinates such as cylindrical or spherical. You can derive the
Laplacian for these other coordinate systems by changing variables and rewriting
the unit vectors starting from the above Cartesian expression. (See Problem 0.10.)
Regardless of the coordinate system, the Laplacian for a vector function can be
obtained from first derivatives though
∇2 E ≡ ∇(∇ · E) − ∇ × (∇ × E) (0.10)
Verification of (0.10) in Cartesian coordinates
From (0.6), we have

∂E z ∂E y ∂E z ∂E x ∂E y ∂E x
µ ¶ µ ¶ µ ¶
∇×E = − x̂ − − ŷ + − ẑ
∂y ∂z ∂x ∂z ∂x ∂y
and
¯ ¯
¯ x̂ ŷ ẑ ¯
∂/∂x ∂/∂y ∂/∂z
¯ ¯
∇ × (∇ × E) = ¯ ³
¯ ¯
¯ ∂E z ∂E y ∂E
¯
∂E z ∂E x ∂E x
´ ³ ´ ³ ´
y ¯
¯
∂y
− ∂z − ∂x − ∂z ∂x
− ∂y ¯
∂ ∂E y ∂E x ∂ ∂E z ∂E x ∂ ∂E y ∂E x ∂ ∂E z ∂E y
· µ ¶ µ ¶¸ · µ ¶ µ ¶¸
= − + − x̂ − − − − ŷ
∂y ∂x ∂y ∂z ∂x ∂z ∂x ∂x ∂y ∂z ∂y ∂z
∂ ∂E z ∂E x ∂ ∂E z ∂E y
· µ ¶ µ ¶¸
+ − − − − ẑ
∂x ∂x ∂z ∂y ∂y ∂z
∂2 E x ∂2 E y
x̂ + ∂y 2 ŷ + ∂∂zE2z ẑ and then rearranging, we
2
After adding and subtracting ∂x 2
get
∂2 E y ∂2 E z 2
∂2 E x ∂ E y ∂2 E z
2
∂2 E x ∂ E y ∂2 E z
" # " # " #
∂2 E x
∇ × (∇ × E) = + + x̂ + + + ŷ + + + ẑ
∂x 2 ∂x∂y ∂x∂z ∂x∂y ∂y 2 ∂y∂z ∂x∂z ∂y∂z ∂z 2
" 2
∂ E y ∂2 E y ∂2 E y
" # # " #
∂2 E x ∂2 E x ∂2 E x ∂2 E z ∂2 E z ∂2 E z
− + + x̂ − + + ŷ − + + ẑ
∂x 2 ∂y 2 ∂z 2 ∂x 2 ∂y 2 ∂z 2 ∂x 2 ∂y 2 ∂z 2
After some factorization, we obtain

∂E x ∂E y ∂E z
¸ " 2 #
∂ ∂ ∂ ∂ ∂2 ∂2 £
· ¸·
¤
∇ × (∇ × E) = x̂ + ŷ + ẑ + + − + + E x x̂ + E y ŷ + E z ẑ
∂x ∂y ∂z ∂x ∂y ∂z ∂x 2 ∂y 2 ∂z 2
= ∇ (∇ · E) − ∇2 E
where on the final line we invoked (0.4), (0.5), and (0.8).
We will also encounter several integral theorems3 involving vector functions.

The divergence theorem for a vector function F is
I Z
F · n̂ d a = ∇ · F d v (0.11)
S V
3 For succinct treatments of the divergence theorem and Stokes’ theorem, see M. R. Spiegel,
Schaum’s Outline of Advanced Mathematics for Engineers and Scientists, p. 154 (New York: McGraw-
Hill 1971).
The integration on the left-hand side is over the closed surface S, which contains
the volume V associated with the integration on the right-hand side. The unit
vector n̂ points outward, normal to the surface. The divergence theorem is espe-
cially useful in connection with Gauss’ law, where the left-hand side is interpreted
as the number of field lines exiting a closed surface.
Example 0.3
2
¡ ¢
Check the divergence theorem (0.11) for the vector function F x, y, z = ¯ y¯ x̂+x y ŷ+
2
x z ẑ. Take as the volume a cube contained by the six planes |x| = ±1, ¯ y ¯ = ±1, and
|z| = ±1.
Solution: First, we evaluate the left side of (0.11) for the function:
I Z1 Z1 Z1 Z1 Z1 Z1
d xd y x 2 z z=1 − d xd y x 2 z z=−1 +
¡ ¢ ¡ ¢ ¡ ¢
F · n̂d a = d xd z x y y=1
Figure 0.3 The function F (red S −1 −1 −1 −1 −1 −1
arrows) plotted for several points Z1 Z1 Z1 Z1 Z1 Z1
on the surface S. −
¡ ¢
d xd z x y y=−1 + d yd z y 2 x=1 −
¡ ¢
d yd z y 2 x=−1
¡ ¢
−1 −1 −1 −1 −1 −1
Z1 Z1 Z1 Z1 3 ¯¯1 ¯1
x ¯ x 2 ¯¯ 8
=2 d xd y x 2 + 2 d xd zx = 4 + 4 =
3 −1
¯ 2 −1 3
¯
−1 −1 −1 −1
Now we evaluate the right side of (0.11):
Z1 Z1 Z1 Z1 · 2 ¸1
x x3 8
Z
d xd yd z x + x 2 = 4 d x x + x2 = 4
£ ¤ £ ¤
∇ · Fd v = + =
2 3 −1 3
V −1 −1 −1 −1
Another important theorem is Stokes’ theorem :

Z I
(∇ × F) · n̂ d a = F · d ` (0.12)
S C
The integration on the left-hand side is over an open surface S (not enclosing a
volume). The integration on the right-hand side is around the edge of the surface.
Again, n̂ is a unit vector that always points normal to the surface . The vector d `
points along the curve C that bounds the surface S. If the fingers of your right
hand point in the direction of integration around C , then your thumb points
in the direction of n̂. Stokes’ theorem is especially useful in connection with
Ampere’s law and Faraday’s law. The right-hand side is an integration of a field
around a loop.
0.2 Complex Numbers

It is often convenient to represent electromagnetic wave phenomena (i.e. light) as
a superposition of sinusoidal functions, each having the form A cos α + β . The
¡ ¢
0.2 Complex Numbers 7
sine function is intrinsically present in this formula via the identity
cos α + β = cos α cos β − sin α sin β

¡ ¢
(0.13)
This is a good formula to commit to memory, as well as the frequently used

identity
sin α + β = sin α cos β + sin β cos α
¡ ¢
(0.14)
With a basic familiarity with trigonometry, we can approach many optical
problems including those involving the addition of multiple waves. However, the
manipulation of trigonometric functions via identities such as (0.13) and (0.14)
can be cumbersome and tedious. Fortunately, complex-number notation offers
an equivalent approach with far less busy work. The modest investment needed to
become comfortable with complex notation is definitely worth it; optics problems
can become cumbersome enough even with the most efficient methods!
The convenience of complex-number notation has its origins in Euler’s for-
Leonhard Euler (1707-1783, Swiss)
mula: was born in Basel, Switzerland. His
e i φ = cos φ + i sin φ (0.15) father, Paul Euler, was friends with
p the well-known mathematician Johann

where i ≡ −1 is an imaginary number. By inverting Euler’s formula (0.15) we Bernoulli, who discovered young Euler's
great talent for mathematics and tu-
can obtain the following representation of the cosine and sine functions: tored him regularly. Euler enrolled at
iφ −i φ the University of Basel at age thirteen.

e +e In 1726 Euler accepted an oer to join
cos φ = ,
2 the Russian Academy of Sciences in
(0.16) St Petersburg, having unsuccessfully
e i φ − e −i φ
sin φ = applied for a professorship at the Uni-
2i versity of Basel. Under the auspices of

the Czars (with the exception of 12-year-
Equation (0.16) shows how ordinary sines and cosines are intimately related to old Peter II), foreign academicians in
hyperbolic cosines and hyperbolic sines. If φ happens to be imaginary such that the Russian Academy were given con-
siderable freedom to pursue scientic
φ = i γ where γ is real, then we have questions with relatively light teaching
duties. Euler spent his early career in
e −γ − e γ Russia, his mid career in Berlin, and
sin i γ = = i sinh γ his later career again in Russia. Euler
2i (0.17)
e −γ + e γ
introduced the concept of a function.
cos i γ = = cosh γ He successfully dened logarithms and

2 exponential functions for complex num-
bers and discovered the connection to
trigonometric functions. The special
case of Euler's formula ei π + 1 = 0 has
Proof of Euler’s formula been voted by modern fans of math-
ematics (including Richard Feynman)
We can prove Euler’s formula using a Taylor’s series expansion: as the Most Beautiful Mathematical
Formula Ever for its single uses of
¯ 2 ¯¯
1 d f ¯¯ 1 2 d f ¯ addition, multiplication, exponentia-
f (x) = f (x 0 ) + (x − x 0 ) + (x − x 0 ) +··· (0.18) tion, equality, and the constants 0, 1,
1! d x ¯x=x0 2! d x 2 ¯x=x0
e, i and π. Euler and his wife, Katha-
By expanding each function appearing in (0.15) in a Taylor’s series about the origin rina Gsell, were the parents of 13 chil-
dren, many of whom died in childhood.
we obtain
(Wikipedia)
φ2 φ4
cos φ = 1 − + −···
2! 4!
φ3 φ5
i sin φ = i φ − i +i −··· (0.19)
3! 5!
φ2 φ3 φ4 φ5
eiφ = 1 + i φ − −i + +i −···
2! 3! 4! 5!
The last line of (0.19) is seen to be the sum of the first two lines, from which Euler’s
formula directly follows.
Example 0.4
Prove (0.13) and (0.14) as well as cos2 φ + sin2 φ = 1 by taking advantage of (0.16).
Solution: We start with (0.13). By direct application of (0.16) and some rearranging,
we have
e i α + e −i α e i β + e −i β e i α − e −i α e i β − e −i β
cos α cos β − sin α sin β = −
2 2 2i 2i
Brook Taylor (1685-1731, English) was
born in Middlesex, England. He studied e i (α+β) + e i (α−β) + e −i (α−β) + e −i (α+β)
=
at Cambridge as a fellow-commoner 4
earning a bachelor degree in 1709
and a doctoral degree in 1714. Soon
e i (α+β) − e i (α−β) − e −i (α−β) + e −i (α+β)
+
thereafter, he developed the branch 4
of mathematics known as calculus of i (α+β) −i (α+β)
e +e
= cos α + β
¡ ¢
nite dierences. He used it to study =
the movement of vibrating strings. As 2
part of that work, he developed the for-
mula known today as Taylor's theorem,
We can prove (0.14) using the same technique:
which was under-appreciated until 1772,
when French mathematician Lagrange e i α − e −i α e i β + e −i β e i β − e −i β e i α + e −i α
referred to it as the main foundation of sin α cos β + sin β cos α = +
2i 2 2i 2
dierential calculus. (Wikipedia)
i (α+β) i (α−β) −i (α−β) −i (α+β)
e +e −e −e
=
4i
e i (α+β) − e i (α−β) + e −i (α−β) − e −i (α+β)
+
4i
e i (α+β) − e −i (α+β)
= sin α + β
¡ ¢
=
2i
Finally, for cos2 φ + sin2 φ = 1 we have

¶2 ¶2
e i φ + e −i φ e i φ − e −i φ
µ µ
2 2
cos φ + sin φ = +
2 2i
e 2i φ + 2 + e −2i φ e 2i φ − 2 + e −2i φ
= − =1
4 4
As was mentioned previously, we will often be interested in waves of the

form A cos (x + α). We can use complex notation to represent this wave simply by
writing n o
A cos α + β = Re Ãe i α
¡ ¢
(0.20)
where the phase β is conveniently contained within the complex factor Ã ≡ Ae i β .

The operation Re { } means to retain only the real part of the argument without
regard for the imaginary part. As an example, we have Re {1 + 2i } = 1. The formula
(0.20) follows directly from Euler’s equation (0.15).
0.2 Complex Numbers 9
It is common (even conventional) to omit the explicit writing of Re { }. Thus,

physicists participate in a conspiracy that Ãe i α actually means A cos α + β . This
¡ ¢
laziness is permissible because it is possible to perform linear operations on

© ª
Re f such as addition, differentiation, or integration while procrastinating the
taking of the real part until the end:
© ª © ª © ª
Re f + Re g = Re f + g
d df
½ ¾
© ª
Re f = Re
dx dx (0.21)
Z ½Z ¾
© ª
Re f d x = Re f dx
As an example, note that Re {1 + 2i } + Re {3 + 4i } = Re {(1 + 2i ) + (3 + 4i )} = 4.

However, we must be careful when performing other operations such as multi- Gerolamo Cardano (1501-1576, Italian)
was the rst to introduce the notion
plication. In this case, it is essential to take the real parts before performing the
of complex numbers (which he called
operation. Notice that ctitious) while developing solutions
to cubic and quartic equations. He was
born in Pavia, Italy, the illegitimate son
© ª © ª © ª
Re f × Re g 6= Re f × g (0.22)
of a lawyer who was an acquaintance
of Leonardo da Vinci. Cardano was for-
As an example, we see Re {1 + 2i } × Re {3 + 4i } = 3, but Re {(1 + 2i ) (3 + 4i )} = −5. tunate to survive infancy as his father
claimed that his mother attempted to
When dealing with complex numbers it is often advantageous to transform
abort him and his older siblings all died
between a Cartesian representation and a polar representation. With the aid of of the plague. Cardano studied at the
Euler’s formula, it is possible to transform any complex number a + i b into the University of Pavia and later at Padua.
He was known for being eccentric and
form ρe i φ , where a, b, ρ, and φ are real. From (0.15), the required connection confrontational, which did not earn him
between ρ, φ and (a, b) is
¡ ¢
many friends. He supported himself
in part as a somewhat successful gam-
iφ bler, but he was often short of money.

ρe = ρ cos φ + i ρ sin φ = a + i b (0.23) Cardano also introduced binomial co-
ecients and the binomial theorem.
The real and imaginary parts of this equation must separately be equal. Thus, we (Wikipedia)
have
a = ρ cos φ
(0.24)
b = ρ sin φ
Quadrant I
These equations can be inverted to yield
p II
ρ= a2 + b2
b (0.25)
φ = tan−1 (a > 0)
a
When a < 0, we must adjust φ by π since the arctangent has a range only from III IV
−π/2 to π/2.
The transformations in (0.24) and (0.25) have a clear geometrical interpreta-
tion in the complex plane, and this makes it easier to remember them. They are
Figure 0.4 A number in the com-
just the usual connections between Cartesian and polar coordinates. As seen in plex plane can be represented
Fig. 0.4, ρ is the hypotenuse of a right triangle having legs with lengths a and b, either by Cartesian or polar repre-
and φ is the angle that the hypotenuse makes with the x-axis. Again, you should sentation.
be careful when a is negative since the arctangent is defined in quadrants I and
IV. An easy way to deal with the situation of a negative a is to factor the minus
sign out before proceeding (i.e. a + i b = − (−a − i b) ). Then the transformation
is made on −a − i b where −a is positive. The overall minus sign out in front is
just carried along unaffected and can be factored back in at the end. Notice that
−ρe i φ is the same as ρe i (φ±π) .
Example 0.5
Write −3 + 4i in polar format.
Solution: We must be careful with the negative real part since it indicates a quad-
rant (in this case II) outside of the domain of the inverse tangent (quadrants I and
IV). Best to factor the negative out and deal with it separately.
−1 (−4) −1 4 −1 4
−3 + 4i = −(3 − 4i ) = − 32 + (−4)2 e i tan 3 = e i π 5e −i tan 3 = 5e i (π−tan 3 )
p
Figure 0.5 Geometric representa- Finally, we consider the concept of a complex conjugate. The conjugate of a
tion of −3 + 4i complex number z = a + i b is denoted with an asterisk and amounts to changing
the sign on the imaginary part of the number:
z ∗ = (a + i b)∗ ≡ a − i b (0.26)
The complex conjugate is useful when computing the absolute value of a complex
number: p p p
|z| = z ∗ z = (a − i b) (a + i b) = a 2 + b 2 = ρ (0.27)
Note that the absolute value of a complex number is the same as its magnitude ρ
as defined in (0.25). The complex conjugate is also useful for eliminating complex
numbers from the denominator of expressions:
a + i b (a + i b) (c − i d ) ac + bd + i (bc − ad )
= = (0.28)
c + i d (c + i d ) (c − i d ) c2 + d2
No matter how complicated an expression, the complex conjugate is calcu-

lated by inserting a minus sign in front of all occurrences of i in the expression,
and placing an asterisk on all complex variables in the expression. For example,
the complex conjugate of ρe i φ is ρe −i φ assuming ρ and φ are real, as can be seen
from Euler’s formula (0.15). As another example consider
¤∗
E 0 exp {i (K z − ωt )} = E 0∗ exp −i K ∗ z − ωt
£ © ¡ ¢ª
(0.29)
assuming z, ω, and t are real, but E 0 and K are complex.

A common way of obtaining the real part of an expression is by adding the
complex conjugate and dividing the result by 2:
1¡
z + z∗
¢
Re {z} = (0.30)
2
0.3 Linear Algebra 11
Notice that the expression for cos φ in (0.16) is an example of this formula. Some-
times when a lengthy expression is added to its own complex conjugate, we let
“C.C.” represent the complex conjugate in order to avoid writing the expression
twice.
In optics we sometimes encounter a complex angle, , such as K z in (0.29). The
imaginary part of K governs exponential decay (or growth) when a light wave
propagates in an absorptive (or amplifying) medium. Similarly, when we compute
the transmission angle for light incident upon a surface beyond the critical angle
for total internal reflection, we encounter the arcsine of a number greater than
one in an effort to satisfy Snell’s law. Even though such an angle does not exist in
the physical sense, a complex value for the angle can be found, which satisfies
(0.16) and describes evanescent waves.
0.3 Linear Algebra

Throughout this book we will often encounter sets of linear equations. (They
are called linear equations because they represent lines in a plane or in space.)
Most often, there are just two equations with two variables to solve. The simplest
example of such a set of equations is
Ax + B y = F and Cx + Dy = G (0.31)
where x and y are variables. A set of linear equations such as (0.31) can be
expressed using matrix notation as
· ¸· ¸ · ¸ · ¸
A B x Ax + B y F
= = (0.32)
C D y Cx +Dy G
As seen above, the 2 × 2 matrix multiplied onto the two-dimensional column

vector results in a two-dimensional vector. The elements of rows are multiplied
onto elements of the column and summed to create each new element in the
result. A matrix can also be multiplied onto another matrix (rows multiplying
columns, resulting in a matrix). The order of multiplication is important; matrix
multiplication is not commutative.
To solve a matrix equation such as (0.32), we multiply both sides by an inverse
matrix, which gives
· ¸−1 · ¸· ¸ · ¸−1 · ¸
A B A B x A B F
= (0.33)
C D C D y C D G
The inverse matrix has the property that

· ¸−1 · ¸ · ¸
A B A B 1 0
= (0.34)
C D C D 0 1
where the right-hand side is called the identity matrix. You can easily check that
the identity matrix leaves unchanged anything that it multiplies, and so (0.33)
simplifies to
· ¸ · ¸−1 · ¸
x A B F
=
y C D G
Once the inverse matrix is found, the matrix multiplication on the right can be
performed and the answers for x and y obtained as the upper and lower elements
of the result.
The inverse of a 2 × 2 matrix is given by
· ¸−1 · ¸
A B 1 D −B
=¯ (0.35)
C D −C A
¯
¯ A B ¯¯
¯
¯ C D ¯
where ¯ ¯
¯ A B ¯¯
¯
¯ C ≡ AD −C B
D ¯
is called the determinant. We can check that (0.35) is correct by direct substitution:
· ¸−1 · ¸ · ¸· ¸
A B A B 1 D −B A B
=
C D C D AD − BC −C A C D
· ¸
1 AD − BC 0
= (0.36)
AD − BC 0 AD − BC
· ¸
1 0
=
0 1
The above review of linear algebra is very basic. In contrast, we next dis-
cuss Sylvester’s theorem, which you probably have not previously encountered.
Sylvester’s theorem is useful when multiplying the same 2 × 2 matrix (with a de-
terminate of unity) together many times (i.e. raising the matrix to a power). This
situation occurs when modeling periodic multilayer mirror coatings or when
considering light rays trapped in a laser cavity as they reflect many times.
Sylvester’s Theorem:4 If the determinant of a 2×2 matrix is one, (i.e. AD −BC = 1)
then
¸N
A sin N θ − sin (N − 1) θ B sin N θ
· · ¸
A B 1
= (0.37)
C D sin θ C sin N θ D sin N θ − sin (N − 1) θ
James Joseph Sylvester (1814-1897,
English) made fundamental contribu-
tions to matrix theory, invariant theory,
where
1
number theory, partition theory and cos θ = (A + D) (0.38)
combinatorics. He played a leadership 2
role in American mathematics in the
later half of the 19th century as a pro-
fessor at the Johns Hopkins University
4 The theorem presented here is a specific case. See A. A. Tovar and L. W. Casperson, “Generalized
and as founder of the American Journal
of Mathematics. (Wikipedia) Sylvester theorems for periodic applications in matrix optics,” J. Opt. Soc. Am. A 12, 578-590 (1995).
0.4 Fourier Theory 13
Proof of Sylvester’s theorem by induction
When N = 1, the equation is seen to be correct by direct substitution. Next we

assume that the theorem holds for arbitrary N , and we check to see if it holds for
N + 1:
¸N +1
· · ¸· ¸
A B 1 A B
=
C D sin θ C D C sin N θ D sin N θ − sin (N − 1) θ
· ¡ 2
A + BC sin N θ − A sin (N − 1) θ (AB + B D) sin N θ − B sin (N − 1) θ
¢ ¸
1
= ¡ 2
sin θ (AC +C D) sin N θ −C sin (N − 1) θ D + BC sin N θ − D sin (N − 1) θ
¢
Now we inject the condition AD − BC = 1 into the diagonal elements and obtain
· ¡ 2
A + AD − 1 sin N θ − A sin (N − 1) θ B [(A + D) sin N θ − sin (N − 1) θ]
¢ ¸
1
¡ 2
sin θ C [(A + D) sin N θ − sin (N − 1) θ] D + AD − 1 sin N θ − D sin (N − 1) θ
¢
and then
A [(A + D) sin N θ − sin (N − 1) θ] − sin N θ B [(A + D) sin N θ − sin (N − 1) θ]
· ¸
1
sin θ C [(A + D) sin N θ − sin (N − 1) θ] D [(A + D) sin N θ − sin (N − 1) θ] − sin N θ
In each matrix element, the expression
(A + D) sin N θ = 2 cos θ sin N θ = sin (N + 1) θ + sin (N − 1) θ (0.39)

1
occurs, which we have rearranged using cos θ = 2 (A + D) while twice invoking
(0.14). The result is
¸N +1
A sin (N + 1) θ − sin N θ B sin (N + 1) θ
· · ¸
A B 1
=
C D sin θ C sin (N + 1) θ D sin (N + 1) θ − sin N θ
which completes the proof.
0.4 Fourier Theory

In optics, it is common to decompose complicated light fields into a superposi-
tion of pure sinusoidal waves. This is called Fourier analysis.5 This is important
since individual sine waves tend to move differently through optical systems (say,
a piece of glass with frequency dependent index). After propagation through a
system, we can also reassemble sinusoidal waves to see the effect on the over-
all waveform. In fact, it will be possible to work simultaneously with infinitely
many sinusoidal waves, where the frequencies comprising a light field are spread
continuously over a range. Fourier transforms are also helpful for diffraction
problems where many waves (all with the same frequency) interfere spatially.
5 See Murray R. Spiegel, Schaum’s Outline of Advanced Mathematics for Engineers and Scientists,
Chaps. 7-8 (New York: McGraw-Hill 1971).

We begin with a derivation of the Fourier integral theorem. As asserted by

Fourier, a periodic function can be represented in terms of sines and cosines in
the following manner:
∞
X
f (t ) = a n cos (n∆ωt ) + b n sin (n∆ωt ) (0.40)
n=0
This is called a Fourier expansion. It is similar in idea to a Taylor’s series (0.18),

which rewrites a function as a polynomial. In both cases, the goal is to represent
one function in terms of a linear combination of other functions (requiring a
complete basis set). In a Taylor’s series the basis functions are polynomials and
in a Fourier expansion the basis functions are sines and cosines with various
frequencies (multiples of a fundamental frequency).
By inspection, we see that all terms in (0.40) repeat with a maximum period
of 2π/∆ω. In other words, a Fourier series is good for functions where f (t ) =
f (t + 2π/∆ω). The expansion (0.40) is useful even if f (t ) is complex, requiring a n
and b n to be complex.
Using (0.16), we can rewrite the sines and cosines in the expansion (0.40) as
Joseph Fourier (1768-1830, French) ∞
X e i n∆ωt + e −i n∆ωt e i n∆ωt − e −i n∆ωt
was born to a tailor in Auxerre, France.
f (t ) = an + bn
He was orphaned at age eight. Because
n=0 2 2i
of his humble background, which closed
∞ a −ib ∞ a +ib
(0.41)
some doors to his education and career,
X n n i n∆ωt X n n −i n∆ωt
he became a prominent supporter of the
= a0 + e + e
n=1 2 n=1 2
French Revolution. He was rewarded
by an appointment to a position in the
École Polytechnique. In 1798, partici- or more simply as
pated in Napoleon's expedition to Egypt
∞
c n e −i n∆ωt
X
and served as governor over lower Egypt f (t ) = (0.42)
for a time. Fourier made signicant con- n=−∞
tributions to the study of heat transfer
and vibrations (presented in 1822), and
where
a −n − i b −n
it was in this context that he asserted c n<0 ≡
that functions could be represented as a 2
series of sine waves. (Wikipedia) an + i bn (0.43)
c n>0 ≡
2
c0 ≡ a0
Notice that if c −n = c n∗ for all n, then f (t ) is real (i.e. real a n and b n ); otherwise
f (t ) is complex. The real parts of the c n coefficients are connected with the cosine
terms in (0.40), and the imaginary parts of the c n coefficients are connected with
the sine terms in (0.40).
Given a known function f (t ), we can compute the various coefficients c n .
There is a trick for figuring out how to do this. We multiply both sides of (0.42) by
e i m∆ωt , where m is an integer, and integrate over the function period 2π/∆ω:
π/∆ω π/∆ω
Z ∞ Z
i m∆ωt
e i (m−n)∆ωt d t
X
f (t )e dt = cn
n=−∞
−π/∆ω −π/∆ω
¸π/∆ω
∞ e i (m−n)∆ωt
·
X
= cn
n=−∞ i (m − n) ∆ω −π/∆ω (0.44)
X 2πc n e i (m−n)π − e −i (m−n)π
∞ · ¸
=
n=−∞ ∆ω 2i (m − n) π
∞ 2πc sin [(m − n) π]
X n
=
n=−∞ ∆ω (m − n) π
The function sin [(m − n) π] / [(m − n) π] is equal to zero for all n 6= m, and it is
equal to one when n = m (to see this, use L’Hospital’s rule on the zero-over-zero
situation, or just go back and re perform the above integral for n = m). Thus, only
one term contributes to the summation in (0.44). We now have
π/∆ω
∆ω
Z
cm = f (t )e i m∆ωt d t (0.45)
2π
−π/∆ω
from which the coefficients c n can be computed, given a function f (t ). (Note that
m is a dummy index so we can change it back to n if we like.)
This completes the circle. If we know the function f (t ), we can find the
coefficients c n via (0.45), and, if we know the coefficients c n , we can generate the
function f (t ) via (0.42). If we are feeling a bit silly, we might combine these into a
single identity:
π/∆ω
 
 ∆ω
∞ Z
f (t )e i n∆ωt d t  e −i n∆ωt
X
f (t ) = (0.46)
n=−∞ 2π
−π/∆ω
We start with a function f (t ) followed by a lot of computation and obtain the

function back again! (This is not quite as foolish as it first appears, as we will
discuss later.)
As mentioned above, Fourier expansions represent functions f (t ) that are
periodic over the interval 2π/∆ω. This is disappointing since many optical wave-
forms do not repeat (e.g. a single short laser pulse). Nevertheless, we can represent
a function f (t ) that is not periodic if we let the period 2π/∆ω become infinitely
long. In other words, we can accommodate non-periodic functions if we take the
limit as ∆ω goes to zero so that the spacing of terms in the series becomes very
fine. Applying this limit to (0.46) we obtain
Z∞
 
1 ∞ 0
e −i n∆ωt f t 0 e i n∆ωt d t 0 ∆ω
X ¡ ¢
f (t ) = lim (0.47)
2π ∆ω→0 n=−∞
−∞
At this point, a brief review of the definition of an integral is helpful to better

understand the next step that we shall administer to (0.47).
Changing the summation in (0.47) over to an integral
Recall that an integral is really a summation of rectangles under a curve with finely
spaced steps:
Zb b−a
∆ω
g (ω) d ω ≡ lim g (a + n∆ω) ∆ω
X
∆ω→0 n=0
a
b−a
(0.48)
a +b
2∆ω
µ ¶
+ n∆ω ∆ω
X
= lim g
∆ω→0 b−a 2
n=− 2∆ω
The final expression has been manipulated so that the index ranges through both
negative and positive numbers. If we set a = −b and take the limit b → ∞, then the
above expression becomes
Z∞ ∞
g (ω) d ω = lim g (n∆ω) ∆ω
X
(0.49)
∆ω→0 n=−∞
−∞
This concludes our short review of calculus.
Now, (0.47) has the same form as (0.49) if g (n∆ω) represents everything in
the square brackets of (0.47). The result is the Fourier integral theorem:
Z∞ Z∞
 
1 1
e −i ωt  p f t 0 e i ωt d t 0  d ω
¡ ¢ 0
f (t ) = p (0.50)
2π 2π
−∞ −∞
The piece in brackets is called the Fourier transform, and the rest of the operation
is called the inverse Fourier transform. The Fourier integral theorem (0.50) is often
written with the following (potentially confusing) notation:
Z∞
1
f (ω) ≡ p f (t )e i ωt d t
2π
−∞
(0.51)
Z∞
1 −i ωt
f (t ) ≡ p f (ω) e dω
2π
−∞
The transform and inverse transform are also sometimes written as f (ω) ≡
F f (t ) and f (t ) ≡ F −1 f (ω) . Note that the functions f (t ) and f (ω) are en-
© ª © ª
tirely different, even taking on different units (e.g. the latter having extra units of
per frequency). The two functions are distinguished by their arguments, which
also have different units (e.g. time vs. frequency). Nevertheless, it is customary to
use the same letter to denote either function since they form a transform pair.
You should be aware that it is arbitrary which of the expressions in (0.51) is

called the transform and which is called the inverse transform. In other words, the
signs in the exponents of (0.51) may be interchanged (and this convention varies
in published works!). Also, the factor 2π may be placed on either the transform or
the inverse transform, or divided equally between the two as has been done here.
Example 0.6
2 /2T 2
Compute the Fourier transform of E (t ) = E 0 e −t e −i ω0 t followed by the inverse
Fourier transform.
Solution: According to (0.51), the Fourier transform is

Z∞ ³ Z∞
1 −t 2 /2T 2 −i ω0 t
´
i ωt E0 2 2
E (ω) = p E0e e e dt = p e −t /2T +i (ω−ω0 )t d t
2π 2π
−∞ −∞
The integration can be performed with the help of (0.55), which yields
(ω−ω0 )2
π
r
E0 − 2 2
E (ω) = p 2
e 4(1/2T 2 )
= T E 0 e −T (ω−ω0 ) /2
2π 1/2T
Similarly, the inverse Fourier transform of the above function is
Z∞ ³ Z∞
1 −T 2 (ω−ω0 )2 /2
´
−i ωt T E0 T2 2
ω2 +(T 2 ω0 −i t )ω− T2 ω20
E (t ) = p T E0e e dω = p e− 2 dω
2π 2π
−∞ −∞
where again we use (0.55) to obtain

2
T E0
r
π (T 2 ω0 −i t ) 2
− T2 ω20 2 /2T 2 −i ω t
2
E (t ) = p e 4(T /2) = E 0 e −t 0
2π T 2 /2
which brings us back to where we started.
As was previously mentioned, it would seem rather pointless to perform

a Fourier transform on the function f (t ) followed by an inverse Fourier trans-
form, just to end up with f (t ) again. Instead, we will typically apply a frequency-
dependent effect on f (ω) before performing the inverse Fourier transform. In
this case, the final function will be different from f (t ). Keep in mind that f (ω) is
the continuous analog of the discrete coefficients c n (or the a n and b n ). The real
part of f (ω) indicates the amplitudes of the cosine waves necessary to construct
the function f (t ). The imaginary part of f (ω) indicates the amplitudes of the sine
waves necessary to construct the function f (t ).
Finally, we comment on the Dirac delta function, 6 which is defined indirectly
through
Z∞
f t0 δ t0 − t dt0
¡ ¢ ¡ ¢
f (t ) = (0.52)
−∞
6 See G. B. Arfken and H. J. Weber, Mathematical Methods for Physicists 6th ed., Sect. 1.15 (San
Diego: Elsevier Academic Press 2005).

The delta function δ t 0 − t is zero everywhere except at t 0 = t where it is infinite

¡ ¢
in such a way as to make the integral take on the value of the function f (t ). (You
can think of δ t 0 − t d t 0 as an infinitely tall and infinitely thin rectangle centered
¡ ¢
at t 0 = t with an area unity.) The integral only pays attention to the value of f t 0
¡ ¢
at the point t 0 = t .
A remarkable attribute of the delta function can be seen from the Fourier
integral theorem. After rearranging the order of integration, the Fourier integral
theorem (0.50) can be written as
Z∞ Z∞
 
¡ 0¢ 1
e i ω(t −t ) d ω d t 0
0
f (t ) = f t  (0.53)
2π
−∞ −∞
A comparison of (0.52) and (0.53) shows that you may write the delta function
as a uniform superposition of all frequency components:
Z∞
1
e i ω(t −t ) d ω
0
0
δ t −t =
¡ ¢
(0.54)
2π
−∞
Example 0.7
Use (0.54) to prove Parseval’s relation: 7
Z∞ Z∞
¯ f (ω)¯2 d ω = ¯ f (t )¯2 d t
¯ ¯ ¯ ¯
−∞ −∞
which comes up often in the study of optics.
Solution:
Z∞ Z∞
¯ f (ω)¯2 d ω = f (ω) f ∗ (ω) d ω
¯ ¯
−∞ −∞
Z∞  1 Z∞   1 Z∞ ¡ ¢
  
0

= p f (t ) e i ωt d t p f ∗ t 0 e −i ωt d t 0 d ω
 2π   2π 
−∞ −∞ −∞
The order of integration can be changed to give

Z∞ Z∞ Z∞  1 Z∞
 
0

e i ω(t −(−t )) d ω d t d t 0
¯2
¯ f (ω)¯ d ω = f (t ) f ∗ −t 0
¯ ¡ ¢
 2π 
−∞ −∞ −∞ −∞
Z∞ Z∞
f (t ) f ∗ −t 0 δ t 0 − (−t ) d t d t 0
¡ ¢ ¡ ¢
=
−∞ −∞
Z∞ Z∞
f (t ) f ∗ (t ) d t = ¯ f (t )¯2 d t
¯ ¯
=
−∞ −∞
7 For a more general version of the relation, see G. B. Arfken and H. J. Weber, Mathematical
Methods for Physicists 6th ed., Sect. 15.5 (San Diego: Elsevier Academic Press 2005).
Equation (0.54) was used to reach the final result.

Appendix 0.A Table of Integrals and Sums

The following formulas are useful for various problems encountered in the text.
Z∞
π b2 +c
r
−ax 2 +bx+c
e dx = e 4a (Re {a} > 0) (0.55)
a
−∞
Z∞
e i ax π |b| −|ab|
2 2
dx = e (b > 0) (0.56)
1 + x /b 2
0
Z2π
e ±i a cos(θ−θ ) d θ = 2πJ 0 (a)
0
(0.57)
0
Za
a
J 0 (bx) x d x = J 1 (ab) (0.58)
b
0
Z∞ 2
−ax 2 e −b /4a
e J 0 (bx) x d x = (0.59)
2a
0
Z∞
sin2 (ax) π
2
dx = (0.60)
(ax) 2a
0
dy y
Z
¤3/2 = p (0.61)
c y2 + c
£
y2 + c
p
dx 1 −1 c
Z
p = − p sin (0.62)
x x2 − c c |x|
Zπ Zπ
1
sin(ax) sin(bx) d x = cos(ax) cos(bx) d x = δab (a, b integer) (0.63)
2
0 0
N 1 − r N +1
rn =
X
(0.64)
n=0 1−r
N r (1 − r N )
rn =
X
(0.65)
n=1 1−r
∞ 1
rn =
X
(r < 1) (0.66)
n=0 1−r
Exercises 21
Exercises
Exercises for 0.1 Vector Calculus

¡ ¢ ¡ ¢
P0.1 Let r = x̂ + 2ŷ − 3ẑ m and r0 = −x̂ + 3ŷ + 2ẑ m.
(a) Find the magnitude of r.
(b) Find r − r0 .
(c) Find the angle between r and r0 .
p
Answer: (a) r = 14 m; (c) 94◦ .
P0.2 Use the dot product (0.2) to show that the cross product E × B is per-
pendicular to E and to B.
P0.3 Verify the “BAC-CAB” rule: A × (B × C) = B (A · C) − C (A · B).
P0.4 Prove the following identity:
r − r0
¡ ¢
1
∇r =− ,
|r − r0 | |r − r0 |3
where ∇r operates only on r, treating r0 as a constant vector.
P0.5 (r−r0 ) is zero, except at r = r0 where a singularity situation

Prove that ∇r · |r−r 0 |3
occurs.
P0.6 Verify ∇ · (∇ × f) = 0 for any vector function f.

¡ ¢ ¡ ¢ ¡ ¢
P0.7 Verify ∇ × f × g = f ∇ · g − g (∇ · f) + g · ∇ f − (f · ∇) g.
¡ ¢ ¡ ¢
P0.8 Verify ∇ · f × g = g · (∇ × f) − f · ∇ × g .
¡ ¢ ¡ ¢ ¡ ¢
P0.9 Verify ∇ · g f = f · ∇g + g ∇ · f and ∇ × g f = ∇g × f + g ∇ × f.
P0.10 Show that the Laplacian in cylindrical coordinates can be written as
1 ∂ ∂ 1 ∂2 ∂2
µ ¶
2
∇ = ρ + 2 +
ρ ∂ρ ∂ρ ρ ∂φ2 ∂z 2
Solution: (Partial)
Continuing with the approach in Example 0.2, we have
Ã ! Ã !
∂2 f ∂2 ρ ∂ f ∂ρ ∂ ∂ f ∂2 φ ∂ f ∂φ ∂ ∂ f
= + + +
∂x 2 ∂x 2 ∂ρ ∂x ∂ρ ∂x ∂x 2 ∂φ ∂x ∂φ ∂x
Ã ! ¸ Ã 2 !
∂2 ρ ∂ f ∂ρ ∂ ∂ρ ∂ f ∂φ ∂ f ∂ φ ∂f ∂φ ∂ ∂ρ ∂ f ∂φ ∂ f
·µ ¶ µ ¶ ·µ ¶ µ ¶ ¸
= + + + + +
∂x 2 ∂ρ ∂x ∂ρ ∂x ∂ρ ∂x ∂φ ∂x 2 ∂φ ∂x ∂φ ∂x ∂ρ ∂x ∂φ
and
∂2 f ∂2 f ∂2 f
∇2 f = + +
∂x 2 ∂y 2 ∂z 2
Ã ! Ãµ ¶ !
∂2 ρ ∂2 ρ ∂ρ 2 ∂2 f
¶¸ 2
∂f ∂ρ 2 ∂φ ∂ρ ∂φ ∂ρ ∂ f
¶ µ ·µ ¶µ ¶ µ ¶µ
= + + + +2 +
∂x 2 ∂y 2 ∂ρ ∂x ∂y ∂ρ 2 ∂x ∂x ∂y ∂y ∂φ∂ρ
"Ã ! Ã !# "µ ¶ #
∂2 φ ∂2 φ ∂f ∂φ 2 ∂φ 2 ∂2 f ∂2
¶ µ
+ + + + +
∂x 2 ∂y 2 ∂φ ∂x ∂y ∂φ2 ∂z 2
The needed first derivatives are given in Example 0.2. The needed second derivatives are
∂2 ρ 1 x2 sin2 φ
=q −¡ =
∂x 2 ρ
¢3/2
x2 + y 2 x2 + y 2
∂2 φ 2x y 2 sin φ cos φ
=¡ ¢2 =
∂x 2 x2 + y 2 ρ2
∂2 ρ 1 y2 cos2 φ
=q −¡ =
∂y 2 ρ
¢3/2
x2 + y 2 x2 + y 2
∂2 φ 2x y 2 sin φ cos φ
=−¡ ¢2 = −
∂y 2 x2 + y 2 ρ2
Finish the derivation by substituting these derivatives into the above expression.
P0.11 Verify Stokes’ theorem (0.12) for the function given in Example 0.3.
Take¯the¯ surface to be a square in the x y-plane contained by |x| = ±1
and ¯ y ¯ = ±1, as illustrated in Fig. 0.6.
P0.12 Verify the following vector integral theorem for the same volume used
in Example 0.3, but with F = y 2 x x̂ + x y ẑ and G = x 2 x̂:
Z I
[F (∇ · G) + (G · ∇) F] d v = F (G · n̂) d a
Figure 0.6
V S
P0.13 Use the divergence theorem to show that the function in P0.5 is 4π
times the three-dimensional delta function
δ 3 r0 − r ≡ δ x 0 − x δ y 0 − y δ z 0 − z
¡ ¢ ¡ ¢ ¡ ¢ ¡ ¢
which has the property that
1 if V contains r0
Z ½
δ3 r0 − r d v =
¡ ¢
0 otherwise
V
Solution: We have by the divergence theorem
r − r0 r − r0
I ¡ ¢ Z ¡ ¢
¯ ¯3 · n̂d a = ∇r · ¯ ¯3 d v
¯r − r 0 ¯ ¯r − r 0 ¯
S V
Exercises 23
From P0.5, the argument in the integral on the right-hand side is zero except at r = r0 . Therefore,
if the volume V does not contain the point r = r0 , then the result of both integrals must be zero.
Let us construct a volume between an arbitrary surface S 1 containing r = r0 and S 2 , the surface
of a tiny sphere centered on r = r0 . Since the point r = r0 is excluded by the tiny sphere, the result
of either integral in the divergence theorem is still zero. However, we have on the tiny sphere
Z2πZπ Ã !
r − r0
I ¡ ¢
1
¯3 · n̂d a = − r ²2 sin φd φd α = −4π
r ²2
¯
¯r − r 0 ¯
S2 0 0
Therefore, for the outer surface S 1 (containing r = r0 ) we must have the equal and opposite
result:
r − r0
I ¡ ¢
¯ ¯3 · n̂d a = 4π
¯r − r 0 ¯
S1
This implies
r − r0
¡ ¢
4π if V contains r0
Z ½
∇r · ¯ ¯3 d v = 0 otherwise
¯r − r 0 ¯
V
r−r0
¡ ¢
The integrand exhibits the same characteristics as the delta function Therefore, ∇r · =
|r−r0 |3
4πδ3 r − r0 . The delta function is defined in (0.52)
¡ ¢
Exercises for 0.2 Complex Numbers
P0.14 Using only a calculator’s arithmetic and trigonometric functions, com-

pute z 1 − z 2 and z 1 /z 2 in both rectangular and polar form for z 1 = 1 − i
and z 2 = 3 + 4i .
P0.15 Show that

a −ib −1 b
= e −2i tan a
a +ib
regardless of the sign of a, assuming a and b are real.
P0.16 Invert (0.15) to get both formulas in (0.16). HINT: You can get a second
equation by considering Euler’s equation with a negative angle −φ.
P0.17 Show Re {A} × Re {B } = (AB + A ∗ B ) /4 +C .C .
P0.18 If E 0 = |E 0 | e i δE and B 0 = |B 0 | e i δB , and if k, z, ω, and t are all real, prove

n o n o 1¡
Re E 0 e i (kz−ωt ) Re B 0 e i (kz−ωt ) = E 0∗ B 0 + E 0 B 0∗
¢
4
1
+ |E 0 | |B 0 | cos [2 (kz − ωt ) + δE + δB ]
2
p
P0.19 (a) If sin φ = 2, show that cos φ = i 3. HINT: Use sin2 φ + cos2 φ = 1.
p
(b) Show that the angle φ in (a) is π/2 − i ln(2 + 3).
P0.20 Write A cos(ωt ) + 2A sin(ωt + π/4) as simple phase-shifted cosine wave

(i.e. find the amplitude and phase of the resultant cosine wave).
Exercises for 0.4 Fourier Theory
P0.21 Prove that Fourier Transforms have the property of linear superposi-
tion:
F ag (t ) + bh (t ) = ag (ω) + bh (ω)
© ª
where g (ω) ≡ F g (t ) and h(ω) ≡ F {h(t )}.

© ª
ª 1 ¡ω¢
Prove F g (at ) = |a|
©
P0.22 g a .
Prove F g (t − τ) = g (ω)e i ωτ .
© ª
P0.23
2
P0.24 Show that the Fourier transform of E (t ) = E 0 e −(t /T ) cos ω0 t is
T E 0 − (ω+ω0 )2 ω−ω0 )2
µ ¶
−(
E (ω) = p e 4/T 2 + e 4/T 2
2 2
P0.25 Take the inverse Fourier transform of the result in P0.24. Check that it
returns exactly the original function.
P0.26 The following operation is referred to as the convolution of the func-

tions g (t ) and h(t ):
Z∞
¯
g (t ) ⊗ h(t )¯τ ≡ g (t )h(τ − t ) d t
−∞
A convolution measures the overlap of g (t ) and a reversed h(t ) as a

function of the offset τ. The result is a function of τ.
(a) Prove the convolution theorem:
p
F g (t ) ⊗ h(t )¯τ ¯ω = 2πg (ω)h(ω)
© ¯ ª¯
(b) Prove this related form of the convolution theorem:

1
F g (t )h(t ) ¯ω = p g (ω0 ) ⊗ h(ω0 )¯ω
© ª¯ ¯
2π
Solution: Part (a)

 ∞
Z∞  Z∞
¯  
Z ¯¯ 1
g (t ) h (τ − t ) d t e i ωτ d τ

F g (t )h(τ − t ) d t ¯¯ = p (Let τ = t 0 + t )
 ¯ 2π  
−∞ ω −∞ −∞
Z∞ Z∞
1 ¡ 0 ¢
g (t ) h t 0 e i ω t +t d t d t 0
¡ ¢
=p
2π
−∞ −∞
p Z∞ Z∞
1 1 0
g (t ) e i ωt d t p h t 0 e i ωt d t 0
¡ ¢
= 2π p
2π 2π
−∞ −∞
p
= 2πg (ω) h (ω)
Exercises 25
P0.27 Prove the autocorrelation theorem:

 ∞ 
Z  p
F h(t )h ∗ (t − τ)d t = 2π |h(ω)|2
 
−∞
P0.28 (a) Compute the Fourier transform of a Gaussian function, g (t ) =

2 2
e −t /2T . Do the integral by hand using the table in Appendix 0.A.
(b) Compute the Fourier transform of a sine function, h(t ) = sin ω0 t .
Do the integral by hand using sin(x) = (e i x − e −i x )/2i , combined with
the integral formula (0.54).
(c) Use your results to parts (a) and (b) and a convolution theorem from
2 2
P0.26(b) to evaluate the Fourier transform f (t ) = e −t /2T sin ω0 t . (The
answer should be similar to P0.24).
(d) Plot f (t ) and the imaginary part of its Fourier transform for the
parameters ω0 = 1 and T = 8.
Chapter 1
Electromagnetic Phenomena
In 1861, James Maxwell assembled the various known relationships of electricity

and magnetism into a concise1 set of equations:2
ρ
∇·E = (Gauss’s Law) (1.1)
²0
∇·B = 0 (Gauss’s Law for magnetism) (1.2)
∂B
∇×E = − (Faraday’s Law) (1.3)
∂t
B ∂E
∇× = ²0 +J (Ampere’s Law revised by Maxwell) (1.4)
µ0 ∂t
Here E and B represent electric and magnetic fields, respectively. The charge
density ρ describes the charge per volume distributed through space.3 The current
density J describes the motion of charge density (in units of ρ times velocity). The
constant ²0 is called the permittivity, and the constant µ0 is called the permeability.
Taken together, these are known as Maxwell’s equations.
After introducing a key revision of Ampere’s law, Maxwell realized that together
these equations comprise a complete self-consistent theory of electromagnetic
phenomena. Moreover, the equations imply the existence of electromagnetic
waves, which travel at the speed of light. Since the speed of light had been
measured before Maxwell’s time, it was immediately apparent (as was already
suspected) that light is a high-frequency manifestation of the same phenomena
that govern the influence of currents and charges upon each other. Previously,
optics had been considered a topic quite separate from electricity and magnetism.
Once the connection was made, it became clear that Maxwell’s equations form
the theoretical foundations of optics, and this is where we begin our study of light.
1 In Maxwell’s original notation, this set of equations was hardly concise, written without the
convenience of modern vector notation or ∇. His formulation wouldn’t fit easily on a T-shirt!
2 See J. D. Jackson, Classical Electrodynamics, 3rd ed., p. 1 (New York: John Wiley, 1999) or the
back cover of D. J. Griffiths, Introduction to Electrodynamics, 3rd ed. (New Jersey: Prentice-Hall,
1999).
3 Later in the book we use ρ for the radius in cylindrical coordinates, not to be confused with
charge density.
27
28 Chapter 1 Electromagnetic Phenomena
1.1 Gauss’ Law

The force on a point charge q located at r exerted by another point charge q 0
located at r0 is
F = qE(r) (1.5)
where
q 0 r − r0
¡ ¢
E (r) = (1.6)
4π²0 |r − r0 |3
This relationship is known as Coulomb’s law. The force is directed along the
vector r − r0 , which points from charge 0
¯ q 0 ¯to q as seen in Fig. 1.1. The length or
0
magnitude of this vector is given by r − r (i.e. the distance between q¯ and q).
¯ ¯
The familiar inverse square law can be seen by noting that r − r0 ¯r − r0 ¯ is a unit
¡ ¢ ±¯
vector. We have written the force in terms of an electric field E (r), which is defined
throughout space (regardless of whether a second charge q is actually present).
The permittivity ²0 amounts to a proportionality constant.
The total force from a collection of charges is found by summing expression
Origin
(1.5) over all charges q n0 associated with their specific locations r0n . If the charges
are distributed continuously throughout space, having density ρ r0 (units of
¡ ¢
Figure 1.1 The geometry of
Coulomb’s law for a point charge charge per volume), the summation for finding the net electric field at r becomes
an integral:
¡ 0 ¢ r − r0
¡ ¢
1
Z
E (r) = ρ r d v0 (1.7)
4π²0 |r − r0 |3
V
4
This three-dimensional integral gives the net electric field produced by the
charge density ρ distributed throughout the volume V .
Gauss’ law (1.1), the first of Maxwell’s equations, follows directly from (1.7)
with some mathematical manipulation. No new physical phenomenon is intro-
duced in this process.5
Origin
Figure 1.2 The geometry of Derivation of Gauss’ law

Coulomb’s law for a charge dis-
We begin with the divergence of (1.7):
tribution.
r − r0
¡ ¢
1
Z
¡ 0¢
∇ · E (r) = ρ r ∇r · d v0 (1.8)
4π²0 |r − r0 |3
V
The subscript on ∇r indicates that it operates on r while treating r0 , the dummy

variable of integration, as a constant. The integrand contains a remarkable mathe-
matical property that can be exploited, even without specifying the form of the
4 Here d v 0 stands for d x 0 d y 0 d z 0 and r0 = x 0 x̂ + y 0 ŷ + z 0 ẑ (in Cartesian coordinates).
5 Actually, Coulomb’s law applies only to static charge configurations, and in that sense it is
incomplete since it implies an instantaneous response of the field to a reconfiguration of the

charge. The generalized version of Coulomb’s law, one of Jefimenko’s equations, incorporates
the fact that electromagnetic news travels at the speed of light. See D. J. Griffiths, Introduction
to Electrodynamics, 3rd ed., Sect. 10.2.2 (New Jersey: Prentice-Hall, 1999). Ironically, Gauss’ law,
which can be derived from Coulomb’s law, holds perfectly whether the charges remain still or are in
motion.
1.2 Gauss’ Law for Magnetic Fields 29
charge distribution ρ r0 . In modern mathematical language, the vector expression

¡ ¢
in the integral is a three-dimensional delta function (see (0.52):6
r − r0
¡ ¢
≡ 4πδ3 r0 − r ≡ 4πδ x 0 − x δ y 0 − y δ z 0 − z
¡ ¢ ¡ ¢ ¡ ¢ ¡ ¢
∇r · 3
(1.9)
0
|r − r |
A derivation of this formula is addressed in problem P0.13. The delta function

allows the integral in (1.8) to be performed, and the relation becomes simply
ρ (r)
∇ · E (r) =
²0
which is the differential form of Gauss’ law (1.1).
The (perhaps more familiar) integral form of Gauss’ law can be obtained by
integrating (1.1) over a volume V and applying the divergence theorem (0.11) to
the left-hand side:
1 Figure 1.3 Gauss’ law in integral
I Z
E (r) · n̂ d a = ρ (r) d v (1.10) form relates the flux of the elec-
²0
S V tric field through a surface to the
This form of Gauss’ law shows that the total electric field flux extruding through a charge contained inside that sur-
face.
closed surface S (i.e. the integral on the left side) is proportional to the net charge
contained within it (i.e. within volume V contained by S).
Example 1.1
Suppose we have an electric field given by E = (αx 2 y 3 x̂ + βz 4 ŷ) cos ωt . Use Gauss’
law (1.1) to find the charge density ρ(x, y, z, t ).
Solution:
∂ ∂ ∂
µ ¶
ρ = ²0 ∇ · E = ²0 x̂ + ŷ + ẑ (αx 2 y 3 x̂ + βz 4 ŷ) cos ωt = 2²0 αx y 3 cos ωt
∂x ∂y ∂z
Carl Friedrich Gauss (17771855, Ger-

man) was born in Braunschweig, Ger-
1.2 Gauss’ Law for Magnetic Fields many to a poor family. Gauss was a
child prodigy, and he made his rst sig-
nicant advances to mathematics as a
In order to ‘feel’ a magnetic force, a charge q must be moving at some velocity (call
teenager. In grade school, he purport-
it v). The magnetic field arises itself from charges that are in motion. We consider edly was asked to add all integers from
the magnetic field to arise from a distribution of moving charges described by a 1 to 100, which he did in seconds to the
astonishment of his teacher. (Presum-
current density J r0 throughout space. The current density has units of charge
¡ ¢
ably, Friedrich immediately realized that
times velocity per volume (or equivalently, current per cross sectional area). The the numbers form fty pairs equal to
101.) Gauss made important advances
magnetic force law analogous to Coulomb’s law is in number theory and dierential geome-
try. He developed the law discussed here
F = qv × B (1.11) as one of Maxwell's equations in 1835,
but it was not published until 1867, af-
6 For a derivation of Gauss’ law from Coulomb’s law that does not rely directly on the Dirac delta ter Gauss' death. Ironically, Maxwell
was already using Gauss' law by that
function, see J. D. Jackson, Classical Electrodynamics 3rd ed., pp. 27-29 (New York: John Wiley,
time. (Wikipedia)
1999).
where
µ0 r − r0
Z ¡ ¢
¡ 0¢
B (r) = J r × d v0 (1.12)
4π |r − r0 |3
V
The latter equation is known as the Biot-Savart law. The permeability µ0 dictates
the strength of the magnetic field, given the current distribution.
As with Coulomb’s law, we can apply mathematics to the Biot-Savart law
to obtain another of Maxwell’s equations. Nevertheless, the essential physics
is already inherent in the Biot-Savart law.7 Using the result from P0.4, we can
rewrite (1.12) as8
Jean-Baptiste Biot (1774-1862,
µ0 µ0 J r0
¡ ¢
1
Z Z
French) was born in Paris. He attended ¡ 0¢ 0
the École Polytechnique where mathe-
B (r) = − J r × ∇r dv = ∇× d v0 (1.13)
4π |r − r0 | 4π |r − r0 |
matician Gaspard Monge recognized his V V
academic potential. After graduating,
Biot joined the military and then took
Since the divergence of a curl is identically zero (see P0.6), we get straight away
part in an insurrection on the side of
the Royalists. He was captured, and the second of Maxwell’s equations (1.2)
his career might of have met a tragic
ending there had Monge not success-
∇·B = 0
fully pleaded for his release from jail.
Biot went on to become a professor
of physics at the College de France. which is known as Gauss’ law for magnetic fields. (Two equations down; two to
Among other contributions, Biot partic-
go.)
ipated in the rst hot-air balloon ride
with Gay-Lussac and correctly deduced The similarity between ∇ · B = 0 and ∇ · E = ρ/²0 , Gauss’ law for electric fields,
that meteorites that fell on L'Aigle,
is immediately apparent. In integral form, Gauss’ law for magnetic fields looks the
France in 1803 came from space. Later
Biot collaborated with the younger Felix same as (1.10), only with zero on the right-hand side. If one were to imagine the
Savart (1791-1841) on the theory of existence of magnetic monopoles (i.e. isolated north or south ‘charges’), then the
magnetism and electrical currents. They
formulated their famous law in 1820.
right-hand side would not be zero. The law implies that the total magnetic flux
(Wikipedia) extruding through any closed surface balances, with as many field lines pointing
inwards as pointing outwards.
Example 1.2
The field surrounding a magnetic dipole is given by
B = β 3xz x̂ + 3y z ŷ + 3z 2 − r 2 ẑ r 5
£ ¡ ¢ ¤±
p
where r ≡ x 2 + y 2 + z 2 . Show that this field satisfies Gauss’ law for magnetic
fields (1.2).
7 Like Coulomb’s law, the Biot-Savart law is incomplete since it also implies an instantaneous
response of the magnetic field to a reconfiguration of the currents. The generalized version of the
Biot-Savart law, another of Jefimenko’s equations, incorporates the fact that electromagnetic news
travels at the speed of light. Ironically, Gauss’ law for magnetic fields and Maxwell’s version of
Ampere’s law, derived from the Biot-Savart law, hold perfectly whether the Currents are steady or
vary in time. The Jefimenko equations, analogs of Coulomb and Biot-Savart, also embody Faraday’s
law, the only of Maxwell’s equations that cannot be derived from the usual forms of Coulomb’s law
and the Biot-Savart law. See D. J. Griffiths, Introduction to Electrodynamics, 3rd ed., Sect. 10.2.2
(New Jersey: Prentice-Hall, 1999).
8 Note that ∇ ignores the variable of integration r0 .
r
1.3 Faraday’s Law 31
Solution:
∂ ³ xz ´ ∂ ³ y z ´ ∂ 3z 2
· µ ¶¸
1
∇·B = β 3 +3 + − 3
∂x r 5 ∂y r 5 ∂z r 5 r
z 5xz ∂r z 5y z ∂r 6z 15z 2 ∂r 3 ∂r
· µ ¶ µ ¶ µ ¶¸
=β 3 5 − 6 +3 5 − 6 + 5− 6 +
r r ∂x r r ∂y r r ∂z r 4 ∂z
∂r ∂r ∂r 3 ∂r
· µ ¶ ¸
12z 15z
=β − 6 x +y +z + 4
r5 r ∂x ∂y ∂z r ∂z
.p
The necessary derivatives are ∂r /∂x = x x 2 + y 2 + z 2 = x/r , ∂r /∂y = y/r , and
∂r /∂z = z/r , which lead to
· ¸
12z 15z 3z
∇·B = β − + =0 Michael Faraday (17911867, English)
r5 r5 r5 was one of the greatest experimental
physicists in history. Born on the out-
skirts of London, his family was not well
o, his father being a blacksmith. The
young Michael Faraday only had access
1.3 Faraday’s Law to a very basic education, and so he

was mostly self taught and never did
acquire much skill in mathematics. As
Michael Faraday discovered that changing magnetic fields induce electric fields. a teenager, he obtained a seven-year
apprenticeship with a book binder, dur-
This distinct physical effect, called induction, can be observed when a magnet is
ing which time he read many books,
waved by a loop of wire. Faraday’s law says that a change in magnetic flux through including books on science and electric-
a circuit loop (see Fig. 1.4) induces a voltage around the loop according to ity. Given his background, Faraday's
entry into the scientic community was
very gradual, from servant to assistant
∂
I Z
E · d` = − B · n̂ d a (1.14) and eventually to director of the labo-
∂t ratory at the Royal Institution. Faraday
C S is perhaps best known for his work that
established the law of induction and
The right side describes a change in the magnetic flux through a surface and the for the discovery that magnetic elds
can interact with light, known as the
left side describes the voltage around the loop containing the surface. Faraday eect. He also made many ad-
We apply Stokes’ theorem (0.12) to the left-hand side of Faraday’s law and vances to chemistry during his career
including guring out how to liquify
obtain
several gases. Faraday was a deeply re-
ligious man, serving as a Deacon in his
∂ ∂B
Z Z Z µ ¶
church. (Wikipedia)
(∇ × E) · n̂ d a = − B · n̂ d a or ∇×E+ · n̂ da = 0 (1.15)
∂t ∂t
S S S
Since this equation is true regardless of what surface is chosen, it implies
∂B
∇×E = −
∂t
which is the differential form of Faraday’s law (1.4) (three of Maxwell’s equations
down; one to go).
N
Example 1.3
For the electric field given in Example 1.1, E = (αx 2 y 3 x̂+βz 4 ŷ) cos ωt , use Faraday’s Magnet
law (1.3) to find B(x, y, z, t ).
Figure 1.4 Faraday’s law.
Solution:
¯ ¯
¯ x̂ ŷ ẑ ¯¯
∂B ¯
∂ ∂ ∂ ¯
= −∇ × E = − cos ωt ¯
¯
∂x ∂y ∂z ¯¯
∂t ¯ αx 2 y 3 βz 4 0 ¯
¯
∂ ∂ ¡ 4¢ ∂ ∂ ¡ 2 3¢
·
= − cos ωt x̂ (0) − x̂ βz − ŷ (0) + ŷ αx y
∂y ∂z ∂x ∂z
∂ ¡ 4¢ ∂ ¡ 2 3¢
¸
+ẑ βz − ẑ αx y
∂x ∂y
= 4βz 3 x̂ + 3αx 2 y 2 ẑ cos ωt
¡ ¢
Integrating in time, we get
¢ sin ωt
B = 4βz 3 x̂ + 3αx 2 y 2 ẑ
¡
ω
plus possibly a constant field.
1.4 Ampere’s Law

The Biot-Savart law (1.12) can also be used to derive Ampere’s law. Ampere’s law
is merely the inversion of the Biot-Savart law (1.12) so that J appears by itself,
unfettered by integrals or the like.
Inversion of Biot-Savart Law
We take the curl of (1.12):

André-Marie Ampère (1775-1836, µ0
Z ·
¡ 0¢
¡
r − r0
¢¸
French) was born in Lyon, France. The ∇ × B (r) = ∇r × J r × d v0 (1.16)
young André-Marie was tutored in Latin 4π |r − r0 |3
V
by his father, which gave him access to
the mathematical works of Euler and
We next apply the differential vector rule from P0.7 while noting that J r0 does not
¡ ¢
Bernoulli to which he was drawn at
an early age. When Ampère reached depend on r so that only two terms survive. The curl of B (r) then becomes
young adulthood, French revolutionaries
µ0 r − r0 ¤ r − r0
¡ ¢¸ ¡ ¢¶
executed his father. In 1799, Ampère
Z µ ·
J r0 ∇r ·
£ ¡ 0¢
d v0
¡ ¢
married Julie Carron, who died of ill- ∇ × B (r) = − J r · ∇r (1.17)
ness a few years later. These tragedies 4π |r − r0 |3 |r − r0 |3
V
weighed heavy on Ampère through-
out his life, especially because he was
According to (1.9), the first term in the integral is 4πJ r0 δ3 r0 − r , which is easily
¡ ¢ ¡ ¢
away from his wife during much of their
short life together, while he worked as
integrated. To make progress on the second term, we observe that the gradient can
a professor of physics and chemistry in
Bourg. After her death, Ampère was
be changed to operate on the primed variables without affecting the final result
appointed professor of mathematics at (i.e. ∇r → −∇r0 ). In addition, we take advantage of a vector integral theorem (see
the University of Lyon and then in 1809 P0.12) to arrive at
at the Ècole Polytechnique in Paris. Af-
ter hearing that a current-carrying wire
µ0 r − r0 £ ¡ 0 ¢¤ 0 µ0 r − r0 £ ¡ 0 ¢ ¤ 0
Z ¡ ¢ I ¡ ¢
could attract a compass needle in 1820, ∇ × B (r) = µ0 J (r) − ∇r0 · J r d v + J r · n̂ d a
Ampère quickly developed the theory of 4π |r − r0 |3 4π |r − r0 |3
V S
electromagnetism. (Wikipedia)
(1.18)
1.5 Maxwell’s Adjustment to Ampere’s Law 33
The last term in (1.18) vanishes if we assume that the current density J is com-
pletely contained within the volume V so that it is zero at the surface S. Thus, the
expression for the curl of B (r) reduces to
µ0 r − r0 £
Z ¡ ¢
∇ × B (r) = µ0 J (r) − ∇r0 · J r0 d v 0
¡ ¢¤
(1.19)
4π |r − r0 |3
V
The latter term in (1.19) vanishes if
∇·J ∼
=0 (steady-state approximation) (1.20)
in which case we have succeeded in isolating J and obtained Ampere’s law.
Without Maxwell’s correction, Ampere’s law
∇ × B = µ0 J (1.21)
only applies to quasi steady-state situations. The physical interpretation of Am-

pere’s law is more apparent in integral form. We integrate both sides of (1.21) over
an open surface S, bounded by contour C and apply Stokes’ theorem (0.12) to the
left-hand side: I Z
B (r) · d ` = µ0 J (r) · n̂ d a ≡ µ0 I (1.22)
C S
This law says that the line integral of B around a closed loop C is proportional to
the total current flowing through the loop (see Fig. 1.5). The units of J are current
per area, so the surface integral containing J yields the current I in units of charge
per time.
1.5 Maxwell’s Adjustment to Ampere’s Law

Maxwell was the first to realize that Ampere’s law was incomplete as written in Figure 1.5 Ampere’s law.
(1.21) since there exist situations where ∇ · J 6= 0 (especially the case for optical
phenomena). Maxwell figured out that (1.20) should be replaced with
∂ρ
∇·J = − (1.23)
∂t
This is called the continuity equation for charge and current densities. Simply
stated, if there is net current flowing into a volume there ought to be charge piling
up inside. For the steady-state situation inherently considered by Ampere, the
current into and out of a volume is balanced so that ∂ρ ∂t = 0.
±
Derivation of the Continuity Equation
Consider a volume of space enclosed by a surface S through which current is

flowing. The total current exiting the volume is
I
I = J · n̂ d a (1.24)
S
where n̂ is the outward normal to the surface. The units on this equation are that
of current, or charge per time, leaving the volume.
Since we have considered a closed surface S, the net current leaving the enclosed
volume V must be the same as the rate at which charge within the volume vanishes:
∂
Z
I =− ρ dv (1.25)
∂t
V
Upon equating these two expressions for current, as well as applying the diver-
gence theorem (0.11) to the former, we get
∂ρ ∂ρ
Z Z Z µ ¶
∇ · Jd v = − d v or ∇·J+ dv = 0 (1.26)
∂t ∂t
V V V
Since (1.26) is true regardless of which volume V we choose, it implies (1.23).
Maxwell’s main contribution (aside from organizing other people’s formulas9

and recognizing them as a complete set of coupled differential equations—a big
deal) was the injection of the continuity equation (1.23) into the derivation of
James Clerk Maxwell (18311879, Ampere’s law (1.19). This yields
Scottish) was born to a wealthy family
in Edinburgh, Scotland. Originally, his
µ0 ∂ ¡ 0 ¢ r − r0
Z ¡ ¢
name was John Clerk, but he added his
∇ × B = µ0 J + ρ r d v0 (1.27)
mother's maiden name when he inher- 4π ∂t |r − r0 |3
ited an estate from her family. Maxwell V
was a bright and inquisitive child and
displayed an unusual gift for mathe- Then substitution of (1.7) into this formula gives
matics at an early age. He attended
Edinburgh University and then Trin-
B ∂E
ity College at Cambridge University. ∇× = J + ²0
Maxwell started his career as a professor µ0 ∂t
at Aberdeen University, but lost his job
a few years later during restructuring,
the last of Maxwell’s equations (1.4).
at which time Maxwell took a post at
King's College of London. Maxwell is This revised Ampere’s law includes the additional term ²0 ∂E/∂t , which is
best known for his fundamental contri- known as the displacement current (density). The displacement current exists
butions to electricity and magnetism
and the kinetic theory of gases. He
even in the absence of any actual charge density ρ.10 It indicates that a changing
studied numerous other subjects, includ- electric field behaves like a current in the sense that it produces magnetic fields.
ing the human perception of color and
color-blindness, and is credited with pro-
The similarity between Faraday’s law and the corrected Ampere’s law (1.4) is
ducing the rst color photograph. He apparent. No doubt this played a part in motivating Maxwell’s work.
originally postulated that electromag-
In summary, in the previous section we saw that the basic physics in Ampere’s
netic waves propagated in a mechanical
`luminiferous ether'. He founded the law is present in the Biot-Savart law. Infusing it with charge conservation (1.23)
Cavendish laboratory at Cambridge in yields the corrected form of Ampere’s law.
1874, which has produced 28 Nobel
prizes to date. Maxwell, one of Ein-
stein's heros, died of stomach cancer in 9 Although Gauss developed his law in 1835, it was not published until after his death in 1867,
his forties. (Wikipedia)
well after Maxwell published his laws of electromagnetism, so in practice Maxwell accomplished
much more than merely fixing Ampere’s law.
10 Based on (1.27), one might think that the displacement current ² ∂E/∂t ought to be zero in a
0
region of space with no charge density ρ. However, in (1.27) ρ appears in a volume integral over a
region of space sufficiently large (consistent with a previous supposition) to include any charges
responsible for the field E; presumably, all fields arise from sources.
1.5 Maxwell’s Adjustment to Ampere’s Law 35
Example 1.4
(a) Use Gauss’s law to find the electric field in a gap that interrupts a current-
carrying wire, as shown in Fig. 1.6.
(b) Find the strength of the magnetic field on contour C using Ampere’s law applied
to surface S 1 . C
(c) Show that the displacement current in the gap leads to the identical magnetic
field when using surface S 2 .
I I
Solution: (a) We’ll assume that the cross-sectional area of the wire A is much wider
than the gap separation. Then the electric field in the gap will be uniform, and the
integral on the left-hand side of (1.10) reduces to E A since there is essentially no
field other than in the gap. If the accumulated charge on the ‘plate’ is Q, then the
right-hand side of (1.10) integrates to Q/²0 , and the electric field turns out to be
E = Q/(²0 A). Figure 1.6 Charging capacitor.
(b) Let the contour C be a circle at radius r . The magnetic field points around the
circumference with constant strength. The left-hand side of (1.22) becomes 2πr B
while the right-hand side is
∂Q
Z
µ0 J · n̂d a = µ0 I = µ0
∂t
S
This gives for the magnetic field

µ0 ∂Q
B=
2πr ∂t
(c) If instead we use the displacement current ²0 ∂E/∂t in place of J in in the right-
hand side of right-hand side of (1.22), we get for that piece
∂E ∂E ∂Q
Z µ ¶
µ0 ε0 · n̂d a = µ0 ε0 A = µ0
∂t ∂t ∂t
S
which is the same as before.
Example 1.5
2 3
For the electric field E = (αx ¡ y x̂ + βz 4 ŷ) cos ¢ωt (see Example 1.1) and the as-
sociated magnetic field B = 4βz x̂ + 3αx 2 y 2 ẑ sinωωt (see Example 1.3), find the
3
current density J (x, y, z, t ).
Solution:
¯ ¯
* ¯ x̂ ŷ ẑ ¯
B ∂E sin ωt ¯ ∂ ¯
∂ ∂
¯
J = ∇× − ²0 = ¯ + ²0 ω(αx 2 y 3 x̂ + βz 4 ŷ) sin ωt
¯
¯ ∂x ∂y ∂z
µ0 ∂t µ0 ω ¯¯ 3 2 2 ¯
¯
4βz 0 3αx y
sin ωt £
6αx 2 y x̂ − 6αx y 2 ŷ + 4βz 3 ŷ + ²0 ω(αx 2 y 3 x̂ + βz 4 ŷ) sin ωt
¤
=
µ0 ω
6αx 2 y 4βz 3 6αx y 2
·µ ¶ µ ¶ ¸
= ²0 ωαx 2 y 3 + x̂ + ²0 ωβz 4 + − ŷ sin ωt
µ0 ω µ0 ω µ0 ω
1.6 Polarization of Materials

We are essentially finished with our analysis of Maxwell’s equations except for a
brief discussion of current density J and charge density ρ. The current density
can be decomposed into three general categories. First, as you might expect,
currents can arise from free charges in motion such as electrons in a metal. We
denote this type of current as Jfree . Second, individual atoms can exhibit internal
currents that give rise to paramagnetic and diamagnetic effects, denoted by Jm .
These are seldom important in optics problems, and so we will ignore these types
of currents. Third, molecules in a material can elongate and become dipoles in
response to an applied electric field. We denote this type of current, which arises
from the polarization of the medium, by Jp .
The polarization current Jp is associated with a dipole distribution function
P, called the polarization (in units of dipoles per volume, or charge times length
per volume). Physically, if the dipoles (depicted in Fig. 1.7) change their strength
or orientation as a function of time in some coordinated fashion, an effective
current density arises in the medium. Since the time-derivative of an individual
dipole moment renders charge multiplied by velocity, the time-derivative of a
distribution of ‘sloshing’ dipoles gives a current density equal to
Figure 1.7 A polarized medium ∂P
with ∇ · P = 0. Jp = (1.28)
∂t
We thus write the total current in an optical medium (ignoring magnetic effects)
as
∂P
J = Jfree + (1.29)
∂t
Now let’s turn our attention to charge density ρ. We seldom consider the
propagation of electromagnetic waveforms through electrically charged materials.
We therefore will write ρ free = 0. One might be tempted in this case to set the
overall charge density ρ to zero, but this would be wrong. The polarization of a
neutral material, described by P, can vary spatially, leading to local concentrations
of positive or negative charges.
We let ρ p denote the charge density created by variations in the polarization
P(r). To determine an expression for ρ p , we write the continuity equation (1.23)
as applied to the currents and charges associated with this polarization:
∂ρ p
∇ · Jp = − (1.30)
∂t
Substitution of (1.28) into this equation immediately yields
ρ p = −∇ · P (1.31)
To better appreciate local charge buildup due to variation in the medium

polarization, consider the divergence theorem (0.11) applied to P (r):
I Z
− P (r) · n̂ d a = − ∇ · P (r) d v (1.32)
S V
1.7 The Wave Equation 37
The left-hand side of (1.32) is a surface integral, which after integrating gives
units of charge. Physically, it is the sum of the charges touching the inside of
surface S (multiplied by a minus since by convention dipole vectors point from
the negatively charged end of a molecule to the positively charged end). When
∇ · P is zero, there are equal numbers of positive and negative charges touching
S from within, as depicted in Fig. 1.7. When ∇ · P is not zero, the positive and
negative charges touching S are not balanced, as depicted in Fig. 1.8. Essentially,
excess charge ends up within the volume because the non-uniform alignment of
dipoles causes them to be cut preferentially at the surface.11
Since we will ignore free charges (for optical media), we write the charge
density according to (1.31) as
ρ = −∇ · P (1.33)
In summary, in electrically neutral non-magnetic media, Maxwell’s equations
(in terms of the medium polarization P) are12
∇·P
∇·E = − (Gauss’s law) (1.34) Figure 1.8 A polarized medium
²0 with ∇ · P 6= 0.
∇·B = 0 (Gauss’s law for magnetism) (1.35)
∂B
∇×E = − (Faraday’s law) (1.36)
∂t
B ∂E ∂P
∇× = ²0 + + Jfree (Ampere’s law; fixed by Maxwell) (1.37)
µ0 ∂t ∂t
1.7 The Wave Equation

When Maxwell unified electromagnetic theory, he immediately noticed that waves
are solutions to this set of equations. In fact his desire to find a set of equations
that allowed for waves aided his effort to find the correct equations. After all, it
was already known that light traveled as waves. Kirchhoff had previously noticed
±p
that 1 ²0 µ0 gives the correct speed of light c = 3×108 m/s (which had previously
been measured). Faraday and Kerr had observed that strong magnetic and electric
fields affect light propagating in crystals. The time was right to suspect that light
was an electromagnetic phenomena at high frequency.
11 The figures may give you the impression that you could always just draw a surface that avoids
cutting any dipoles. However, the function P (r) is continuous, while the figures depict crudely just
a few dipoles. In a continuous material you can’t draw a surface that avoids cutting dipoles.
12 It is not uncommon to see the macroscopic Maxwell equations written in terms of two auxiliary
fields: H and D. The field H is useful in magnetic materials. In these materials, the combination
B µ0 in Ampere’s law is replaced by H ≡ B/µ0 − M, where Jm = ∇ × M is the current associated
±
with the material’s magnetization. Since we only consider nonmagnetic materials (M = 0), there
is little point in using H. The field D, called the displacement, is defined as D ≡ ²0 E + P. This
combination of E and P occurs in Coulomb’s law and Ampere’s law. For the purposes of this book,
it is conceptually more clear to retain the polarization P as a separate field in these two equations.
At first glance, Maxwell’s equations might not immediately suggest (to the
inexperienced eye) that waves are solutions. However, we can manipulate the
equations (first order differential equations that couple E to B) into the familiar
wave equation (decoupled second order differential equations for either E or B).
You should become familiar with this derivation. In what follows, we will derive
the wave equation for E. The derivation of the wave equation for B is very similar
(see problem P1.6).
Derivation of the Wave Equation
Taking the curl of (1.3) gives
∂
∇ × (∇ × E) + (∇ × B) = 0 (1.38)
∂t
We may eliminate ∇ × B by substitution from (1.4), which gives
∂2 E ∂J
∇ × (∇ × E) + µ0 ²0 = −µ0 (1.39)
∂t 2 ∂t
Next we apply the differential vector identity (0.10), ∇ × (∇ × E) = ∇ (∇ · E) − ∇2 E,

and use Gauss’ law (1.1) to replace the term ∇ · E, which brings us to
∂2 E ∂J ∇ρ
∇2 E − µ0 ²0 = µ0 + (1.40)
∂t 2 ∂t ²0
Substitution from (1.29) and (1.33) gives the more-useful-for-optics form
∂2 E ∂Jfree ∂2 P 1
∇2 E − µ0 ²0 = µ0 + µ0 − ∇ (∇ · P) (1.41)
∂t 2 ∂t ∂t 2 ²0
The left-hand side of (1.41) is the familiar wave equation. However, the right-
hand side contains a number of source terms, which arise when various currents
and/or polarizations are present. The first term on the right-hand side of (1.41)
describes currents of free charges, which are important for determining the reflec-
tion of light from a metallic surface or for determining the propagation of light in
a plasma. The second term on the right-hand side describes dipole oscillations,
which behave similar to currents. The final term on the right-hand side of (1.41)
is important in anisotropic media such as crystals. In this case, the polarization
P responds to the electric field along a direction not necessarily parallel to E,
due to the influence of the crystal lattice (addressed in chapter 5). In summary,
when light propagates in a material, at least one of the terms on the right-hand
side of (1.41) will be non zero. As an example, in glass, Jfree = 0 and ∇ · P = 0, but
∂2 P ∂t 2 6= 0 since the medium polarization responds to the light field, giving rise
±
to refractive index (discussed in chapter 2).
Example 1.6
1.7 The Wave Equation 39
Show that the electric field
E = (αx 2 y 3 x̂ + βz 4 ŷ) cos ωt
and the associated charge density (see Example 1.1)
ρ = 2²0 αx y 3 cos ωt
together with the associated current density (see Example 1.5)
6αx 2 y 4βz 3 6αx y 2

·µ ¶ µ ¶ ¸
J= ²0 ωαx 2 y 3 + x̂ + ²0 ωβz 4 + − ŷ sin ωt
µ0 ω µ0 ω µ0 ω
satisfy the wave equation (1.40).
Solution: We have
∂2 E £ ¡ 3
∇2 E − µ0 ²0 = α 2y + 6x 2 y x̂ + 12βz 2 ŷ cos ωt
¢ ¤
∂t 2
+ µ0 ²0 ω2 (αx 2 y 3 x̂ + βz 4 ŷ) cos ωt

= α 2y 3 + 6x 2 y + µ0 ²0 ω2 x 2 y 3 x̂ + β 12z 2 + µ0 ²0 ω2 z 4 ŷ cos ωt
£ ¡ ¢ ¡ ¢ ¤
Similarly,
∂J ∇ρ £¡
µ0 = µ0 ²0 ω2 αx 2 y 3 + 6αx 2 y x̂ + µ0 ²0 ω2 βz 4 + 12z 2 − 6αx y 2 ŷ cos ωt
¢ ¡ ¢ ¤
+
∂t ²0
+ 2αy 3 x̂ + 6αx y 2 ŷ cos ωt
£ ¤
= α µ0 ²0 ω2 x 2 y 3 + 6x 2 y + 2y 3 x̂ + µ0 ²0 ω2 βz 4 + 12z 2 ŷ cos ωt
£ ¡ ¢ ¡ ¢ ¤
The two expressions are equivalent, and the wave equation is satisfied.13
The magnetic field B satisfies a similar wave equation, decoupled from E (see
P1.6). However, the two waves are not independent. The fields for E and B must
be chosen to be consistent with each other through Maxwell’s equations. After
solving the wave equation (1.41) for E, one can obtain the consistent B from E via
Faraday’s law (1.36).
In vacuum all of the terms on the right-hand side in (1.41) are zero, in which
case the wave equation reduces to
∂2 E
∇2 E − µ0 ²0 =0 (vacuum) (1.42)
∂t 2
Solutions to this equation can take on every imaginable functional shape (speci-
fied at a given instant—the evolution thereafter being controlled by (1.42)). More-
over, since the differential equation is linear, any number of solutions can be
added together to create other valid solutions. Consider the subclass of solutions
13 The expressions in Example 1.6 hardly look like waves. The (quite unlikely) current and charge
distributions, which fill all space, would have to be artificially induced rather than arise naturally in
response to a field disturbance on a medium.
that propagate in a particular direction. These waveforms preserve shape while

traveling with speed
±p
c ≡1 ²0 µ0 = 2.9979 × 108 m/s (1.43)
In this case, E depends on the argument û·r−c t , where û is a unit vector specifying
the direction of propagation. The shape is preserved since features occurring at a
given position recur ‘downstream’ at a distance c t after a time t . By checking this
solution in (1.42), one confirms that the speed of propagation is c (see P1.8). As
mentioned previously, one may add together any combination of solutions (even
with differing directions of propagation) to form other valid solutions.
Exercises 41
Exercises
Exercises for 1.1 Gauss’ Law
P1.1 Consider an infinitely long hollow cylinder with inner radius a and
outer radius b as shown in Fig. 1.9. Assume that the cylinder has a
charge density ρ = k/s 2 for a < s < b and no charge elsewhere, where s
is the radial distance from the axis of the cylinder. Use Gauss’s Law in
integral form to find the electric field produced by this charge for each
of the three regions: s < a, a < s < b, and s > b.
a
HINT: For each region first draw an appropriate ‘Gaussian surface’ and
integrate the charge density over the volume to figure out the enclosed
b
charge. Then use Gauss’s law in integral form and the symmetry of the
problem to solve for the electric field. Figure 1.9 A charged cylinder with
charge located between a and b.
Exercises for 1.3 Faraday’s Law
Suppose that an electric field is given by E(r, t ) = E0 cos k · r − ωt + φ ,

¡ ¢
P1.2
where k⊥E0 and φ is a constant phase. Show that
k × E0
cos k · r − ωt + φ
¡ ¢
B(r, t ) =
ω
is consistent with (1.3).
Exercises for 1.4 Ampere’s Law
P1.3 A conducting cylinder with the same geometry as P1.1 carries a current
density J = k/s ẑ along the axis of the cylinder for a < s < b, where s is
the radial distance from the axis of the cylinder. Using Ampere’s Law in
integral form, find the magnetic field due to this current. Find the field
for each of the three regions: s < a, a < s < b, and s > b.
HINT: For each region first draw an appropriate ‘Amperian loop’ and
integrate the current density over the surface to figure out how much
current passes through the loop. Then use Ampere’s law in integral
form and the symmetry of the problem to solve for the magnetic field.
Exercises for 1.6 Polarization of Materials
P1.4 Memorize Maxwell’s equations (1.1)–(1.4) together with (1.29) and

(1.33). Be prepared to reproduce them from memory on an exam,
and write them on your homework from memory to indicate comple-
tion. Also very briefly summarize the physical principles described by
each of Maxwell’s equations, and the assumptions that go into writing

(1.29) and (1.33).
P1.5 Check that the E and B fields in P1.2, satisfy the rest of Maxwell’s equa-
tions (1.1), (1.2), and (1.4). What are the implications for J and ρ?
Exercises for 1.7 The Wave Equation
P1.6 Derive the wave equation for the magnetic field B in vacuum (i.e. J = 0
and ρ = 0).
P1.7 Show that the magnetic field in P1.2 is consistent with the wave equa-
tion derived in P1.6.
P1.8 Verify that E(û·r−c t ) satisfies the vacuum wave equation (1.42), where
E has an arbitrary functional form.
(a) Show that E (r, t ) = E0 cos k(û · r − c t ) + φ is a solution to the vac-

¡ ¢
P1.9
uum wave equation (1.42), where û is an arbitrary unit vector and k is
a constant with units of inverse length.
(b) Show that each wave front forms a plane, which is why such solu-
tions are often called ‘plane waves’. HINT: A wavefront is a surface in
space where the argument of the cosine (i.e. the phase of the wave) has
a constant value. Set the cosine argument to an arbitrary constant and
see what positions are associated with that phase.
(c) Determine the speed v = ∆r /∆t that a wave front moves in the û
direction. HINT: Set the cosine argument to a constant, solve for r, and
differentiate r with respect to t .
(d) By analysis, determine the wavelength λ in terms of k. HINT: Find
the distance between identical wave fronts by changing the cosine
argument by 2π at a given instant in time.
Screen D (e) Use (1.34) to show that E0 and û must be perpendicular to each
Laser other in vacuum.
A
L1.10 Measure the speed of light using a rotating mirror. Provide an estimate
B C of the experimental uncertainty in your answer (not the percentage
Rotating Delay Path
Mirror error from the known value). (video)
Figure 1.10 Geometry for lab 1.10. Figure 1.10 shows a simplified geometry for the optical path for light
in this experiment. Laser light from A reflects from a rotating mirror
at B towards C . The light returns to B , where the mirror has rotated,
sending the light to point D. Notice that a mirror rotation of θ deflects
the beam by 2θ.
Exercises 43
Retro-reflecting
Collimation Telescope
Rotating Long Corridor
mirror
Front of laser can
serve as screen
for returning light
Laser
Figure 1.11 A schematic of the setup for lab 1.10.
P1.11 Ole Roemer made the first successful measurement of the speed of light Ole Roemer (16441710, Danish) was
a man of many interests. In addition to
in 1676 by observing the orbital period of Io, a moon of Jupiter with a
measuring the speed of light, he created
period of 42.5 hours. When Earth is moving toward Jupiter, the period a temperature scale which with slight
modication became the Fahrenheit
is measured to be shorter than 42.5 hours because light indicating the
scale, introduced a system of standard
end of the moon’s orbit travels less distance than light indicating the weights and measures, and was heavily
beginning. When Earth is moving away from Jupiter, the situation is involved in civic aairs (city planning,
etc.). Scientists initially became inter-
reversed, and the period is measured to be longer than 42.5 hours. ested in Io's orbit because its eclipse
(when it went behind Jupiter) was an
(a) If you were to measure the time for 40 observed orbits of Io when
event that could be seen from many
Earth is moving directly toward Jupiter and then several months later places on earth. By comparing accurate
measurements of the local time when Io
measure the time for 40 observed orbits when Earth is moving directly
was eclipsed by Jupiter at two remote
away from Jupiter, what would you expect the difference between these places on earth, scientists in the 1600s
two measurements be? Take the Earth’s orbital radius to be 1.5×1011 m. were able to determine the longitude
dierence between the two places.
To simplify the geometry, just assume that Earth moves directly toward
or away from Jupiter over the entire 40 orbits (see Fig. 1.12).
Earth
(b) Roemer did the experiment described in part (a), and experimen-
Io
tally measured a 22 minute difference. What speed of light would one Sun
deduce from that value?
Jupiter
P1.12 In an isotropic medium (i.e. ∇ · P = 0), the polarization can often be Earth
written as function of the electric field: P = ²0 χ (E ) E, where χ (E ) =
Figure 1.12 Geometry for P1.11
χ1 + χ2 E + χ3 E 2 · · · . The higher order coefficients in the expansion (i.e.
χ2 , χ3 , ...) are typically small, so only the first term is important at low
intensities. The field of nonlinear optics deals with intense light-matter
interactions, where the higher order terms of the expansion become
important. This can lead to phenomena such as harmonic generation.
Starting with Maxwell’s equations, derive the wave equation for nonlin-
ear optics in an isotropic medium:
¢ ∂2 E ∂2 χ2 E + χ3 E 2 + · · · E ∂J
¡ ¢
2
∇ E − µ0 ²0 1 + χ1 = µ0 ²0 + µ0
¡
∂t 2 ∂t 2 ∂t
We retain the possibility of current here since, for example, in a gas
some of the molecules might ionize in the presence of a strong field,
giving rise to currents.
Chapter 2
Plane Waves and Refractive Index
Now we turn our focus to sinusoidal solutions of Maxwell’s equations, called

plane waves. Restricting our attention to plane waves may seem limiting at first,
since (as mentioned in chapter 1) any waveform shape that travels in a particular
direction can satisfy the wave equation in vacuum, as long as it moves at the
speed c and has the requisite connections between E and B. It turns out, however,
that an arbitrary waveform can always be constructed from a linear superposition
of sinusoidal waves. Thus, there is no loss of generality if we focus our attention
on plane-wave solutions.
In a material, the electric field of a plane wave induces oscillating dipoles,
and these oscillating dipoles in turn alter the electric field. We use the index of
refraction to describe this effect. Plane waves of different frequencies experience
different refractive indices, which causes them to travel at different speeds in
materials. Thus, an arbitrary waveform, which is composed of multiple sinusoidal
waves, invariably changes shape as it travels in a material, as the different sinu-
soidal waves change relationship with respect to one another. This dispersion
phenomenon is a primary reason why physicists and engineers choose to work
with sinusoidal waves. Every waveform except for individual sinusoidal waves
changes shape as it travels in a material.
When describing plane waves, it is convenient to employ complex numbers
to represent physical quantities. This is particularly true for problems involving
absorption, which takes place in metals and, to a lesser degree (usually), in di-
electrics (e.g. glass). When the electric field is represented using complex notation,
the index of refraction also becomes a complex number. You should make sure
you are comfortable with the material in section 0.2 before proceeding.
2.1 Plane Wave Solutions to the Wave Equation

Consider the wave equation for an electric field waveform propagating in vacuum
(1.42):
∂2 E
∇2 E − µ0 ²0 2 = 0 (2.1)
∂t
45
46 Chapter 2 Plane Waves and Refractive Index
We are interested in solutions to (2.1) that have the functional form (see P1.9)
E(r, t ) = E0 cos k · r − ωt + φ
¡ ¢
(2.2)
AM
Frequency (Hz)
Here φ represents an arbitrary (constant) phase term. The vector k, called the
Radio
wave vector, may be written as
FM 2π
k ≡ k û = û (vacuum) (2.3)
λvac
where k has units of inverse length, û is a unit vector defining the direction of
Microwave
Radar
propagation, and λvac is the length by which r must vary (in the direction of û) to
cause the cosine to go through a complete cycle. This distance is known as the
(vacuum) wavelength. The frequency of oscillation is related to the wavelength via
2πc
ω= (vacuum) (2.4)
λvac
Infrared
The frequency ω has units of radians per second. Frequency is also often ex-
pressed as ν ≡ ω/2π in units of inverse seconds or Hz. Notice that k and ω cannot
Visible
be chosen independently; the wave equation requires them to be related through
the dispersion relation
Ultraviolet ω
k= (vacuum) (2.5)
c
Typical values for λvac are given in Fig. 2.1. Sometimes the spatial period of the
wave is expressed as 1/λvac , in units of cm−1 , called the wave number.
A magnetic wave accompanies any electric wave, and it obeys a similar wave
X-rays equation (see P1.6). The magnetic wave corresponding to (2.2) is
B(r, t ) = B0 cos k · r − ωt + φ ,
¡ ¢
(2.6)
It is important to note that B0 , k, ω, and φ are not independently chosen in (2.6).

Wavelength (m)
Gamma Rays
In order to satisfy Faraday’s law (1.3), the arguments of the cosine in (2.2) and
(2.6) must be identical. Therefore, in vacuum the electric and magnetic fields
travel in phase. In addition, Faraday’s law requires (see P1.2)
k × E0
B0 = (2.7)
ω
The above cross product means that B0 , is perpendicular to both E0 and k. Mean-
Figure 2.1 The electromagnetic while, Gauss’ law ∇ · E = 0 forces k to be perpendicular to E0 . It follows that the
spectrum magnitudes of the fields are related through B 0 = kE 0 /ω or B 0 = E 0 /c, in view of
(2.5).
The influence of the magnetic field only becomes important (in comparison
to the electric field) for charged particles moving near the speed of light. This
typically takes place only for extremely intense lasers (> 1018 W/cm2 , see P2.12)
where the electric field is sufficiently strong to cause electrons to oscillate with
velocities near the speed of light. We will be interested in optics problems that take
2.1 Plane Wave Solutions to the Wave Equation 47
place at far less intensity where the effects of the magnetic field can typically be
safely ignored. Throughout the remainder of this book, we will focus our attention
mainly on the electric field with the understanding that we can at any time deduce
the (less important) magnetic field from the electric field via Faraday’s law.
Figure 2.2 depicts the electric field (2.2) and the associated magnetic field
(2.6) like transverse waves on a string. However, they are actually large planar
sheets of uniform field strengths (difficult to draw) that move in the direction of k.
The name plane wave is given since a constant argument in (2.2) at any moment
describes a plane, which is perpendicular to k. A plane wave fills all space and
may be thought of as a series of infinite sheets, each with a different uniform field
strength, moving in the k direction.
At this point, we rewrite our plane wave solution using complex number nota- Figure 2.2 Depiction of electric
tion. Although this change in notation will not make the task at hand any easier and magnetic fields associated
with a plane wave.
(and may even appear to complicate things), we introduce it here in preparation
for later sections, where it will save considerable labor. (For a review of complex
notation, see section 0.2.)
Using complex notation we rewrite (2.2) as
n o
E(r, t ) = Re Ẽ0 e i (k·r−ωt ) (2.8)
where we have hidden the phase term φ inside of Ẽ0 as follows:1
Ẽ0 ≡ E0 e i φ (2.9)
The next step we take is to become intentionally sloppy. Physicists throughout

the world have conspired to avoid writing Re { } in an effort (or lack thereof if
you prefer) to make expressions less cluttered. Nevertheless, only the real part of
the field is physically relevant even though expressions and calculations contain
both real and imaginary terms. This sloppy notation is okay since the real and
imaginary parts of complex numbers never intermingle when adding, subtracting,
differentiating, or integrating. We can delay taking the real part of the expression
until the end of the calculation. Also, when hiding a phase φ inside of the field
amplitude as in (2.8), we drop the tilde (might as well since we are already being
sloppy); we will automatically assume that the field amplitude is complex and
contains phase information. Putting this all together, our plane wave solution in
complex notation is written simply as
E(r, t ) = E0 e i (k·r−ωt ) (2.10)
It is possible to construct any electromagnetic disturbance from a linear superpo-

sition of such waves, which we will do in chapter 7.
1 We have assumed that each vector component of the field propagates with the same phase. To
be more general, one could write Ẽ0 ≡ x̂E 0x e i φx + ŷE 0y e i φy + ẑE 0z e i φz .

Example 2.1
Verify that the complex plane wave (2.10) is a solution to the wave equation (2.1).
Solution: The first term gives
∂2 ∂2 ∂2
· ¸
∇2 E0 e i (k·r−ωt ) = E0 + + e i (k x x+k y y+k z z−ωt )
∂x 2 ∂y 2 ∂z 2
(2.11)
³ ´
= −E0 k x2 + k y2 + k z2 e i (k·r−ωt )
= −k 2 E0 e i (k·r−ωt )
and the second term gives
1 ∂2 ³ i (k·r−ωt )
´ ω2
E 0 e = − E0 e i (k·r−ωt ) (2.12)
c 2 ∂t 2 c2
Upon insertion into (2.1) we obtain the vacuum dispersion relation (2.5), which
specifies the connection between the wavenumber k and the frequency ω, empha-
sizing that k and ω cannot be chosen independently.
2.2 Index of Refraction

Now let’s examine how plane waves behave in dielectric media (e.g. glass). We
assume an isotropic,2 homogeneous,3 and non-conducting medium (i.e. Jfree = 0).
In this case, we expect E and P to be parallel to each other so ∇·P = 0 from (1.34).4
The general wave equation (1.41) for the electric field reduces in this case to
∂2 E ∂2 P
∇2 E − ²0 µ0 = µ0 2 (2.13)
∂t 2 ∂t
Since we are considering sinusoidal waves, we consider solutions of the form
E = E0 e i (k·r−ωt )
(2.14)
P = P0 e i (k·r−ωt )
By writing this, we are making the (reasonable) assumption that if an electric

field stimulates a medium at frequency ω, then the polarization in the medium
also oscillates at frequency ω. This assumption is typically rather good except
for extreme electric fields, which can generate frequency harmonics through
nonlinear effects (see P1.12). Recall that by our prior agreement, the complex
amplitudes of E0 and P0 carry phase information. Thus, while E and P in (2.14)
oscillate at the same frequency, they can be out of phase with respect to each
2 Isotropic means the material behaves the same for propagation in any direction. Many crystals
are not isotropic as we’ll see in Chapter 5.

3 Homogeneous means the material is everywhere the same throughout space.
4 This follows for a wave of the form (2.14) if P and k are perpendicular.
2.2 Index of Refraction 49
other. This phase discrepancy is most pronounced for materials that absorb
energy at the plane wave frequency.
Substitution of the trial solutions (2.14) into (2.13) yields
− k 2 E0 e i (k·r−ωt ) + ²0 µ0 ω2 E0 e i (k·r−ωt ) = −µ0 ω2 P0 e i (k·r−ωt ) (2.15)
To go further, we need to make an explicit connection between E0 and P0 (exter-

nal to Maxwell’s equations). In a linear medium, the polarization amplitude is
proportional to the strength of the applied electric field:
P0 (ω) = ²0 χ (ω) E0 (ω) (2.16)
This is known as a constitutive relation. We have introduced a dimensionless pro-

portionality factor χ(ω) called the susceptibility, which depends on the frequency
of the field. We account for the possibility that E and P oscillate out of phase by
allowing χ(ω) to be a complex number.
By inserting (2.16) into (2.15) and canceling the field terms, we obtain the
dispersion relation in dielectrics:
ωp
k 2 = ²0 µ0 1 + χ (ω) ω2 1 + χ (ω)
£ ¤
or k= (2.17)
c
p
where we have used c ≡ 1/ ²0 µ0 . In general, χ(ω) is a complex number, which
leads to a complex index of refraction, defined by5
N (ω) ≡ n(ω) + i κ(ω) = 1 + χ(ω)

p
(2.18)
where n and κ are respectively the real and imaginary parts of the index. (Note
that κ is not k.) According to (2.17), the magnitude of the wave vector is also
complex according to
N ω (n + i κ) ω
k= = (2.19)
c c
The use of complex index of refraction only makes sense in the context of complex
representation of plane waves.
The complex index N takes into account absorption as well as the usual
oscillatory behavior of the wave. We see this by explicitly placing (2.19) into
(2.14):
κω
¡ nω ¢
E(r, t ) = E0 e −Im{k}·r e i (Re{k}·r−ωt ) = E0 e − c û·r e i c û·r−ωt (2.20)
As before, û is a real unit vector specifying the direction of k. Again, when looking
at (2.20), by special agreement in advance, we should just think of the real part,
namely6 ³ nω
κω
´
E(r, t ) = E0 e − c û·r cos û · r − ωt + φ (2.21)
c
5 Electrodynamics books often use the electric displacement D ≡ ² E + P = ²E. See M. Born and
0
E. Wolf, Principles of Optics, 7th ed., p. 3 (Cambridge University Press, 1999). The permittivity
² encapsulates the constitutive relation that connects P with E. In a linear medium we have
p
² ≡ ²0 (1 + χ), so that the index of refraction is given by N = ²/²0 .
6 For the sake of simplicity in writing (2.21) we assume linearly polarized light. That is, all vector
components of E0 have the same complex phase φ. We will consider other possibilities, such as
circularly polarized light, in chapter 6.
where an overall phase φ was formerly held in the complex vector Ẽ0 . (The tilde
had been suppressed.) Figure 2.3 shows a graph of (2.21). The imaginary part of
the index κ causes the wave to decay as it travels. The real part of the index n is
associated with the oscillations of the wave. By inspection of the cosine argument
in (eq:2.3.20), we see that the speed of the (diminishing) sinusoidal wave fronts is
v phase (ω) = c /n(ω) (2.22)
0 It is apparent that n(ω) is the ratio of the speed of the light in vacuum to the speed
of the wave in the material.
In a dielectric, the vacuum relations (2.3) and (2.4) are modified to read
2π
0 10 20 Re {k} ≡ û, (2.23)
λ
where
Figure 2.3 Electric field of a decay-
λ ≡ λvac /n. (2.24)
ing plane wave. For convenience
in plotting, the direction of prop- While the frequency ω is the same, whether in a material or in vacuum, the
agation is chosen to be in the z
wavelength λ varies with the real part of the index n.
direction (i.e. û = ẑ).
Example 2.2
When n = 1.5, κ = 0.1, and ν = 5 × 1014 Hz, find (a) the wavelength inside the
material, and (b) the propagation distance over which the amplitude of the wave
diminishes by the factor e −1 (called the skin depth).
Solution: (a)
λvac 2πc 3 × 108 m/s

¡ ¢
c
λ= = = = ¡ ¢ = 400 nm
n nω nν 1.5 5 × 1014 Hz
(b)
κω c c 3 × 108 m/s
e− c z = e −1 ⇒ z= = = ¢ = 950 nm
κω 2πκν 2π (0.1) 5 × 1014 Hz
¡
Obtaining n and κ from the complex susceptibility χ
From (2.18) we have
(n + i κ)2 = n 2 − κ2 + i 2nκ = 1 + Re χ + i Im χ = 1 + χ
© ª © ª
(2.25)
The real parts and the imaginary parts in the above equation are separately equal:
n 2 − κ2 = 1 + Re χ and 2nκ = Im χ
© ª © ª
(2.26)
2.3 The Lorentz Model of Dielectrics 51
From the latter equation we have
κ = Im χ /2n
© ª
(2.27)
When this is substituted into the first equation of (2.26) we get a quadratic in n 2
¡ © ª¢2
4
© ª¢ 2 Im χ
n − 1 + Re χ n −
¡
=0 (2.28)
4
The positive7 real root to this equation is
v
u¡ © ª¢ q¡ © ª¢2 ¡ © ª¢2
t 1 + Re χ + 1 + Re χ + Im χ
u
n= (2.29)
2
The imaginary part of the index is then obtained from (2.27).
When absorption is small we can neglect the imaginary part of χ(ω), and
(2.29) reduces to
n (ω) = 1 + χ (ω)
p
(negligible absorption) (2.30)
2.3 The Lorentz Model of Dielectrics

To compute the index of refraction in either a dielectric or a conducting material,
we require a model that describes the response of electrons in the material to
the passing electric field wave. Of course, the model in turn influences how the
electric field propagates, which is what influences the material in the first place!
The model therefore must be solved together with the propagating field in a
self-consistent manner.
Hendrik Lorentz developed a very successful model in the late 1800s, which Hendrik Antoon Lorentz (18531928,
treats each (active) electron in the medium as a classical particle obeying Newton’s Dutch) was born in Arnhem, Nether-
lands, the son a successful nurseryman.
second law (F = ma). In the case of a dielectric medium, electrons are subject to Hendrick's mother died when he was
an elastic restoring force that keeps each electron bound to its respective atom four years old. He studied classical lan-
guages and then entered the University
and a damping force that dissipates energy and gives rise to absorption.
of Leiden where he was strongly inu-
The Lorentz model determines the susceptibility χ (ω) (the connection be- enced by astronomy professor Frederik
tween the electric field E0 and the polarization P0 ) and hence the index of re- Kaiser, whose niece Hendrik married.
Hendrik was persuaded to become a
fraction. The model assumes that all atoms (or molecules) in the medium are physicist and wrote a doctoral disserta-
identical, each with one (or a few) active electrons responding to the external tion entitled On the theory of reection
and refraction of light, in which he
field. The atoms are uniformly distributed throughout space with N identical rened Maxwell's electromagnetic the-
active electrons per volume (units of number per volume). The polarization of ory. Lorentz correctly hypothesized that
the atoms were composed of charged
the material is then
particles, and that their movement was
P = N q e rmicro (2.31) the source of light. He also derived the
transformations of space and time, later
Recall that polarization has units of dipoles per volume. Each dipole has strength used in Einstein's theory of relativity.
q e rmicro , where rmicro is a microscopic displacement of the electron from equilib- Lorentz won the Nobel prize in 1902
for his contributions to electromagnetic
rium. theory. (Wikipedia)
7 It is possible to have n < 0 for so called meta materials, not considered here.
At the time of Lorentz, atoms were thought to be clouds of positive charge

wherein point-like electrons sat at rest unless stimulated by an applied electric
field. In our modern quantum-mechanical viewpoint, rmicro corresponds to an
average displacement of the electronic cloud, which surrounds the nucleus (see
Fig. 2.4). The displacement rmicro of the electron charge in an individual atom
depends on the local strength of the applied electric field E at the position of the
atom. Since the diameter of the electronic cloud is tiny compared to a wavelength
of (visible) light, we may consider the electric field to be uniform across any
individual atom.
Unperturbed
The Lorentz model uses Newton’s equation of motion to describe an electron
displacement from equilibrium within an atom. In accordance with the classical
laws of motion, the electron mass m e times its acceleration is equal to the sum of
+ the forces on the electron:
m e r̈micro = q e E − m e γṙmicro − k Hooke rmicro (2.32)
The electric field pulls on the electron with force q e E.8 A drag force (or friction)
In an electric field −m e γṙmicro opposes the electron motion and accounts for absorption of energy.
Without this term, it is only possible to describe optical index at frequencies away
from where absorption takes place. Finally, −k Hooke rmicro is a force accounting
- for the fact that the electron is bound to the nucleus. This restoring force can be
+
thought of as an effective spring that pulls the displaced electron back towards
equilibrium with a force proportional to the amount of displacement, so this
term is essentially the familiar Hooke’s law. With some rearranging, (2.32) can be
written as
qe
r̈micro + γṙmicro + ω20 rmicro = E (2.33)
Figure 2.4 A distorted electronic me
cloud becomes a dipole.
where ω0 ≡ k Hooke /m e is the natural oscillation frequency (or resonant fre-
p
quency) associated with the electron mass and the ‘spring constant.’
There is a subtle problem with our analysis, which we will continue to neglect
in this section, but which should be mentioned. The field E in (2.32) is the net
field, which is influenced by the presence of all of the dipoles. The actual field that
a dipole ‘feels’, however, does not include its own field. That is, we should remove
from E the field produced by each dipole in its own vicinity. This significantly
modifies the result if the density of the material is sufficiently high. This effect is
described by the Clausius-Mossotti formula, which is treated in appendix 2.B.
In accordance with our examination of a single sinusoidal wave, we insert
(2.14) into (2.33) and obtain
qe
r̈micro + γṙmicro + ω20 rmicro = E0 e i (k·r−ωt ) (2.34)
me
Note that within a given atom the excursions of rmicro are so small that k·r remains
essentially constant, since k·r varies with displacements on the scale of an optical
8 The electron also experiences a force due to the magnetic field of the light, F = q v
e micro × B,
but this force is tiny for typical optical fields.
2.3 The Lorentz Model of Dielectrics 53
wavelength, which is huge compared to the size of an atom. The inhomogeneous

solution to (2.34) is (see P2.1)
qe E0 e i (k·r−ωt )
µ ¶
rmicro = (2.35)
m e ω20 − i ωγ − ω2
The electron position rmicro oscillates (not surprisingly) with the same frequency
ω as the driving electric field. This solution illustrates the convenience of com-
plex notation. The imaginary part in the denominator implies that the electron
oscillates with a phase different from the electric field oscillations; the damping
term γ (the imaginary part in the denominator) causes the two to be out of phase
somewhat. The complex algebra in (2.35) accomplishes quite easily what would
otherwise be cumbersome (i.e. working out a trigonometric phase).
We are now able to write the polarization in terms of the electric field. By
substituting (2.35) into (2.31) and rearranging, we obtain
ω2p
Ã !
P = ²0 E0 e i (k·r−ωt ) (2.36)
ω20 − i ωγ − ω2
where the plasma frequency ωp has been introduced:9

s 1
N q e2
ωp ≡ (2.37)
²0 m e
A comparison of (2.36) with (2.16) reveals the (complex) susceptibility:
ω2p
χ (ω) = (2.38) 0
-10 -5 0 5 10
ω20 − i ωγ − ω2
The index of refraction is then found by substituting the susceptibility (2.38) into
Figure 2.5 Real and imaginary
(2.18). The real and imaginary parts of the index are solved by equating separately parts of the index for a single
the real and imaginary parts of (2.18), namely Lorentz oscillator dielectric with
ωp = 10γ.
ω2p
(n + i κ)2 = 1 + χ (ω) = 1 + (2.39)
ω20 − i ωγ − ω2
A graph of n and κ is given in Fig. 2.5.
Most materials actually have more than one species of active electron, and
different active electrons behave differently. The generalization of (2.39) in this
case is
f j ω2p j
2
(n + i κ) = 1 + χ (ω) = 1 +
X
(2.40)
j ω0 j − i ωγ j − ω
2 2
where f j is the aptly named oscillator strength for the j th species of active electron.
Each species also has its own plasma frequency ωp j , natural frequency ω0 j , and
damping coefficient γ j .
9 In a plasma, charges move freely so that both the Hooke restoring force and the dragging term
can be neglected (i.e. ω0 ∼

= 0, γ ∼
= 0). For a plasma, ωp is the dominant parameter.
Lorentz introduced this model well before the development of quantum

mechanics. Even though the model pays no attention to quantum physics, it
works surprisingly well for describing frequency-dependent optical indices and
absorption of light. As it turns out, the Schrödinger equation applied to two levels
in an atom reduces in mathematical form to the Lorentz model in the limit of
low-intensity light. Quantum mechanics also explains the oscillator strength,
which before the development of quantum mechanics had to be inserted ad hoc
to make the model agree with experiments. The friction term γ turns out not to be
associated with something internal to atoms but rather with collisions between
atoms, which on average give rise to the same behavior.
2.4 Index of Refraction of a Conductor

In a conducting medium, the outer electrons of atoms are free to move without
being tethered to any particular atom. However, the electrons are still subject to a
damping force due to collisions that remove energy and give rise to absorption.
Such collisions are associated with resistance in a conductor. As it turns out,
we can obtain a simple formula for the refractive index of a conductor from the
Lorentz model in section 2.3. We simply remove the restoring force that binds
electrons to their atoms. That is, we set ω0 = 0 in (2.39), which gives
1
ω2p
(n + i κ)2 = 1 − (2.41)
i ωγ + ω2
This underscores the fact that ∂P/∂t is a current very much like Jfree . When
we remove the restoring force k Hooke = m e ω20 from the atomic model, the elec-
0
trons effectively become free, and it is not surprising that they exactly mimic the
-20 0 20 40 behavior of a free current Jfree . A graph of n and κ in the conductor model is given
in Fig. 2.6. Below, we provide the derivation for (2.41) in the context of Jfree rather
Figure 2.6 Real and imaginary than as a limiting case of the dielectric model.10
parts of the index for conductor
with ωp = 50γ.
Derivation of Refractive Index for a Conductor
We will include the current density Jfree while setting the medium polarization P
to zero. The wave equation is
∂2 ∂
∇2 E − ²0 µ0 E = µ0 Jfree (2.42)
∂t 2 ∂t
We assume that the current is made up of individual electrons traveling with
velocity vmicro :
Jfree = N q e vmicro (2.43)
As before, N is the number density of free electrons (in units of number per vol-
ume). Recall that current density Jfree has units of charge times velocity per volume
10 G. Burns, Solid State Physics, Sect. 9-5 (Orlando: Academic Press, 1985).
2.5 Poynting’s Theorem 55
(or current per cross sectional area), so (2.43) may be thought of as a definition of
current density in a fundamental sense.
Again, the electrons satisfy Newton’s equation of motion, similar to (2.32) except
without a restoring force:
m e r̈micro = q e E − m e γṙmicro (2.44)
For a sinusoidal electric field E = E0 e i (k·r−ωt ) , the solution to this equation is
q e E0 e i (k·r−ωt )
µ ¶
vmicro ≡ ṙmicro = (2.45)
me γ−iω
where again we assume that the electron oscillation excursions described by rmicro
are small compared to the wavelength so that r can be treated as a constant in
(2.44). The current density (2.43) in terms of the electric field is then
N q e2 E0 e i (k·r−ωt )
µ ¶
Jfree = (2.46)
me γ−iω
We substitute this together with the electric field into the wave equation (2.42) and
get
ω2 µ0 N q e2 E0 e i (k·r−ωt )
µ ¶
2 i (k·r−ωt ) i (k·r−ωt )
− k E0 e + 2 E0 e = −i ω (2.47)
c me γ−iω
This simplifies down to the dispersion relation
ω2p
Ã !
2 ω2
k = 2 1− (2.48)
c i γω + ω2
which agrees with (2.41). We have made the substitution ω2p = N q e2 /²0 m e in accor-
ω2 1+χ
dance with (2.37). As usual, k 2 = ( ) = ω (n+i κ) , so the susceptibility and the
2 2
c2 c2
index may be extracted from (2.48).
Note that in the low-frequency limit (i.e. ω ¿ γ), the current density (2.46)
+ +
reduces to Ohm’s law J = σE, where σ = N q e2 /m e γ is the DC conductivity. In
the high-frequency limit (i.e. ω À γ), the behavior changes over to that of a
+
free plasma, where collisions, which are responsible for resistance, become less
important since the excursions of the electrons during oscillations become very
+
small. This formula captures the general behavior of metals, but actual values of
the index vary from this somewhat (see P2.6 ). +
+
In either the conductor or dielectric model, the damping term removes energy
from electron oscillations. The damping term gives rise to an imaginary part
of the index, which causes an exponential attenuation of the plane wave as it
propagates. Figure 2.7 The electrons in a
conductor can easily move in
response to the applied field.
2.5 Poynting’s Theorem
Until now, we have described light as the propagation of an electromagnetic
disturbance. However, we typically observe light by detecting absorbed energy
rather than the field amplitude directly. In this section we examine the connection
between propagating electromagnetic fields (such as the plane waves discussed
in this chapter) and the energy transported by such fields.
In the late 1800s John Poynting developed (from Maxwell’s equations) the the-
oretical foundation that describes light energy transport. You should appreciate
and remember the ideas involved, especially the definition and meaning of the
Poynting vector, even if you forget the specifics of its derivation.
Derivation of Poynting’s Theorem
We require just two of Maxwell’s Equations: (1.3) and (1.4). We take the dot product
of B/µ0 with the first equation and the dot product of E with the second equation.
Then by subtracting the second equation from the first we obtain
∂E B ∂B
µ ¶
B B
· (∇ × E) − E · ∇ × + ²0 E · + · = −E · J (2.49)
µ0 µ0 ∂t µ0 ∂t
The first two terms can be simplified using the vector identity P0.8. The next two
John Henry Poynting (18521914, terms are the time derivatives of ²0 E 2 /2 and B 2 /2µ0 , respectively. The relation
English) was the youngest son of a Uni-
tarian minister who operated a school
(2.49) then becomes
near Manchester England where John
∂ ²0 E 2 B 2
µ ¶ µ ¶
received his childhood education. He B
∇· E× + + = −E · J (2.50)
later attended Owen's College in Manch- µ0 ∂t 2 2µ0
ester and then went on to Cambridge
University where he distinguished him- This is Poynting’s theorem. Each term in this equation has units of power per
self in mathematics and worked under
James Maxwell in the Cavendish Lab-
volume.
oratory. Poynting joined the faculty of
the University of Birmingham (then
called Mason Science College) where It is conventional to write Poynting’s theorem as follows:11
he was a professor of physics from 1880
until his death. Besides developing his
famous theorem on the conservation
∂
∇·S+ (u field + u medium ) = 0 (2.51)
of energy in electromagnetic elds, he ∂t
performed innovative measurements of
Newton's gravitational constant and where
discovered that the Sun's radiation B
draws in small particles towards it, the S ≡ E× (2.52)
Poynting-Robertson eect. Poynting
µ0
was the principal author of a multi-
volume undergraduate physics textbook,
is called the Poynting vector, which has units of power per area, called irradiance.
which was in wide use until the 1930s. The expression
(Wikipedia) ²0 E 2 B 2
u field ≡ + (2.53)
2 2µ0
is the energy per volume stored in the electric and magnetic fields. Derivations of
the electric field energy density and the magnetic field energy density are given in
Appendices 2.C and 2.D. (See (2.79) and (2.86).) The derivative
∂u medium
≡ E·J (2.54)
∂t
11 See D. J. Griffiths, Introduction to Electrodynamics, 3rd ed., Sect. 8.1.2 (New Jersey: Prentice-Hall,
1999).
2.5 Poynting’s Theorem 57
describes the power per volume delivered to the medium from the field. Equa-
tion (2.54) is reminiscent of the familiar circuit power law, Power = Voltage ×
Current. Power is delivered when a charged particle traverses a distance while
experiencing a force. This happens when currents flow in the presence of electric
fields.
Poynting’s theorem is essentially a statement of the conservation of energy,
where S describes the flow of energy. To appreciate this, consider Poynting’s
theorem (2.51) integrated over a volume V (enclosed by surface S). If we also
apply the divergence theorem (0.11) to the term involving ∇ · S we obtain
∂
I Z
S · n̂ d a = − (u field + u medium ) d v (2.55)
∂t
S V
Notice that the volume integral over energy densities u field and u medium gives
the total energy stored in V , whether in the form of electromagnetic field energy
density or as energy density that has been given to the medium. The integration
of the Poynting vector over the surface gives the net Poynting vector flux directed
outward. Equation (2.55) indicates that the outward Poynting vector flux matches
the rate that total energy disappears from the interior of V . Conversely, if the
Poynting vector is directed inward (negative), then the net inward flux matches
the rate that energy increases within V . The vector S defines the flow of energy
through space. Its units of power per area are just what is needed to describe the
brightness of light impinging on a surface.
Example 2.3
(a) Find the Poynting vector S and energy density u field for the plane wave field E =
x̂E 0 cos (kz − ωt ) traveling in vacuum. (b) Check that S and u field satisfy Poynting’s
theorem.
Solution: The associated magnetic field is (see P1.2)
ẑk × x̂E 0 kE 0
B= cos (kz − ωt ) = ŷ cos (kz − ωt )
ω ω
(a) The Poynting vector is
E×B kE 0
S= = x̂E 0 cos (kz − ωt ) × ŷ cos (kz − ωt )
µ0 ωµ0
= ẑc²0 E 02 cos2 (kz − ωt )
where we have used ω = kc and µ0 = 1/(c 2 ²0 ). The energy density is
²0 E 2 B 2 ²0 E 02 kE 02
u field = + = cos2 (kz − ωt ) + cos2 (kz − ωt )
2 2µ0 2 2µ0 ω2
= ²0 E 02 cos2 (kz − ωt )
Notice that S = cu. The energy density traveling at speed c gives rise to the power
per area passing a surface (perpendicular to z).
(b) We have
∂
∇ · S = c²0 E 02 cos2 (kz − ωt ) = −2kc²0 E 02 cos (kz − ωt ) sin (kz − ωt )
∂z
whereas
∂u field ∂
= ²0 E 02 cos2 (kz − ωt ) = 2ω²0 E 02 cos (kz − ωt ) sin (kz − ωt )
∂t ∂t
Poynting’s theorem (2.50) is satisfied since ω = kc.
It is common to replace the rapidly oscillating function cos2 (kz − ωt ) with its time
average 1/2, but this would have inhibited our ability to take the above derivatives.
2.6 Irradiance of a Plane Wave

Consider the electric plane-wave field E(r, t ) = E0 e i (k·r−ωt ) . The magnetic field
that accompanies this electric field can be found from Maxwell’s equation (1.3),
and it turns out to be (compare with problem P1.2)
k × E0 i (k·r−ωt )
B(r, t ) = e (2.56)
ω
When k is complex, B is out of phase with E, and this occurs when absorption
takes place. When there is no absorption, then k is real, and B and E carry the
same complex phase.
Before computing the Poynting vector (2.52), which involves multiplication,
we must remember our unspoken agreement that only the real parts of the fields
are relevant. We necessarily remove the imaginary parts before multiplying (see
(0.22)). To obtain the real parts of the fields, we add their respective complex
conjugates and divide the result by 2 (see (0.30)). The real field associated with
the plane-wave electric field is
1h ∗
i
E(r, t ) = E0 e i (k·r−ωt ) + E∗0 e −i (k ·r−ωt ) (2.57)
2
and the real field associated with (2.56) is
1 k × E0 i (k·r−ωt ) k∗ × E∗0 −i (k∗ ·r−ωt )

· ¸
B(r, t ) = e + e (2.58)
2 ω ω
We have merely exercised our previous (conspiratorial) agreement that only the
real parts of (2.39) and (2.56) are to be retained.
Now we are ready to calculate the Poynting vector. The algebra is a little messy
in general, so we restrict the analysis to the case of an isotropic medium for the
sake of simplicity.
2.6 Irradiance of a Plane Wave 59
Calculation of the Poynting Vector for a Plane Wave
Using (2.57) and (2.56) in (2.52) gives
B
S ≡ E×
µ0
1 k × E0 i (k·r−ωt ) k∗ × E∗0 −i (k∗ ·r−ωt )
· ¸
1h ∗
i
= E0 e i (k·r−ωt ) + E∗0 e −i (k ·r−ωt ) × e + e
2 2µ0 ω ω
E∗ ×(k×E ) ∗
" #
E0 ×(k×E0 ) 2i (k·r−ωt )
+ 0 ω ¡ e i (k−k )·r
0
1 ω¡ e ¢
= E0 × k∗ ×E∗ ∗ ·r E ∗ × k∗ ×E∗
¢
∗
4µ0 + ω
0
e (
i k−k ) + 0
ω
0
e (k ·r−ωt )
−2i
1 k k ∗
· ¸
2i (k·r−ωt ) −2 κω û·r
= E0 × (û × E0 ) e + E0 × (û × E0 ) e c + C.C.
4µ0 ω ω
(2.59)
The letters ‘C.C.’ stand for the complex conjugate of what precedes in the square
brackets. The direction of k is specified with the real unit vector û. We have also
used (2.19) to rewrite i (k − k∗ ) as −2 (κω/c) û.
The assumption of an isotropic medium (not a crystal) means that ∇ · E(r, t ) = 0
and therefore û · E0 = 0. We can use this fact together with the BAC-CAB rule P0.3
to reduce the above expression to
û k k¡
· ¸
2i (k·r−ωt ) ∗ −2 κω û·r
¢
S= (E0 · E0 ) e + E0 · E 0 e c + C.C. (2.60)
4µ0 ω ω
The final expression shows that (in an isotropic medium) the flow of energy is in
the direction of û (or k). This agrees with our intuition that energy flows in the
direction that the wave propagates.
Very often, we are interested in the time-average of the Poynting vector, de-
noted by 〈S〉t . There are no electronics that can keep up with the rapid oscillation
of visible light (i.e. > 1014 Hz). Therefore, what is always measured is the time-
averaged absorption of energy. Under time averaging, the first term in (2.60)
vanishes since it rapidly oscillates positive and negative. The time-averaged
Poynting vector (including the term C.C.) becomes
û k + k ∗ ¡ κω
E0 · E∗0 e −2 c û·r
¢
〈S〉t =
4µ0 ω (2.61)
n²0 c ³ ¯2 ´ κω
|E 0 x |2 + ¯E 0 y ¯ + |E 0 z |2 e −2 c û·r
¯
= û
2
We have used (2.19) to rewrite k + k ∗ as 2 (nω/c). We have also used (1.43) to
rewrite 1/µ0 c as ²0 c.
The expression (2.61) is formally called the irradiance (with the direction û
included). However, we often speak of the intensity of a field I , which amounts to
the same thing, but without regard for the direction û. The definition of intensity
is thus less specific, and it can be applied, for example, to standing waves where
the net irradiance is technically zero (i.e. counter-propagating plane waves with
zero net energy flow). Nevertheless, atoms in standing waves ‘feel’ the oscillating
field. In general, the intensity is written as
Radiant Power (of a source): Elec-
n²0 c n²0 c ³ ¯2 ´
tromagnetic energy. Units: W = J/s E0 · E∗0 = |E 0 x |2 + ¯E 0 y ¯ + |E 0 z |2
¯
I= (2.62)
2 2
Radiant Solid-Angle Intensity
¯2 ignored absorption (i.e. κ ≈ 0). Alternatively, we

(of a source): Radiant power per where in this case we ¯have
steradian emitted from a point-
could consider |E 0 x |2 , ¯E 0 y ¯ , and |E 0 z |2 to include the factor exp(−2(κω/c)û · r)
like source (4π steradians in a
sphere). Units: W/Sr so that they correspond to the local electric field. Equation (2.62) agrees with S in
Example 2.3 where n = 1 and E0 = x̂E 0 is real; the cosine squared averages to 1/2.
Radiance or Brightness (of a
source): Radiant solid-angle in-
tensity per unit projected area of
an extended source. The projected
Appendix 2.A Radiometry, Photometry, and Color
area foreshortens by cos θ, where
θ is the observation angle rela- The field of study that quantifies the energy in electromagnetic radiation (in-
tive to the surface normal. Units: cluding visible light) is referred to as radiometry. Table 2.1 lists several concepts
W/(Sr · cm2 ) important in radiometry. The radiance at a detector and the exitance from a
Radiant Emittance or Exitance source are both direct measurements of the average Poynting flux, and the other
(from a source): Radiant Power quantities in the table are directly related to the Poynting flux through geometric
emitted per unit surface area of an
factors. One of the challenges in radiometry is that light sensors always have a
extended source (the Poynting flux
leaving). Units: W/cm2 wavelength-dependent sensitivity to light, whereas the quantities in Table 2.1
treat light of all wavelengths on equal footing. Disentangling the detector re-
Irradiance (to a receiver) Often
called intensity: Electromagnetic sponse from the desired signal in a radiometric measurement takes considerable
power delivered per area to a care.
receiver: Poynting flux arriving. Photometry refers to the characterization of light energy in the context of the
Units: W/cm2 response of the human eye. In contrast to radiometry, photometry takes great care
to mimic the wavelength-dependent effects of the eye-brain detection system so
Table 2.1 Radiometric quantities that photometric quantities are an accurate reflection of our everyday experience
and units. with light. The concepts used in photometry are similar to those in radiometry,
except that the radiometric quantities are multiplied by the spectral response of
our eye-brain system.
2000
Scotoptic Our eyes contain two types of photoreceptors—rods and cones. The rods
1700 W/lm @507 nm
1500 are very sensitive and provide virtually all of our vision in dim light conditions
Photoptic (e.g. when you are away from artificial light at night). Under these conditions we
1000 683 W/lm @555 nm
experience scotoptic vision, with a response curve that peaks at λvac = 507 nm
500 and is insensitive to wavelengths longer than 640 nm12 (see Fig. 2.8). As the
light gets brighter the less-sensitive cones take over, and we experience photoptic
0
400 500 600 700 800 vision, with a response curve that peaks at λvac = 555 nm and drops to near zero
wavelength (nm) for wavelengths longer than λvac = 700 nm or shorter than λvac = 400 nm (see
Fig. 2.8). Photometric quantities are usually measured using the bright-light
Figure 2.8 The response of a “stan-
(photoptic) response curve since that is what we typically experience in normally
dard” human eye under relatively
bright conditions (photoptic) and lit spaces.
in dim conditions (scotoptic). 12 Since rods do not detect the longer red wavelengths, it is possible to have artificial red illumina-
tion without ruining your dark-adapted vision. For example, an airplane can have red illumination
on the instrument panel without interfering with a pilot’s ability to achieve full dark-adapted vision
to see things outside the cockpit.
2.A Radiometry, Photometry, and Color 61
Photometric units, which may seem a little obscure, were first defined in terms
of an actual candle with prescribed dimensions made from whale tallow. The
basic unit of luminous power is called the lumen, defined to be (1/683) W of light Luminous Power (of a source):
Visible light energy emitted per
with wavelength λvac = 555 nm, the peak of the eye’s bright-light response. More time from a source. Units: lumens
radiant power is required to achieve the same number of lumens for wavelengths (lm) lm=(1/683) W @ 555 nm
away from the center of the eye’s spectral response. Photometric units are often
Luminous Solid-Angle Intensity
used to characterize room lighting as well as photographic, projection, and display (of a source) Luminous power per
equipment. For example, both a 60 W incandescent bulb and a 13 W compact steradian emitted from a point-
fluorescent bulb emit a little more than 800 lumens of light. The difference in like source. Units: candelas (cd),
cd = lm/Sr.
photometric output versus radiometric output reflects the fact that most of the
energy radiated from an incandescent bulb is emitted in the infrared, where Luminance (of a source): Lumi-
nous solid-angle intensity per pro-
our eyes are not sensitive. Table 2.2 gives the names of the various photometric
jected area of an extended source.
quantities, which parallel the entries for radiometric quantities in Table 2.1. We (The projected area foreshortens
include a variety of units that are sometimes encountered. by cos θ, where θ is the observa-
Cones come in three varieties, each of which is sensitive to light in different tion angle relative to the surface
wavelength bands. Figure 2.9 plots the normalized sensitivity curves13 for short normal.) Units: cd/cm2 = stilb,
cd/m2 = nit, nit = 3183 lambert =
(S), medium (M), and long (L) wavelength cones. Because your brain gets separate 3.4 footlambert
signals from each type of cone, this system gives you the ability to measure
Luminous Emittance or Exitance
basic information about the spectral content of light. We interpret this spectral
(from a source): Luminous Power
information as the color of the light. When the three types of cones are stimulated emitted per unit surface area of an
equally the light appears white, and when they are stimulated differently the extended source. Units: lm/cm2
light appears colored. Light with different spectral distributions can produce the Illuminance (to a receiver): Inci-
exact same color sensation, so our perception of color only gives very general dent luminous power delivered
information about the spectral content of light. For example, light coming from per area to a receiver. Units: lux;
lm/m2 = lux, lm/cm2 = phot,
a television has a different spectral composition than the light incident on the
lm/ft2 = footcandle
camera that recorded the image, but both can produce the same color sensation.
This ambiguity can lead to a potentially dangerous situation in the lab because
Table 2.2 Photometric quantities
lasers from 670 nm to 800 nm all appear the same color. (They all stimulate the
and units.
L and M cones in essentially the same ratio.) However, your eye’s response falls
off quickly in the near-infrared, so a dangerous 800 nm high-intensity beam can
appear about the same brightness as an innocuous 670 nm laser pointer.
Because we have have three types of cones, our perception of color can be
well-represented using a three-dimensional vector space referred to as a color
S
space.14 A color space can be defined in terms of three “basis” light sources
M
13 A. Stockman, L. Sharpe, and C. Fach, “The spectral sensitivity of the human short-wavelength L
cones,” Vision Research, 39, 2901-2927 (1999); A. Stockman, and L. Sharpe, “Spectral sensitivities
of the middle- and long-wavelength sensitive cones derived from measurements in observers of
known genotype,” Vision Research, 40, 1711-1737 (2000).
14 The methods we use to represent color are very much tied to human physiology. Other species
400 500 600 700 800
have photoreceptors that sense different wavelength ranges or do not sense color at all. For instance,
wavelength (nm)
Papilio butterflies have six types of cone-like photoreceptors and certain types of shrimp have
twelve. Reptiles have four-color vision for visible light, and pit vipers (a subgroup of snakes) have Figure 2.9 Normalized cone sensi-
an additional set of “eyes” that look like pits on the front of their face. These pits are essentially
tivity functions
pinhole cameras sensitive to infrared light, and give these reptiles crude night-vision capabilities.
(Not surprisingly, pit vipers hunt most actively at night time.) On the other hand, some insects can
perceive markings on flowers that are only visible in the ultraviolet. Each of these species would
referred to as primaries. Different colors (i.e. the “vectors” in the color space) are
created by mixing the primary light in different ratios. If we had three primaries
that separately stimulated each type of rod (S, M, and L), we could recreate any
color sensation exactly by mixing those primaries. However, by inspecting Fig. 2.9
you can see that this ideal set of primaries cannot be found because of the overlap
between the S, M, and L curves. Any light that will stimulate one type of cone will
also stimulate another. This overlap makes it impossible to display every possible
color with three primaries. (Although it is possible to quantify all colors with three
primaries, even if the primaries can’t display the colors—we’ll see how shortly.)
The range of colors that can be displayed with a given set of primaries is referred
to as the gamut of that color space. As your experience with computers suggests,
we are able engineer devices with a very broad gamut, but there are always colors
that cannot be displayed.
The CIE1931 RGB15 color space is a very commonly encountered color space
based on a series of experiments performed by W. David Wright and John Guild
in 1931. In these experiments, test subjects were asked to match the color of a
monochromatic test light source by mixing monochromatic primaries at 700 nm
(R), 546.1 nm (G), and 435.8 nm (B ). The relative amount of R, G, and B light
required to match the color at each test wavelength was recorded as the color
matching functions r¯(λ), ḡ (λ), and b̄(λ), shown in Fig. 2.10. Note that the color
matching functions sometimes go negative. This is most noticeable for r¯(λ), but
all three have negative values. These negative values indicate that the test color
was outside the gamut of the primaries (i.e. the color of the test source could not
be matched by adding primaries). In these cases, the observers matched the test
0 light as closely as possible by mixing primaries, and then they added some of the
primary light to the test light until the colors matched. The amount of primary
400 500 600 700
light that had to be added to the test light was recorded as a negative number. In
Test Wavelength (nm)
this way they were able to quantify the color, even though it couldn’t be displayed
Figure 2.10 The CIE 1931 RGB using their primaries.
color matching functions. It turns out that the eye responds essentially linearly with respect to color
perception. That is, if an observer perceives one light source to have components
(R 1 ,G 1 , B 1 ) and another light source to have components (R 2 ,G 2 , B 2 ), a mixture
of the two lights will have components (R 1 + R 2 ,G 1 + G 2 , B 1 + B 2 ). This linearity
allows us to calculate the color components of an arbitrary light source with
spectrum I (λ) by integrating the spectrum against the color matching functions:
Z Z Z
R = I (λ)r¯d λ G = I (λ)ḡ d λ B = I (λ)b̄d λ (2.63)
If R, G, or B turn out to be negative for a given I (λ), then that color of light
falls outside the gamut of these particular primaries. However, the negative
coordinates still provide a valid abstract representation of that color.
find the color spaces we use to record and recreate color sensations very inaccurate.
15 This is not the RGB space you may have probably used on a computer—that space is referred
to as sRGB. CIE is an abbreviation for the French “Commission Internationale de l’Éclairage,” an

international commission that defines lighting and color standards.
2.B Clausius-Mossotti Relation 63
The RGB color space is an additive color model, where the primaries are added
together to produce color and the absence of light gives black. Subtractive color
models produce color using a background that reflects all visible light equally so
that it appears white (e.g. a piece of paper or canvas) and then placing absorbing
pigments over the background to remove portions of the reflected spectrum.
Some color spaces use four basis vectors. For example, color printers use
the subtractive CMYK color space (Cyan, Magenta, Yellow, and Black), and some
television manufacturers add a fourth type of primary (usually yellow) to their
display. The fourth basis vector increases the range of colors that can be displayed
by these systems (i.e. it increases the gamut). However, the fourth basis vector
makes the color space overdetermined and only helps in displaying colors—we
can abstractly represent all colors using just three coordinates (in an appropriately
chosen basis).
Example 2.4
The CIE1931 XYZ color space is derived from the CIE1931 RGB space by the trans-
formation
    
X 0.49 0.31 0.20 R
 Y = 1  0.17697 0.81240 0.01063   G  (2.64)
0.17697
Z 0.00 0.01 0.99 B
where X , Y , and Z are the color coordinates in the new basis. The matrix elements
in (2.64) were carefully chosen to give this color space some desirable properties:
none of the new coordinates (X , Y , or Z ) are ever negative; the Y gives the pho-
tometric brightness of the light and the X and Z coordinates describe the color
part (i.e. the chromatisity) of the light; and the coordinates (1/3,1/3,1/3) give the
color white. The XYZ coordinates do not represent new primaries, but rather linear
combinations of the original primaries. Find the representation in the CIE1931
RGB basis for each of the basis vectors in the XYZ space.
Solution: We first invert the transformation matrix to find

    
R 0.4185 −0.1587 −0.08283 X
 G  =  −0.09117 0.2524 0.01571   Y 
B 0.0009209 −0.002550 0.1786 Z
Then we can see that X = 0.4185R − 0.09117G + 0.0009209B , Y = −0.1587R +

0.2524G − 0.002550B , and Z = −0.08283R + 0.01571G + 0.1786B . Because the XYZ
primaries contain negative amounts of the physical RGB primaries, the XYZ basis
is not physically realizable. However, it is extensively used because it can abstractly
represent all colors using a triplet of positive numbers.
Appendix 2.B Clausius-Mossotti Relation

Equation (2.35) has the form rmicro = αE/q e , where α is called the atomic (or
molecular) polarizability. We take absorption to be negligible so that α is real. E
is the macroscopic field in the medium, which includes a contribution from all of
the dipoles. To avoid double-counting the dipole’s own field, we should replace E
with
Eactual ≡ E − Edipole (2.65)
and write
q e rmicro = αEactual (2.66)
That is, we ought not to allow the dipole’s own field to act on itself as we previously
(inadvertently) did. Here Edipole is the average field that a dipole contributes to its
quota of space in the material.
Since N is the number of dipoles per volume, each dipole occupies a volume
1/N . As will be shown below, the average field due to a dipole16 centered in such
a volume (symmetrically chosen) is
N q e rmicro
Edipole = − (2.67)
3²o
Substitution of (2.67) and (2.66) into (2.65) yields
N αEactual E
Eeffective = E + ⇒ Eactual = α
(2.68)
3²o 1− N
3²o
Then ( 2.66) becomes

αE
q e rmicro = α
(2.69)
1− N
3²o
Now according to (2.16) the susceptibility is defined via P = ²0 χE, where E is
the macroscopic field. Also, the polarization is always based on the combined be-
havior of all of the dipoles P = N q e rmicro (see (2.31)). Therefore, the susceptibility
is
N α(ω)
²o
χ (ω) = N α(ω)
(2.70)
1 − 3²o
This is known as the Clausius-Mossotti relation. In section 2.3, we only included
the numerator of (2.70). The extra term in the denominator becomes important
when N is sufficiently large, which is the case for liquid or solid densities.
Since we neglect absorption, from (2.25) we have χ = n 2 −1, and we may write
N α/²0
n2 − 1 = (2.71)
1 − N α/3²0
In this case, we may invert the relation to write N α/²0 in terms of the index:17
Nα n2 − 1
=3 2 (2.72)
²0 n +2
16 In principle, the detailed fields of nearby dipoles should also be considered rather than repre-
senting their influence with the macroscopic field. However, if they are symmetrically distributed
the result is the same. See J. D. Jackson, Classical Electrodynamics, 3rd ed., Sect. 4.5 (New York: John
Wiley, 1999).
17 This form of Clausius-Mossotti relation, in terms of the refractive index, was renamed the
Lorentz-Lorenz formula, but probably undeservedly so, since it is essentially the same formula.
2.B Clausius-Mossotti Relation 65
Example 2.5
Xenon vapor at STP (density 4.46×10−5 mol/cm3 ) has index n = 1.000702 measured
at wavelength 589nm. Use (a) the Clausius-Mossotti relation (2.70) and (b) the
uncorrected formula (i.e. numerator only) to predict the index for liquid xenon
with density 2.00×10−2 mol/cm3 Compare with the measured value of n = 1.332.18
Solution: At the low density, we may may safely neglect the correction in the
denominator of (2.71) and simply write Natm α/²0 = 1.0007022 − 1 = 1.404 × 10−3 .
The liquid density Nliquid is 2.00 × 10−2 /4.46 × 10−5 = 449 times greater. Therefore,
Nliquid α/²0 = 449 × 1.404 × 10−3 = 0.630. (a) According to Clausius-Mossotti (2.71),
the index is r
0.630
n = 1+ = 1.341
1 − 0.630/3
(b) On the other hand, without the correction in the denominator, we get
p
n = 1 + 0.630 = 1.277
The Clausius-Mossotti formula gets much closer to the measured value.
Average Field Produced by a Dipole
Consider a dipole comprised of point charges ±q e separated by spacing rmicro = ẑd .

If the dipole is centered on the origin, then by Coulomb’s law the field surrounding
the point charges is
q e r − ẑd /2 q e r + ẑd /2
E= 3
−
4π²0 |r − ẑd /2| 4π²0 |r + ẑd /2|3
We wish to compute the average field within a cubic volume V = L 3 that symmet- Figure 2.11 The field lines sur-
rically encompasses the dipole.19 We take the volume dimension L to be large rounding a dipole.
compared to the dipole dimension d . Integrating the field over this volume yields
ZL/2 ZL/2 ZL/2 " #

qe x x̂ + y ŷ + (z − d /2) ẑ x x̂ + y ŷ + (z + d /2) ẑ
Z
Ed v = dx dy dz £ ¤3/2 − £ ¤3/2
4π²0 2 2
x + y + (z − d /2) 2 x + y 2 + (z + d /2)2
2
- L/2 - L/2 - L/2
ZL/2 ZL/2
 
qe 1 1
= −ẑ dx dy q −q 
2π²0
- L/2 - L/2 x 2 + y 2 + (L − d )2 /4 x 2 + y 2 + (L + d )2 /4
18 D. H. Garside, H. V. Molgaard, and B. L. Smith, “Refractive Index and Lorentz-Lorenz function
of Xenon Liquid and Vapour,” J. Phys. B: At. Mol. Phys. 1, 449-457 (1968).
19 Authors often obtain the same result using a spherical volume with the (usually unmentioned)
conceptual awkwardness that spheres cannot be closely packed to form a macroscopic medium
without introducing voids.
The terms multiplying x̂ and ŷ vanish since they involve odd functions integrated
over even limits on either x or y, respectively. On the remaining term, the integra-
tion on z has been executed. Before integrating the remaining expression over x
and y, we make the following approximation based on L >> d :
1 ∼ 11
q =p q
x 2 + y 2 + (L ± d )2 /4 x 2 + y 2 + L 2 /4
1 ± x 2 +yLd2 +L
/2
2 /4
Ld
· ¸
∼ 1 /4
=p 1∓ 2
x 2 + y 2 + L 2 /4 x + y 2 + L 2 /4
which will make integration considerably easier.20 Then integration over the y
dimension brings us to21
ZL/2 ZL/2 ZL/2

qe d Ld y qe d L2d x
Z
Ed v = −ẑ dx ¤3/2 = −ẑ ¢p
4π²0 4π²0
¡
x 2 + L 2 /4 x 2 + L 2 /2
£
-L/2 -L/2
x 2 + y 2 + L 2 /4 -L/2
The final integral is the same as twice the integral from 0 to L/2.
p Then, with x > 0,
we can employ the variable change s = x 2 +L 2 /4 ⇒ 2d x = d s/ s − L 2 /4 and obtain
LZ2 /2
qe d L2d s q e d 4π
Z
Ed v = −ẑ p = −ẑ
4π²0 s s 2 − L 4 /16 4π²0 3
L 2 /4
Reinstalling rmicro = ẑd and dividing by the volume 1/N , allotted to individual
dipoles, brings us to the anticipated result (2.67).
Appendix 2.C Energy Density of Electric Fields

In this appendix we show that the term ²0 E 2 /2 in (2.53) corresponds to the energy
density of an electric field.22 The electric potential φ(r) (in units of energy per
20 One might be tempted to begin this calculation with the well-known dipole field
 
qe r − ẑd /2 r + ẑd /2 qe d
E= −h i ∼ = [3rˆ (ẑ · r̂) − ẑ]
 
4π²0 r 3 2 3/2 d 2 3/2 4π²0 r 3
h i
1 − ẑ · rˆ dr + d 2 d
1 + ẑ · r̂ r + 2
4r 4r
which relies on the approximation

h i−3/2 3d ẑ · r̂
1 ± ẑ · r̂d /r + d 2 /4r 2 ∼
= [1 ± ẑ · r̂d /r ]
−3/2 ∼
= 1∓
2r
This dipole-field expression, while useful for describing the field surrounding the dipole, contains
no information about the fields internal to the diple. Note that we integrate z through the origin,
which would violate the above assumption r À d . Alternatively, the influence of the internal fields
on our integral could be accomplished using a delta function as is done in J. D. Jackson, Classical
Electrodynamics, 3rd ed., p. 149 (New York: John Wiley, 1999).
21 Two useful integral formulas are (0.61) and (0.61).
22 J. R. Reitz, F. J. Milford, and R. W. Christy, Foundations of Electromagnetic Theory 3rd ed., Sect.
6-3 (Reading, Massachusetts: Addison-Wesley, 1979).

2.C Energy Density of Electric Fields 67
charge, or volts) describes the potential energy that a charge would experience
if placed at any given point in the field. The electric field and the potential are
connected through
E (r) = −∇φ (r) (2.73)
The energy U necessary to assemble a distribution of charges (owing to attraction
or repulsion) can be written in terms of a summation over all of the charges (or
charge density ρ (r)) located within the potential:
1
Z
U= φ (r) ρ (r) d v (2.74)
2
V
We consider the potential to arise from the charges themselves. The factor 1/2
is necessary to avoid double counting. To appreciate this factor consider just
two point charges: We only need to count the energy due to one charge in the
presence of the other’s potential to obtain the energy required to bring the charges
together.
A substitution of (1.1) for ρ (r) into (2.74) gives
²0
Z
U= φ (r) ∇ · E (r) d v (2.75)
2
V
Next, we use the vector identity in P0.9 and get
²0 ²0
Z Z
∇ · φ (r) E (r) d v −
£ ¤
U= E (r) · ∇φ (r) d v (2.76)
2 2
V V
An application of the divergence theorem (0.11) on the first integral and a substi-
tution of (2.73) into the second integral yields
²0 ²0
I Z
U= φ (r) E (r) · n̂d a + E (r) · E (r) d v (2.77)
2 2
S V
We can consider the volume V (enclosed by S) to be as large as we like, say

a sphere of radius R, so that all charges are contained well within it. Then the
surface integral over S vanishes as R → ∞ since φ ∼ 1/R and E ∼ 1/R 2 , whereas
d a ∼ R 2 . Then the total energy is expressed solely in terms of the electric field:
Z
U= u E (r) d v (2.78)
All
Space
where
²0 E 2
u E (r) ≡ (2.79)
2
is interpreted as the energy density of the electric field.
Appendix 2.D Energy Density of Magnetic Fields

In a derivation similar to that in appendix 2.C, we consider the energy associated
with magnetic fields.23 The magnetic vector potential A (r) (in units of energy
per charge×velocity) describes the potential energy that a charge moving with
velocity v would experience if placed in the field. The magnetic field and the
vector potential are connected through
B (r) = ∇ × A (r) (2.80)
The energy U necessary to assemble a distribution of currents can be written in
terms of a summation over all of the currents (or current density J (r)) located
within the vector potential field:
1
Z
U= J (r) · A (r) d v (2.81)
2
V
As in (2.74), the factor 1/2 is necessary to avoid double counting the influence of
the currents on each other.
Under the assumption of steady currents (no variations in time), we may
substitute Ampere’s law (1.21) into (2.81), which yields
1
Z
U= [∇ × B (r)] · A (r) d v (2.82)
2µ0
V
Next we employ the vector identity P0.8 from which the previous expression
becomes
1 1
Z Z
U= B (r) · [∇ × A (r)] d v − ∇ · [A (r) × B (r)] d v (2.83)
2µ0 2µ0
V V
Upon substituting (2.80) into the first equation and applying the Divergence
theorem (0.11) on the second integral, this expression for total energy becomes
1 1
Z I
U= B (r) · B (r) d v − [A (r) × B (r)] · n̂ d a (2.84)
2µ0 2µ0
V S
As was done in connection with (2.77), if we choose a large enough volume (a

sphere with radius R → ∞), the surface integral vanishes since A ∼ 1/R and
B ∼ 1/R 2 , whereas d a ∼ R 2 . The total energy (2.84) then reduces to
Z
U= u B (r) d v (2.85)
All
Space
where
B2
u B (r) ≡ (2.86)
2µ0
is the energy density for a magnetic field.
23 J. R. Reitz, F. J. Milford, and R. W. Christy, Foundations of Electromagnetic Theory 3rd ed., Sect.
12-2 (Reading, Massachusetts: Addison-Wesley, 1979.

Exercises 69
Exercises
Exercises for 2.3 The Lorentz Model of Dielectrics
P2.1 Verify that (2.35) is a solution to (2.34).
P2.2 Derive the Sellmeier equation
Aλ2vac
n2 = 1 +
λ2vac − λ20,vac
from (2.39) for a gas with negligible absorption (i.e. γ ∼ = 0, valid far
from resonance ω0 ), where λ0,vac corresponds to frequency ω0 and A is
a constant. Many materials (e.g. glass, air) have strong resonances in
the ultraviolet. In such materials, do you expect the index of refraction
for blue light to be greater than that for red light? Make a sketch of n as
a function of wavelength for visible light down to the ultraviolet (where
λ0,vac is located).
P2.3 In the Lorentz model, take N = 1028 m−3 for the density of bound
electrons in an insulator (note that N is number per volume, not just
number), and a single transition at ω0 = 6 × 1015 rad/sec (in the UV),
and damping γ = ω0 /5 (quite broad). Assume E0 is 104 V/m.
For three frequencies ω = ω0 −2γ, ω = ω0 , and ω = ω0 +2γ find the mag-
nitude and phase (relative to the phase of E0 e i (k·r−ωt ) ) of the following
quantities. Give correct SI units with each quantity. You don’t need to
worry about vector directions.
(a) The charge displacement amplitude rmicro (2.35)
(b) The polarization P(ω)
(c) The susceptibility χ(ω). What would the susceptibility be for twice
the E-field strength as before?
For the following no phase is needed:
(d) Find n and κ at the three frequencies. You will have to solve for the
real and imaginary parts of (n + i κ)2 = 1 + χ(ω).
(e) Find the three speeds of light in terms of c. Find the three wave-
lengths λ.
(f) Find how far light penetrates into the material before only 1/e of the
amplitude of E remains. Find how far light penetrates into the material
before only 1/e of the intensity I remains.
P2.4 (a) Use a computer graphing program and the Lorentz model to plot n
and κ as a function of ω frequency for a dielectric (i.e. obtain graphs
such as the ones in Fig. 2.5). Use these parameters to keep things
simple: ωp = 1, ω0 = 10, and γ = 1; plot your function from ω = 0 to

ω = 20.
(b) Plot n and κ as a function of frequency for a material that has
three resonant frequencies: ω0 1 = 10, γ1 = 1, f 1 = 0.5; ω0 2 = 15, γ2 = 1,
f 2 = 0.25; and ω0 3 = 25, γ3 = 3, f 3 = 0.25. Use ωp = 1 for all three
resonances, and plot the results from ω = 0 to ω = 30. Comment on
your plots.
Exercises for 2.4 Index of Refraction of a Conductor
P2.5 For silver, the complex refractive index is characterized by n = 0.13

and κ = 4.0.24 Find the distance that light travels inside of silver before
the field is reduced by a factor of 1/e. Assume a wavelength of λvac =
633 nm. What is the speed of the wave crests in the silver (written as a
number times c)? Are you surprised?
P2.6 Use (2.27), (2.29), and (2.48) to estimate the index of silver at λ =
633nm. The density of free electrons in silver is N = 5.86×1028 m−3 and
the DC conductivity is σ = 6.62 × 107 C2 / (J · m · s).25 Compare with the
actual index given in P2.5.
Answer: n + i κ = 0.02 + i 4.50
P2.7 The dielectric model and the conductor model give identical results
for n in the case of a low-density plasma where there is no restoring
force (i.e. ω0 = 0) and no dragging term (i.e. γ = 0). Use this to model
the ionosphere (the uppermost part of the atmosphere that is ionized
by solar radiation to form a low-density plasma).
(a) If the index of refraction of the ionosphere is n = 0.9 for an FM
station at ν = ω/2π = 100 MHz, calculate the number of free electrons
per cubic meter.
(b) What is the complex refractive index of the ionosphere for radio
waves at 1160 kHz (KSL radio station)? Is this frequency above or below
the plasma frequency? Assume the same density of free electrons as in
part (a).
For your information, AM radio reflects better than FM radio from the
ionosphere (like visible light from a metal mirror). At night, the lower
layer of the ionosphere goes away so that AM radio waves reflect from
a higher layer.
P2.8 Use a computer to plot n and κ as a function of frequency for a con-

ductor (obtain plots such as the ones in Fig. 2.6). Use these parame-
ters to keep things simple, let γ = 0.02ωp and plot your function from
ω = 0.6ωp to ω = 2ωp .
24 Handbook of Optical Constants of Solids, Edited by E. D. Palik (Elsevier, 1997).
25 G. Burns, Solid State Physics, p. 194 (Orlando: Academic Press, 1985).
Exercises 71
Exercises for 2.6 Irradiance of a Plane Wave
P2.9 In the case of a linearly-polarized plane wave, where the phase of each
vector component of E0 is the same, re-derive (2.61) directly from the
real field (2.21). For simplicity, you may ignore absorption (i.e. κ ∼
= 0).
HINT: The time-average of cos2 k · r − ωt + φ is 1/2.
¡ ¢
P2.10 (a) Find the intensity (in W/cm2 ) produced by a short laser pulse (lin-
early polarized) with duration ∆t = 2.5 ×10−14 s and energy E = 100 mJ,
focused in vacuum to a round spot with radius r = 5 µm.
(b) What is the peak electric field (in V/Å)?
HINT: The SI units of electric field are N/C = V/m.
(c) What is the peak magnetic field (in T = kg/(s · C)?
P2.11 (a) What is the intensity (in W/cm2 ) on the retina when looking directly
at the sun? Assume that the eye’s pupil has a radius r pupil = 1 mm.
Take the Sun’s irradiance at the earth’s surface to be 1.4 kW/m2 , and
neglect refractive index (i.e. set n = 1). HINT: The Earth-Sun distance
is d o = 1.5 × 108 km and the pupil-retina distance is d i = 22 mm. The
radius of the Sun r Sun = 7.0 × 105 km is de-magnified on the retina
according to the ratio d i /d o .
(b) What is the intensity at the retina when looking directly into a
1 mW HeNe laser? Assume that the smallest radius of the laser beam
is r waist = 0.5 mm positioned d o = 2 m in front of the eye, and that the
entire beam enters the pupil. Compare with part (a).
P2.12 Show that the magnetic field of an intense laser with λ = 1 µm becomes
important for a free electron oscillating in the field at intensities above
1018 W/cm2 . This marks the transition to relativistic physics. Neverthe-
less, for convenience, use classical physics in making the estimate.
HINT: At lower intensities, the oscillating electric field dominates, so
the electron motion can be thought of as arising solely from the electric
field. Use this motion to calculate the magnetic force on the mov-
ing electron, and compare it to the electric force. The forces become
comparable at 1018 W/cm2 .
Exercises for 2.A Radiometry, Photometry, and Color
P2.13 The CIE1931 RGB color matching function r¯(λ), ḡ (λ), and b̄(λ) can be
transformed using (2.64) to obtain color matching functions for the
XYZ basis: x̄(λ), ȳ(λ), and z̄(λ), plotted in Fig 2.12. As with the RGB
color matching functions, the XYZ color matching functions can be
used to calculate the color coordinates in the XYZ basis for an arbitrary
spectrum:
Z Z Z
X = I (λ)x̄d λ Y = I (λ) ȳd λ Z = I (λ)z̄d λ (2.87)
(The function ȳ(λ) was chosen to be exactly the scoptic response curve
(shown in Fig. 2.8), so that Y describes the photometric brightness of
the light.)
Obtain a copy of the XYZ color matching functions from www.cvrl.org
400 500 600 700
and calculate the XYZ color coordinates for the spectrum
wavelength (nm)
2
Figure 2.12 Color matching func- I (λ) = I 0 e −(λ−500 nm)
tions for the CIE XYZ color space
P2.14 The color space you’ve probably encountered most is sRGB, used to rep-
resent color on computer displays. The sRGB coordinates are related
to the XYZ coordinates by the transformation
R̃ 3.2406 −1.5372 −0.4986 X

    
 G̃  =  −0.9689 1.8758 0.0415   Y 
B̃ 0.0557 −0.2040 1.0570 Z
where the XYZ coordinates need to be scaled to values similar to those

accepted by the sRGB device (commonly 0 to 255) and then the sRGB
coordinates R̃, G̃, and B̃ need to be scaled or clipped to fit in the appro-
priate range. (This scaling and clipping result from the fact that your
monitor cannot display arbitrarily bright light.)
Obtain a copy of the XYZ color matching functions from www.cvrl.org
and use it to calculate the sRGB components for monochromatic light
from λ0 = 400 nm to λ0 = 700 nm in 1 nm intervals. Make a plot of
the individual sRGB values and also use the coordinates to display a
rainbow. HINT: Matlab has all the functions you need to display the
rainbow.
Chapter 3
Reflection and Refraction
As we know from everyday experience, when light arrives at an interface between

materials it is partially reflected and partially transmitted. In this chapter, we
examine what happens to plane waves when they propagate from one material
(characterized by index n or even by complex index N ) to another material. We
will derive expressions to quantify the amount of reflection and transmission. The
results depend on the angle of incidence (i.e. the angle between k and the normal
to the surface) as well as on the orientation of the electric field (called polarization
– not to be confused with P, also called polarization). In this chapter, we consider
only isotropic materials (e.g. glass); in chapter 5 we consider anisotropic materials
(e.g. a crystal).
As we develop the connection between incident, reflected, and transmitted
light waves,1 several familiar relationships will emerge naturally (e.g. Snell’s law
and Brewster’s angle). The formalism also describes polarization-dependent
phase shifts upon reflection (especially interesting in the case of total internal
reflection or in the case of reflections from absorbing surfaces such as metals).
For simplicity, we initially neglect the imaginary part of the refractive index.
Each plane wave is thus characterized by a real wave vector k. We will write each
plane wave in the form E(r, t ) = E0 exp [i (k · r − ωt )], where, as usual, only the real
part of the field corresponds to the physical field. The restriction to real refractive
indices is not as serious as it might seem. The use of the letter n instead of N
hardly matters. The math is all the same, which demonstrates the power of the
complex notation. We can simply update our expressions in the end to include
complex refractive indices, but in the mean time it is easier to think of absorption
as negligible.
3.1 Refraction at an Interface

Consider a planar boundary between two materials with different indexes. The
index n i characterizes the material on the left, and the index n t characterizes the
1 See M. Born and E. Wolf, Principles of Optics, 7th ed., Sect. 1.5 (Cambridge University Press,
1999).
73
74 Chapter 3 Reflection and Refraction
material on the right as depicted in the Fig. 3.1. When a plane wave traveling in
the direction ki is incident the boundary from the left, it gives rise to a reflected
vector traveling in the direction kr and a transmitted plane wave traveling in the
direction kt . The incident and reflected waves exist only to the left of the material
interface, and the transmitted wave exists only to the right of the interface. The
angles θi , θr , and θt give the angles that each respective wave vector (ki , kr , and
kt ) makes with the normal to the interface.
For simplicity, we’ll assume that both of the materials are isotropic here.
(Chapter 5 discusses refraction for anisotropic materials.) In this case, ki , kr , and
kt all lie in a single plane, referred to as the plane of incidence, (i.e. the plane
represented by the surface of this page). We are free to orient our coordinate
system in many different ways (and every textbook seems to do it differently!).2
We choose the y–z plane to be the plane of incidence, with the z-direction normal
to the interface and the x-axis pointing into the page.
The electric field vector for each plane wave is confined to a plane perpendic-
ular to its wave vector. We are free to decompose the field vector into arbitrary
components as long as they are perpendicular to the wave vector. It is customary
to choose one of the electric field vector components to be that which lies within
the plane of incidence. We call this p-polarized light, where p stands for parallel to
z-axis the plane of incidence. The remaining electric field vector component is directed
normal to the plane of incidence and is called s-polarized light.. The s stands for
x-axis senkrecht, a German word meaning perpendicular.
directed into page
Using this system, we can decompose the electric field vector Ei into its p-
(p)
polarized component E i and its s-polarized component E i(s) , as depicted in
Fig. 3.1. The s component E i(s) is represented by the tail of an arrow pointing
into the page, or the x-direction in our convention. The other fields Er and Et
Figure 3.1 Incident, reflected, and
are similarly split into s and p components as indicated in Fig. 3.1. All field
transmitted plane wave fields at a
material interface.
components are considered to be positive when they point in the direction of
their respective arrows.3 Note that the s-polarized components are parallel for
all three plane waves, whereas the p-polarized components are not (except at
normal incidence) because each plane wave travels in a different direction.
By inspection of Fig. 3.1, we can write the various wave vectors in terms of the
ŷ and ẑ unit vectors:
ki = k i ŷ sin θi + ẑ cos θi
¡ ¢
kr = k r ŷ sin θr − ẑ cos θr
¡ ¢
(3.1)
kt = k t ŷ sin θt + ẑ cos θt
¡ ¢
Also by inspection of Fig. 3.1 (following the conventions for the electric fields
indicated by the arrows), we can write the incident, reflected, and transmitted
2 For example, our convention is different than that used by E. Hecht, Optics, 3rd ed., Sect. 4.6.2
(Massachusetts: Addison-Wesley, 1998).

3 Many textbooks draw the arrow for E (p) in the direction opposite of ours. However, that choice
r
leads to an awkward situation at normal incidence (i.e. θi = θr = 0) where the arrows for the incident
and reflected fields are parallel for the s-component but anti parallel for the p-component.
3.1 Refraction at an Interface 75
fields in terms of x̂, ŷ, and ẑ:

h i
Ei = E i ŷ cos θi − ẑ sin θi + x̂E i(s) e i [ki ( y sin θi +z cos θi )−ωi t ]
(p) ¡ ¢
h i
Er = E r ŷ cos θr + ẑ sin θr + x̂E r(s) e i [kr ( y sin θr −z cos θr )−ωr t ]
(p) ¡ ¢
(3.2)
h i
Et = E t ŷ cos θt − ẑ sin θt + x̂E t(s) e i [kt ( y sin θt +z cos θt )−ωt t ]
(p) ¡ ¢
Each field has the form (2.8), and we have utilized the k-vectors (3.1) in the
exponents of (3.2).
Now we are ready to connect the fields on one side of the interface to the
fields on the other side. This is done using boundary conditions. As explained
in appendix 3.A, Maxwell’s equations require that the component of E that are Figure 3.2 Animation of s- and
parallel to the interface must be the same on either side of the boundary. In p-polarized fields incident on an
our coordinate system, the x̂ and ŷ components are parallel to the interface, and interface as the angle of incidence
z = 0 defines the interface. This means that at z = 0 the x̂ and ŷ components is varied.
of the combined incident and reflected fields must equal the corresponding
components of the transmitted field:
h i h i
E i ŷ cos θi + x̂E i(s) e i (ki y sin θi −ωi t ) + E r ŷ cos θr + x̂E r(s) e i (kr y sin θr −ωr t )
(p) (p)
h i
= E t ŷ cos θt + x̂E t(s) e i (kt y sin θt −ωt t ) (3.3)
(p)
Since this equation must hold for all conceivable values of t and y, we are com-
pelled to set all the phase factors in the complex exponentials equal to each other.
The time portion of the phase factors requires the frequency of all waves to be the
same:
ωi = ωr = ωt ≡ ω (3.4)
(We could have guessed that all frequencies would be the same; otherwise wave
fronts would be annihilated or created at the interface.) Similarly, equating the
spatial terms in the exponents of (3.3) requires
Willebrord Snell (or Snellius) (1580
k i sin θi = k r sin θr = k t sin θt (3.5) 1626, Dutch) was an astronomer and
mathematician born in Leiden, Nether-
Now recall from (2.19) the relations k i = k r = n i ω/c and k t = n t ω/c. With these lands. In 1613 he succeeded his father
as professor of mathematics at the
relations, (3.5) yields the law of reflection University of Leiden. He was an accom-
plished mathematician, developing a
θr = θi (3.6) new method for calculating π as well as
an improved method for measuring the
circumference of the earth. He is most
and Snell’s law
famous for his rediscovery of the law of
n i sin θi = n t sin θt (3.7) refraction in 1621. (The law was known
(in table form) to the ancient Greek
The three angles θi , θr , and θt are not independent. The reflected angle matches mathematician Ptolemy, to Persian en-
the incident angle, and the transmitted angle obeys Snell’s law. The phenomenon gineer Ibn Sahl (900s), and to Polish
philosopher Witelo (1200s).) Snell au-
of refraction refers to the fact that θi and θt are different. thored several books, including one on
Because the exponents are all identical, (3.3) reduces to two relatively simple trigonometry, published a year after his
death. (Wikipedia)
equations (one for each dimension, x̂ and ŷ):
E i(s) + E r(s) = E t(s) (3.8)

and ³ ´
(p) (p) (p)
Ei + Er cos θi = E t cos θt (3.9)
We have derived these equations from the boundary condition (3.52) on the
parallel component of the electric field. This set of equations has four unknowns
(p) (p)
(E r , E r(s) , E t , and E t(s) ), assuming that we pick the incident fields. We require
two further equations to solve the system. These are obtained using the separate
boundary condition on the parallel component of magnetic fields given in (3.56)
(also discussed in appendix 3.A).
From Faraday’s law (1.3), we have for a plane wave (see (2.56))
k×E n
B= = û × E (3.10)
ω c
where û ≡ k/k is a unit vector in the direction of k. We have also utilized (2.19)
for a real index. This expression is useful for writing Bi , Br , and Bt in terms of the
electric field components that we have already introduced. When injecting (3.1)
and (3.2) into (3.10), the incident, reflected, and transmitted magnetic fields turn
out to be
ni h ¢i
−x̂E i + E i(s) −ẑ sin θi + ŷ cos θi e i [ki ( y sin θi +z cos θi )−ωi t ]
(p) ¡
Bi =
c
n r h (p) ¢i
x̂E r + E r(s) −ẑ sin θr − ŷ cos θr e i [kr ( y sin θr −z cos θr )−ωr t ]
¡
Br = (3.11)
c h
nt i
−x̂E t + E t(s) −ẑ sin θt + ŷ cos θt e i [kt ( y sin θt +z cos θt )−ωt t ]
(p) ¡ ¢
Bt =
c
Next, we apply the boundary condition (3.56), namely that the components of B
parallel to the interface (i.e. in the x̂ and ŷ dimensions) are the same4 on either
side of the plane z = 0. Since we already know that the exponents are all equal
and that θr = θi and n i = n r , the boundary condition gives
ni h (p)
i n h
i (p)
i n h
t (p)
i
−x̂E i + E i(s) ŷ cos θi + x̂E r − E r(s) ŷ cos θi = −x̂E t + E t(s) ŷ cos θt
c c c
(3.12)
As before, (3.12) reduces to two relatively simple equations (one for the x̂ dimen-
sion and one for the ŷ dimension):
³ ´
(p) (p) (p)
ni E i − E r = nt E t (3.13)
and ³ ´
n i E i(s) − E r(s) cos θi = n t E t(s) cos θt (3.14)
These two equations together with (3.8) and (3.9) allow us to solve for the reflected
Er and transmitted fields Et for the s and p polarization components. However,
(3.8), (3.9), (3.13), and (3.14) are not yet in their most convenient form.
4 We assume the permeability µ is the same everywhere—no magnetic effects.
0
3.2 The Fresnel Coefficients 77
3.2 The Fresnel Coefficients

Augustin Fresnel first derived the results of the previous section. Since he lived
well before Maxwell’s time, he did not have the benefit of Maxwell’s equations as
we have. Instead, Fresnel thought of light as transverse mechanical waves prop-
agating within materials. (Fresnel was naturally a proponent of ‘luminiferous
ether’.) Instead of relating the parallel components of the electric and magnetic
fields across the boundary between the materials, Fresnel used the principle that,
as a transverse mechanical wave propagates from one material to the other, the
two materials should not slip past each other at the interface. This “gluing” of the
materials at the interface also forbids the possibility of the materials detaching
from one another (creating gaps) or passing through one another as they expe-
rience wave vibrations. This mechanical approach to light worked splendidly
and explained polarization effects along with the variations in reflectance and
transmittance as a function of the incident angle of the light.
Fresnel wrote the relationships between the various plane waves depicted
in Fig. 3.1 in terms of coefficients that compare the reflected and transmitted
field amplitudes to those of the incident field. He then calculated the ratio of
the reflected and transmitted field components to the incident field components Augustin Fresnel (17881829, French)
was born in Broglie, France, the son of
for each polarization. In the following example, we illustrate this procedure for
an architect. As a child, he was slow to
s-polarized light. It is left as a homework exercise to solve the equations for develop and still could not read when
p-polarized light (see P3.1). he was eight years old, but by age six-
teen he excelled and entered the École
Polytechnique where he earned distinc-
tion. As a young man, Fresnel began a
Example 3.1 successful career as an engineer, but he
lost his post in 1814 when Napoleon re-
Calculate the ratio of transmitted field to the incident field and the ratio of the turned to power. (Fresnel had supported
reflected field to incident field for s-polarized light. the Bourbons.) This dicult year was
when Fresnel turned his attention to
optics. Fresnel became a major propo-
Solution: We use (3.8) nent of the wave theory of light and
E i(s) + E r(s) = E t(s) four years later wrote a paper on dirac-
tion for which he was awarded a prize
and (3.14), which with the help of Snell’s law is written by the French Academy of Sciences. A
year later he was appointed commis-
(s) (s) sin θi cos θt (s) sioner of lighthouses, which motivated
Ei − Er = E (3.15)
sin θt cos θi t the invention of the Fresnel lens (still
used in many commercial applications).
If we add these two equations, we get Fresnel was under appreciated before
his untimely death from tuberculosis.
sin θi cos θt Many of his papers did not make it into

· ¸
(s)
2E i = 1 + E (s) (3.16) print until years later. Fresnel made
sin θt cos θi t huge advances in the understanding of
reection, diraction, polarization, and
(s)
and after dividing by E i and doing a little algebra, it turns into birefringence. In 1824 Fresnel wrote
to Thomas Young, All the compli-
E t(s) 2 sin θt cos θi ments that I have received from Arago,

= Laplace and Biot never gave me so
Ei(s)
sin θt cos θi + sin θi cos θt much pleasure as the discovery of a
theoretic truth, or the conrmation of
To get the ratio of reflected to incident, we subtract (3.15) from (3.8) to obtain a calculation by experiment. Augustin
Fresnel is a hero of one of the authors
sin θi cos θt of this textbook. (Wikipedia)
· ¸
2E r(s) = 1 − E (s) (3.17)
sin θt cos θi t
and then divide (3.17) by (3.16). After a little algebra, we arrive at

E r(s) sin θt cos θi − sin θi cos θt
=
E i(s) sin θt cos θi + sin θi cos θt
The ratio of the reflected and transmitted field components to the incident
field components are specified by the following coefficients, called the Fresnel
coefficients:
E r(s) sin θt cos θi − sin θi cos θt sin (θi − θt ) n i cos θi − n t cos θt

rs ≡ = =− = (3.18)
E i(s) sin θt cos θi + sin θi cos θt sin (θi + θt ) n i cos θi + n t cos θt
E (s) 2 sin θt cos θi 2 sin θt cos θi 2n i cos θi

t s ≡ t(s) = = = (3.19)
Ei sin θt cos θi + sin θi cos θt sin (θi + θt ) n i cos θi + n t cos θt
(p)
Er cos θt sin θt − cos θi sin θi tan (θi − θt ) n i cos θt − n t cos θi
r p ≡ (p) = =− = (3.20)
Ei cos θt sin θt + cos θi sin θi tan (θi + θt ) n i cos θt + n t cos θi
(p)
E 2 cos θi sin θt 2 cos θi sin θt 2n i cos θi
t p ≡ t(p) = = =
Ei cos θt sin θt + cos θi sin θi sin (θi + θt ) cos (θi − θt ) n i cos θt + n t cos θi
(3.21)
All of the above forms of the Fresnel coefficients are potentially useful, depending
1 on the problem at hand. Remember that the angles in the coefficient are not inde-
pendently chosen, but are subject to Snell’s law (3.7). (The right-most expression
for each coefficient is obtained from the first form using Snell’s law).
The Fresnel coefficients pin down the electric field amplitudes on the two
0 sides of the boundary. They also keep track of phase shifts at a boundary. In
Fig. 3.3 we have plotted the Fresnel coefficients for the case of an air-glass inter-
-0.5 face. Notice that the reflection coefficients are sometimes negative in this plot,
which corresponds to a phase shift of π upon reflection (remember e i π = −1).
-1 Later we will see that when absorbing materials are encountered, more compli-
0 20 40 60 80
cated phase shifts can arise due to the complex index of refraction.
Figure 3.3 The Fresnel coefficients

plotted versus θi for the case of an 3.3 Reflectance and Transmittance
air-glass interface with n i = 1 and
n t = 1.5. We are often interested in knowing the fraction of intensity that transmits through
or reflects from a boundary. Since intensity is proportional to the square of the
amplitude of the electric field, we can write the fraction of the light reflected from
the surface, or reflectance, in terms of the Fresnel coefficients:
¯ (s) ¯2 ¯ (p) ¯2
¯ ¯ ¯ ¯
(s) E (p) E
Ir ¯ r ¯
2 Ir ¯ r ¯ ¯ ¯2
R s ≡ (s) = ¯ ¯2 = |r s | and R p ≡ (p) = ¯ ¯2 = ¯r p ¯ (3.22)
Ii ¯ (s) ¯ Ii ¯ (p) ¯
¯E i ¯ ¯E i ¯
These expressions are applied individually to each polarization component (s or

p). The intensity reflected for each of these orthogonal polarizations is additive
3.3 Reflectance and Transmittance 79
because the two electric fields are orthogonal and cannot interfere with each
other. The total reflected intensity is therefore
(p) (p)
I r(total) = I r(s) + I r = R s I i(s) + R p I i (3.23)
where the incident intensity is given by (2.62):
·¯ ¯ ¸
(total) (s) (p) 1 ¯ (s) ¯2 ¯ (p) ¯2
¯ ¯
Ii = I i + I i = n i ²0 c ¯E i ¯ + ¯E i ¯ (3.24)
2
Since intensity is power per area, we can rewrite (3.23) as incident and re-
flected power:
(p) (p)
P r(total) = P r(s) + P r = R s P i(s) + R p P i (3.25)
Using this expression and requiring that energy be conserved (i.e. P i(total) = P r(total) +
P t(total) ), we find that the portion of the power that transmits is
(p) (p)
P t(total) = P i(s) + P i − P r(s) + P r
¡ ¢ ¡ ¢
1
¢ (p) (3.26)
= (1 − R s ) P i(s) + 1 − R p P i
¡
0.8
From this expression we see that the transmittance (i.e. the fraction of the light
0.6
that transmits) for either polarization is
Ts ≡ 1 − Rs and Tp ≡ 1 − Rp (3.27) 0.4
Figure 3.4 shows typical reflectance and transmittance values for an air-glass 0.2
interface.
0
You might be surprised at first to learn that 0 20 40 60 80
¯ ¯2
T s 6= |t s |2 and T p 6= ¯t p ¯ (3.28)
Figure 3.4 The reflectance and
However, recall that the transmitted intensity (in terms of the transmitted fields)
transmittance plotted versus θi for
depends also on the refractive index. The Fresnel coefficients t s and t p relate the the case of an air-glass interface
bare electric fields to each other, whereas the transmitted intensity (similar to with n i = 1 and n t = 1.5.
(3.24)) is ·¯
1 ¯ ¸
¯ (s) ¯2 ¯ (p) ¯2
¯ ¯
(total) (s) (p)
It = I t + I t = n t ²0 c ¯E t ¯ + ¯E t ¯ (3.29)
2
Therefore, we expect T s and T p to depend on the ratio of the refractive indices n t
and n i as well as on the squares of t s and t p .
There is another more subtle reason for the inequalities in (3.28). Consider
a lateral strip of light associated with a plane wave incident upon the material
interface in Fig. 3.5. Upon refraction into the second medium, the strip is seen
to change its width by the factor cos θt / cos θi . This is a geometrical effect, owing
to the change in propagation direction at the interface. The change in direction
alters the intensity (power per area) but not the power. In computing the trans-
mittance, we must remove this geometrical effect from the ratio of the intensities,
which leads to the following transmittance coefficients:
n t cos θt
Ts = |t s |2
n i cos θi
(valid when no total internal reflection) (3.30)
n t cos θt ¯¯ ¯¯2
Tp = tp Figure 3.5 Light refracting into a
n i cos θi
surface
Note that (3.30) is valid only if a real angle θt exists; it does not hold when the
incident angle exceeds the critical angle for total internal reflection, discussed in
section 3.5. In that situation, we must stick with (3.27).
Example 3.2
Show analytically for p-polarized light that R p + Tp = 1, where R p is given by (3.22)
and T p is given by (3.30).
Solution: From (3.20) we have
¯ cos θt sin θt − cos θi sin θi ¯2

¯ ¯
Rp = ¯
¯ ¯
cos θt sin θt + cos θi sin θi ¯
cos2 θt sin2 θt − 2 cos θi sin θi cos θt sin θt + cos2 θi sin2 θi
=
(cos θt sin θt + cos θi sin θi )2
From (3.21) and (3.30) we have

¸2
n t cos θt 2 cos θi sin θt
·
Tp =
n i cos θi cos θt sin θt + cos θi sin θi
sin θi cos θt 4 cos2 θi sin2 θt
· ¸
=
sin θt cos θi (cos θt sin θt + cos θi sin θi )2
4 cos θi sin θt sin θi cos θt
=
Then
cos2 θt sin2 θt + 2 cos θi sin θi cos θt sin θt + cos2 θi sin2 θi

Rp + Tp =
David Brewster (17811868, Scot- (cos θt sin θt + cos θi sin θi )2
=
tish) was born in Jedburgh, Scottland. (cos θt sin θt + cos θi sin θi )2
His father was a teacher and wanted
David to become a clergyman. At age =1
twelve, David went to the University of
Edinburgh for that purpose, but his incli-
nation for natural science soon became
apparent. He became licensed to preach,
but his interests in science distracted
him from that profession, and he spent 3.4 Brewster’s Angle
much of his time studying diraction.
Taking an empirical approach, Brewster
Notice r p and R p go to zero at a certain angle in Figs. 3.3 and 3.4, indicating that
independently discovered many of the
same things usually credited to Fresnel. no p-polarized light is reflected at this angle. This behavior is quite general, as we
He even made a dioptric apparatus for can see from the second form of the Fresnel coefficient formula for r p in (3.20),
lighthouses before Fresnel developed
his. Brewster became somewhat fa-
which has tan (θi + θt ) in the denominator. Since the tangent ‘blows up’ at π/2,
mous in his day for the development the reflection coefficient goes to zero when
of the kaleidoscope and stereoscope
for enjoyment by the general public. π
Brewster was a prolic science writer θi + θt = (requirement for zero p-polarized reflection) (3.31)
and editor throughout his life. Among
2
his works is an important biography of
Isaac Newton. He was knighted for his By inspecting Fig. 3.1, we see that this condition occurs when the reflected and
accomplishments in 1831. (Wikipedia) transmitted wave vectors, kr and kt , are perpendicular to each other. If we insert
3.5 Total Internal Reflection 81
(3.31) into Snell’s law (3.7), we can solve for the incident angle θi that gives rise to
this special circumstance:
³π ´
n i sin θi = n t sin − θi = n t cos θi (3.32)
2
The angle that satisfies this equation, in terms of the refractive indices, is
readily found to be
nt
θB = tan−1 (3.33)
ni
We have replaced the specific θi with θB in honor of Sir David Brewster who first
discovered the phenomenon. The angle θB is called Brewster’s angle. At Brewster’s Oscillating 0
angle, no p-polarized light reflects (see L 3.4). Physically, the p-polarized light Dipole
cannot reflect because kr and kt are perpendicular. A reflection would require
the microscopic dipoles at the surface of the second material to radiate along
270 90
their axes, which they cannot do. Maxwell’s equations ‘know’ about this, and so
everything is nicely consistent.
180
3.5 Total Internal Reflection
Figure 3.6 The intensity radiation
pattern of an oscillating dipole as
From Snell’s law (3.7), we can compute the transmitted angle in terms of the
a function of angle. Note that the
incident angle:
dipole does not radiate along the
ni
µ ¶
θt = sin−1 sin θi (3.34) axis of oscillation, giving rise to
nt Brewster’s angle for reflection.
The angle θt is real only if the argument of the inverse sine is less than or equal to
one. If n i > n t , we can find a critical angle at which the argument begins to exceed
one:
nt
θc ≡ sin−1 (3.35)
ni
When θi > θc , then there is total internal reflection and we can directly show that
R s = 1 and R p = 1 (see P3.9).5 To demonstrate this, one computes the Fresnel
coefficients (3.18) and (3.20) while employing the following substitutions:
ni
sin θt = sin θi (Snell’s law) (3.36)
nt
and v
u 2
un
cos θt = i t i2 sin2 θi − 1 (θi > θc ) (3.37)
nt
(see P0.19).
In this case, θt is a complex number. However, we do not assign geometrical
significance to it in terms of any direction. Actually, we don’t even need to know
the value for θt ; we need only the values for sin θt and cos θt , as specified in (3.36)
and (3.37). Even though sin θt is greater than one and cos θt is imaginary, we can
use their values to compute r s , r p , t s , and t p . (Complex notation is wonderful!)
5 M. Born and E. Wolf, Principles of Optics, 7th ed., Sect. 1.5.4 (Cambridge University Press, 1999).
Upon substitution of (3.36) and (3.37) into the Fresnel reflection coefficients
(3.18) and (3.20) we obtain
r
ni n i2
nt cos θi − i n t2
sin2 θi − 1
rs = r (θi > θc ) (3.38)
ni n i2
nt cos θi + i n t2
sin θi − 1
2
and r
ni n i2
cos θi − i nt n t2
sin2 θi − 1
rp = − r (θi > θc ) (3.39)
ni n i2
cos θi + i nt n t2
sin θi − 1
2
These Fresnel coefficients can be manipulated (see P3.9) into the forms
  v 
u 2
 n t
u n
t i sin2 θ − 1

r s = exp −2i tan−1  i (θi > θc ) (3.40)
 n i cos θi n 2 
t
and
  v 
u 2
 n i
u n
t i sin2 θ − 1

r p = − exp −2i tan−1  i (θi > θc ) (3.41)
 n t cos θi n 2 
t
Each coefficient has a different phase (note n i /n t vs. n t /n i in the expressions),

Figure 3.7 Animation of light which means that the s- and p-polarized fields experience different ¯ ¯ phase shifts
waves incident on an interface upon reflection. Nevertheless, we definitely have |r s | = 1 and ¯r p ¯ = 1. We rightly
both below and beyond the critical conclude that 100% of the light reflects. The transmittance is zero as dictated by
angle. (3.27). We emphasize that one should not employ (3.29) or (3.30) in the case of
total internal reflection, as the imaginary θt makes the geometric factor in this
equation invalid.
Even with zero transmittance, the boundary conditions from Maxwell’s equa-
tions (as worked out in appendix 3.A) require that the fields be non-zero on the
Incident
Wave Evanescent transmitted side of the boundary, meaning t s 6= 0 and t p 6= 0. While this situation
Wave may seem like a contradiction at first, it is an accurate description of what actually
happens. The coefficients t s and t p characterize evanescent waves that exist on
the transmitted side of the interface. The evanescent wave travels parallel to the
interface so that no energy is conveyed away from the interface deeper into the
medium on the transmission side.
To compute the explicit form of the evanescent wave,6 we plug (3.36) and
Figure 3.8 A wave experiencing (3.37) into the transmitted field (3.2):
total internal reflection creates an
evanescent wave that propagates h i
Et = E t ŷ cos θt − ẑ sin θt + x̂E t(s) e i [kt ( y sin θt +z cos θt )−ωt ]
(p) ¡ ¢
parallel to the interface. (The
s
reflected wave is not shown.)   v u 2   n2
i sin2 θ −1
n −k t z n
h i
(p)  t i
u n i (s)  n t2
i i k t y n i sin θi −ωt
= tp Ei
 ŷi sin2
θ i − 1 − ẑ sin θ i
 + x̂t s E i
e e t
n t2 nt
(3.42)
6 G. R. Fowles, Introduction to Modern Optics, 2nd ed., Sect 2.9 (New York: Dover, 1975).
3.6 Reflections from Metal 83
Figure 3.8 plots the evanescent wave described by (3.42) along with the associ-
ated incident wave. The phase of the evanescent wave indicates that it propagates
parallel to the boundary (in the y-dimension). Its strength decays exponentially
away from the boundary (in the z-dimension). We leave the calculation of t s and
t p as an exercise (P3.10).
3.6 Reflections from Metal

In this section we generalize our analysis to materials with complex refractive
index N ≡ n + i κ. As a reminder, the imaginary part of the index controls atten-
uation of a wave as it propagates within a material. The real part of the index
governs the oscillatory nature of the wave. It turns out that both the imaginary
and real parts of the index strongly influence the reflection of light from a sur-
face. The reader may be grateful that there is no need to re-derive the Fresnel
coefficients (3.18)–(3.21) for the case of complex indices. The coefficients remain
valid whether the index is real or complex – just replace the real index n with the
complex index N . However, we do need to be a bit careful when applying them.
1
We restrict our discussion to reflections from a metallic or other absorbing
material surface. As we found in the case of total internal reflection, we actually do
0.98
not need to know the transmitted angle θt to employ Fresnel reflection coefficients
(3.18) and (3.20). We need only acquire expressions for cos θt and sin θt , and we
can obtain those from Snell’s law (3.7). To minimize complications, we let the 0.96
incident refractive index be n i = 1 (which is often the case). Let the index on
the transmitted side be written as Nt = N . Then by Snell’s law, the sine of the 0.94
transmitted angle is
sin θi 0
sin θt = (3.43)
N
This expression is of course complex since N is complex, which is just fine.7 The
cosine of the same angle is
-0.5 p
1
q q
cos θt = 1 − sin2 θt = N 2 − sin2 θi (3.44)
N
The positive sign in front of the square root is appropriate since it is clearly the
right choice if the imaginary part of the index approaches zero. -p
Upon substitution of these expressions, the Fresnel reflection coefficients 0 20 40 60 80
(3.18) and (3.20) become

p Figure 3.9 The reflectances (top)
cos θi − N 2 − sin2 θi
rs = p (3.45) with associated phases (bottom)
cos θi + N 2 − sin2 θi for silver, which has index n = 0.13
and and κ = 4.05. Note the minimum
of R p corresponding to a kind of
p
N 2 − sin2 θi − N 2 cos θi
rp = p (3.46) Brewster’s angle.
N 2 − sin2 θi + N 2 cos θi
7 See M. Born and E. Wolf, Principles of Optics, 7th ed., Sect. 14.2 (Cambridge University Press,
1999).
These expressions are tedious to evaluate. When evaluating the expressions, it is

usually desirable to put them into the form
r s = |r s | e i φs (3.47)
and
r p = ¯r p ¯ e i φp
¯ ¯
(3.48)
We refrain from putting (3.45) and (3.46) into this form using the general ex-
pressions; we would get a big mess. It is a good idea to let your calculator or
a computer do it after a specific value for N ≡ n + i κ is chosen. An important
point to notice is that the phases upon reflection can be very different for s and
p-polarization components (i.e. φp and φs can be very¯ different).
¯ This is true in
general, even when the reflectivity is high (i.e. |r s | and r p on the order of unity).
¯ ¯
Brewster’s angle exists also for surfaces with complex refractive index. How-
ever, in general the expressions (3.46) and (3.48) do not go to zero at any incident
angle θi . Rather, the reflection of p-polarized light can go through a minimum at
some angle θi , which we refer to as Brewster’s angle (see Fig. ¯ ¯3.9). This minimum
is best found numerically since the general expression for ¯r p ¯ in terms of n and κ
and as a function of θi can be unwieldy.
Appendix 3.A Boundary Conditions For Fields at an Inter-

face
We are interested in the continuity of fields across a boundary from one medium
with index n 1 to another medium with index n 2 . We will show that the compo-
nents of electric field and the magnetic field parallel to the interface surface must
be the same on either side (adjacent to the interface). This result is independent
of the refractive index of the materials; in the case of the magnetic field we assume
the permeability µ0 is the same on both sides. To derive the boundary conditions,
we consider a surface S (a rectangle) that is perpendicular to the interface between
the two media and which extends into both media, as depicted in Fig. 3.10.
First we examine the integral form of Faraday’s law (1.14)
d
∂
I Z
S E · d` = − B · n̂ d a (3.49)
C ∂t S
applied to the rectangular contour depicted in Fig. 3.10. We perform the path
d
integration on the left-hand side around the loop as follows:
I
E · d ` = E 1|| d − E 1⊥ `1 − E 2⊥ `2 − E 2|| d + E 2⊥ `2 + E 1⊥ `1 = E 1|| − E 2|| d (3.50)
¡ ¢
Figure 3.10 Interface of two mate-
rials.
Here, E 1|| refers to the component of the electric field in the material with index
n 1 that is parallel to the interface. E 1⊥ refers to the component of the electric field
in the material with index n 1 which is perpendicular to the interface. Similarly,
E 2|| and E 2⊥ are the parallel and perpendicular components of the electric field
3.A Boundary Conditions For Fields at an Interface 85
in the material with index n 2 . We have assumed that the rectangle is small enough
that the fields are uniform within the half rectangle on either side of the boundary.
Next, we shrink the loop down until it has zero surface area by letting the
lengths `1 and `2 go to zero. In this situation, the right-hand side of Faraday’s law
(3.49) goes to zero Z
B · n̂ d a → 0 (3.51)
S
and we are left with

E 1|| = E 2|| (3.52)
This simple relation is a general boundary condition, which is met at any material
interface. The component of the electric field that lies in the plane of the interface
must be the same on both sides of the interface.
We now derive a similar boundary condition for the magnetic field using the
integral form of Ampere’s law:8
∂E
I Z µ ¶
B · d ` = µ0 J + ²0 · n̂ d a (3.53)
∂t
C S
As before, we are able to perform the path integration on the left-hand side for
the geometry depicted in the figure, which gives
I
B · d ` = B 1|| d −B 1⊥ `1 −B 2⊥ `2 −B 2|| d +B 2⊥ `2 +B 1⊥ `1 = B 1|| − B 2|| d (3.54)
¡ ¢
The notation for parallel and perpendicular components on either side of the
interface is similar to that used in (3.50).
Again, we can shrink the loop down until it has zero surface area by letting the
lengths `1 and `2 go to zero. In this situation, the right-hand side of (3.53) goes to
zero (ignoring the possibility of surface currents):
∂E
Z µ ¶
J + ²0 · n̂ d a → 0 (3.55)
∂t
S
and we are left with

B 1|| = B 2|| (3.56)
This is a general boundary condition that must be satisfied at the material inter-
face.
8 This form can be obtained from (1.4) by integration over the surface S in Fig. 3.10 and applying
Stokes’ theorem (0.12) to the magnetic field term.

Exercises
Exercises for 3.2 The Fresnel Coefficients
P3.1 Derive the Fresnel coefficients (3.20) and (3.21) for p-polarized light.
P3.2 Verify that each of the alternative forms given in (3.18)–(3.21) are equiv-
alent (given Snell’s law). Show that at normal incidence (i.e. θi = θt = 0)
the Fresnel coefficients reduce to
nt − ni 2n i
lim r s = lim r p = − and lim t s = lim t p =
θi →0 θi →0 nt + ni θi →0 θi →0 nt + ni
P3.3 Undoubtedly the most important interface in optics is when air meets
glass. Use a computer to make the following plots for this interface as a
function of the incident angle. Use n i = 1 for air and n t = 1.6 for glass.
Explicitly label Brewster’s angle on all of the applicable graphs.
(a) r p and t p (plot together on same graph)
(b) R p and T p (plot together on same graph)
(c) r s and t s (plot together on same graph)
(d) R s and T s (plot together on same graph)
Exercises for 3.3 Reflectance and Transmittance
L3.4 (a) In the laboratory, measure the reflectance for both s and p polarized
light from a flat glass surface at about ten points. You can normalize
the detector by placing it in the incident beam of light before the glass
surface. Especially watch for Brewster’s angle (described in section 3.4).
Figure 3.11 illustrates the experimental setup. (video)
High sensitivity
detector
Slide detector
with the beam
Uncoated glass
Polarizer on rotation stage
Laser
Figure 3.11 Experimental setup for lab 3.4.
(b) Use a computer to calculate the theoretical air-to-glass reflectance

as a function of incident angle (i.e. plot R s and R p as a function of θi ).
Exercises 87
Take the index of refraction for glass to be n t = 1.54 and the index for air
to be one. Plot this theoretical calculation as a smooth line on a graph.
Plot your experimental data from (a) as points on this same graph (not
points connected by lines).
P3.5 A pentaprism is a five-sided reflecting prism used to deviate a beam of

light by 90◦ without inverting an image (see Fig. 3.12). Pentaprisms are
used in the viewfinders of SLR cameras.
(a) What prism angle β is required for a normal-incidence beam from
the left to exit the bottom surface at normal incidence?
(b) If all interfaces of the pentaprism are uncoated glass with index
n = 1.5, what fraction of the intensity would get through this system for
a normal incidence beam? Compute for p-polarized light, and include
transmission through the first and final surfaces as well as reflection at
the two interior surfaces.
NOTE: The transmission you calculate will be very poor. The reflect-
ing surfaces on pentaprisms are usually treated with a high-reflection
coating and the transmitting surfaces are treated with anti-reflection Figure 3.12
coatings.
P3.6 Show analytically for s-polarized light that R s + T s = 1, where R s is

given by (3.22) and T s is given by (3.30).
Exercises for 3.4 Brewster’s Angle
P3.7 Find Brewster’s angle for glass n = 1.5.
Exercises for 3.5 Total Internal Reflection
P3.8 Diamonds have an index of refraction of n = 2.42 which allows total in-
ternal reflection to occur at relatively shallow angles of incidence. Gem
cutters choose facet angles that ensure most of the light entering the B
top of the diamond will reflect back out to give the stone its expensive
sparkle. One such cut, the “Eulitz Brilliant" cut, is shown in Fig. 3.13. A 33.4°
18.0° 40.5°
(a) What is the critical angle for diamond?
(b) One way to spot fake diamonds is by noticing reduced brilliance in 50.6° 40.5°
the sparkle. What fraction of p-polarized light (intensity) would make

it from point A to point B in the diagram for a diamond? If a piece
of fused quartz (n = 1.46) was cut in the Eulitz Brilliant shape, what
fraction of p-polarized light (intensity) would make it from point A to Figure 3.13 A Eulitz Brilliant cut
diamond.
point B in the diagram?
(c) What is the phase shift due to reflection for s-polarized light at the
first internal reflection depicted in the figure (incident angle 40.5◦ ) in
diamond? What is the phase shift in fused quartz?
P3.9 Derive (3.40) and (3.41) and show that R s = 1 and R p = 1. HINT: See
problem P0.15.
P3.10 Compute t s and t p in the case of total internal reflection. Put your
answer in polar form (i.e. t = |t |e i φ ).
P3.11 Use a computer to plot the air-to-water transmittance as a function

of incident angle (i.e. plot (3.27) as a function of θi ). Also plot the
water-to-air transmittance on a separate graph. Plot both T s and T p on
each graph. The index of refraction for water is n = 1.33. Take the index
of air to be one.
P3.12 Light (λvac = 500 nm) reflects internally from a glass surface (n = 1.5)
surrounded by air. The incident angle is θi = 45◦ . An evanescent wave
travels parallel to the surface on the air side. At what distance from the
surface is the amplitude of the evanescent wave 1/e of its value at the
surface?
Exercises for 3.6 Reflections from Metal
P3.13 Using a computer, plot |r s |, |r p | versus θi for silver (n = 0.13 and κ =

4.05). Make a separate plot of the phases φs and φp from (3.47) and
(3.48). Clearly label each plot, and comment on how the phase shifts
are different from those experienced when reflecting from glass. 9
80
s P3.14 Find Brewster’s angle for silver (n = 0.13 and κ = 4.0) by calculating R p
and finding its minimum. You will want to use a computer program to
p
do this (Matlab, Maple, Mathematica, etc.).
Figure 3.14 Geometry for P3.15 P3.15 The complex index for silver is given by n = 0.13 and κ = 4.0. Find r s
and r p when reflecting from vacuum (n = 1, κ = 0) at θi = 80◦ and put
them into the forms (3.47) and (3.48).
9 Are you surprised that the real part of the index can be less than one?
Chapter 4
Multiple Parallel Interfaces
In chapter 3, we studied the transmission and reflection of light at a single in-

terface between two (isotropic homogeneous) materials with indices n i and n t .
We found that the percent of light reflected versus transmitted depends on the
incident angle and on whether the light is s- or p-polarized. The Fresnel coef-
ficients r s , t s , r p , t p (3.18)–(3.21) connect the reflected and transmitted fields to
the incident field, , depending on the polarization of the incident light . Similarly,
either R s and T s or R p and T p determine the fraction of incident power that either
reflects or transmits (see (3.25) and (3.27)).
In this chapter we consider the overall transmission and reflection through
multiple parallel interfaces. We start with a two-interface system, where a layer of
material is inserted between the initial and final materials. This situation occurs
frequently in optics. For example, lenses are often coated with a thin layer of
material in an effort to reduce reflections. Metal mirrors usually have a thin oxide
layer or a protective coating between the metal and the air. We can develop
reflection and transmission coefficients r tot and t tot , which apply to the overall
double-boundary system, similar to the Fresnel coefficients for a single boundary.
Likewise, we can compute an overall reflectance and transmittance R tot and T tot .
These can be used to compute the ‘tunneling’ of evanescent waves across a gap
between two parallel surfaces when the critical angle for total internal reflection
is exceeded.
The formalism we develop for the double-boundary problem is useful for
describing a simple instrument called a Fabry-Perot etalon (or interferometer if
the instrument has the capability of variable spacing between the two surfaces).
Such an instrument, which is constructed from two partially reflective parallel
surfaces, is useful for distinguishing closely spaced wavelengths.
Finally, in this chapter we will extend our analysis to multilayer coatings,
where an arbitrary number of interfaces exist between many material layers.
Multilayers are often used to make highly reflective mirror coatings from dielectric
materials (as opposed to metallic materials). Such mirror coatings can reflect
with efficiencies greater than 99.9% at certain wavelengths. In contrast, metallic
mirrors typically reflect with ∼ 96% efficiency, which can be a significant loss
89
90 Chapter 4 Multiple Parallel Interfaces
if there are many mirrors in an optical system. Dielectric multilayer coatings

also have the advantage of being more durable and less prone to damage from
high-intensity lasers.
4.1 Double-Interface Problem Solved Using Fresnel Coef-

ficients
Consider a slab of material sandwiched between two other materials as depicted
in Fig. 4.1. Because there are multiple reflections inside the middle layer, we have
dropped the subscripts i, r, and t used in chapter 3 and instead use the symbols
and to indicate forward and backward traveling waves, respectively. Let n 1
stand for the refractive index of the middle layer. For consistency with notation
that we will later use for many-layer systems, let n 0 and n 2 represent the indices
of the other two regions. For simplicity, we assume that indices are real. As with
the single-boundary problem, we are interested in finding the overall transmitted
(p) (p)
fields E 2(s) and E 2 and the overall reflected fields E 0(s) and E 0 in terms of the
(p)
incident fields E 0(s) and E 0 .
Both forward and backward traveling plane waves exist in the middle region.
Our intuition rightly tells us that in this region there are many reflections, bounc-
ing both forward and backward between the two surfaces. It might therefore seem
that we need to keep track of an infinite number of plane waves, each correspond-
ing to a different number of bounces. Fortunately, the many forward-traveling
plane waves all travel in the same direction. Similarly, the backward-traveling
plane waves are all parallel. These plane-wave fields then join neatly into a single
net forward-moving and a single net backward-moving plane wave within the
x-axis directed into page
Figure 4.1 Waves propagating through a dual interface between materials.

4.1 Double-Interface Problem Solved Using Fresnel Coefficients 91
middle region.1
As of yet, we do not know the amplitudes or phases of the net forward and net
backward traveling plane waves in the middle layer. We denote them by E 1(s) and
(p) (p)
E 1(s) or by E 1 and E 1 , separated into their s and p components as usual. Similarly,
(p) (p)
E 0(s) and E 0 as well as E 2(s) and E 2 are understood to include light that ‘leaks’
through the boundaries from the middle region. Thus, we need only concern
ourselves with the five plane waves depicted in Fig. 4.1.
The various plane-wave fields are connected to each other at the boundaries
via the single-boundary Fresnel coefficients (3.18)–(3.21). At the first surface we
define
sin θ1 cos θ0 − sin θ0 cos θ1 cos θ1 sin θ1 − cos θ0 sin θ0
r s01 ≡ r p01 ≡
sin θ1 cos θ0 + sin θ0 cos θ1 cos θ1 sin θ1 + cos θ0 sin θ0
(4.1)
2 sin θ1 cos θ0 2 cos θ0 sin θ1
t s01 ≡ t p01 ≡
The notation 0 1 indicates the first surface from the perspective of starting
on the incident side and propagating towards the middle layer. The Fresnel
coefficients for the backward traveling light approaching the first interface from
within the middle layer are given by
r s10 = −r s01 r p10 = −r p01

2 sin θ0 cos θ1 2 cos θ1 sin θ0 (4.2)
t s10 ≡ t p10 ≡
where 1 0 again indicates connections at the first interface, but from the per-
spective of beginning inside the middle layer. Finally, the single-boundary coeffi-
cients for light approaching the second interface are
sin θ2 cos θ1 − sin θ1 cos θ2 cos θ2 sin θ2 − cos θ1 sin θ1

r s12 ≡ r p12 ≡
(4.3)
2 sin θ2 cos θ1 2 cos θ1 sin θ2
t s12 ≡ t p12 ≡
In a similar fashion, the notation 1 2 indicates connections made at the second

interface from the perspective of beginning in the middle layer.
To solve for the connections between the five fields depicted in Fig.4.1, we
will need four equations for either s or p polarization (taking the incident field as
a given). To simplify things, we will consider s-polarized light in the upcoming
analysis. The equations for p-polarized light look exactly the same; just replace
the subscript s with p. Through the remainder of this section and the next, we
will continue to economize by writing the equations only for s-polarized light
with the understanding that they apply equally well to p-polarized light.
1 The sum of parallel plane waves P E e i (k·r−ωt ) , where the phase of each wave is contained in
j
j
P
E j , can be written as ( E j )e i (k·r−ωt ) , which is effectively one plane wave.
j
The forward-traveling wave in the middle region arises from both a transmis-
sion of the incident wave and a reflection of the backward-traveling wave in the
middle region at the first interface. Using the Fresnel coefficients, we can write
E 1(s) as the sum of fields arising from E 0(s) and E 1(s) as follows:
E 1(s) = t s01 E 0(s) + r s10 E 1(s) (4.4)
The factor t s01 and r s10 are the single-boundary Fresnel coefficients selected
appropriately from (4.1). Similarly, the overall reflected field E 0(s) , is given by the
reflection of the incident field and the transmission of the backward-traveling
field in the middle region according to
E 0(s) = r s01 E 0(s) + t s10 E 1(s) (4.5)
Two connections done; two to go.

Before we continue, we need to specify an origin so that we can calculate
phase shifts associated with propagation in the middle region. Propagation was
not an issue in the single-boundary problem studied back in chapter 3. However,
in the double-boundary problem, the thickness of the middle region dictates
phase variations that strongly influence the result. We take the origin to be
located on the first interface, as shown in Fig. 4.1. Since all fields in (4.4) and (4.5)
are evaluated at the origin (y, z) = (0, 0), there are no phase factors needed.
We will connect the plane-wave fields across the second interface at the point
r = ẑd . The appropriate phase-adjusted2 field at (y, z) = (0, d ) is E 1(s) e i k1 ·r =
E 1(s) e i k1 d cos θ1 , since E 1(s) is the field at the origin (y, z) = (0, 0). The transmitted
field in the final medium arises only from the forward-traveling field in the middle
region, and at our selected point it is
E 2(s) = t s12 E 1(s) e i k1 d cos θ1 (4.6)
Note that E 2(s) stand for the transmitted field at the point (y, z) = (0, d ); its local
phase can be built into its definition so no need to write an explicit phase.
The backward-traveling plane wave in the middle region arises from the
reflection of the forward-traveling plane wave in that region:
E 1(s) e −i k1 d cos θ1 = r s12 E 1(s) e i k1 d cos θ1 (4.7)
Like before, E 1(s) is referenced to the origin (y, z) = (0, 0). Therefore, the factor
e i k1 ·r = e −i k1 d cos θ1 is needed at (y, z) = (0, d ).
The relations (4.4)–(4.7) permit us to find overall transmission and reflection
coefficients for the two-interface problem.
Example 4.1
Derive the transmission coefficient that connects the final transmitted field to the
incident field for the double-interface problem according to t stot ≡ E 2(s) /E 0(s) .
2 In the middle region, k = k 1 ŷ sin θ1 + ẑ cos θ1 and k1 = k 1 ŷ sin θ1 − ẑ cos θ1 .
¡ ¢ ¡ ¢
1
4.2 Two-Interface Transmittance at Sub Critical Angles 93
Solution: From (4.6) we may write
E 2(s) −i k1 d cos θ1
E 1(s) = e (4.8)
t s12
Substitution of this into (4.7) gives
r s12
E 1(s) = E 2(s) e i k1 d cos θ1 (4.9)
t s12
Next, substituting both (4.8) and (4.9) into (4.4) yields the connection we seek
between the incident and transmitted fields:
E 2(s) −i k1 d cos θ1 r 12
12
e = t s01 E 0(s) + r s10 E 2(s) s12 e i k1 d cos θ1 (4.10)
ts ts
After rearranging, we arrive at the more useful form
E 2(s) t s01 e i k1 d cos θ1 t s12

t stot ≡ = (4.11) (p can be switched for s)
E 0(s) 1 − r s10 r s12 e 2i k1 d cos θ1
The coefficient t stot derived in Example 4.1 connects the amplitude and phase
of the incident field to the amplitude and phase of the transmitted field in a
manner similar to the single-boundary Fresnel coefficients. The numerator of
(4.11) reminds us of the physics of the situation: the field transmits through the
first interface, acquires a phase due to propagating through the middle layer, and
transmits through the second interface. The denominator of (4.11) modifies the
result to account for feedback from multiple reflections in the middle region.3
The overall reflection coefficient is found to be (see P4.1)
E 0(s) t s01 e i k1 d cos θ1 r s12 e i k1 d cos θ1 t s10

r stot ≡ = r 01
s + (4.12) (p can be switched for s)
E 0(s) 1 − r s10 r s12 e i 2k1 d cos θ1
Again the equation reminds us of the basic physics, and we did not completely
simplify the expression to make this more apparent. There is an initial reflection
from the first interface. That light is joined by light that transmits through the first
interface (looking at only the numerator of the second term), propagates through
the middle layer, reflects from the second interface, propagates back through the
middle layer, and transmits back through the first interface. The denominator of
the second term accounts for the effects of multiple-reflection feedback.
4.2 Two-Interface Transmittance at Sub Critical Angles

Often we are interested in the intensity of the light that goes through or reflects
from the double-interface setup. Because the transmission coefficient (4.11) has
3 Our derivation method avoids the need for explicit accounting of multiple reflections. For an
alternative approach arriving at the same result via an infinite geometric series, see M. Born and E.
Wolf, Principles of Optics, 7th ed., Sect. 7.6.1 (Cambridge University Press, 1999) or G. R. Fowles,
Introduction to Modern Optics, 2nd ed., Sect 4.1 (New York: Dover, 1975).
a simpler form than the reflection coefficient (4.12), it will be easier to calcu-
late the total transmittance T stot and obtain the reflectance, if desired, from the
relationship
T stot + R stot = 1 (4.13)
When the transmitted angle θ2 is real, we may write the fraction of the transmitted
power as in (3.30):
n 2 cos θ2 ¯¯ tot ¯¯2

T stot = t
n 0 cos θ0 s
(p can be switched for s) ¯ 01 ¯2 ¯ 12 ¯2 (θ2 real) (4.14)
n 2 cos θ2 ¯t ¯ ¯t ¯
s s
=
n 0 cos θ0 ¯e −i k1 d cos θ1 − r s10 r s12 e i k1 d cos θ1 ¯2
¯ ¯
(Before squaring, we multiplied the top and bottom of (4.11) by e −i k1 d cos θ1 to

make the denominator more symmetric for later convenience.) Equation (4.14)
remains valid even if the angle θ1 is complex. Thus, it can be applied to the case of
evanescent waves ‘tunneling’ through a gap where θ0 lies beyond the critical angle
for total internal reflection from the middle layer. This will be studied further in
section 4.3.
When there are no evanescent waves in any of the regions (i.e. θ0 and θ1 both
do not exceed critical angle) we can simplify (4.14) into the following useful form
(see P4.3):4
T smax
(p can be switched for s) T stot = (θ1 and θ2 real) (4.15)
1 + F s sin2 Φ2
¡ ¢
where
T s01 T s12
T smax ≡ ³ p ´2 (4.16)
1 − R s10 R s12
Φ ≡ 2k 1 d cos θ1 + δr s10 + δr s12 (4.17)

and p 10 12
4Rs Rs
Fs ≡ ³ p 10 12 ´2 (4.18)
1 − Rs Rs
The quantity T smax is the maximum possible transmittance of power through the
two surfaces. The single-interface transmittances (T s01 and T s12 ) and reflectances
(R s10 and R s12 ) are calculated from the single-interface Fresnel coefficients in
the usual way as described in chapter 3. The numerator of T smax represents the
combined transmittances for the two interfaces without considering feedback
due to multiple reflections. The denominator enhances this value to account for
reinforcing feedback in the middle layer.
The phase delay experienced by the plane wave in the middle region is de-
scribed by Φ. The term 2k 1 d cos θ1 represents the phase delay acquired during
round-trip propagation in the middle region. The terms δr s10 and δr s12 account
4.2 Two-Interface Transmittance at Sub Critical Angles 95
for possible phase shifts upon reflection from each interface. They are defined
indirectly from the single-boundary Fresnel reflection coefficients:
¯ ¯ i δ 10 ¯ ¯ i δ 1 2
r s10 = ¯r s10 ¯ e r s and r s12 = ¯r s12 ¯ e r s (4.19)
If all the indices in the double-boundary system are real, then δr s10 and δr s12
can only be zero or π (i.e. the coefficients can only be positive or negative real
numbers).
F s is called the coefficient of finesse (not to be confused with reflecting finesse
defined in section 4.6), which determines how strongly the transmittance is
influenced when Φ is varied (for example, through varying d or the wavelength
λvac ).
Example 4.2
Consider a ‘beam splitter’ designed for s-polarized light incident on a substrate of
Partial Anti-reflection
glass (n = 1.5) at 45◦ as shown in Fig. 4.2. A thin coating of zinc sulfide (n = 2.32)
reflection coating
is applied to the front of the glass to cause about half of the light to reflect. A coating
magnesium fluoride (n = 1.38) coating is applied to the back surface of the glass to
minimize reflections at that surface.5 Each coating constitutes a separate double- 46%
interface problem. The front coating is deferred to problem P4.5. In this example, 54%
find the highest transmittance possible through the antireflection film at the back
of the ‘beam splitter’ and the smallest possible d 2 that accomplishes this for light
with wavelength λvac = 633 nm. Glass
Solution: For the back coating, we have n 0 = 1.5, n 1 = 1.38, and n 2 = 1. We can
find θ0 and θ1 from θ2 = 45◦ using Snell’s law
Figure 4.2 Side view of a beam-
sin 45◦
µ ¶
n 1 sin θ1 = sin θ2 ⇒ θ1 = sin −1
= 30.82◦ splitter.
1.38
µ ◦¶
−1 sin 45
n 0 sin θ0 = sin θ2 ⇒ θ0 = sin = 28.13◦
1.5
Next we calculate the single-boundary Fresnel coefficients:
sin (θ1 − θ2 ) sin (30.82◦ − 45◦ )

r s12 = − =− = 0.253
sin (θ1 + θ2 ) sin (30.82◦ + 45◦ )
sin (θ1 − θ0 ) sin (30.82◦ − 28.13◦ )

r s10 = − =− = −0.0549
sin (θ1 + θ0 ) sin (30.82◦ + 28.13◦ )
These coefficients give us the phase shift due to reflection
δr s10 = π , δr s12 = 0
5 We ignore possible feedback between the front and rear coatings. Since the antireflection
films are usually imperfect, beam splitter substrates are often slightly wedged so that unwanted
reflections from the second surface travel in a different direction.
The single-boundary reflectances are given by

¯2
R s10 ≡ ¯r s10 ¯ = |−0.0549|2 = 0.0030
¯
¯2
R s12 ≡ ¯r s12 ¯ = |0.253|2 = 0.0640
¯
and the transmittances are
T s01 = T s10 = 1 − R s10 = 1 − 0.0030 = 0.997

T s12 = 1 − R s12 = 1 − 0.0640 = 0.936
Finally, we calculate the coefficient of finesse

p p
4 R s10 R s12 4 (0.0030) (0.0640)
F=³ =
´2 ¡ p ¢2 = 0.0570
1 − (0.0030) (0.0640)
p
1 − R s10 R s12
and the maximum transmittance

T s01 T s12 (0.997) (0.936)
T smax = ³ ´2 = ¡ p ¢2 = 0.960
1 − (0.0030) (0.0640)
p
10 12
1 − Rs Rs
Putting everything together, we have
0.960
T stot =
θ1 +π
³ ´
1 + 0.0570 sin2 2k1 d2 cos
2
The maximum transmittance occurs when the sine is zero. In that case, T stot =
0.960, meaning that 96% of the light is transmitted. We find the thickness by setting
the argument of the sine to π
2k 1 d 2 cos θ1 + π = 2π
Since k 1 = 2πn 1 /λvac , we have
λvac 633 nm
d2 = = = 134 nm
4n 1 cos θ1 4(1.38) cos 30.82◦
Without the coating, (i.e. d 2 = 0), the transmittance through the antireflection
coating would be 0.908, so the coating does give an improvement.
4.3 Beyond Critical Angle: Tunneling of Evanescent Waves

When θ0 is greater than the critical angle, an evanescent wave forms in the middle
region. In this case the formula (4.15) for the transmittance cannot be used.
Figure 4.3 Animation showing
However, the formula (4.14) still holds if the angle θ2 is real (i.e. if the critical
frustrated total internal reflection. angle in the absence of the middle layer is not exceeded). Thus, we can use (4.14)
to describe frustrated total internal reflection. In this case an evanescent wave
occurs in the middle region, and if the second surface is sufficiently close to the
first surface, the evanescent wave stimulates the second surface to produce a
4.3 Beyond Critical Angle: Tunneling of Evanescent Waves 97
transmitted wave propagating at an angle θ2 . This behavior is sometimes referred

to as ‘tunneling’.
We do not need to deal directly with the complex angle θ1 . Rather, we just need
sin θ1 and cos θ1 in order to calculate the single-boundary Fresnel coefficients.
From Snell’s law we have
n0 n2
sin θ1 = sin θ0 = sin θ2 (4.20)
n1 n1
and for the middle layer we write

q
cos θ1 = i sin2 θ1 − 1 (4.21)
Note that beyond the critical angle, sin θ1 is greater than one. We illustrate how to
apply (4.14) via a specific example:
Example 4.3
Calculate the transmittance of p-polarized light through the region between two
closely spaced 45◦ right prisms, as shown in Fig. 4.4, as a function of λvac and
the prism spacing d . Take the index of refraction of the prisms to be n = 1.5
surrounded by index n = 1, and use θ0 = θ2 = 45◦ . Neglect possible reflections
from the exterior surfaces of the prisms.
Solution: From (4.20) and (4.21) we have

p Figure 4.4 Frustrated total internal
sin θ1 = 1.5 sin 45◦ = 1.061 and cos θ1 = i 1.0612 − 1 = i 0.3536 reflection in two prisms.
We must compute various expressions involving Fresnel coefficients that appear

in (4.14):
¯2
¯
2 cos θ0 sin θ1
¯2 ¯¯ 2 p1 (1.061)
¯ 01 ¯2 ¯¯
¯ ¯ ¯
¯ 2
¯t p ¯ = ¯ ¯ = ¯¯ ¯ = 5.76
¯
cos θ1 sin θ1 + cos θ0 sin θ0 ¯ ¯ (i 0.3536) (1.061) + p1 p1 ¯
2 2
¯2
¯
2 cos θ1 sin θ2
¯2 ¯¯ 2 (i 0.3536) p1
¯ 12 ¯2 ¯¯
¯ ¯ ¯
¯ 2
¯t p ¯ = ¯ ¯ = ¯¯ ¯ = 0.640
¯
cos θ2 sin θ2 + cos θ1 sin θ1 ¯ ¯ p1 p1 + (i 0.3536) (1.061) ¯
2 2
cos θ1 sin θ1 − cos θ0 sin θ0 (i 0.3536) (1.061) − p1 p1

2 2
r p12 = − =− = e −i 1.287
cos θ1 sin θ1 + cos θ0 sin θ0 (i 0.3536) (1.061) + p1 p1
2 2
For the last step in the r p12 calculation, see problem P0.15. Also note that r p12 =
r p10 = −r p01 since n 0 = n 2 . We also need
d
µ ¶
2π 2π
k 1 d cos θ1 = d cos θ1 = d (i 0.3536) = i 2.22
λvac λvac λvac
1 We are now ready to compute the total transmittance (4.14). The factors out in
front vanish since θ0 = θ2 and n 0 = n 2 , and we have
¯ 01 ¯2 ¯ 12 ¯2
¯ ¯ ¯ ¯
¯t p ¯ ¯t p ¯
T ptot = ¯
¯e −i k1 d cos θ1 − r 10 r 12 e i k1 d cos θ1 ¯2
¯
p p
(5.76)(0.640)
=¯ h ³ í h ³ í ¯2
¯ −i i 2.22 d d
−i 1.287 e −i 1.287 e i i 2.22 λvac ¯
¯
¯e λvac − e
¯ ¯
0
0 0.5 1 1.5 2 3.69
=µ ³ ´ ³ ´ ¶µ ³ ´ ³ ´ ¶
2.22 λ d −2.22 λ d −i 2.574 2.22 λ d −2.22 λ d +i 2.574
e vac −e vac e vac −e vac
(4.22)
Figure 4.5 Plot of (4.22)
3.69
= ³
d
´ ³
d
´ ³ ´
4.44 λvac
−4.44 λvac e i 2.574 +e −i 2.574
e +e −2 2
3.69
= ³
d
´ ³
d
´
4.44 λvac
−4.44 λvac
e +e − 2 cos(2.574)
3.69
= ³
d
´ ³
d
´
4.44 λvac
−4.44 λvac
e +e + 1.69
Figure 4.5 shows a plot of the transmittance (4.22) calculated in Example 4.3.
Notice that the transmittance is 100% when the two prisms are brought together
as expected (T ptot (d /λvac = 0) = 1). When the prisms are about a wavelength apart,
the transmittance is significantly reduced, and as the distance gets large compared
to a wavelength, the transmittance quickly goes to zero (T ptot (d /λvac À 1) ≈ 0).
Maurice Paul Auguste Charles Fabry
(1867-1945, French) was born in Mar-
seille, France. At age 18, he entered the 4.4 Fabry-Perot
École Polytechnique in Paris where he
studied for two years. Following that, he
In the 1890s, Charles Fabry realized that a double interface could be used to
spent a number of years teaching state
secondary school while simultaneously distinguish wavelengths of light that are very close together. He and a talented
working on a doctoral dissertation on experimentalist colleague, Alfred Perot, constructed an instrument and began
interference phenomona. After com-
pleting his doctorate, he began working
to use it to make measurements on various spectral sources. The Fabry-Perot
as a lecturer and laboratory assistant instrument6 consists of two identical (parallel) surfaces separated by spacing d .
at the University of Marseille where a
decade later he was appointed a pro-
We can use our analysis in section 4.2 to describe this instrument. For simplicity,
fessor of physics. Soon after his arrival we choose the refractive index before the initial surface and after the final surface
to the University of Marseille, Fabry
to be the same (i.e. n 0 = n 2 ). We assume that the transmission angles are such
began a long and fruitful collaboration
with Alfred Perot (1863-1925). Fabry that total internal reflection is avoided. The transmission through the device
focused on theoretical analysis and mea- depends on the exact spacing between the two surfaces, the reflectivity of the
surements while his colleague did the
design work and construction of their surfaces, as well as on the wavelength of the light.
new interferometer, which they continu- If the spacing d separating the two parallel surfaces is adjustable (scanned),
ally improved over the years. During his
career, Fabry made signicant contribu-
the instrument is called a Fabry-Perot interferometer. If the spacing is fixed while
tions to spectroscopy and astrophysics the angle of the incident light is varied, the instrument is called a Fabry-Perot
and is credited with co-discovery of the
etalon. An etalon can therefore be as simple as a piece of glass with parallel
ozone layer. See J. F. Mulligan, Who
were Fabry and Perot?, Am. J. Phys. 6 M. Born and E. Wolf, Principles of Optics, 7th ed., Sect. 7.6.2 (Cambridge University Press, 1999).
66. 797-802 (1998).
4.4 Fabry-Perot 99
surfaces. Sometimes, a thin optical membrane called a pellicle is used as an

etalon (occasionally inserted into laser cavities to discriminate against certain
wavelengths). However, to achieve sharp discrimination between closely-spaced
wavelengths, a large spacing d is desirable.
As we previously derived (4.11), the transmittance through a double boundary
is
T max
T tot = (4.23)
1 + F sin2 Φ2
¡ ¢
In the case of identical interfaces, the transmittance and reflectance coefficients

are the same at each surface (i.e. T = T 01 = T 12 and R = R 10 = R 12 ). In this
case, the maximum transmittance and the finesse coefficient simplify to
T2
T max = (4.24) Jean-Baptiste Alfred Perot (1863-
(1 − R)2 1925, French) was born in Metz, France.
He attended the Ecole Polytechnique
and
4R and then the University of Paris, where
F= 2
(4.25) he earned a doctorate in 1888. He be-
(1 − R) came a professor in in Marseille in 1894
In principle, these equations should be evaluated for either s- or p-polarized light. where he began his collaboration with
Fabry. Perot contributed his consider-
However, a Fabry-Perot interferometer or etalon is usually operated near normal able talent of instrument fabrication to
incidence so that there is little difference between the two polarizations. the endeavor. Perot spent much of his
later career making precision astronom-
When using a Fabry-Perot instrument, one observes the transmittance T tot as ical and solar measurements. See J. F.
the parameter Φ is varied. The parameter Φ can be varied by altering d , θ1 , or λ Mulligan, Who were Fabry and Perot?,
as prescribed by Am. J. Phys. 66. 797-802 (1998).
4πn 1 d
Φ= cos θ1 + δr (4.26)
λvac
To increase the sensitivity of the instrument, it is desirable to have the transmit-
tance T tot vary strongly when Φ is varied. By inspection of (4.23), we see that T tot
d
varies strongest if the finesse coefficient F is large. We achieve a large finesse
coefficient by increasing the reflectance R.
The basic setup of a Fabry-Perot instrument is shown in Fig. 4.6. In order to
achieve a relatively high reflectivity R (and therefore large F ), special coatings can
be applied to the surfaces, for example, a thin layer of silver to achieve a partial
reflection, say 90%. Typically, two glass substrates are separated by distance d ,
Incident
with the coated surfaces facing each other as shown in the figure. The substrates light R T
Ag A
are aligned so that the interior surfaces are parallel to each other. It is typical for coatings
each substrate to be slightly wedge-shaped so that unwanted reflections from the
outer surfaces do not interfere with the double boundary situation between the Figure 4.6 Typical Fabry-Perot
two plates. setup. If the spacing d is variable,
it is called an interferometer; oth-
Technically, each coating constitutes its own double-boundary problem (or
erwise, it is called an etalon.
multiple-boundary as the case may be). We can ignore this detail and simply
think of the overall setup as a single two-interface problem. Regardless of the
details of the coatings, we can say that each coating has a certain reflectance R
and transmittance T . However, as light goes through a coating, it can also be
attenuated because of absorption. In this case, we have
R +T + A = 1 (4.27)
where A represents the amount of light absorbed at a coating. The attenuation A

reduces the amount of light that makes it through the instrument, but it does not
impact the nature of the interferences within the instrument.
The total transmittance T tot (4.23) through an ideal Fabry-Perot instrument is
depicted in Fig. 4.7 as a function of Φ. The various curves correspond to different
values of F . Typical values of Φ can be extremely large. For example, suppose
that the instrument is used at near-normal incidence (i.e. cos θ1 ∼ = 1) with a
wavelength of λvac = 500 nm and an interface separation of d 0 = 1 cm. From
0 (4.26) the value of Φ (ignoring the constant phase terms δr ) is approximately
0 2p 4p
4π(1 cm)
Φ0 = = 80, 000π
Figure 4.7 Transmittance as the 500 nm
phase Φ is varied. The different As we vary d , λ, or θ1 by small amounts, we can easily cause Φ to change by 2π as
curves correspond to different
depicted in Fig. 4.7. The figure shows small changes in Φ above a value Φ0 , which
values of the finesse coefficient.
represents a large multiple of 2π.
Φ0 represents a large multiple of
2π. The reflection phase δr in (4.26) depends on the exact nature of the coatings
in the Fabry-Perot instrument. However, we do not need to know the value of δr
(depending on both the complex index of the coating material and its thickness).
Whatever the value of δr , we only care that it is constant. Experimentally, we can
always compensate for the δr by ‘tweaking’ the spacing d , whose exact value is
likely not controlled for in the first place. Note that the required ‘tweak’ on the
Actuated
Substrate spacing need only be a fraction of a wavelength, which is typically tiny compared
Detector
to the overall spacing d .
Collimated
Interferometer
Light
Angle
Adjustment
Aperture
4.5 Setup of a Fabry-Perot Instrument
Figure 4.8 shows the typical experimental setup for a Fabry-Perot interferometer.
Trig Sig
A collimated beam of light is sent through the instrument. The beam is aligned so
Oscilloscope that it is normal to the surfaces. It is critical for the two surfaces of the interferom-
Figure 4.8 Setup for a Fabry-Perot eter to be extremely close to parallel. When aligned correctly, the transmission
interferometer. of a collimated beam will ‘blink’ all together as the spacing d is changed (by tiny
amounts). A mechanical actuator can be used to vary the spacing between the
plates while the transmittance is observed on a detector. To make the alignment
of the instrument somewhat less critical, a small aperture can be placed in front
Transmittance
of the detector so that it observes only a small portion of the beam.

The transmittance as a function of plate separation is shown in Fig. 4.9. In this
case, Φ varies via changes in d (see (4.26) with cos θ1 = 1 and fixed wavelength).
As the spacing is increased by only a half wavelength, the transmittance changes
0 through a complete period. The various peaks in the figure are called fringes.
0
The setup for a Fabry-Perot etalon is similar to that of the interferometer
except that the spacing d remains fixed. Often the two surfaces in the etalon are
Figure 4.9 Transmittance as the held parallel to each other by a precision spacer. An advantage to the Fabry-Perot
separation d is varied (F = 100).
etalon (as opposed to the interferometer) is that no moving parts are needed. To
d 0 represents a large distance for
make measurements with an etalon, the angle of the light is varied rather than the
which Φ is a multiple of 2π.
plate separation. After all, to see fringes, we just need to cause Φ in (4.23) to vary
Etalon
in some way. According to (4.26), we can do that as easily by varying θ1 as we can Point
Source
by varying d . One way to obtain a range of angles is to observe light from a ‘point
source’, as depicted in Fig. 4.10. Different portions of the beam go through the
device at different angles. When aligned straight on, the transmitted light forms a Angle
Adjustment Screen
‘bull’s-eye’ pattern on a screen.
In Fig. 4.11 we graph the transmittance T tot (4.23) as a function of angle Figure 4.10 A diverging monochro-
(holding λvac = 500 nm and d = 1 cm fixed). Since cos θ1 is not a linear function, matic beam traversing a Fabry-
the spacing of the peaks varies with angle. As θ1 increases from zero, the cosine Perot etalon. (The angle of diver-
steadily decreases, causing Φ to decrease. Each time Φ decreases by 2π we get a gence is exaggerated.)
new peak. Not surprisingly, only a modest change in angle is necessary to cause
the transmittance to vary from maximum to minimum, or vice versa.
The bull’s-eye pattern in Fig. 4.10 can be understood as the curve in Fig. 4.11
rotated about a circle. Depending on the exact spacing between the plates, the
Transmission
radii (or angles) where the fringes occur can be different. For example, the center
spot could be dark.
Spectroscopic samples often are not compact point-like sources. Rather, they
are extended diffuse sources. The point-source setup shown in Fig. 4.10 won’t
work for extended sources unless all of the light at the sample is blocked except 0
0 5 10 15
for a tiny point. This is impractical if there remains insufficient illumination at
the final screen for observation.
Figure 4.11 Transmittance
In order to preserve as much light as possible, we can sandwich the etalon through a Fabry-Perot etalon
between two lenses. We place the diffuse source at the focal plane of the first lens. (F = 10) as the angle θ1 is varied. It
We place the screen at the focal plane of the second lens. This causes an image of is assumed that the distance d is
the source to appear on the screen.7 Each point of the diffuse source is mapped chosen such that Φ is a multiple of
to a corresponding point on the screen. Moreover, the light associated with any 2π when the angle is zero.
particular point of the source travels as a unique collimated beam in the region Diffuse
Source Screen
between the lenses. Each collimated beam traverses the etalon with a unique Lens Etalon Lens
angle. Thus, light associated with each emission point traverses the etalon with
higher or lower transmittance, according to the differing angles. The result is that
a bull’s eye pattern becomes superimposed on the image of the diffuse source.
The lens and retina of your eye can be used for the final lens and screen.
Figure 4.12 Setup of a Fabry-Perot

4.6 Distinguishing Nearby Wavelengths in a Fabry-Perot etalon for looking at a diffuse
Instrument source.
Thus far, we have examined how the transmittance through a Fabry-Perot instru-
ment varies with surface separation d and angle θ1 . However, the main purpose
of a Fabry-Perot instrument is to measure small changes in the wavelength of
light, which similarly affect the value of Φ (see (4.26)).8
7 If the diffuse source has the shape of Mickey Mouse, then an image of Mickey Mouse appears
on the screen. Imaging techniques are discussed in chapter 9.

Consider a Fabry-perot interferometer where the transmittance through the

instrument is plotted as a function of surface separation d . Let the spacing d 0
correspond to the case when Φ is a multiple of 2π for the wavelength λvac = λ0 .
Next, suppose we adjust the wavelength of the light from λvac = λ0 to λvac =
λ0 + ∆λ while observing the transmittance. As we do this, the value of Φ changes.
Figure 4.13 shows what happens as we scan the spacing d of the interferometer in
the neighborhood of d 0 . The dashed line corresponds to a different wavelength.
As the wavelength changes, the plate separation at which a particular fringe
occurs also changes.
We now derive the connection between a change in wavelength and the
amount that Φ changes, which gives rise to the fringe shift seen in Fig. 4.13. At
Transmittance
the wavelength λ0 , we have
4πn 1 d 0 cos θ1
Φ0 = + δr (4.28)
λ0
which we previously supposed is an integer times 2π. At a new wavelength (all
0
0 else remaining the same) we have
4πn 1 d 0 cos θ1
Φ= + δr (4.29)
Figure 4.13 Transmittance as the λ0 + ∆λ
spacing d is varied for two differ-
ent wavelengths (F = 100). The The change in wavelength ∆λ is usually very small compared to λ0 , so we can
solid line plots the transmittance represent the denominator with the first two terms of a Taylor-series expansion:
of light with a wavelength of λ0 ,
and the dashed line plots the 1 1 ∼ 1 − ∆λ/λ0
= = (4.30)
transmittance of a wavelength λ0 + ∆λ λ0 (1 + ∆λ/λ0 ) λ0
shorter than λ0 . Note that the
fringes shift positions for different Then the difference between Φ0 and Φ can be rewritten as
wavelengths. 4πn 1 d 0 cos θ1
∆Φ ≡ Φ0 − Φ = ∆λ (4.31)
λ20
If the change in wavelength is enough to cause ∆Φ = 2π, the fringes in Fig. 4.13
shift through a whole period, and the picture looks the same.
This brings up an important limitation of the instrument. If the fringes shift
by too much, we might become confused as to what exactly has changed, owing
to the periodic nature of the fringes. If two wavelengths aren’t sufficiently close,
the fringes of one wavelength may be shifted past several fringes of the other
wavelength, and we will not be able to tell by how much they differ.
This introduces the concept of free spectral range, which is the wavelength
change ∆λFSR that causes the fringes to shift through one period. We find this by
setting (4.31) equal to 2π. After rearranging, we get
λ2vac
∆λFSR = (4.32)
2n 1 d 0 cos θ1
The free spectral range tends to be extremely narrow; a Fabry-Perot instrument is
not well suited for measuring wavelength ranges wider than this. In summary, the
free spectral range is the largest change in wavelength permissible while avoiding
confusion. To convert this wavelength difference ∆λFSR into a corresponding
frequency difference, one differentiates ν = c/λvac to get
c∆λFSR
|∆νFSR | = (4.33)
λ2vac
Example 4.4
A Fabry-Perot interferometer has plate spacing d 0 = 1 cm and index n 1 = 1. If it
is used in the neighborhood of λvac = 500 nm, find the free spectral range of the
instrument.
Solution: From (4.32), the free spectral range is
λ2vac (500 nm)2

∆λFSR = = ∆λFSR = = 0.0125 nm
2n 1 d 0 cos θ1 2 (1) (1 cm) cos 0◦
This means that we should not use the instrument to distinguish wavelengths that
are separated by more than this small amount.
We next consider the smallest change in wavelength that can be noticed,

or resolved with a Fabry-Perot instrument. For example, if two very near-by
wavelengths are sent through the instrument simultaneously, we can distinguish
them only if the separation between their corresponding fringe peaks is at least
as large as the width of individual peaks. This situation of two barely resolvable
fringe peaks is illustrated in Fig. 4.14 for a diverging beam traversing an etalon.
We will look for the wavelength change that causes a peak to shift by its own
width. We define the width of a peak by its full width at half maximum (FWHM).
Again, let Φ0 be a multiple of 2π so that a peak in transmittance occurs when
Φ = Φ0 . In this case, we have from (4.23) that
Figure 4.14 Transmittance of a
T max diverging beam through a Fabry-
T tot = ´ = T max (4.34) Perot etalon. Two nearby wave-
sin2 Φ20
³
1+F
lengths are sent through the in-
strument simultaneously, (top)
since sin (Φ0 /2) = 0. If Φ varies from Φ0 to Φ0 ± ΦFWHM /2, then, by definition, the
barely resolved and (bottom) eas-
transmittance drops to one half. Therefore, we may write ily resolved.
T max T max
T tot = ´= (4.35)
Φ0 ±ΦFWHM /2
³
1 + F sin2 2
2
In solving for (4.35) for ΦFWHM , we see that this equation requires
2 ΦFWHM
µ ¶
F sin =1 (4.36)
4
where we have taken advantage of the fact that Φ0 is a multiple of 2π. Next,
we suppose that ΦFWHM is rather small so that we may represent the sine by its
argument. This approximation is okay if the finesse coefficient F is rather large

(say, 100). With this approximation, (4.36) simplifies to
4
ΦFWHM ∼
=p . (4.37)
F
The ratio of the period between peaks 2π to the width ΦFWHM of individual peaks
is called the reflecting finesse (or just finesse).
p
2π π F
f ≡ = (4.38)
ΦFWHM 2
This parameter is often used to characterize the performance of a Fabry-Perot

instrument. Note that a higher finesse f implies sharper fringes in comparison to
the fringe spacing.
The free spectral range ∆λFSR compared to the minimum wavelength ∆λFWHM
is the same ratio f . Therefore, we have
∆λFSR λ2vac
∆λFWHM = = p (4.39)
f πn 1 d 0 cos θ1 F
As a final note, the ratio of λ0 to ∆λmin , where ∆λmin is the minimum change
of wavelength that the instrument can distinguish in the neighborhood of λ0 is
called the resolving power. For a Fabry-Perot instrument it is
λ0
RP ≡ (4.40)
∆λFWHM
Fabry-Perot instruments tend to have very high resolving powers since they re-
spond to very small differences in wavelength.
Example 4.5
If the Fabry-Perot interferometer in Example 4.4 has reflectivity R = 0.85, find the
finesse, the minimum distinguishable wavelength separation, and the resolving
power.
Solution: From (4.25), the finesse coefficient is

4R 4 (0.85)
F= = = 151
(1 − R)2 (1 − (0.85))2
and by (4.38) the finesse is

p p
π F π 151
f = = = 19.3
2 2
The minimum resolvable wavelength change is then
∆λFSR 0.0125 nm
∆λFWHM = = = 0.00065 nm (4.41)
f 19
4.7 Multilayer Coatings 105
The instrument can distinguish two wavelengths separated by this tiny amount,
which gives an impressive resolving power of
λvac 500 nm
RP = = = 772, 000
∆λFWHM 0.00065 nm
For comparison, the resolving power of a typical grating spectrometer is much less
(a few thousand). However, a grating spectrometer has the advantage that it can
simultaneously observe wavelengths over hundreds of nanometers, whereas the
Fabry-Perot instrument is confined to the extremely narrow free spectral range.
4.7 Multilayer Coatings

As we saw in Example 4.2, a single coating cannot always accomplish a desired
effect, especially if the goal is to make a highly reflective mirror. For example, if
we want to make a mirror surface using a dielectric (i.e. nonmetallic) coating,
a single layer is insufficient to reflect the majority of the light. In P4.5 we find
that a single dielectric layer deposited on glass can reflect at most about 46%
of the light, even when we used a material with very high index. We would like
to do much better (e.g. >99%), and this can be accomplished with multilayer
dielectric coatings. Multilayer dielectric coatings can perform considerably better
than metal surfaces such as silver and have the advantage of being less prone to
damage.
In this section, we develop the formalism for dealing with arbitrary numbers
of parallel interfaces (i.e. multilayer coatings).9 Rather than incorporate the single-
interface Fresnel coefficients into the problem as we did in section 4.1, we will
find it easier to return to the fundamental boundary conditions for the electric
and magnetic fields at each interface between the layers.
We examine p-polarized light incident on an arbitrary multilayer coating (all
interfaces parallel to each other). It is left as an exercise to re-derive the formalism
for s-polarized light (see P4.13). The upcoming derivation is valid also for complex
refractive indices, although our notation suggests real indices. The ability to deal
with complex indices is very important if, for example, we want to make mirror
coatings work in the extreme ultraviolet wavelength range where virtually every
material is absorptive. Consider the diagram of a multilayer coating in Fig. 4.15
for which the angle of light propagation in each region may be computed from
Snell’s law:
n 0 sin θ0 = n 1 sin θ1 = · · · = n N sin θN = n N +1 sin θN +1 (4.42)
where N denotes the number of layers in the coating. The subscript 0 represents
the initial medium outside of the multilayer, and the subscript N + 1 represents
the final material, or the substrate on which the layers are deposited.
9 G. R. Fowles, Introduction to Modern Optics, 2nd ed., Sect 4.4 (New York: Dover, 1975); E. Hecht,
Optics, 3rd ed., Sect. 9.7.1 (Massachusetts: Addison-Wesley, 1998).

z-direction
Figure 4.15 Light propagation through multiple layers.
In each layer, only two plane waves exist, each of which is composed of light
arising from the many possible bounces from various layer interfaces. The arrows
pointing right indicate plane wave fields in individual layers that travel roughly
in the forward (incident) direction, and the arrows pointing left indicate plane
wave fields that travel roughly in the backward (reflected) direction. In the final
(p)
region, there is only one plane wave traveling with a forward direction (E N +1 )
which gives the overall transmitted field.
As we have studied in chapter 3 (see (3.9) and (3.13)), the boundary conditions
for the parallel components of the E field and for the parallel components of the
B field lead respectively to
cos θ0 E 0 + E 0 = cos θ1 E 1 + E 1
¡ (p) (p)
¢ ¡ (p) (p)
¢
(4.43)
and
¡ (p) (p)
¢ ¡ (p) (p)
¢
n 0 E 0 − E 0 = n 1 E 1 − E 1 (4.44)
Similar equations give the field connection for s-polarized light (see (3.8) and
(3.14)).
We have applied these boundary conditions at the first interface only. Of
course there are many more interfaces in the multilayer. For the connection
between the j th layer and the next, we may similarly write
³ ´
cos θ j E j e i k j ` j cos θ j + E j e −i k j ` j cos θ j = cos θ j +1 E j +1 + E j +1
(p) (p)
¡ (p) (p)
¢
(4.45)
and ³ ´
n j E j e i k j ` j cos θ j − E j e −i k j ` j cos θ j = n j +1 E j +1 − E j +1
(p) (p)
¡ (p) (p)
¢
(4.46)
Here we have set the origin within each layer at the left surface. Then when
making the connection with the subsequent layer at the right surface, we must
specifically take into account the phase k j · ` j ẑ = k j ` j cos θ j . This corresponds
¡ ¢
to the phase acquired by the plane wave field in traversing the layer with thickness
` j . The right-hand sides of (4.45) and (4.46) need no phase adjustment since the
( j + 1)th field is evaluated on the left side of its layer.
4.7 Multilayer Coatings 107
At the final interface, the boundary conditions are

³ ´
cos θN E N e i k N `N cos θN + E N e −i k N `N cos θN = cos θN +1 E N +1
(p) (p) (p)
(4.47)
and ³ ´
n N E N e i k N `N cos θN − E N e −i k N `N cos θN = n N +1 E N +1
(p) (p) (p)
(4.48)
since there is no backward-traveling field in the final medium.

At this point we are ready to solve (4.43)–(4.48). We would like to eliminate
(p) (p) (p)
all fields besides E 0 , E 0 , and E N +1 . Then we will be able to find the overall
reflectance and transmittance of the multilayer coating. In solving (4.43)–(4.48),
we must proceed with care, or the algebra can quickly get out of hand. Fortunately,
you have probably had training in linear algebra, and this is a case where that
training pays off.
We first write a general matrix equation that summarizes the mathematics in
(4.43)–(4.48), as follows:
cos θ j e i β j cos θ j e −i β j cos θ j +1 cos θ j +1

· ¸· (p) ¸ · ¸· (p) ¸
E j E j +1
= (4.49)
n j ei βj −n j e −i β j
(p) (p)
Ej n j +1 −n j +1 E j +1
where ½
0 j =0
βj ≡ (4.50)
k j ` j cos θ j 1≤ j ≤N
and
(p)
E N +1 ≡ 0 (4.51)
(It would be good to take a moment to convince yourself that this set of matrix
equations properly represents (4.43)–(4.48) before proceeding.) We rewrite (4.49)
as
¸−1 ·
cos θ j e i β j cos θ j e −i β j cos θ j +1 cos θ j +1
· (p) ¸ · ¸· (p) ¸
E j E j +1
=
n j ei βj −n j e −i β j
(p) (p)
Ej n j +1 −n j +1 E j +1
(4.52)
Keep in mind that (4.52) represents a distinct matrix equation for each differ-
ent j . We can substitute the j = 1 equation into the j = 0 equation to get
¸−1
cos θ0 cos θ0 cos θ2 cos θ2
· (p) ¸ · · ¸· (p) ¸
E 0 (p) E 2
(p) = M1 (p) (4.53)
E0 n0 −n 0 n2 −n 2 E2
where we have grouped the matrices related to the j = 1 layer together via
¸−1
cos θ1 cos θ1 cos θ1 e i β1 cos θ1 e −i β1
· ¸·
(p)
M1 ≡ (4.54)
n1 −n 1 n 1 e i β1 −n 1 e −i β1
We can continue to substitute into this equation progressively higher order equa-
tions (i.e. for j = 2, j = 3, ... ) until we reach the j = N layer. All together this will
give
¸−1 Ã N
!·
cos θ0 cos θ0 cos θN +1 cos θN +1
· (p) ¸ · (p) ¸ ¸·
E 0 E N +1
Y (p)
(p) = Mj
E0 n0 j =1
−n 0 0 n N +1 −n N +1
(4.55)
where the matrices related to the j th layer are grouped together according to
¸−1
cos θ j cos θ j cos θ j e i β j cos θ j e −i β j
· ¸·
(p)
Mj ≡
nj −n j n j ei βj −n j e −i β j
(4.56)
cos β j −i sin β j cos θ j /n j
· ¸
=
−i n j sin β j / cos θ j cos β j
The matrix inversion in the first line was performed using (0.35). The symbol Π
signifies the product of the matrices with the lowest subscripts on the left:
N
Y (p) (p) (p) (p)
M j ≡ M1 M2 · · · M N (4.57)
j =1
(p)
As a finishing touch, we divide (4.55) by the incident field E 0 as well as perform
the matrix inversion on the right-hand side to obtain
(p)
· ¸ · ± (p) ¸
1 (p) E N +1 E 0
(p)
± (p) =A (4.58)
E0 E 0 0
where
¸Ã N
!·
cos θ0 cos θN +1
· (p) (p) ¸ · ¸
(p) a 11 a 12 1 n0 0 Y (p)
A ≡ = Mj
− cos θ0
(p) (p)
a 21 a 22 2n 0 cos θ0 j =1
n0 0 n N +1
(4.59)
In the final matrix in (4.59) we have replaced the entries in the right column with
zeros. This is permissable since it operates on a column vector with zero in the
bottom component.
Equation (4.58) represents two equations, which must be solved simultane-
(p) (p) (p) (p)
ously to find the ratios E 0 /E 0 and E N +1 /E 0 . Once the matrix A (p) is computed,
this is a relatively simple task:
(p)
E N +1 1
tp ≡ (p)
= (p)
(Multilayer) (4.60)
E 0 a 11
(p) (p)
E0 a 21
rp ≡ (p)
= (p)
(Multilayer) (4.61)
E 0 a 11
The convenience of this notation lies in the fact that we can deal with an
arbitrary number of layers N with varying thickness and index. The essential
information for each layer is contained succinctly in its respective 2 × 2 char-
acteristic matrix M . To find the overall effect of the many layers, we need only
4.8 Repeated Multilayer Stacks 109
multiply the matrices for each layer together to find A from which we compute
the reflection and transmission coefficients for the whole system.
The derivation for s-polarized light is similar to the above derivation for p-
polarized light. The equation corresponding to (4.58) for s-polarized light turns
out to be · ¸ · (s) ± (s) ¸
1 (s) E N +1 E 0
(s)
± (s) =A (4.62)
E 0 E 0 0
where
¸Ã N
!·
n 0 cos θ0
· (s) (s)
¸ · ¸
(s) a 11 a 12 1 1 Y (s) 1 0
A ≡ = Mj
(s)
a 21 (s)
a 22 2n 0 cos θ0 n 0 cos θ0 −1 j =1
n N +1 cos θN +1 0
(4.63)
and
cos β j −i sin β j /(n j cos θ j )
· ¸
M (s) = (4.64)
j −i n j cos θ j sin β j cos β j
The transmission and reflection coefficients are found (as before) from
E N(s)+1 1
ts ≡ (s)
= (s) (Multilayer) (4.65)
E 0 a 11
E 0(s) (s)
a 21
rs ≡ = (Multilayer) (4.66)
E 0(s) a 11
(s)
4.8 Repeated Multilayer Stacks

Many different types of multilayer coatings are possible. For example, a Brewster’s-
angle polarizer has a coating designed to transmit with high efficiency p-polarized
light while simultaneously reflecting s-polarized light with high efficiency. The
backside of the substrate is left uncoated where p-polarized light passes with
100% efficiency at Brewster’s angle.
...
substrate
Sometimes multilayer coatings are made with repeated stacks of layers. In
general, if the same series of layers in (4.69) is repeated many times, say q times,
Sylvester’s theorem (see appendix 0.3) can come in handy:
¸q
A sin qθ − sin q − 1 θ
· · ¡ ¢ ¸
A B 1 B sin qθ
= (4.67) Figure 4.16 A repeated multilayer
D sin qθ − sin q − 1 θ
¡ ¢
C D sin θ C sin qθ
structure with alternating high
where and low indexes where each layer
1 is a quarter wavelength in thick-
cos θ ≡
(A + D) . (4.68)
2 ness. This structure can achieve
This formula relies on the condition AD − BC = 1, which is true for matrices of very high reflectance.
the form (4.56) and (4.64) or any product of them. Here, A, B , C , and D represent
the elements of a matrix composed of a block of matrices corresponding to a
repeated pattern within the stack.
In general, high-reflection coatings are designed with alternating high and
low refractive indices. For high reflectivity, each layer should have a quarter-
wave thickness. Since the layers alternate high and low indices, at every other
boundary there is a phase shift of π upon reflection from the interface. Hence,
the quarter wavelength spacing is appropriate to give constructive interference in
the reflected direction.
Example 4.6
Derive the reflection and transmission coefficients for p polarized light interacting
with a high reflector constructed using a λ/4 stack.
Solution: For a λ/4 stack we need

π
βj =
2
This amounts to a thickness requirement of
λvac
`j =
4n j cos θ j
In this situation, the matrix (4.56) for each layer simplifies to
−i cos θ j /n j
· ¸
(p) 0
Mj =
−i n j / cos θ j 0
The matrices for a high and a low refractive index layer are multiplied together in
the usual manner. Each layer pair takes the form
θH θL
# " n cos θ
− i cos − i cos
" #" #
0 nH 0 nL
− nL cos θH 0
= H L
i nH i nL cos θL
− cos θH 0 − cos θL 0 0 − nn Hcos θ L H
To extend to q = N /2 identical layer pairs, we have

" n cos θ #q
N − nL cos θH 0
(p)
Y
Mj = H L
cos θL
j =1 0 − nn Hcos θH
L
n L cos θH q
 ³ ´ 
− n cos θ 0
H L
= ³
cos θL q
´ 
0 − nn Hcos θ L H
Substituting this into (4.59), we obtain

n L cos θH q cos θN +1 n H cos θL q
 ³ ´ ³ ´ 
n N +1
1 − n cos θ cos θ + − n cos θ n0 0
A (p) =  ³ n Hcos θ L ´q cos θ 0 ³ L H ´
n H cos θL q n N +1
 (4.69)
2 − nL cos θH N +1
− − 0
H cos θ0
L n cos θ L H n0
With A (p) in hand, we can now calculate the transmission coefficient from (4.60)
1
tp = ³ (λ/4 stack, p-polarized) (4.70)
θH q cos θN +1 cos θL q
´ ³ ´
n N +1
− nnL cos
cos θ cos θ0 + − nn Hcos θ n0
H L L H
and the reflection coefficient from (4.61)

θH q cos θN +1 n H cos θL q
³ ´ ³ ´
n N +1
− nnL cos − −
H cos θL cos θ0 n L cos θH n0
rp = ³ (λ/4 stack, p-polarized) (4.71)
n L cos θH q cos θN +1 n H cos θL q
´ ³ ´
n N +1
− n cos θ cos θ0 + − n cos θ n0
H L L H
4.8 Repeated Multilayer Stacks 111
The quarter-wave multilayer considered in Example 4.6 can achieve extraor-

dinarily high reflectivity. In the limit of q → ∞, we have t p → 0 and r p → −1 (see
Fig. 4.17), giving 100% reflection with a π phase shift. 0
t
-0.5
r
-1
0 5 10
q
Figure 4.17 The transmission and

reflection coefficients for a quarter
wave stack as q is varied (n L = 1.38
and n H = 2.32).
Exercises
Exercises for 4.1 Double-Interface Problem Solved Using Fresnel Coefficients
P4.1 Use (4.4)–(4.7) to derive r stot given in (4.12).
P4.2 Consider a 1 micron thick coating of dielectric material (n = 2) on

a piece of glass (n = 1.5). Use a computer to plot the magnitude of
the overall Fresnel coefficient (4.11) from air into the glass at normal
incidence. Plot as a function of wavelength for wavelengths between
200 nm and 800 nm (assume the index remains constant over this
range).
Exercises for 4.2 Two-Interface Transmittance at Sub Critical Angles
P4.3 Verify that in the case that θ1 and θ2 are real that (4.14) simplifies to
(4.15).
P4.4 A light wave impinges at normal incidence on a thin glass plate with
index n and thickness d .
(a) Show that the transmittance through the plate as a function of
wavelength is
1
T tot = 2
1 + ( 4n 2 ) sin2 2πnd
2
³ ´
n −1
λvac
HINT: Find
n −1
r 12 = r 10 = −r 01 =
n +1
and then use
T 01 = 1 − R 01
T 12 = 1 − R 12
(b) If n = 1.5, what is the maximum and minimum transmittance
through the plate?
(c) If the plate thickness is d = 150 µm, what wavelengths transmit with
maximum efficiency? Express your answer as a formula involving an
integer j .
P4.5 Show that the maximum reflectance possible from the front coating in
Example 4.2 is 46%. Find the smallest possible d 1 that accomplishes
this for light with wavelength λvac = 633 nm.
Exercises 113
Exercises for 4.3 Beyond Critical Angle: Tunneling of Evanescent Waves
P4.6 Re-compute (4.22) in the case of s-polarized light. Write the result in
the same form as the last expression in (4.22).
L4.7 Consider s-polarized microwaves (λvac = 3 cm) encountering an air

gap separating two paraffin wax prisms (n = 1.5). The 45◦ right-angle
prisms are arranged with the geometry shown in Fig. 4.4. The presence
of the second prism ‘frustrates’ the total internal reflection.
Microwave Paraffin Paraffin Paraffin Microwave

Source Lens Prisms Lens Detector
Figure 4.19
Separation (cm)
(a) Use a computer to plot the transmittance through the gap (i.e. the
result of P4.6) as a function of separation d (normal to gap surface). Figure 4.18 Theoretical vs. mea-
Neglect reflections from other surfaces of the prisms. sured microwave transmission
through wax prisms. Mismatch is
(b) Measure the transmittance of the microwaves through the prisms
presumably due to imperfections
as function of spacing d (normal to the surface) and superimpose the
in microwave collimation and/or
results on the graph of part (a). Figure 4.18 shows a plot of typical data extraneous reflections.
taken with this setup. (video)
Exercises for 4.6 Distinguishing Nearby Wavelengths in a Fabry-Perot Instru-

ment
P4.8 A Fabry-Perot interferometer has silver-coated plates each with re-

flectance R = 0.9, transmittance T = 0.05, and absorbance A = 0.05.
The plate separation is d = 0.5 cm with interior index n 1 = 1. Suppose
that the wavelength being observed near normal incidence is 587 nm.
(a) What is the maximum and minimum transmittance through the
interferometer?
(b) What are the free spectral range ∆λFSR and the fringe width ∆λFWHM ?
(c) What is the resolving power?
P4.9 Generate a plot like Fig. 4.11(a), showing the fringes you get in a Fabry-
Perot etalon when θ1 is varied. Let Tmax = 1, F = 10, λ = 500 nm,
d = 1 cm, and n 1 = 1.
(a) Plot T vs. θ1 over the angular range used in Fig. 4.11(a).
(c) Suppose d was slightly different, say 1.00001 cm. Make a plot of T
vs θ1 for this situation.
P4.10 Consider the configuration depicted in Fig. 4.10, where the center of the
diverging light beam λvac = 633 nm approaches the plates at normal
incidence. Suppose that the spacing of the plates (near d = 0.5 cm) is
just right to cause a bright fringe to occur at the center. Let n 1 = 1. Find
the angle for the m th circular bright fringe surrounding the central spot
(the 0th fringe corresponding to the center). HINT: cos θ ∼ = 1−θ 2 /2. The
p
answer has the form a m; find the value of a.
L4.11 Characterize a Fabry-Perot etalon in the laboratory using a HeNe laser

(λvac = 633 nm). Assume that the bandwidth ∆λHeNe of the HeNe laser
is very narrow compared to the fringe width of the etalon ∆λFWHM .
Assume two identical reflective surfaces separated by 5.00 mm. Deduce
the free spectral range ∆λFSR , the fringe width ∆λFWHM , the resolving
power, and the reflecting finesse (small f ). (video)
Diverging Lens
Laser
Filter
Fabry-Perot CCD
Etalon Camera
Figure 4.20
L4.12 Use the same Fabry-Perot etalon to observe the Zeeman splitting of the
N yellow line λ = 587.4 nm emitted by a krypton lamp when a magnetic
Filter
field is applied. As the line splits and moves through half of the free
spectral range, the peak of the decreasing wavelength and the peak of
CCD the increasing wavelength meet on the screen. When this happens, by
S Fabry-Perot Camera
Etalon how much has each wavelength shifted? (video)
Figure 4.21
Exercises for 4.7 Multilayer Coatings
P4.13 (a) Write (4.43) through (4.48) for s-polarized light.

(b) From these equations, derive (4.62)–(4.64).
P4.14 Show that (4.65) for a single layer (i.e. two interfaces), is equivalent to
(4.11). WARNING: This is more work than it may appear at first.
Exercises for 4.8 Repeated Multilayer Stacks
P4.15 (a) What should be the thickness of the high and the low index layers in
a periodic high-reflector mirror? Let the light be p-polarized and strike
Exercises 115
the mirror surface at 45◦ . Take the indices of the layers be n H = 2.32
and n L = 1.38, deposited on a glass substrate with index n = 1.5. Let
the wavelength be λvac = 633 nm.
(b) Find the reflectance R with 1, 2, 4, and 8 periods in the high-low
stack.
P4.16 Find the high-reflector matrix for s-polarized light that corresponds to
(4.69).
P4.17 Design an anti-reflection coating for use in air (assume the index of air
is 1):
(a) Show that for normal incidence and λ/4 films (thickness= 14 the
wavelength of light inside the material), the reflectance of a single layer
(n 1 ) coating on a glass is
!2
n g − n 12
Ã
R=
n g + n 12
(b) Show that for a two coating setup (air-n 1 -n 2 -glass; n 1 and n 2 are
each a λ/4 film), that !2
n 22 − n g n 12
Ã
R= 2
n 2 + n g n 12
(c) If n g = 1.5, and you have a choice of these common coating ma-
terials: ZnS (n = 2.32), CeF (n = 1.63) and MgF (n = 1.38), find the
combination that gives you the lowest R for part (b). (Be sure to specify
which material is n 1 and which is n 2 .) What R does this combination
give?
P4.18 Consider a two-coating ‘anti-reflection optic’ (each coating set for

λ/4, as in problem P4.17) using n 1 = 1.6 and n 2 = 2.1 applied to a
glass substrate n g = 1.5 at normal incidence. Suppose the coating
thicknesses are optimized for λ = 550 nm (in the middle of the visible
range) and ignore possible variations of the indices with λ. Use the
matrix techniques and a computer to plot R(λair ) for 400 to 700 nm
(visible range). Do this for a single bilayer (one layer of each coating),
two bilayers, four bilayers, and 25 bilayers.
Chapter 5
Propagation in Anisotropic Media
To this point, we have considered only isotropic media where the susceptibility
χ(ω) (and hence the index of refraction) is the same for all propagation directions
and polarizations. In anisotropic materials, such as crystals, it is possible for
light to experience a different index of refraction depending on the orientation
(i.e. polarization) of the electric field E. This difference in the index of refraction
occurs when the direction and strength of the induced dipoles depends on the
lattice structure of the material in addition to the propagating field.1 The unique
properties of anisotropic materials make them important elements in many
optical systems.
In section 5.1 we discuss how to connect E and P in anisotropic media using a
susceptibility tensor. In section 5.2 we apply Maxwell’s equations to a plane wave
traveling in a crystal. The analysis leads to Fresnel’s equation, which relates the
components of the k-vector to the components of the susceptibility tensor. In
section 5.3 we apply Fresnel’s equation to a uniaxial crystal (e.g. quartz, sapphire)
where χx = χ y 6= χz . In the context of a uniaxial crystal, we show that the Poynting
vector and the k-vector are generally not parallel.
More than a century before Fresnel, Christian Huygens successfully described
birefringence in crystals using the idea of elliptical wavelets. His method gives
the direction of the Poynting vector associated with the extraordinary ray in a
crystal. It was Huygens who coined the term ‘extraordinary’ since one of the
rays in a birefringent material appeared not to obey Snell’s law. Actually, the
k-vector always obeys Snell’s law, but in a crystal, the k-vector points in a different
direction than the Poynting vector, which delivers the energy seen by an observer.
Huygens’ approach is outlined in Appendix 5.D.
5.1 Constitutive Relation in Crystals

In a anisotropic crystal, asymmetries in the lattice can cause the medium polar-
ization P to respond in a different direction than the electric field E (i.e. P 6= ²0 χE).
1 Not all crystals are anisotropic. For instance, crystals with a cubic lattice structure (such as
NaCl) are highly symmetric and respond to electric fields the same in any direction.
117
118 Chapter 5 Propagation in Anisotropic Media
However, at low intensities the response of materials is still linear (or propor-
tional) to the strength of the electric field. The linear constitutive relation which
connects P to E in a crystal can be expressed in its most general form as
Px χxx χx y χxz Ex
    
 P y  = ²0  χ y x χy y χy z   E y  (5.1)
Pz χzx χz y χzz Ez
The matrix in (5.1) is called the susceptibility tensor. To visualize the behavior
of electrons in such a material, we imagine each electron bound as though by
tiny springs with different strengths in different dimensions to represent the
anisotropy (see Fig. 5.1). When an external electric field is applied, the electron
experiences a force that moves it from its equilibrium position. The ‘springs’
Figure 5.1 A physical model of an
(actually the electric force from ions bound in the crystal lattice) exert a restoring
electron bound in a crystal lat-
tice with the coordinate system force, but the restoring force is not equal in all directions—the electron tends to
specially chosen along the princi- move more along the dimension of the weaker spring. The displaced electron
pal axes so that the susceptibility creates a microscopic dipole, but the asymmetric restoring force causes P to be in
tensor takes on a simple form. a direction different than E as depicted in Fig. 5.2.
To understand the geometrical interpretation of the many coefficients χi j ,
assume, for example, that the electric field is directed along the x-axis (i.e. E y =
E z = 0) as depicted in Fig. 5.2. In this case, the three equations encapsulated in
(5.1) reduce to
P x = ²0 χxx E x
P y = ²0 χ y x E x
P z = ²0 χzx E x
Notice that the coefficient χxx connects the strength of P in the x̂ direction with
the strength of E in that same direction, just as in the isotropic case. The other two
coefficients (χ y x and χzx ) describe the amount of polarization P produced in the
ŷ and ẑ directions by the electric field component in the x-dimension. Likewise,
the other coefficients with mixed subscripts in (5.1) describe the contribution to
P in one dimension made by an electric field component in another dimension.
Figure 5.2 The applied field E As you might imagine, working with nine susceptibility coefficients can get
and the induced polarization P in
complicated. Fortunately, we can greatly reduce the complexity of the description
general are not parallel in a crystal
by a judicious choice of coordinate system. In Appendix 5.A we explain how
lattice.
conservation of energy requires that the susceptibility tensor (5.1) for typical
non-aborbing crystals be real and symmetric (i.e. χi j = χ j i ).2
Appendix 5.B shows that, given a real symmetric tensor, it is always possible
to choose a coordinate system for which off-diagonal elements vanish. This is
true even if the lattice planes in the crystal are not mutually orthogonal (e.g.
rhombus, hexagonal, etc.). We will imagine that this rotation of coordinates
2 By ‘typical’ we mean that the crystal does not exhibit optical activity. Optically active crystals
have a complex susceptibility tensor, even when no absorption takes place. Conservation of energy
in this more general case requires that the susceptibility tensor be Hermitian (χi j = χ∗j i ).
5.2 Plane Wave Propagation in Crystals 119
has been accomplished. In other words, we can let the crystal itself dictate the
orientation of the coordinate system, aligned to the principal axes of the crystal
for which the off-diagonal elements of (5.1) are zero
With the coordinate system aligned to the principal axes, the constitutive
relation for a non absorbing crystal simplifies to
Px χx 0 0 Ex
    
 P y  = ²0  0 χy 0  Ey  (5.2)
Pz 0 0 χz Ez
or without the matrix notation (since it no longer offers much convenience)
P = x̂²0 χx E x + ŷ²0 χ y E y + ẑ²0 χz E z (5.3)
By assumption, χx , χ y , and χz are all real. (We have dropped the double subscript;
χx stands for χxx , etc.)
5.2 Plane Wave Propagation in Crystals

We consider a plane wave with frequency ω propagating in a crystal. In a manner
similar to our previous analysis of plane waves propagating in isotropic materials,
we write as trial solutions
E = E0 e i (k·r−ωt )
B = B0 e i (k·r−ωt ) (5.4)
i (k·r−ωt )
P = P0 e
where restrictions on E0 , B0 , P0 , and k are yet to be determined. As usual, the
phase of each wave is included in the amplitudes E0 , B0 , and P0 , whereas k is real
in accordance with our assumption of no absorption.
We can make a quick observation about the behavior of these fields by apply-
ing Maxwell’s equations directly. Gauss’s law for electric fields requires
∇ · (²0 E + P) = k · (²0 E + P) = 0 (5.5)
and Gauss’s law for magnetism gives
∇·B = k·B = 0 (5.6)
We immediately notice the following peculiarity: From its definition, the Poynting
vector S ≡ E × B/µ0 is perpendicular to both E and B, and by (5.6) the k-vector is
perpendicular to B. However, by (5.5) the k-vector is not necessarily perpendicu-
lar to E, since in general k · E 6= 0 if P points in a direction other than E. Therefore,
k and S are not necessarily parallel in a crystal. In other words, the flow of energy
and the direction of the phase-front propagation can be different in anisotropic
media.
Our main goal here is to relate the k-vector to the susceptibility parameters χx ,
χ y , and χz . To do this, we plug our trial plane-wave fields into the wave equation
(1.41). Under the assumption Jfree = 0, we have
∂2 E ∂2 P
∇2 E − µ0 ²0 = µ0 + ∇ (∇ · E) (5.7)
∂t 2 ∂t 2
Derivation of the dispersion relation in crystals
We begin by substituting the trial solutions (5.4) into the wave equation (5.7). After
carrying out the derivatives we find
k 2 E − ω2 µ0 (²0 E + P) = k (k · E) (5.8)
Inserting the constitutive relation (5.3) for crystals into (5.8) yields
k 2 E − ω2 µ0 ²0 1 + χx E x x̂ + 1 + χ y E y ŷ + 1 + χz E z ẑ = k (k · E)
£¡ ¢ ¡ ¢ ¡ ¢ ¤
(5.9)
This relationship is unwieldy because of the mix of electric field components that
appear in the expression. This was not a problem when we investigated isotropic
materials for which the k-vector is perpendicular to E, making the right-hand side
of the equations zero. However, there is a trick for dealing with this.
Relation (5.9) actually contains three equations, one for each dimension. Explicitly,
these equations are
ω2 ¡
· ¸
2
k − 2 1 + χx E x = k x (k · E)
¢
(5.10)
c
ω2 ¡
· ¸
k 2 − 2 1 + χ y E y = k y (k · E)
¢
(5.11)
c
and
ω2 ¡
· ¸
k2 − χ
¢
1 + z E z = k z (k · E) (5.12)
c2
We have replaced the constants µ0 ²0 with 1/c 2 in accordance with (1.43). We
multiply (5.10)–(5.12) respectively by k x , k y , and k z and also move the factor in
square brackets in each equation to the denominator on the right-hand side. Then
by adding the three equations together we get
k x2 (k · E) k y2 (k · E) k z2 (k · E)
i+h i+h i = k x E x + k y E y + k z E z = (k · E)
ω2 1+χ ω2 ω2 1+χ
k2 − ( x) (1+χ y ) k2 − ( z)
h
c2 k2 − c2 c2
(5.13)
Now k · E appears in every term and can be divided away. This gives the dispersion
relation (unencumbered by field components):
2
k x2 ky k z2 ω2
¢¤ + £ ¢¤ + £ ¢¤ = 2 (5.14)
k 2 c 2 /ω2 − 1 + χx k 2 c 2 /ω2 − 1 + χ y k 2 c 2 /ω2 − 1 + χz
£ ¡ ¡ ¡
c
As a final touch, we have multiplied the equation through by ω2 /c 2

5.2 Plane Wave Propagation in Crystals 121
The dispersion relation (5.14) allows us to find a suitable k, given values for ω,
χx , χ y , and χz . Actually, it only restricts the magnitude of k; we must still decide
on a direction for the wave to travel (i.e. we must choose the ratios between k x , k y ,
and k z ). To remind ourselves of this fact, we introduce a unit vector that points in
the direction of k
¡ ¢
k = k x x̂ + k y ŷ + k z ẑ = k u x x̂ + u y ŷ + u z ẑ = k û (5.15)
With this unit vector inserted, the dispersion relation (5.14) for plane waves in a
crystal becomes
u x2 u 2y u z2 ω2
¢¤ + £ ¢¤ + £ ¢¤ = 2 2 (5.16)
k 2 c 2 /ω2 − 1 + χx k 2 c 2 /ω2 − 1 + χ y k 2 c 2 /ω2 − 1 + χz
£ ¡ ¡ ¡
k c
We may define refractive index as the ratio of the speed of light in vacuum
c to the speed of phase propagation in a material ω/k (see P1.9). The relation
introduced for isotropic media (i.e. (2.19) for real index) remains appropriate.
That is
kc
n= (5.17)
ω
This familiar relationship between k and ω, in the case of a crystal, depends on
the direction of propagation in accordance with (5.16).
Inspired by (2.30), we will find it helpful to introduce several refractive-index
parameters:
n x ≡ 1 + χx
p
q
n y ≡ 1 + χy (5.18)
n z ≡ 1 + χz
p
With these definitions (5.17)-(5.18), the dispersion relation (5.16) becomes
u x2 u 2y u z2 1
¡
2 2
¢ + ¡
2 2
¢ + ¡
2 2
¢= 2 (5.19)
n − nx n − ny n − nz n
This is called Fresnel’s equation (not to be confused with the Fresnel coefficients
studied in chapter 3). The relationship contains the yet unknown index n that
varies with the direction of the k-vector (i.e. the direction of the unit vector û).
After multiplying through by all of the denominators (and after a fortuitous
cancelation owing to u x2 + u 2y + u z2 = 1), Fresnel’s equation (5.19) can be rewritten
as a quadratic in n 2 . The two solutions are
p
2 B ± B 2 − 4AC
n = (5.20)
2A
where
A ≡ u x2 n x2 + u 2y n 2y + u z2 n z2 (5.21)
³ ´ ³ ´
B ≡ u x2 n x2 n 2y + n z2 + u 2y n 2y n x2 + n z2 + u z2 n z2 n x2 + n 2y
¡ ¢
(5.22)
C ≡ n x2 n 2y n z2 (5.23)
The upper and lower signs (+ and −) in (5.20) give two positive solutions for n 2 .
The positive square root of these solutions yields two physical values for n. It turns
out that each of the two values for n is associated with a polarization direction
of the electric field, given a propagation direction k. A broader analysis carried
out in appendix 5.C renders the orientation of the electric fields, whereas here we
only show how to find the two values of n. We refer to the two indices as the slow
and fast index, since the waves associated with each propagate at speed v = c/n.
In the special cases of propagation along one of the principal axes of the
crystal, the index n takes on two of the values n x , n y , or n z , depending on which
are orthogonal to the direction of propagation.
Example 5.1
Calculate the two possible values for the index of refraction when k is in the ẑ
direction (in the crystal principal frame).
Solution: With u z = 1 and u x = u y = 0 we have

³ ´
A = n z2 ; B = n z2 n x2 + n 2y ; C = n x2 n 2y n z2
The square-root term is then

p q ¡
B 2 − 4AC = n z4 n x4 + 2n x2 n 2y + n 4y − 4n x2 n 2y n z4
¢
q ¡ ¢2
= n z4 n x2 − n 2y
³ ´
= n z2 n x2 − n 2y
Inserting this expression into (5.20), we find the two values for the index:
n = nx , n y
The index n x is experienced by light whose electric field points in the x-dimension,
and the index n y is experienced by light whose electric field points in the y-
dimension (see appendix 5.C ).
Before moving on, let us briefly summarize what has been accomplished so
far. Given values for χx , χ y , and χz associated with light in a crystal at a given
frequency, you can define the indices n x , n y , and n z , according to (5.18). Next, a
direction for the k-vector is chosen (i.e. u x , u y , and u z ). This direction generally
has two values for the index of refraction associated with it, found using Fresnel’s
equation (5.20). Each index is associated with a specific polarization direction
for the electric field as outlined in appendix 5.C. Every propagation direction û
has its own natural set of polarization components for the electric field. The two
polarization components travel at different speeds, even though the frequency is
the same. This is known as birefringence.
Figure 5.3 Spherical coordinates.
5.3 Biaxial and Uniaxial Crystals 123
5.3 Biaxial and Uniaxial Crystals

All anisotropic crystals have certain special propagation directions where the two
values for n from Fresnel’s equation are equal. These directions are referred to as
the optic axes of the crystal. The optic axes do not necessarily coincide with the
principal axes x̂, ŷ, and ẑ. When propagation is along an optic axis, all polarization
components experience the same index of refraction. If the values of n x , n y , and
n z are all unique, a crystal will have two optic axes, and hence is referred to as a
biaxial crystal.
It is often convenient to use spherical coordinates to represent the compo-
nents of û (see to Fig. 5.3):
u x = sin θ cos φ
u y = sin θ sin φ (5.24)
2.22 2.35
u z = cos θ
Here θ is the polar angle measured from the z-axis of the crystal and φ is the
azimuthal angle measured from the x-axis of the crystal. These equations em-
phasize the fact that there are only two degrees of freedom when specifying
propagation direction (θ and φ). It is important to remember that these angles
must be specified in the frame of the crystal’s principal axes, which are often not
aligned with the faces of a cut crystal in an optical setup.
By convention, we order the crystal axes for biaxial crystals so that n x < n y <
n z . Under this convention, the two optic axes occur in the x-z plane (φ = 0) at
two values of the polar angle θ, measured from the z-axis (see P5.3): 2.35 2.41
v
u 2
nx u n z − n 2y
cos θ = ± t 2 (Optic axes directions, biaxial crystal) (5.25)
n y n z − n x2
While finding the direction of the optic axes in a biaxial crystal is not too bad, an
expression for the two indices of refraction is messy. The smaller value is com-
monly referred to as the ‘fast’ index and the larger value the ‘slow’ index. Figure 5.4
shows the two refractive indices (i.e. the solutions to Fresnel’s equation (5.20)) for
a biaxial crystal plotted with color shading on the surface of a sphere. Each point 0 0.19
on the sphere represents a different θ and φ. The two optic axes are apparent
in the plot of the difference between n slow and n fast . When propagating in these Figure 5.4 The fast and slow re-
fractive indices (and their differ-
directions, either polarization experiences the same index. For the remainder of
ence) as a function of direction
this chapter, we will focus on the simpler case of uniaxial crystals. for potassium niobate (KNbO3 ) at
In uniaxial crystals two of the coefficients χx , χ y , and χz are the same. In λ = 500 nm (n x = 2.22, n y = 2.35,
this case, there is only one optic axis for the crystal (hence the name uniaxial). and n z = 2.41) .
By convention, in uniaxial crystals we label the dimension that has the unique
susceptibility as the z-axis (i.e. χx = χ y 6= χz ). This makes the z-axis the optic axis.
The unique index of refraction is called the extraordinary index
n z = ne (5.26)
and the other index is called the ordinary index
n x = n y = no (5.27)
These names were coined by Huygens, one of the early scientists to study light
in crystals (see appendix 5.D). A uniaxial crystal with n e > n o is referred to as a
positive crystal, and one with n e < n o is referred to as a negative crystal.
To calculate the index of refraction for a wave propagating in a uniaxial crystal,
we use definitions (5.26) and (5.27) along with the spherical representation of û
(5.24) in Fresnel’s equation (5.20) to find the following two values for n (see P5.4):
n = no (uniaxial crystal) (5.28)
and
no ne
n = n e (θ) ≡ q (uniaxial crystal) (5.29)
1.56 1.68
n o2 sin2 θ + n e2 cos2 θ
The index n e (θ) in (5.29) is also commonly referred to as the extraordinary index
along with the constant n e = n z . While this has the potential for some confusion,
the practice is so common that we will perpetuate it here. We will write n e (θ)
when the angle dependent quantity specified by (5.29) is required, and write n e
in formulas where the constant (5.26) is called for (as in the right hand side of
(5.29)). Notice that n e (θ) depends only on θ (the polar angle measured from the
optic axis ẑ) and not φ (the azimuthal angle). Figure 5.5 shows the two refractive
indices (5.28) and (5.29) as a function θ and φ. Since n e (θ) has no φ dependence
and n o is constant, the variation is much simpler than for the biaxial case.
1.68 1.68
As outlined in appendix 5.C, the index n o corresponds to an electric field
component that points perpendicular to the plane containing û and ẑ (e.g. if
û is in the x-z plane, n o is associated with light polarized in the y-dimension).
On the other hand, the index n e (θ) corresponds to field polarization that lies
within the plane containing û and ẑ. In this case, the polarization component
is directed partially along the optic axis (i.e. it has a z-component). That is why
(5.29) gives for the refractive index a mixture of n o and n e . If θ = 0, then the
k-vector is directed exactly along the optic axis, and n e (θ) reduces to n o so that
both polarization components experience same index n o .
0 0.12
Figure 5.5 The extraordinary and 5.4 Refraction at a Uniaxial Crystal Surface
ordinary refractive indices (and
their difference) as a function of Next we consider refraction as light enters a uniaxial crystal. Snell’s law (3.7)
direction for beta barium borate describes the connection between the k-vectors incident upon and transmitted
(BBO) at λ = 500 nm (n o = 1.68 through the surface. We must consider separately the portion of the light that ex-
and n e = 1.56).
periences the ordinary index from the portion that experiences the extraordinary
index. Because of the different indices, the ordinary and extraordinary polarized
light refract into the crystal at two different angles; they travel at two different
velocities in the crystal; and they have two different wavelengths in the crystal.
5.5 Poynting Vector in a Uniaxial Crystal 125
If we assume that the index outside of the crystal is one, Snell’s law for the
ordinary polarization is
sin θi = n o sin θt (ordinary polarized light) (5.30)
where n o is the ordinary index inside the crystal. The extraordinary polarized
light also obeys Snell’s law, but now the index of refraction in the crystal depends
on direction of propagation inside the crystal relative to the optic axis. Snell’s law
for the extraordinary polarization is
sin θi = n e (θ 0 ) sin θt (extraordinary polarized light) (5.31)
where θ 0 is the angle between the optic axis inside the crystal and the direction of
propagation in the crystal (given by θt in the plane of incidence). When the optic
axis is at an arbitrary angle with respect to the surface the relationship between
θ 0 and θt is cumbersome. We will examine Snell’s law only for the specific case
when the optic axis is perpendicular to the crystal surface, for which θt = θ 0 .
y-axis
Example 5.2
Derive Snell’s law for a uniaxial crystal with optic axis perpendicular to the surface.
Solution: Refer to Fig. 5.6. With the optic axis perpendicular to the surface, if the z-axis
light hits the crystal surface straight on, the index of refraction is n o , regardless of
x-axis (directed into page)
the orientation of polarization since θ 0 = 0. When the light strikes the surface at an
angle, s-polarized light continues to experience the index n o , while p-polarized
light experiences the extraordinary index n e (θ). 3
When we insert (5.29) into Snell’s law (5.31) with θ 0 = θt , the expression can be
inverted to find the transmitted angle θt in terms of θi (see P5.5):
Figure 5.6 Propagation of light in a
n e sin θi
tan θt = q (extraordinary polarized, optic axis ⊥ surface) (5.32) uniaxial crystal with its optic axis
n o n e2 − sin2 θi perpendicular to the surface.
As strange as this formula may appear, it is Snell’s law, but with an angularly
dependent index.
5.5 Poynting Vector in a Uniaxial Crystal

When an object is observed through a crystal (acting as a window), the energy
associated with ordinary and extraordinary polarized light follow different paths,
3 The correspondence between s and p and ordinary and extraordinary polarization components
is specific to the orientation of the optic axis in this example. For arbitrary orientations of the
optic axis with respect to the surface, the ordinary and extraordinary components will generally be
mixtures of s and p polarized light.
giving rise to two different images. This phenomenon is one of the more com-
monly observed manifestations of birefringence. Since the Poynting vector dic-
tates the direction of energy flow, it is the direction of S that determines the
separation of the double image seen when looking through a birefringent crystal.
Snell’s law dictates the connection between the directions of the incident
and transmitted k-vectors. The Poynting vector S for purely ordinary polarized
light points in the same direction as the k-vector, so the direction of energy flow
for ordinary polarized light also obeys Snell’s law. However, for extraordinary
polarized light, the Poynting vector S is not parallel to k (recall the discussion in
connection with (5.5) and (5.6)). Thus, the energy flow associated with extraordi-
nary polarized light does not obey Snell’s law. When Christiaan Huygens saw this
in the 1600s, he exclaimed “how extraordinary!” Huygens’ method for describing
the phenomenon is outlined appendix 5.D.
To analyze this situation, it is necessary to derive an expression for extraordi-
nary polarized light similar to Snell’s law, but which applies to S rather than to k.
This describes the direction that the energy associated with extraordinary rays
takes upon entering the crystal. To calculate the direction that the extraordinary
polarized S takes upon entering a crystal, we first calculate the direction of k
inside the crystal using Snell’s law (5.31). Then we use the expression (5.62) for E
along with B = (k × E)/ω, to evaluate S = E × B/µ0 . In general, this process is best
done numerically, since Snell’s law (5.31) for extraordinary polarized light usually
does not have simple analytic solutions.
Example 5.3
Find a relationship between direction of the Poynting Vector in a uniaxial crystal
and the angle of incidence in the special case where the optic axis is perpendicular
to the surface.
Solution: To find the direction of energy flow, we must calculate S = E × B/µ0 . We

will need to know E associated with n e (θ). We can obtain E from the procedures
outlined in appendix 5.C. Equivalently, we can obtain it from the constitutive
relation (5.3) with the definitions (5.18):
²0 E + P = ²0 1 + χx E x x̂ + 1 + χ y E y ŷ + 1 + χz E z ẑ
£¡ ¢ ¡ ¢ ¡ ¢ ¤
(5.33)
= ²0 n o2 E x x̂ + n o2 E y ŷ + n e2 E z ẑ
¡ ¢
Let the k-vector lie in the y-z plane. We may write it as k = k ŷ sin θt + ẑ cos θt .
¡ ¢
Then the ordinary component of the field points in the x-direction, while the
extraordinary component lies in the y-z plane.
Equation (5.33) requires
k · (²0 E + P) = k ŷ sin θt + ẑ cos θt · ²0 n o2 E x x̂ + n o2 E y ŷ + n e2 E z ẑ

¡ ¢ ¡ ¢
= ²0 k n o2 E y sin θt + n e2 E z cos θt
¡ ¢
(5.34)
=0
5.5 Poynting Vector in a Uniaxial Crystal 127
Therefore, the y and z components of the extraordinary field are related through
n o2 E y
Ez = − tan θt (5.35)
n e2
We may write the extraordinary polarized electric field as
n2
µ ¶
E = E y ŷ − ẑ o2 tan θt e i (k·r−ωt ) (extraordinary polarized) (5.36)
ne
The associated magnetic field (see (2.56)) is
n o2
³ ´
k × E k ŷ sin θt + ẑ cos θt × E y ŷ − ẑ ne2 tan θt i (k·r−ωt )
¡ ¢
B= = e
ω ω
µ 2
kE y n o
¶
= −x̂ sin θt tan θt + cos θt e i (k·r−ωt )
ω n e2
(extraordinary polarized) (5.37)
The time-averaged Poynting vector then becomes
¿ ½ ¾À
B
〈S〉t = Re {E} × Re
µ0 t
k|E y |2
Ã ! Ã !
n 2 n2 D E
=− ŷ − ẑ o2 tan θt × o2 sin θt tan θt + cos θt x̂ cos2 (k · r − ωt + φE y )
µ0 ω ne ne t
2
Ã !Ã !
k|E y | n o 2 n 2
= sin θt tan θt + cos θt ẑ + ŷ o2 tan θt
2µ0 ω n e2 ne
(extraordinary polarized) (5.38)
Let us label the direction of the Poynting vector with the angle θS . By definition,
the tangent of this angle is the ratio of the two vector components of S:
Sy n o2
tan θS ≡ = tan θt (extraordinary polarized) (5.39)
Sz n e2
While the k-vector is characterized by the angle θt , the Poynting vector is char-
acterized by the angle θS . Combining (5.32) and (5.39), we can connect θS to the
incident angle θi :
n o sin θi
tan θS = q (extraordinary polarized) (5.40)
n e n e2 − sin2 θi
As we noted in the last example, we have the case where ordinary polarized light is
s-polarized light, and extraordinary polarized light is p-polarized light due to our
specific choice of orientation for the optic axis in this section. In general, the s- and
p-polarized portions of the incident light can each give rise to both extraordinary
and ordinary rays.
Appendix 5.A Symmetry of Susceptibility Tensor

Here we show that the assumption of a non-absorbing (and non optically active)
medium implies that the susceptibility tensor is symmetric. We assume that P is
due to a single species of electron, so that we have P = N p. Here N is the number
of microscopic dipoles per volume and p = q e r, where q e is the charge on the
electron and r is the microscopic displacement of the electron. The force on this
electron due to the electric field is given by F = Eq e . With these definitions, we
can use (5.1) to write a connection between the force due to a static E and the
electron displacement:
x χxx χx y χxz Fx
    
²0
 χy x
N qe  y  = χy y χy z   F y  (5.41)
qe
z χzx χz y χzz Fz
The column vector on the left represents the components of the displacement
r. We next invert (5.41) to find the force of the electric field on an electron as a
function of its displacement4
Fx k xx kx y k xz x
    
 Fy  =  kyx ky y kyz   y  (5.42)
Fz k zx kz y k zz z
where
−1
k xx kx y k xz χxx χx y χxz
  
2
N qe
 kyx ky y kyz  ≡  χy x χy y χy z  (5.43)
²0
k zx kz y k zz χzx χz y χzz
The total work done on an electron in moving it to its displaced position is
given by Z
W= F(r0 ) · d r0 (5.44)
path
While there are many possible paths for getting the electron to any specific dis-
placement (each path specified by a different history of the electric field), the
work done along any of these paths must be the same if the system is conservative
(i.e. no absorption). For example, for a final displacement of r = x x̂ + y ŷ we could
have the following two paths:
(x,y,0) Path 2 (x,y,0)
Path 1
(0,0,0) (0,0,0)
4 This inversion assumes the field changes slowly so the forces on the electron are always es-
sentially balanced. This is not true for optical fields, but the proof gives the right flavor for why
conservation of energy results in the symmetry. A more formal proof that doesn’t make this as-
sumption can be found in Principles of Optics, 7th Ed., Born and Wolf, pp. 790-791.
5.B Rotation of Coordinates 129
We can use (5.42) in (5.44) to calculate the total work done on the electron
along path 1:
Z x Z y
W= F x (x 0 , y 0 = 0, z 0 = 0)d x 0 + F y (x 0 = x, y 0 , z 0 = 0)d y 0
0 0
Z x Z y
0 0
= k xx x d x + (k y x x + k y y y 0 ) d y 0
0 0
k xx 2 ky y 2
= x + kyx x y + y
2 2
If we take path 2, the total work is
Z y Z x
0 0 0 0
W= F y (x = 0, y , z = 0)d y + F x (x 0 , y 0 = y, z 0 = 0)d x 0
Z0 y Z x 0
0 0
= ky y y d y + (k xx x + k x y y) d x 0
0
0 0
ky y 2 k xx 2
= y + kx y x y + x
2 2
Since the work must be the same for these two paths, we clearly have k x y = k y x .
Similar arguments for other pairs of dimensions ensure that the matrix of k
coefficients is symmetric. From linear algebra, we learn that if the inverse of a
matrix is symmetric then the matrix itself is also symmetric. When we combine
this result with the definition (5.43), we see that the assumption of no absorption
requires the susceptibility matrix to be symmetric.
Appendix 5.B Rotation of Coordinates

In this appendix, we go through the labor of showing that (5.1) can always be
written as (5.3) via rotations of the coordinate system, given that the susceptibility
tensor is symmetric (i.e. χi j = χ j i ). We have
P = ²0 χE (5.45)
where
Ex Px χxx χx y χxz
     
E ≡  Ey  P ≡  Py  χ ≡  χx y χy y χy z  (5.46)
Ez Pz χxz χy z χzz
Our task is to find a new coordinate system x 0 , y 0 , and z 0 for which the susceptibil-
ity tensor is diagonal. That is, we want to choose x 0 , y 0 , and z 0 such that
P0 = ²0 χ0 E0 , (5.47)
where
χ0 0 0
     
E x0 0 P x0 0 0 0
 x0 x χ0y 0 y 0
E0 ≡  E y0 0  P0 ≡  P y0 0  0
χ ≡ 0 (5.48)
    

E z0 0 P z0 0 0 0 χ0z 0 z 0
To arrive at the new coordinate system, we are free to make pure rotation trans-
formations. In a manner similar to (6.29), a rotation through an angle γ about the
z-axis, followed by a rotation through an angle β about the resulting y-axis, and
finally a rotation through an angle α about the new x-axis, can be written as
 
R 11 R 12 R 13
R ≡  R21 R22 R23 
R 31 R 32 R 33
cos β 0 sin β cos γ sin γ 0
   
1 0 0
=  0 cos α sin α   0 1 0   − sin γ cos γ 0 
0 − sin α cos α − sin β 0 cos β 0 0 1
cos β cos γ cos β sin γ sin β
 
=  − cos α sin γ − sin α sin β cos γ cos α cos γ − sin α sin β sin γ sin α cos β 
sin α sin γ − cos α sin β cos γ − sin α cos γ − cos α sin β sin γ cos α cos β
(5.49)
The matrix R produces an arbitrary rotation of coordinates in three dimensions.

Specifically, we can write:
E0 = RE
(5.50)
P0 = RP
These transformations can be inverted to give
E = R−1 E0
(5.51)
P = R−1 P0
where
cos β cos γ − cos α sin γ − sin α sin β cos γ sin α sin γ − cos α sin β cos γ
 
R −1
=  cos β sin γ cos α cos γ − sin α sin β sin γ − sin α cos γ − cos α sin β sin γ 
sin β sin α cos β cos α cos β
 
R 11 R 21 R 31
=  R 12 R 22 R 32  = RT (5.52)
R 13 R 23 R 33
Note that the inverse of the rotation matrix is the same as its transpose, an impor-
tant feature that we exploit in what follows.
Upon inserting (5.51) into (5.45) we have
R−1 P0 = ²0 χR−1 E0 (5.53)
or
P0 = ²0 RχR−1 E0 (5.54)
5.C Electric Field in Crystals 131
From this equation we see that the new susceptibility tensor we seek for (5.47) is
χ0 ≡ RχR−1
R 11 R 12 R 13 χxx χx y χxz R 11 R 21 R 31
   
=  R 21 R 22 R 23   χx y χy y χ y z   R 12 R 22 R 32 
R 31 R 32 R 33 χxz χy z χzz R 13 R 23 R 33
χx 0 x 0 χ0x 0 y 0 χ0x 0 z 0
 0 
=  x 0 y 0 χ0y 0 y 0 χ0y 0 z 0 
 χ0
(5.55)

χ0x 0 z 0 χ0y 0 z 0 χ0z 0 z 0
We have expressly indicated that the off-diagonal terms of χ0 are symmetric (i.e.
χ0i j = χ0j i ). This can be verified by performing the multiplication in (5.55). It is a
consequence of χ being symmetric and R−1 being equal to RT
The three off-diagonal elements of χ0 (appearing both above and below the
diagonal) are found by performing the matrix multiplication in the second line
of (5.55). The specific expressions for these three elements are not particularly
enlightening. The important point is that we can make all three of them equal to
zero since we have three degrees of freedom in the angles α, β, and γ. Although,
we do not expressly solve for the angles, we have demonstrated that it is always
possible to set
χ0x 0 y 0 = 0
χ0x 0 z 0 = 0 (5.56)
χ0y 0 z 0 =0
This justifies (5.3).
Appendix 5.C Electric Field in Crystals

To determine the direction of the electric field associated with the each value
of n, we return to (5.10), (5.11), and (5.12) in the analysis in section 5.2. These
equations can be written in matrix format as5
ω2
 
1 + χx − k y2 − k z2
¡ ¢
kx k y kx kz
 
c2  Ex
ω2

2 2
χ
¡ ¢
kx k y  Ey  = 0
1 + y − kx − kz k y kz

 c2 
ω2 Ez
1 + χz − k x2 − k y2
¡ ¢
kx kz k y kz c2
(5.57)
where we have used k x2 + k y2 + k z2 = k 2 . We can divide every element by k 2 and

employ the definitions (5.15), (5.17), and (5.18) to make this matrix equation look
5 A. Yariv and P. Yeh, Optical Waves in Crystals, Sect. 4.2 (New York: Wiley, 1984).
slightly nicer:
n x2
 
− u 2y − u z2 ux u y ux uz
 Ex
 
n2
 n 2y  Ey  = 0

 ux u y n2
− u x2 − u z2 u y uz  (5.58)
n z2 Ez
ux uz u y uz n2
− u x2 − u 2y
For (5.58) to have a non-trivial solution (i.e. non zero fields), the determinant
of the matrix must be zero. Imposing this requirement is an equivalent way to
derive Fresnel’s equation (5.19) for n.
Given a direction for û and a value for n (from Fresnel’s equation), we can use
(5.58) to determine the direction of the electric field associated with that index.
It is left as an exercise to show that when all three components are nonzero (i.e.
u x 6= 0, u y 6= 0, and u z 6= 0), the appropriate field direction for a value of n is given
by
 ux 
 n 2 − n x2 
Ex
 
 
 uy 
 Ey ∝ 2 (5.59)
   
 n − n 2y


Ez 
 uz


n 2 − n z2
This is a proportionality rather than an equation because Maxwell’s equations
only specify the direction of E—we are free to choose the amplitude. Because
Fresnel’s equation gives two values for n, (5.59) specifies two distinct polarization
components associated with each propagation direction û. These polarization
components form a natural basis for describing light propagation in a crystal.
When light is composed of a mixture of these two polarizations, the two polariza-
tion components experience different indices of refraction.
If any of the components of û (i.e. u x , u y , or u z ) is precisely zero, the corre-
sponding entry in (5.59) yields a zero-over-zero situation. This happens when
at least one of the dimensions in (5.58) becomes decoupled from the others. In
these cases, you can and re-solve (5.58) for the polarization directions as in the
following example.
Example 5.4
Determine the directions of the two polarization components associated with light
propagating in the û = ẑ direction. (Compare with Example 5.1.)
Solution: In this case we have u x = u y = 0, so as noted above, we have to go back

to (5.58) and re-solve. In our case, the set of equations becomes
n x2
 
−1 0 0  
 n2
n 2y  Ex
 Ey
 0 −1 0   =0 (5.60)
 n2
2
nz Ez
0 0 n2
5.C Electric Field in Crystals 133
Notice that all three dimensions are decoupled in this system (i.e. there are no
off-diagonal terms). In Example 5.1 we found that the two values of n associated
with û = ẑ are n x and n y . If we use n = n x in our set of equations, we have
 
0 0 0  
 n 2y  Ex
 0 −1 0  Ey  = 0

 n x2
(a) Polarization Direction for Slow Index
n z2 Ez
 
0 0 n x2
Assuming n x and n y are unique so that n y /n x 6= 1, these equations require E y =

E z = 0 but allow E x to be non-zero. This proves our earlier assertion that the index
n x is associated with light polarized in the x-dimension in the special case of û = ẑ.
Similarly, when n y is inserted into (5.60), we find that it is associated with light
polarized in the y-dimension.
We can use (5.59) to study the behavior of polarization direction as the direc-
tion of propagation varies. Figure 5.7 shows plots of the polarization direction (i.e. (b) Polarization Direction for Fast Index
normalized E x , E y , and E z ) in Potassium Niobate as the propagation direction

is varied. The plot is created by inserting the spherical representation of û (5.24)
into Fresnel’s equation (5.20) for a chosen sign of the ±, and then inserting the re-
sulting n into (5.59) to find the associated electric field. As we saw in Example 5.4,
at θ = 0 the light associated with the slow index is polarized along the y-axis and
the light associated with the fast index is polarized along the x-axis.
In Fig. 5.7(c) we have plotted the angle between the two polarization com-
ponents. At θ = 0, the two polarization components are 90◦ apart, as you might
expect. However, notice that in other propagation directions the two linear polar-
(c) Angle Between Polarization Components
ization components are not precisely perpendicular. Even so, the two polarization
components of E are orthogonal in a mathematical sense,6 so that they still com-
prise a useful basis for decomposing the light field.
Determining the Fields in a Uniaxial Crystal.
To find the directions of the electric field for light that experiences the normal
index of refraction in a uniaxial crystal, we insert n = n o into the requirement
(5.58), and solve for the allowed fields (see P5.9) to find Figure 5.7 Polarization direction
associated with the two values of n
− sin φ
 
in Potassium Niobate (KNbO3 ) at
Eo (û) ∝  cos φ  (5.61) λ = 500 nm (n x = 2.22, n y = 2.34,
0 and n z = 2.41) and φ = π/4. Frame
(c) shows the angle between the
This field component is associated with the ordinary wave because just as in an
two polarization components.
isotropic medium such as glass, the index of refraction for light with this polariza-
tion does not vary with θ. The polarization component associated with n e (θ) is
6 The two components of the electric displacement vector D = ² E + P remain perpendiular.
0
found by using (5.59):

sin θ cos φ
 
 n 2 (θ) − n 2 
 e o 
 
 
 sin θ sin φ 
Ee (û) ∝  2 (5.62)
 
 n e (θ) − n o2


 
 
cos θ
 
 
n e2 (θ) − n e2
Notice that this polarization component is partially directed along the optic axis
(i.e. it has a z-component), and it is not perpendicular to k since û · Ee (û) 6= 0 (see
P5.10). It is, however, perpendicular to the ordinary polarization component, since
Ee · Eo = 0.
Notice that when θ = 0, (5.29) reduces to n = n o so that both indices are the same.
On the other hand, if θ = π/2 then (5.29) reduces to n = n e .
Appendix 5.D Huygens’ Elliptical Construct for a Uniaxial

Crystal
In 1690 Christian Huygens developed a way to predict the direction of extraordi-
nary rays in a crystal by examining an elliptical wavelet. The point on the elliptical
wavelet that propagates along the optic axis is assumed to experience the index
n e . The point on the elliptical wavlet that propagates perpendicular to the optic
axis is assumed to experience the index n o . It turns out that Huygens’ approach
agreed with the direction energy propagation (5.40) (as opposed to the direction
of the k-vector). This was quite satisfactory in Huygens’ day (except that he was
largely ignored for a century, owing to Newton’s corpuscular theory) since the
direction of energy propagation is what an observer sees.
y-axis
Consider a plane wave entering a uniaxial crystal with the optic axis perpen-
dicular to the surface. In Huygens’ point of view, each point on a wave front acts
as a wavelet source which combines with neighboring wavelets to preserve the
overall plane wave pattern. Inside the crystal, the wavelets propagate in the shape
of an ellipse. The equation for an elliptical wave front after propagating during a
time t is
y2 z2
+ =1 (5.63)
(c t /n e )2 (c t /n o )2
After rearranging, the equation of the ellipse inside the crystal can also be written
as s
ct y2
z= 1− (5.64)
no (c t n e )2
In order to have the wavelet joint neatly with other wavelets to build a plane wave,
the wave front of the ellipse must be parallel to a new wave front entering the sur-
face at a distance c t / sin θi above the original point. This distance is represented
Figure 5.8 Elliptical wavelet.
5.D Huygens’ Elliptical Construct for a Uniaxial Crystal 135
by the hypotenuse of the right triangle seen in Fig. 5.8. Let the point where the
wave front touches the ellipse be denoted by y, z = (z tan θS , z). The slope (rise
¡ ¢
over run) of the line that connects these two points is then
dz z
=− (5.65)
dy ct / sin θi − z tan θS
At the point where the wave front touches the ellipse (i.e., y, z = (z tan θS , z)), the
¡ ¢
slope of the curve for the ellipse is
dz −yn e2 n e2 y n e2
= =− =− tan θS (5.66)
n o2 z n o2
r
dy y2
n o ct 1 − (c t /n 2
e)
We would like these two slopes to be the same. We therefore set them equal to
each other:
n e2 z c t n e2 tan θS n e2
− tan θS = − ⇒ = 2 tan2 θS + 1 (5.67)
n o2 ct / sin θi − z tan θS z n o2 sin θi no
If we evaluate (5.63) for the point y, z = (z tan θS , z), we obtain

¡ ¢
s
ct n e2
= no tan2 θS + 1 (5.68)
z n o2
Upon substitution of this into (5.67) we arrive at

s
n e2 tan θS n e2 n e4 tan2 θS n e2
= tan2 θS + 1 ⇒ = tan2 θS + 1 (5.69)
n o2 sin θi no2
n o2 sin2 θi n o2
n e2 n o2 n o sin θi
· ¸
⇒ − 1 tan2 θS = ⇒ tan θS = (5.70)
sin θi 2
n e2
q
n e n e2 − sin2 θi
This agrees with (5.40) as anticipated. Again, Huygens’ approach obtained the
correct direction of the Poynting vector associated with the extraordinary wave.
Exercises
Exercises for 5.2 Plane Wave Propagation in Crystals
P5.1 Solve Fresnel’s equation (5.19) to find the two values of n associated
with a given û. Show that both solutions yield a positive index of
refraction
HINT: Show that (5.19) can be manipulated into the form
h³ ´ i
0= u x2 + u 2y + u z2 − 1 n 6
h³ ´ ³ ´ ³ í
+ n x2 + n 2y + n z2 − u x2 n 2y + n z2 − u 2y n x2 + n z2 − u z2 n x2 + n 2y n 4
¡ ¢
h³ ´ i
− n x2 n 2y + n x2 n z2 + n 2y n z2 − u x2 n 2y n z2 − u 2y n x2 n z2 − u z2 n x2 n 2y n 2 + n x2 n 2y n z2
The coefficient of n 6 is identically zero since by definition we have

u x2 + u 2y + u z2 = 1.
P5.2 Suppose you have a crystal with n x = 1.5, n y = 1.6, and n z = 2.0. Use
Fresnel’s equation to determine what the two indices of p refraction are
for a k-vector in the crystal along the û = (x̂ + 2ŷ + 3ẑ)/ 14 direction.
Exercises for 5.3 Biaxial and Uniaxial Crystals
P5.3 Given that the optic axes are in the x-z plane, show that the direction
of the optic axes are given by (5.25).
HINT: The two indices are the same when B 2 − 4AC = 0. You will want
to use polar coordinates for the direction unit vector, as in (5.24). Set
φ = 0 so you are in the x-z plane. Use sin2 θ + cos2 θ = 1 to get an
equation that only has cosine terms and solve for cos2 θ.
P5.4 Use definitions (5.26) and (5.27) along with the spherical representa-
tion of û (5.24) in Fresnel’s equation (5.20) to calculate the two values
for the index in a uniaxial crystal (i.e. (5.28) and (5.29)).
HINT: First show that
A = n o2 sin2 θ + n e2 cos2 θ
B = n o2 n e2 + n o4 sin2 θ + n e2 n o2 cos2 θ
C = n o4 n e2
and then use these expressions to evaluate Fresnel’s equation.
P5.5 Derive (5.32).

Exercises 137
P5.6 Suppose you have a quartz plate (a uniaxial crystal) with its optic axis
oriented perpendicular to the surfaces. The indices of refraction for
quartz are n o = 1.54424 and n e = 1.55335. A plane wave with wave-
length λvac = 633 nm passes through the plate. After emerging from
the crystal, there is a phase difference ∆ between the two polarization
components of the plane wave, and this phase difference depends on
incident angle θi . Use a computer to plot ∆ as a function of incident
angle from zero to 90◦ for a plate with thickness d = 0.96 mm .
HINT: For s-polarized light, show that the number of wavelengths that
d
fit in the plate is (s) . For p-polarized light, show that the
(λvac /n o ) cos θt
number of wavelengths that fit in the plate and the extra leg δ outside
d δ
of the plate (see Fig. 5.9) is (p) + λ , where
(λvac /n p ) cos θt vac
h i
(p) Figure 5.9 Diagram for P5.6.
δ = d tan θt(s) − tan θt sin θi
and n p is given by (5.29). Find the difference between these expressions

and multiply by 2π to find ∆.
L5.7 In the laboratory, send a HeNe laser (λvac = 633 nm) through two
crossed polarizers, oriented at 45◦ and 135◦ . Place the quartz plate
described in P5.6 between the polarizers on a rotation stage. Now
equal amounts of s- and p-polarized light strike the crystal as it is
rotated from normal incidence. (video)
Dim spots
Bright spots
Laser
Polarizer Quartz Crystal Polarizer Screen Phase Difference

on a rotation stage
Figure 5.11 Schematic for L 5.7.
If the phase shift between the two paths discussed in P5.6 is an odd
Figure 5.10 Plot for P5.6 and L 5.7.
integer times π, the polarization direction of the light transmitted
through the crystal is rotated by 90◦ , and the maximum transmission
through the second polarizer results. (In this configuration, the crystal
acts as a half wave plate, which we discuss in Chapter 6. If the phase
shift is an even integer times π, then the polarization is rotated by 180◦
and minimum transmission through the second polarizer results. Plot
these measured maximum and minimum points on your computer-
generated graph of the previous problem.
Exercises for 5.C Electric Field in Crystals
P5.8 Show that (5.59) is a solution to (5.58).

P5.9 Show that the field polarization component associated with n = n o in

a uniaxial crystal is directed perpendicular to the plane containing û
and ẑ by substituting this value for n into (5.58) and determining what
combination of field components are allowable.
HINT: Use (5.24) to represent û with φ = 0 (the index is the same for all
φ, so you may as well use one that makes calculation easy). When you
substitute into (5.58) you will find that E y can be any value because of
the location of zeros in the matrix.
To get a requirement on E x and E z , collapse the matrix equation down
to a 2 × 2 system. For non-trivial solutions to exist (i.e. E x 6= 0 or E y 6= 0),
the determinant of the matrix must be zero. Show that this is only the
case if n o = n e (i.e. the crystal is isotropic).
P5.10 Show that the electric field for extraordinary polarized light Ee (û) in a
uniaxial crystal is not perpendicular to k (i.e. û), but that it is perpen-
dicular to the ordinary polarization component Eo (û).
Review, Chapters 1–5
To prepare for an exam, you should understand the following questions and
problems thoroughly enough to be able to work them without referring back to
previous chapters.
True and False Questions
R1 T or F: The optical index of any material (not vacuum) varies with

frequency.
R2 T or F: The frequency of light can change as it enters a crystal (consider

low intensity—no nonlinear effects).
R3 T or F: The entire expression E0 e i (k·r−ωt ) associated with a light field

(both the real part and the imaginary parts) is physically relevant.
R4 T or F: The real part of the refractive index cannot be less than one.
R5 T or F: s-polarized light and p-polarized light experience the same

phase shift upon reflection from a material with complex index.
R6 T or F: When light is incident upon a material interface at Brewster’s

angle, only one polarization can transmit.
R7 T or F: When light is incident upon a material interface at Brewster’s

angle one of the polarizations stimulates dipoles in the material to
oscillate with orientation along the direction of the reflected k-vector.
R8 T or F: The critical angle for total internal reflection exists on both sides
of a material interface.
R9 T or F: From any given location above a (smooth flat) surface of water,

it is possible to see objects positioned anywhere under the water.
R10 T or F: From any given location beneath a (smooth flat) surface of water,
it is possible to see objects positioned anywhere above the water.
R11 T or F: An evanescent wave travels parallel to an interface surface on

the transmitted side.
139
140 Review, Chapters 1–5
R12 T or F: When p-polarized light enters a material at Brewster’s angle, the

intensity of the transmitted beam is the same as the intensity of the
incident beam.
R13 T or F: For incident angles beyond the critical angle for total internal
reflection, the Fresnel coefficients t s and t p are both zero.
R14 T or F: It is always possible to completely eliminate reflections using a

single-layer antireflection coating if you are free to choose the coating
thickness but not its index.
R15 T or F: For a given incident angle and value of n, there is only one
single-layer coating thickness d that will minimize reflections.
R16 T or F: When coating each surface of a lens with a single-layer antire-

flection coating, the thickness of the coating on the exit surface will
need to be different from the thickness of the coating on the entry
surface.
R17 T or F: As light enters a crystal, the Poynting vector always obeys Snell’s
law.
R18 T or F: As light enters a crystal, the k-vector does not obey Snell’s law
for the extraordinary wave.
Problems
R19 (a) Write down Maxwell’s equations.

(b) Derive the wave equation for E under the assumptions that Jfree = 0
and P = ²0 χE. Note: ∇ × (∇ × f) = ∇ (∇ · f) − ∇2 f.
(c) Show by direct substitution that E (r, t ) = E0 e i (k·r−ωt ) is a solution to
the wave equation. Find the resulting connection between k and ω.
Give appropriate definitions for c and n, assuming that χ is real.
(d) If k = k ẑ and E0 = E 0 x̂, find the associated B-field.
(e) The Poynting vector is S = E × B/µ0 , where the fields are real. Derive
an expression for I ≡ 〈S〉t .
z-axis R20 Consider an interface between two isotropic media where the incident
field is defined by
x-axis
directed into page h i
Ei = E i ŷ cos θi − ẑ sin θi + x̂E i(s) e i [ki ( y sin θi +z cos θi )−ωi t ]
(p)
¡ ¢
The plane of incidence is shown in Fig. 5.12

Figure 5.12 (a) By inspection of the figure, write down similar expressions for the
reflected and transmitted fields (i.e. Er and Et ).
141
(b) Find an expression relating Ei , Er , and Et using the boundary condi-

tion at the interface. From this expression obtain the law of reflection
and Snell’s law.
(c) The boundary condition requiring that the tangential component
of B must be continuous leads to
(p) (p) (p)
n i (E i − E r ) = n t E t
n i (E i(s) − E r(s) ) cos θi = n t E t(s) cos θt

Use this and the results from part (b) to derive
(p)
Er tan (θi − θt )
rp ≡ =−
(p)
Ei tan (θi + θt )
You may use the identity
sin θi cos θi − sin θt cos θt tan (θi − θt )

=
sin θi cos θi + sin θt cos θt tan (θi + θt )
R21 The Fresnel equations are
E r(s) sin θt cos θi − sin θi cos θt

rs ≡ =
E i(s) sin θt cos θi + sin θi cos θt
E t(s) 2 sin θt cos θi

ts ≡ =
Ei (s)
sin θt cos θi + sin θi cos θt
(p)
Er cos θt sin θt − cos θi sin θi
rp ≡ =
Ei
(p)
cos θt sin θt + cos θi sin θi
(p)
Et 2 cos θi sin θt
tp ≡ =
Ei
(p)
cos θt sin θt + cos θi sin θi
(a) Find what each of these equations reduces to when θi = 0. Give your
answer in terms of n i and n t .
(b) What percent of light (intensity) reflects from a glass surface (n =
1.5) when light enters from air (n = 1) at normal incidence?
(c) What percent of light reflects from a glass surface when light exits
into air at normal incidence?
R22 Light goes through a glass prism with optical index n = 1.55. The light
enters at Brewster’s angle and exits at normal incidence as shown in
Fig. 5.13.
(a) Derive and calculate Brewster’s angle θB . You may use the results of Figure 5.13
R20 (c).
(b) Calculate φ.
(c) What percent of the light (power) goes all the way through the prism
if it is p-polarized? You may use the Fresnel coefficients given in R21.
(d) What percent for s-polarized light?
R23 A 45◦ - 90◦ - 45◦ prism is a good device for reflecting a beam of light
parallel to the initial beam (see Fig. 5.14). The exiting beam will be
parallel to the entering beam even when the incoming beam is not
normal to the front surface (although it needs to be in the plane of the
drawing).
(a) How large an angle θ can be tolerated before there is no longer total
internal reflection at both interior surfaces? Assume n = 1 outside of
the prism and n = 1.5 inside.
(b) If the light enters and leaves the prism at normal incidence, what
will the difference in phase be between the s and p-polarizations? You
Figure 5.14 may use the Fresnel coefficients given in R21.
R24 A thin glass plate with index n = 1.5 is oriented at Brewster’s angle so
that p-polarized light with wavelength λvac = 500 nm goes through
with 100% transmittance.
(a) What is the minimum thickness that will make the reflection of
s-polarized light be maximum?
(b) What is the total transmittance T stot for this thickness assuming
s-polarized light?
R25 Consider a Fabry-Perot interferometer. Note: R 1 = R 2 = R.

(a) Show that the free spectral range for a Fabry-Perot interferometer is
λ2
∆λFSR =
2nd cos θ
(b) Show that the fringe width is
λ2
∆λFWHM = p
π F nd cos θ
4R
where F ≡ (1−R)2
.
(c) Derive the reflecting finesse f = ∆λFSR /∆λFWHM .
R26 For a Fabry-Perot etalon, let R = 0.90, λvac = 500 nm, n = 1, and d =
5.0 mm.
(a) Suppose that a maximum transmittance occurs at the angle θ = 0.
What is the nearest angle where the transmittance will be half of the
maximum transmittance? You may assume that cos θ ∼ = 1 − θ 2 /2.
143
(b) You desire to use a Fabry-Perot etalon to view the light from a large
diffuse source rather than a point source. Draw a diagram depicting
where lenses should be placed, indicating relevant distances. Explain
briefly how it works.
R27 You need to make an antireflective coating for a glass lens designed to
work at normal incidence.
The matrix equation relating the incident field to the reflected and
transmitted fields (at normal incidence) is
−i
E0 cos k 1 ` sin k 1 ` E 2
· ¸ · ¸ · ¸· ¸
1 1 n1 1
+ =
n0 −n 0 E 0 −i n 1 sin k 1 ` cos k 1 ` n2 E 0
(a) What is the minimum thickness the coating should have?

HINT: It is less work if you can figure this out without referring to the Figure 5.15
above equation. You may assume n 1 < n 2 .
(b) Find the index of refraction n 1 that will make the reflectivity be zero.
R28 Second harmonic generation (the conversion of light with frequency ω

into light with frequency 2ω) can occur when very intense laser light
travels in a material. For good harmonic production, the laser light
and the second harmonic light need to travel at the same speed in the
material. In other words, both frequencies need to have the same index
of refraction so that harmonic light produced down stream joins in
phase with the harmonic light produced up stream, referred to as phase
matching. This ensures a coherent building of the second harmonic
field rather than destructive cancellations.
Unfortunately, the index of refraction is almost never the same for dif-
ferent frequencies in a given material, owing to dispersion. However,
we can achieve phase matching in some crystals where one frequency
propagates as an ordinary wave and the other propagates as an extraor-
dinary wave. We cause the two indices to be precisely the same by
tuning the angle of the crystal.
Consider a ruby laser propagating and generating the second harmonic
in a uniaxial KDP crystal (potassium dihydrogen phosphate). The
indices of refraction are given by n o and
no ne
q
n o2 sin2 θ + n e2 cos2 θ
where θ is the angle made with the optic axis. At the frequency of a
ruby laser, KDP has indices n o (ω) = 1.505 and n e (ω) = 1.465. At the
frequency of the second harmonic, the indices are n o (2ω) = 1.534 and
n e (2ω) = 1.487.
Show that phase matching can be achieved if the laser is polarized so

that it experiences only the ordinary index and the second harmonic
light is polarized perpendicular to that. At what angle θ does this phase
matching occur?
Selected Answers
R21: (b) 4% (c) 4%.

R22: (b) 33◦ , (c) 95%, (d) 79%.
R23: (a) 4.8◦ , (b) 74◦ .
R24: (a) 100 nm. (b) 0.55.
R26: (a) 0.074◦ .
P27: (b) 1.24.
R28: 51.12◦ .
Chapter 6
Polarization of Light
When the direction of the electric field of light oscillates in a regular, predictable
fashion, we say that the light is polarized. Polarization describes the direction
of the oscillating electric field, a distinct concept from dipoles per volume in a
material P – also called polarization. In this chapter, we develop a formalism for
describing polarized light and the effect of devices that modify polarization. If the
electric field oscillates in a plane, we say that it is linearly polarized. The electric
field can also spiral around while a plane wave propagates, and this is called
elliptical polarization. There is a convenient way for keeping track of polarization
using a two-dimensional Jones vector.
Many devices can affect polarization such as polarizers and wave plates. Their
effects on a light field can be represented by 2 × 2 Jones matrices that operate on
the Jones vector representing the light. A Jones matrix can describe, for example,
a linear polarizer oriented at an arbitrary angle with respect to the coordinate
system. Likewise, a Jones matrix can describe the manner in which a wave plate
introduces a relative phase between two components of the electric field. A wave
plate can be used to convert, for example, linearly polarized light into circularly
polarized light.
In this chapter, we will also see how reflection and transmission at a material Figure 6.1 Animation showing
interface influences field polarization. The Fresnel coefficients studied in chap- different polarization states of
light.
ters 3 and 4 can be conveniently incorporated into the 2×2 matrix formulation for
handling polarization. As we saw previously, the amount of light reflected from a
surface depends on the type of polarization, s or p. In addition, upon reflection,
s-polarized light can acquire a phase lag or phase advance relative to p-polarized
light. This is especially true at metal surfaces, which have complex indices of
refraction. Ellipsometry, outlined in appendix 6.A, is the science of characterizing
optical properties of materials through an examination of these effects.
Throughout this chapter, we consider light to have well characterized polar-
ization. However, most common sources of light (e.g. sunlight or a light bulb)
have an electric-field direction that varies rapidly and randomly. Such sources
are commonly referred to as unpolarized. It is common to have a mixture of
unpolarized and polarized light, called partially polarized light. The Jones vector
145
146 Chapter 6 Polarization of Light
formalism used in this chapter is inappropriate for describing the unpolarized

portions of the light. In appendix 6.B we describe a more general formalism for
dealing with light having an arbitrary degree of polarization.
6.1 Linear, Circular, and Elliptical Polarization

Consider the plane-wave solution to Maxwell’s equations given by
E (r, t ) = E0 e i (k·r−ωt ) (6.1)
The wave vector k specifies the direction of propagation. We neglect absorption

so that the refractive index is real and k = nω/c = 2πn/λvac (see (2.19)–(2.24)). In
an isotropic medium we know that k and E0 are perpendicular, but even after the
direction of k is specified, we are still free to have E0 point anywhere in the two
dimensions perpendicular to k. If we orient our coordinate system with the z-axis
in the direction of k, we can write (6.1) as
E (z, t ) = E x x̂ + E y ŷ e i (kz−ωt )
¡ ¢
(6.2)
As always, only the real part of (6.2) is physically relevant. The complex amplitudes
of E x and E y keep track of the phase of the oscillating field components. In
general the complex phases of E x and E y can differ, so that the wave in one of the
dimensions lags or leads the wave in the other dimension.
The relationship between E x and E y describes the polarization of the light.
+ For example, if E y is zero, the plane wave is said to be linearly polarized along the
x-dimension. Linearly polarized light can have any orientation in the x–y plane,
and it occurs whenever E x and E y have the same complex phase (or a phase
differing by an integer times π). For our purposes, we will take the x-dimension
to be horizontal and the y-dimension to be vertical unless otherwise noted.
As an example, suppose E y = i E x , where E x is real. The y-component of the
field is then out of phase with the x-component by the factor i = e i π/2 . Taking the
real part of the field (6.2) we get
h i h i
E (z, t ) = Re E x e i (kz−ωt ) x̂ + Re e i π/2 E x e i (kz−ωt ) ŷ
= E x cos (kz − ωt ) x̂ + E x cos (kz − ωt + π/2) ŷ (left circular) (6.3)
y
= E x cos (kz − ωt ) x̂ − sin (kz − ωt ) ŷ
£ ¤
x
z In this example, the field in the y-dimension lags behind the field in the x-
Figure 6.2 The combination of
dimension by a quarter cycle. That is, the behavior seen in the x-dimension
two orthogonally polarized plane happens in the y-dimension a quarter cycle later. The field never goes to zero
waves that are out of phase results simultaneously in both dimensions. In fact, in this example the strength of the
in elliptically polarized light. Here electric field is constant, and it rotates in a circular pattern in the x-y dimensions.
we have left circularly polarized For this reason, this type of field is called circularly polarized. Figure 6.2 graph-
light created as specified by (6.3). ically shows the two linear polarized pieces in (6.3) adding to make circularly
polarized light.
6.2 Jones Vectors for Representing Polarization 147
If we view a circularly polarized light field throughout space at a frozen instant

in time (as in Fig. 6.2), the electric field vector spirals as we move along the z-
dimension. If the sense of the spiral (with time frozen) matches that of a common
wood screw oriented along the z-axis, the polarization is called right handed. (It
makes no difference whether the screw is flipped end for end.) If instead the field
spirals in the opposite sense, then the polarization is called left handed. The field
shown in Fig. 6.2 is an example of left-handed circularly polarized light.
An equivalent way to view the handedness convention is to imagine the light
impinging on a screen as a function of time. The field of a right-handed circularly
polarized wave rotates counter clockwise at the screen, when looking along the k
direction (towards the front side of the screen). The field rotates clockwise for a
left-handed circularly polarized wave.
Linearly polarized light can become circularly or, in general, elliptically po-
larized after reflection from a metal surface if the incident light has both s- and
p-polarized components. A good experimentalist working with light needs to
know this. Reflections from multilayer dielectric mirrors can also exhibit these
phase shifts.
6.2 Jones Vectors for Representing Polarization

In 1941, R. Clark Jones introduced a two-dimensional matrix algebra that is useful
for keeping track of light polarization and the effects of optical elements that
influence polarization.1 The algebra deals with light having a definite polarization,
such as plane waves. It does not apply to un-polarized or partially polarized light
(e.g. sunlight). For partially polarized light, a four-dimensional algebra known as
Stokes calculus is used (see Appendix 6.B).
In preparation for introducing Jones vectors, we explicitly write the complex
phases of the field components in (6.2) as R. Clark Jones (19162004, American)
was born in Toledo Ohio. He was one
of six high school seniors to receive a
³ ´
E (z, t ) = |E x |e i φx x̂ + |E y |e i φy ŷ e i (kz−ωt ) (6.4) Harvard College National Prize Fellow-
ship. He earned both his undergraduate
(summa cum laude 1938) and Ph.D.
and then factor (6.4) as follows: degrees from Harvard (1941). After
³ ´ working several years at Bell Labs, he
iδ
E (z, t ) = E eff A x̂ + B e ŷ e i (kz−ωt ) (6.5) spent most of his professional career
at Polaroid Corporation in Cambridge
MA, until his retirement in 1982. He
where is well-known for a series of papers on
q polarization published during the period
|E x |2 + ¯E y ¯ e i φx
¯ ¯2
1941-1956. He also contributed greatly
E eff ≡ (6.6)
to the development of infrared detectors.
|E x | He was an avid train enthusiast, and
A≡ q (6.7) even wrote papers on railway engineer-
63, 519-522
¯ ¯2
|E x |2 + ¯E y ¯ ing. See J. Opt. Soc. Am.
¯ ¯ (1972). Also see SPIE oemagazine,
¯E y ¯ p. 52 (Aug. 2004).
B≡q ¯ ¯2 (6.8)
|E x |2 + ¯E y ¯
1 E. Hecht, Optics, 3rd ed., Sect. 8.12.2 (Massachusetts: Addison-Wesley, 1998).
δ ≡ φ y − φx (6.9)
Please notice that A and B are real non-negative dimensionless numbers that
satisfy A 2 + B 2 = 1. If E y is zero, then B = 0 and everything is well-defined. On the
other hand, if E x happens to be zero, then its phase e i φx is indeterminant. In this
case we let E eff = |E y |e i φy , B = 1, and δ = 0.
The overall field strength E eff is often unimportant in a discussion of polariza-
Linearly polarized along x
tion. It represents the strength of an effective linearly polarized field that would
· ¸
1 correspond to the same intensity as (6.4). Specifically, from (6.5) and (2.62) we
0
have
1 1
Linearly polarized along y I = 〈S〉t = nc²0 E · E∗ = nc²0 |E eff |2 (6.10)
· ¸ 2 2
0
The phase of E eff represents an overall phase shift that one can trivially adjust by
1
physically moving the light source (a laser, say) forward or backward by a fraction
Linearly polarized at angle α of a wavelength.
(measured from the x-axis)
The portion of (6.5) that is relevant to our discussion of polarization is the
cos α vector A x̂+B e i δ ŷ, referred to as the Jones vector. This vector contains the essential
· ¸
sin α information regarding field polarization. Notice that the Jones vector is a kind
Right circularly polarized of unit vector, in that (A x̂ + B e i δ ŷ) · (A x̂ + B e i δ ŷ)∗ = 1. (The asterisk represents
· ¸ the complex conjugate.) When writing a Jones vector we dispense with the x̂ and
1 1
p ŷ notation and organize the components into a column vector (for later use in
2 −i
matrix algebra) as follows:
Left circularly polarized · ¸
A
(6.11)
B eiδ
· ¸
1 1
p
2 i
This vector can describe the polarization state of any plane wave field. Table 6.1
lists some Jones vectors representing various polarization states.
Table 6.1 Jones Vectors for several
common polarization states.
6.3 Elliptically Polarized Light
In general, the Jones vector (6.11) represents a polarization state between linear
and circular. This ‘between’ state is known as elliptically polarized light. As
the wave travels, the field vector makes a spiral motion. If we observe the field
vector at a point as the field goes by, the field vector traces out an ellipse oriented
perpendicular to the direction of travel (i.e. in the x–y plane). One of the axes of
the ellipse occurs at the angle
−1 2AB cos δ
µ ¶
1
α = tan (6.12)
2 A2 − B 2
with respect to the x-axis (see P6.8). This angle sometimes corresponds to the
minor axis and sometimes to the major axis of the ellipse, depending on the exact
values of A, B , and δ. The other axis of the ellipse (major or minor) then occurs at
α ± π/2 (see Fig. 6.3).
We can deduce whether (6.12) corresponds to the major or minor axis of the
ellipse by comparing the strength of the electric field when it spirals through the
6.4 Linear Polarizers and Jones Matrices 149
direction specified by α and when it spirals through α ± π/2. The strength of the
electric field at α is given by (see P6.8)
p
E α = |E eff | A 2 cos2 α + B 2 sin2 α + AB cos δ sin 2α (E max or E min ) (6.13)
and the strength of the field when it spirals through the orthogonal direction
(α ± π/2) is given by
p
E α±π/2 = |E eff | A 2 sin2 α + B 2 cos2 α − AB cos δ sin 2α (E max or E min ) (6.14)
After computing (6.13) and (6.14), we decide which represents E min and which
E max according to
E max ≥ E min (6.15)
We could predict in advance which of (6.13) or (6.14) corresponds to the major
axis and which corresponds to the minor axis. However, making this prediction is
as complicated as simply evaluating (6.13) and (6.14) and determining which is
greater.
Elliptically polarized light is often characterized by the ellipticity, given by the
ratio of the minor axis to the major axis:
E min
e≡ (6.16)
E max
The ellipticity e ranges between zero (corresponding to linearly polarized light) Figure 6.3 The electric field of el-
liptically polarized light traces an
and one (corresponding to circularly polarized light). Finally, the helicity or
ellipse in the plane perpendicular
handedness of elliptically polarized light is as follows (see P6.2): to its propagation direction. The
two plots are for different values
0<δ<π → left-handed helicity (6.17) of A, B , and δ. The angle α can de-
scribe the major axis (top) or the
π < δ < 2π → right-handed helicity (6.18) minor axis (bottom), depending
on the values of these parameters.
6.4 Linear Polarizers and Jones Matrices
In 1928, Edwin Land invented Polaroid at the age of nineteen. He did it by stretch-
ing a polymer sheet and infusing it with iodine. The stretching caused the polymer
chains to align along a common direction, whereupon the sheet was cemented
to a substrate. The infusion of iodine caused the individual chains to become
conductive, like microscopic wires.
When light impinges upon a Polaroid sheet, the component of electric field
that is parallel to the polymer chains causes a current Jfree to oscillate in that
dimension. The resistance to the current quickly dissipates the energy (i.e. the re-
fractive index is complex) and the light is absorbed. The thickness of the Polaroid
sheet is chosen sufficiently large to ensure that virtually none of the light with
electric field component oscillating along the chains makes it through the device.
The component of electric field that is orthogonal to the polymer chains
encounters electrons that are essentially bound to the narrow width of individual
Arbitrary incident polymer molecules. For this polarization component, the wave passes through
polarization the material much like it does through typical dielectrics such as glass (i.e. the
refractive index is real). Today, there is a wide variety of technologies for making
Transmission Axis polarizers, many very different from Polaroid.
A polarizer can be represented as a 2×2 matrix that operates on Jones vectors.2
The function of a polarizer is to pass only the component of electric field that
is oriented along the polarizer transmission axis. Thus, if a polarizer is oriented
with its transmission axis along the x-dimension, then only the x-component
of polarization transmits; the y-component is killed. If the polarizer is oriented
Transmitted polarization with its transmission axis along the y-dimension, then only the y-component of
component
the field transmits, and the x-component is killed. These two scenarios can be
Figure 6.4 Light transmitting represented with the following Jones matrices:
through a Polaroid sheet. The · ¸
1 0
conducting polymer chains run (polarizer with transmission along x-axis) (6.19)
vertically in this drawing, and 0 0
light polarized along the chains · ¸
0 0
is absorbed. Light polarized per- (polarizer with transmission along y-axis) (6.20)
pendicular to the polymer chains
0 1
passes through the polarizer. These matrices operate on any Jones vector representing the polarization of
incident light. The result gives the Jones vector for the light exiting the polarizer.
Example 6.1
Use the Jones matrix (6.19) to calculate the effect of a horizontal polarizer on
light that is initially horizontally polarized, vertically polarized, and arbitrarily
polarized.
Solution: First we consider a horizontally polarized plane wave traversing a polar-

izer with its transmission axis oriented also horizontally (x-dimension):
· ¸· ¸ · ¸
1 0 1 1
= (horizontal polarizer on horizontally polarized field)
0 0 0 0
As expected, the polarization state is unaffected by the polarizer. (We have ignored
possible attenuation from surface reflections.)
Now consider vertically polarized light traversing the same horizontal polarizer. In
this case, we have:
· ¸· ¸ · ¸
1 0 0 0
= (horizontal polarizer on vertical linear polarization)
0 0 1 0
As expected, the polarizer extinguishes the light.
Finally, when a horizontally oriented polarizer operates on light with an arbitrary
Jones vector (6.11), we have
· ¸· ¸ · ¸
1 0 A A
= (horizontal polarizer on arbitrary polarization)
0 0 B eiδ 0
Only the horizontal component of polarization is transmitted through the polar-
izer.
6.4 Linear Polarizers and Jones Matrices 151
While you might readily agree that the matrices given in (6.19) and (6.20)
can be used to get the right result for light traversing a horizontal or a vertical
polarizer, you probably aren’t very impressed as of yet. In the next few sections,
we will derive Jones matrices for a number of optical elements that can modify
polarization: polarizers at arbitrary angle, wave plates at arbitrary angle, and
reflection or transmissions at an interface. Table 6.2 shows Jones matrices for
each of these devices. Before deriving these specific Jones matrices, however, we
take a moment to appreciate why the Jones matrix formulation is useful.
The real power of the formalism becomes clear as we consider situations Linear polarizer
where light encounters multiple polarization elements in sequence. In these situ- ·
cos2 θ sin θ cos θ
¸
ations, we use a product of Jones matrices to represent the effect of the compound sin θ cos θ sin2 θ
systems. We can represent this situation by
Half wave plate
· 0 ¸ · ¸
A A ·
cos 2θ sin 2θ
¸
= Jsystem (6.21)
B0 B eiδ sin 2θ − cos 2θ
Quarter wave plate

where the unprimed Jones vector represents light going into the system and the
primed Jones vector represents light emerging from the system. In general, A 0 cos2 θ + i sin2 θ (1 − i ) sin θ cos θ
· ¸
and B 0 will turn out to be complex. However, if desired they can be changed into (1 − i ) sin θ cos θ sin2 θ + i cos2 θ
the usual form by writing Right circular polarizer

· ¸
1 1 i
A0 |A 0 |
· ¸ · ¸
i φ A0
=e 2 −i 1
|B 0 |e i δ
0
B0
Left circular polarizer
where φ A 0 is an unimportant overall phase, and δ0 is the phase difference between · ¸
1 1 −i
B 0 and A 0 . 2 i 1
The matrix Jsystem is a Jones matrix formed by the series polarization devices.
Reflection from an interface
If there are N devices in the system, the compound matrix is calculated as · ¸
−r p 0
Jsystem ≡ JN JN −1 · · · J2 J1 (6.22) 0 rs
Transmission through an
where Jn is the matrix for the n th polarizing optical element encountered in the interface
system. Notice that the matrices operate on the Jones vector in the order that · ¸
tp 0
the light encounters the devices. Therefore, the matrix for the first device (J1 ) is 0 ts
written on the right, and so on until the last device encountered, which is written
on the left, farthest from the Jones vector.
When part of the light is absorbed by passing through one or more polarizers Table 6.2 Common Jones Matri-
ces. The angle θ is measured with
in a system, the Jones vector of the exiting light does not necessarily remain
respect to the x-axis and specifies
normalized to magnitude one (see Example 6.1). Since the components of a Jones the transmission axis of a linear
vector represent the electric field, we find the factor by which the intensity of the polarizer or the fast axis of a wave
light decreases by dotting the vector with its complex conjugate. In accordance plate.
with (6.10), the intensity of the exiting light is
1 ³ ´ ³ ´∗
I = nc²0 |E eff |2 A 0 x̂ + B 0 e i δ ŷ · A 0 x̂ + B 0 e i δ ŷ
0 0
2 (6.23)
1 ³¯ ¯ ¯ ¯ ´
2 2
= nc²0 |E eff |2 ¯ A 0 ¯ + ¯B 0 ¯
2
¯ ¯2 ¯ ¯2
Notice that the intensity is attenuated by the factor ¯ A 0 ¯ + ¯B 0 ¯ after propagating
through the system. Recall that E eff represents the effective strength of the field
before it enters the polarizer (or other device), so that the initial Jones vector is
normalized to one (see (6.10)). As a reminder, we normally remove an overall
phase factor from the Jones vector so that A 0 is real and non-negative, and we
choose δ0 so that B 0 is real and non-negative. However, if we don’t bother doing
this, the absolute value signs on A 0 and B 0 in (6.23) ensure that we get the correct
value for intensity.
6.5 Jones Matrix for Polarizers at Arbitrary Angles

Incident Light In this section we develop a Jones matrix for describing an ideal polarizer with
its transmission axis at an arbitrary angle θ from the x-axis. We will do this in a
general context so that we can take advantage of the present work when discussing
Transmission wave plates in the next section. To help keep things on a more conceptual level,
Axis
we revert back to using electric field components directly. We will make the
connection with Jones calculus at the end.
The polarizer acts on a plane wave with arbitrary polarization. The electric
field of our plane wave may be written as
Transmitted
E (z, t ) = E x x̂ + E y ŷ e i (kz−ωt )
¡ ¢
component
(6.24)
Let the transmission axis of the polarizer be specified by the unit vector ê1
Figure 6.5 Light transmitting and the absorption axis of the polarizer be specified by ê2 (orthogonal to the
through a polarizer oriented with
transmission axis). The vector ê1 is oriented at an angle θ from the x-axis, as
transmission axis at angle θ from
x-axis.
shown in Fig. 6.6. We need to write the electric field components in terms of the
new basis specified by ê1 and ê2 . By inspection of the geometry, the x-y unit
vectors are connected to the new coordinate system via:
x̂ = cos θê1 − sin θê2

(6.25)
ŷ = sin θê1 + cos θê2
Substitution of (6.25) into (6.24) yields for the electric field
E (z, t ) = (E 1 ê1 + E 2 ê2 ) e i (kz−ωt ) (6.26)
where
E 1 ≡ E x cos θ + E y sin θ
(6.27)
E 2 ≡ −E x sin θ + E y cos θ
Now we introduce the effect of the polarizer on the field: E 1 is transmitted
unaffected, while E 2 is extinguished. To account for the effect of the device, we
multiply E 2 by a parameter ξ. In the case of the polarizer, ξ is zero, but when we
consider wave plates we will use other values for ξ. After traversing the polarizer,
the field becomes
Figure 6.6 Electric field compo- Eafter (z, t ) = (E 1 ê1 + ξE 2 ê2 ) e i (kz−ωt ) (6.28)
nents written in the ê1 –ê2 basis.
6.6 Jones Matrices for Wave Plates 153
We now have the field after the polarizer, but it would be nice to rewrite it in terms
of the original x–y basis. By inverting (6.25), or again by inspection of Fig. 6.6, we
see that
ê1 = cos θx̂ + sin θŷ
(6.29)
ê2 = − sin θx̂ + cos θŷ
Substitution of these relationships into (6.28) together with the definitions (6.27)
for E 1 and E 2 yields
Eafter (z, t ) = E x cos θ + E y sin θ cos θx̂ + sin θŷ

£¡ ¢¡ ¢
+ξ −E x sin θ + E y cos θ − sin θx̂ + cos θŷ e i (kz−ωt )

¡ ¢¡ ¢¤
= E x cos2 θ + ξ sin2 θ + E y (sin θ cos θ − ξ sin θ cos θ) x̂e i (kz−ωt )

£ ¡ ¢ ¤
+ E x (sin θ cos θ − ξ sin θ cos θ) + E y sin2 θ + ξ cos2 θ ŷe i (kz−ωt )

£ ¡ ¢¤
(6.30)
Notice that if ξ = 1 (i.e. no polarizer), then we get back exactly what we started
with (i.e. (6.30) reduces to (6.24)).
To get to the Jones matrix for the polarizer, we note that (6.30) is a linear mix-
ture of E x and E y which can be represented with matrix algebra. If we represent
the electric field as a two-dimensional column vector with its x component in the
top and its y component in the bottom (like a Jones vector), then we can rewrite
(6.30) as
cos2 θ + ξ sin2 θ sin θ cos θ − ξ sin θ cos θ

· ¸· ¸
Ex
Eafter (z, t ) = e i (kz−ωt )
sin θ cos θ − ξ sin θ cos θ sin2 θ + ξ cos2 θ Ey
(6.31)
The matrix here is a properly normalized Jones matrix, even though we did not
bother factoring out E eff to make a properly normalized Jones vector, as specified
in (6.5). We can now write down the Jones matrix for a polarizer by inserting ξ = 0
into the matrix:
cos2 θ sin θ cos θ
· ¸
(polarizer with transmission axis at angle θ) (6.32)
sin θ cos θ sin2 θ
Notice that when θ = 0 this matrix reduces to that of a horizontal polarizer (6.19),
and when θ = π/2, it reduces to that of a vertical polarizer (6.20).
6.6 Jones Matrices for Wave Plates

We next consider wave plates (or retarders), which are usually made from birefrin-
gent crystals. The index of refraction in the crystal depends on the orientation of
the electric field polarization. A wave plate has the appearance of a thin window
through which the light passes. The crystal is cut such that the wave plate has
a fast and a slow axis, which are 90◦ apart in the plane of the window. If the
light is polarized along the fast axis, it experiences index n fast . The orthogonal
polarization component experiences higher index n slow .
When a plane wave passes through a wave plate, the component of the electric
field oriented along the fast axis travels faster than its orthogonal counterpart,
which introduces a relative phase between the two polarization components.
As light passes through a wave plate of thickness d , the phase difference that
accumulates between the fast and the slow polarization components is
2πd
k slow d − k fast d = (n slow − n fast ) (6.33)
λvac
By adjusting the thickness of the wave plate, one can introduce any desired phase
difference.
Slow axis The most common types of wave plates are the quarter-wave plate and the
half-wave plate. The quarter-wave plate introduces a phase difference of
k slow d − k fast d = π/2 + 2πm (quarter-wave plate) (6.34)

Fast axis
between the two polarization components, where m is an integer. This means
that the polarization component along the slow axis is delayed spatially by a
Waveplate quarter wavelength (or five quarters, etc.).
The half-wave plate introduces a phase difference of
Transmitted polarization
components have altered k slow d − k fast d = π + 2πm (half-wave plate) (6.35)
relative phase
where m is an integer. This means that the polarization component along the
Figure 6.7 Wave plate interacting
with a plane wave.
slow axis is delayed spatially by a half wavelength (or three halves, etc.). When
m = 0 in either (6.34) or (6.35), the wave plate is said to be zero order.
The derivation of the Jones matrix for the two wave plates is essentially the
same as the derivation for the polarizer in the previous section. Let ê1 correspond
to the fast axis, and let ê2 correspond to the slow axis, as illustrated in Fig. 6.7. We
proceed as before. However, instead of setting ξ equal to zero in (6.31), we must
choose values for ξ appropriate for each wave plate. Since nothing is absorbed,
ξ should have a magnitude equal to one. The important feature is the phase of
ξ. As seen in (6.33), the field component along the slow axis accumulates excess
phase relative to the component along the fast axis, and we let ξ account for this.
In the case of the quarter-wave plate, the appropriate factor from (6.34) is
ξ = e i π/2 = i (quarter-wave plate) (6.36)
This describes a relative phase delay for the light emerging with polarization along
the slow axis. Substituting (6.36) into (6.30) yields the Jones matrix for a quarter
wave plate:
cos2 θ + i sin2 θ sin θ cos θ − i sin θ cos θ

· ¸
quarter-wave plate Jones matrix (6.37)
sin θ cos θ − i sin θ cos θ sin2 θ + i cos2 θ
For the half-wave plate, the appropriate factor applied to the slow axis is
ξ = e i π = −1 (half-wave plate) (6.38)

6.6 Jones Matrices for Wave Plates 155
and the Jones matrix becomes:
cos2 θ − sin2 θ 2 sin θ cos θ

· ¸ · ¸
cos 2θ sin 2θ
= (6.39) half-wave plate Jones matrix
2 sin θ cos θ sin2 θ − cos2 θ sin 2θ − cos 2θ
Remember that θ refers to the angle that the fast axis makes with respect to the
x-axis.
Before moving on, consider the following two examples that illustrate how
wave plates are often used:
Example 6.2
Calculate the Jones matrix for a quarter-wave plate at θ = 45◦ , and determine its
effect on horizontally polarized light.
Solution: At θ = 45◦ , the Jones matrix for the quarter-wave plate (6.37) reduces to
Figure 6.8 Animation showing
e i π/4
· ¸
1 −i ◦ effects of polarizers and wave
p (quarter-wave plate, fast axis at θ = 45 ) (6.40)
2 −i 1 plates on polarized light.
The overall phase factor e i π/4 in front is not important since it merely accompanies
the overall phase of the beam, which can be adjusted arbitrarily by moving the
light source forwards or backwards through a fraction of a wavelength.
Now we calculate the effect of the quarter-wave plates (oriented at θ = 45◦ ) operat-
ing on horizontally polarized light:
· ¸· ¸ · ¸
1 1 −i 1 1 1
p =p (6.41)
2 −i 1 0 2 −i
The previous example shows that a quarter-wave plate (properly oriented) can
turn linearly polarized light into right-circularly polarized light (see Table 6.1).
On the other hand, as seen in the next example, a half-wave plate can rotate the
polarization angle of linearly polarized light by varying degrees while preserving
the linear polarization.
Example 6.3
Calculate the effect of a half wave plate at an arbitrary θ on horizontally polarized
light.
Solution: Carrying out the multiplication, we obtain

· ¸· ¸ · ¸
cos 2θ sin 2θ 1 cos 2θ
= (6.42)
sin 2θ − cos 2θ 0 sin 2θ
The resulting Jones vector describes linearly polarized light an angle of α = 2θ from
the x-axis (see Table 6.1).
6.7 Polarization Effects of Reflection and Transmission

When light encounters a material interface, the amount of reflected and trans-
mitted light depends on the polarization. The Fresnel coefficients (3.18)–(3.21)
dictate how much of each polarization is reflected and how much is transmitted.
In addition, the Fresnel coefficients keep track of phases intrinsic in the reflec-
tion phenomenon. This is true also for reflections from multilayer coatings with
effective Fresnel coefficients (4.60), (4.61), (4.65) and (4.66).
To the extent that the s and p components of the field behave differently,
the overall polarization state is altered. For example, a linearly-polarized field
upon reflection can become elliptically polarized (see L 6.9). Even when a wave
reflects at normal incidence so that the s and p components are indistinguishable,
right-circular polarized light becomes left-circular polarized. This is the same
effect that causes a right-handed person to appear left-handed when viewed in a
mirror.
We can use Jones calculus to keep track of how reflection and transmission
influences polarization. However, before proceeding, we emphasize that in this
context we do not strictly adhere to a single coordinate system as we did in
chapter 3, for example in Fig. 3.1. Instead, we consider each plane wave, whether
incident, reflected or transmitted, to propagate in the z-direction of its own frame,
x-axis regardless of the relative angles between the incident and reflected wave. This
directed into page loose manner of defining coordinate systems, depicted in Fig. 6.9, has a great
advantage. The x and y dimensions in each individual frame is aligned parallel
to their respective s and p field component. We will adopt the convention that
p-polarized light in all cases is associated with the x-dimension (horizontal, say).
The s-polarized component then lies along the y-dimension (vertical).
We are now in a position to see why there is a handedness inversion upon
Figure 6.9 Incident, reflected and reflection from a mirror. Notice in Fig. 6.9 that for the incident light, the s-
transmitted plane waves, each
component of the field crossed into the p-component of the field yields a vector
propagating along the z-axis of its
own reference frame.
pointing along beam’s propagation direction. However, for the reflected light,
the s-component crossed into the p-component points opposite to that beam’s
propagation direction.
The Jones matrix corresponding to reflection from a surface is
· ¸
−r p 0
(Jones matrix for reflection) (6.43)
0 rs
By convention, we place the minus sign on the coefficient r p to take care of

handedness inversion. We could put the minus sign on r s instead; the important
point is that the two polarizations acquire a relative phase differential of π when
the propagation direction flips.3
The Fresnel coefficients specify the ratios of the exiting fields to the incident
ones. When (6.43) operates on an arbitrary Jones vector such as (6.11), −r p
3 The minus sign is needed for our specific convention of field directions, as drawn in Fig. 6.9. In
our convention, r s and r p are identical at normal incidence.

6.A Ellipsometry 157
multiplies the horizontal component of the field, and r s multiplies the vertical
component of the field. Especially in the case of reflection from an absorbing
surface such as a metal, the phases of the two polarization components can vary
markedly (see P6.11). Thus, linearly polarized light containing both s- and p-
components in general becomes elliptically polarized when reflected from such a
surface. When light undergoes total internal reflection, again the phases of the s-
and p-components differ markedly, which can cause linearly polarized light to
become elliptically polarized (see P6.12).
Transmission through a material interface can also influence the polarization
of the field, although typically to a lesser degree. However, there is no handedness
inversion, since the light continues on in a forward sense. The Jones matrix for
transmission is
· ¸
tp 0
(Jones matrix for transmission) (6.44)
0 ts
If a beam of light encounters a series of mirrors, the final polarization is

determined by multiplying the sequence of appropriate Jones matrices (6.43) rotated x-axis
(in the plane of incidence)
onto the initial polarization. This procedure is straightforward if the normals
to all of the mirrors lie in a single plane (say parallel to the surface of an optical
original
bench). However, if the beam path deviates from this plane (due to vertical y-axis
tilt on the mirrors), then we must reorient our coordinate system before each
mirror to have a new ‘horizontal’ (p-polarized dimension) and the new ‘vertical’ rotated
(s-polarized dimension). Earlier in this chapter we performed a rotation of a y-axis
original
coordinate system through an angle θ, described in (6.27), which is also useful x-axis
here. The rotation can be accomplished by multiplying the following matrix onto
the incident Jones vector: Figure 6.10 If the plane of inci-
dence does not coincide for suc-
cos θ sin θ cessive elements in an optical
· ¸
(rotation of coordinates through an angle θ) (6.45) system, a rotation matrix must be
− sin θ cos θ
applied to rotate the x-axis to the
This is understood as a rotation about the z-axis. The angle of rotation θ is plane of incidence before comput-
chosen such that the rotated x-axis lies in the plane of incidence for the mirror. ing the effect of each element.
When such a reorientation of coordinates is necessary, the two orthogonal field
components in the initial coordinate system are stirred together to form the field
components in the new system. This does not change the intrinsic characteristics
of the polarization, just its representation.
Appendix 6.A Ellipsometry

Measuring the polarization of light reflected from a surface can yield information
regarding the optical constants of that surface (i.e. n and κ). As done in L 6.9, it is
possible to characterize the polarization of a beam of light using a quarter-wave
plate and a polarizer. However, we often want to know n and κ at a range of
frequencies, and this would require a different quarter-wave plate thickness d
for each wavelength used (see (6.34)). Therefore, many commercial ellipsometers
do not try to extract the helicity of the light, but only the ellipticity. In this case
only polarizers are needed, which can be made to work over a wide range of
wavelengths. If, in addition, a variety of incident angles are measured, it is possible
to extract detailed information about the optical constants n and κ and the
thicknesses of possibly many layers of materials influencing the reflection.
Commercial ellipsometers4 typically employ two polarizers, one before and
one after the sample, where s and p-polarized reflections take place. The first
polarizer ensures that linearly polarized light arrives at the test surface (polarized
at angle α to give both s and p-components). The Jones matrix for the test surface
reflection is given by (6.43), and the Jones matrix for the analyzing polarizer
oriented at angle θ is given by (6.32). The Jones vector for the light arriving at the
detector is then
cos2 θ sin θ cos θ cos α
· ¸· ¸· ¸
−r p 0
sin θ cos θ sin2 θ 0 rs sin α
−r p cos α cos2 θ + r s sin α sin θ cos θ
· ¸
= (6.46)
−r p cos α sin θ cos θ + r s sin α sin2 θ
and the intensity arriving to the detector is
¯2 ¯ ¯2
I ∝ ¯−r p cos α cos2 θ + r s sin α cos θ sin θ ¯ +¯−r p cos α cos θ sin θ + r s sin α sin2 θ ¯
¯
³ ´
¯ ¯2 r p r s∗ + r s r p∗
= ¯r p ¯ cos2 α cos2 θ + |r s |2 sin2 α sin2 θ − sin 2α sin 2θ
4
(6.47)
For ellipsometry measurements, it is customary to express the ratio of Fresnel
coefficients as
r p r s ≡ tan Ψe i ∆
±
(6.48)
In this case, the intensity may be shown to be proportional to (see problem P6.13)
I ∝ 1 − η sin 2θ + ξ cos 2θ (6.49)
where
tan Ψ cos ∆ tan α tan2 Ψ − tan2 α
η≡2 and ξ ≡ (6.50)
tan2 Ψ + tan2 α tan2 Ψ + tan2 α
In commercial ellipsometers, the angle θ of the analyzing polarizer often rotates at
a high speed, and the time dependence of the light reaching a detector is analyzed.
From this type of measurement, the coefficients η and ξ can be extracted with
high precision. Then equations (6.50) can be inverted (see problem P6.13) to
reveal s
1+ξ η
tan Ψ = |tan α| and cos ∆ = p sign(α) (6.51)
1−ξ 1 − ξ2
From a series of these types of measurements, it is possible to extract the values
of n and κ for materials from the expressions for r s and r p (with the aid of a
computer!). A more extensive series of such measurements are needed in the case
of multilayers involving multiple layers with varying thicknesses.
4 See Spectroscopic Ellipsometry Tutorial at J. A. Woollam Co.
6.B Partially Polarized Light 159
Appendix 6.B Partially Polarized Light

We outline here an approach for dealing with partially polarized light, which is a
mixture of polarized and unpolarized light. Most natural light such as sunshine is
unpolarized. The transverse electric field direction in natural light varies rapidly
(and quasi randomly). Such variations imply the superposition of multiple fre-
quencies as opposed to the single frequency assumed in the formulation of Jones
calculus earlier in this chapter. Unpolarized light can become partially polarized
when it, for example, reflects from a surface at oblique incidence, since s and p
components of the polarization might reflect with differing strength.
Stokes vectors are used to keep track of the partial polarization (and atten-
uation) of a light beam as the light progresses through an optical system.5 In
contrast, Jones vectors are designed for pure polarization states. We can consider
any light beam as an intensity sum of completely unpolarized light and perfectly
polarized light: Sir George Gabriel Stokes (18191903,
Irish) was born in Skreen, Ireland. He
I = I pol + I un (6.52) entered Cambridge University at age 18
and graduated four years later with the
It is assumed that both types of light propagate in the same direction. distinction of senior wrangler. In 1849,
The main characteristic of unpolarized light is that it cannot be extinguished he became a professor of mathematics
at Cambridge where he later worked
by a single polarizer (even in combination with a wave plate). Moreover, the with James Clerk Maxwell and Lord
transmission of unpolarized light through an ideal polarizer is always 50%. On the Kelvin to form the Cambridge School
of Mathematical Physics. Stokes was a
other hand, polarized light (be it linearly, circularly, or elliptically polarized) can powerful mathematician as well as good
always be represented by a Jones vector, and it is always possible to extinguish experimentalist, often testing his the-
oretical solutions in the laboratory. In
polarized light with a wave plate and a single polarizer.
addition to his contributions to optics,
We may introduce the degree of polarization as the fraction of the intensity Stokes made important contributions to
that is in a definite polarization state: uid dynamics (e.g. the Navier-Stokes

equations) and to mathematical physics;
Stokes' theorem is employed several
I pol
ξpol ≡ (6.53) places in this in this book.(Wikipedia)
I pol + I un
The degree of polarization takes on values between zero and one. Thus, if the light
is completely unpolarized (such that I pol = 0), the degree of polarization is zero,
and if the beam is fully polarized (such that I un = 0), the degree of polarization is
one.
A Stokes vector, which characterizes a partially polarized beam, is written as
 
S0
 S 
 1 
 S2 
 
S3
The parameter
I
S0 ≡ (6.54)
I in
is a comparison of the beam’s intensity (or power) to a benchmark or ’input’

intensity, I in , measured before the beam enters the optical system under consid-
eration. I represents the intensity at the point of investigation, where one wishes
to characterize the beam. Thus, the value S 0 = 1 represents the input intensity,
and S 0 can drop to values less than one, to account for attenuation of light by
polarizers in the system. (S 0 could increase in the atypical case of amplification.)
The next parameter, S 1 , describes how much the light looks either horizontally
or vertically polarized, and it is defined as
2I hor
S1 ≡ − S0 (6.55)
I in
Here, I hor represents the amount of light detected if an ideal linear polarizer is
placed with its axis aligned horizontally directly in front of the detector (inserted
where the light is characterized). S 1 ranges between negative one and one, taking
on its extremes when the light is linearly polarized either horizontally or vertically,
respectively. If the light has been attenuated, it may still be perfectly horizontally
polarized even if S 1 has a magnitude less than one. (Alternatively, you might wish
to examine S 1 /S 0 , which is guaranteed to range between negative one and one.)
The parameter S 2 describes how much the light looks linearly polarized along
the diagonals. It is given by
2I 45◦
S2 ≡ − S0 (6.56)
I in
Similar to the previous case, I 45◦ represents the amount of light detected if an
ideal linear polarizer is placed with its axis at 45◦ directly in front of the detector
(inserted where the light is characterized). As before, S 2 ranges between negative
one and one, taking on extremes when the light is linearly polarized either at 45◦
or 135◦ .
Finally, S 3 characterizes the extent to which the beam is either right or left
circularly polarized:
2I r-cir
S3 ≡ − S0 (6.57)
I in
Here, I r-cir represents the amount of light detected if an ideal right-circular po-
larizer is placed directly in front of the detector. A right-circular polarizer is
one that passes right-handed polarized light, but blocks left handed polarized
light. One way to construct such a polarizer is a quarter wave plate, followed
by a linear polarizer with the transmission axis aligned 45◦ from the wave-plate
fast axis, followed by another quarter wave plate at −45◦ from the polarizer (see
P6.14).6 Again, this parameter ranges between negative one and one, taking on
the extremes for right and left circular polarization, respectively.
Importantly, if any of the parameters S 1 , S 2 , or S 3 take on their extreme values
(i.e. a magnitude equal to S 0 ), the other two parameters necessarily equal zero. As
an example, if a beam is linearly horizontally polarized with I = I in , then we have
6 The final quarter wave plate is to put the light back into the original circular state – not needed
to measure the Stokes parameter.

I hor = I in , I 45◦ = I in /2, and I r-cir = I in /2. This yields S 0 = 1, S 1 = 1, S 2 = 0, and S 3 = 0.

As a second example, suppose that the light has been attenuated to I = I in /3 but is
purely left circularly polarized. Then we have I hor = I in /6, I 45◦ = I in /6, and I r-cir = 0.
Whereas the Stokes parameters are S 0 = 1/3, S 1 = 0, S 2 = 0, and S 3 = −1/3.
Another interesting case is completely unpolarized light, which transmits 50%
through all of the polarizers discussed above. In this case, I hor = I 45◦ = I r-cir = I /2
and S 1 = S 2 = S 3 = 0.
Example 6.4
Find the Stokes
£ parameters for perfectly polarized light, represented by an arbitrary
Jones vector BA where A and B are complex.7 Depending on the values A and B ,
¤
the polarization can follow any ellipse.
Solution: The input intensity of this polarized beam is I in = I pol = |A|2 + |B |2 , ac-
cording to Eq. (6.23), where we absorb the factor 12 ²0 c |E eff |2 into |A|2 and |B |2
for convenience. The Jones vector for the light that passes through a horizontal
polarizer is · ¸· ¸ · ¸
1 0 A A
=
0 0 B 0
which gives a measured intensity of I hor = |A|2 . Similarly, the Jones vector when
the beam is passed through a polarizer oriented at 45◦ is
A +B
· ¸· ¸ · ¸
1 1 1 A 1
=
2 1 1 B 2 1
leading to an intensity of
|A + B |2 |A|2 + |B |2 + A ∗ B + AB ∗
I 45◦ = =
2 2
Finally, the Jones vector for light passing through a right-circular polarizer (see
P6.14) is
A +iB
· ¸· ¸ · ¸
1 1 i A 1
=
2 −i 1 B 2 −i
giving an intensity of
|A + i B |2 |A|2 + |B |2 + i (A ∗ B − AB ∗ )
I r-cir = =
2 2
Thus, the Stokes parameters become
|A|2 + |B |2
S0 = =1
I in
2|A|2 |A|2 + |B |2 |A|2 − |B |2

S1 = − =
I in I in I in
· ¸ · ¸
7 We will find it easier in this appendix to write A |A|
instead of , where δ is the phase
B |B |e i δ
difference between B and A.
|A|2 + |B |2 + A ∗ B + AB ∗ |A|2 + |B |2 A ∗ B + AB ∗
S2 = − =
I in I in I in
|A|2 + |B |2 + i (A ∗ B − AB ∗ ) |A|2 + |B |2 (A ∗ B − AB ∗ )
S3 = − =i
I in I in I in
It is clear from the linear dependence of S 0 , S 1 , S 2 , and S 3 on intensity (see

Eqs. (6.54)–(6.57)) that the overall Stokes vector may be regarded as the sum of
the individual Stokes vectors for polarized and unpolarized light. That is, we may
(pol)
write S j = S j + S (un)
j
, j = 0, 1, 2, 3.
This is certainly true for
I I pol + I un
S0 = = (6.58)
I in I in
and in the other cases the unpolarized portion of the light does not contribute to
the Stokes parameters. Half of the unpolarized light survives any of the test filters,
which cancels neatly with the unpolarized portion of S 0 in Eqs. (6.55)–(6.57).
With the aid of the results in Example 6.4, a completely general form of the
Stokes vector may then be written as
   
S0 I pol + I un
 S  1  |A|2 − |B |2 
 1 
= (6.59)
 
 S 2  I in  A ∗ B + AB ∗ 
  
S3 i (A ∗ B − AB ∗ )
where the Jones vector for the polarized portion of the light is
· ¸
A
B
and the intensity of the polarized portion of the light is
I pol = |A|2 + |B |2 (6.60)
Again, we have hidden the factor 12 ²0 c |E eff |2 for the polarized portion of the light
inside |A|2 and |B |2 .
We would like to express the degree ofq polarization in terms of the Stokes
parameters. We first note that the quantity S 12 + S 22 + S 32 can be expressed as
s
¶2 ¶2 ¶2
|A|2 − |B |2 (A ∗ B + AB ∗ ) i (A ∗ B − AB ∗ )
q µ µ µ
S 12 + S 22 + S 32 = + +
I in I in I in
|A|2 + |B |2 (6.61)
=
I in
I pol
=
I in
Substituting (6.58) and (6.61) into the expression for the degree of polarization
(6.53) yields
1q 2
ξpol ≡ S 1 + S 22 + S 32 (6.62)
S0
If the light is polarized such that it perfectly transmits through or is perfectly
extinguished by one of the three test polarizers associated with S 1 , S 2 , or S 3 , then
the degree of polarization will be unity. Obviously, it is possible to have pure
polarization states that are not aligned with the axes of any one of these test
polarizers. In this situation, the degree of polarization is still one, although the
values S 1 , S 2 , and S 3 may all three contribute to (6.62).
Finally, it is possible to represent polarizing devices as matrices that operate
on the Stokes vectors in much the same way that Jones matrices operate on
Jones vectors. Since Stokes vectors are four-dimensional, the matrices used are
four-by-four. These are known as Mueller matrices.8
Derivation: Mueller Matrix for a Linear Polarizer
We know that the 50% of the unpolarized light transmits through a polarizer,
ending up as polarized light with Jones vector
r
A 01 I un cos θ
· ¸ · ¸
=
B 10 2 sin θ
(see table 6.1). As usual, let θ give the angle of the transmission axis relative to the
horizontal. The Jones matrix (6.23) acts on the polarized portion of the light as
follows
Hans Mueller (Swiss) was a shepherd
cos2 θ cos θ sin θ cos θ
· 0 ¸ · ¸· ¸ · ¸
A2 A
= = [A cos θ + B sin θ] until his late teens. As a physics pro-
B 20 cos θ sin θ sin2 θ B sin θ fessor at MIT, he built on the work of
Stokes and in 1943 formulated a ma-
trix method for manipulating Stokes
A 01 A 02
h i h i
One might be tempted to add B 10
and B 20
, but this would be wrong, since vectors. He was an engaging lecturer
into the 1950s and was known for his
the two beams are not coherent. As mentioned previously, unpolarized light exciting demonstrations. He was a stu-
necessarily contains multiple frequencies, and so the fields from the polarized and dent of Arnold Sommerfeld, and did
unpolarized beams destructively interfere as often as they constructively interfere. seminal work on ferroelectricity (he is
reported to have coined the term). See
In this case, we simply add intensities rather than fields. That is, we have
Laszlo Tisza,Adventures of a Theoret-
ical Physicist, Part II: America, Phys.
¯ A ¯ = ¯ A ¯ + ¯ A ¯ = I un + |A cos θ + B sin θ|2 cos2 θ
· ¸
11, 120-168 (2009).
¯ 0 ¯2 ¯ 0 ¯2 ¯ 0 ¯2
Perspect.
1 2
2
I un
· ¸
+ |A| cos θ + |B |2 sin2 θ + A ∗ B + AB ∗ sin θ cos θ cos2 θ
2 2
¡ ¢
=
2
S 0 cos 2θ
· ¸
sin 2θ
= I in + S1 + S 2 cos2 θ
2 2 2
Similarly,
¯B ¯ = ¯B ¯ + ¯B ¯ = I in S 0 + cos 2θ S 1 + sin 2θ S 2 sin2 θ

· ¸
¯ 0 ¯2 ¯ 0 ¯2 ¯ 0 ¯2
1 2
2 2 2
Since the light has gone through a linear polarizer, we are guaranteed that A 0 and
B 0 have the same phase. Therefore, A 0∗ B 0 = A 0 B 0∗ = |A 0 ||B 0 |. In view of (6.59), these
results lead to
¯ 0 ¯2 ¯ 0 ¯2
¯ A ¯ + ¯B ¯ S 0 cos 2θ sin 2θ
0
S0 = = + S1 + S2
I in 2 2 2
¯ 0 ¯2 ¯ 0 ¯2 ·
¯ A ¯ − ¯B ¯ S 0 cos 2θ sin 2θ
¸
0
S 2 cos2 θ − sin2 θ
¡ ¢
S1 = = + S1 +
I in 2 2 2
cos 2θ cos2 2θ sin 4θ
= S0 + S1 + S2
2
¯ 0¯ ¯ 0¯ ¯ 0¯ ¯ 0¯2 4
¯ A ¯ ¯B ¯ + ¯ A ¯ ¯B ¯ ·
S 0 cos 2θ sin 2θ
¸
S 20 = =2 + S1 + S 2 cos θ sin θ
I in 2 2 2
sin 2θ sin 4θ sin2 2θ
= S0 + S1 + S2
¯ 20 ¯ ¯ 0 ¯ ¯ 40 ¯ ¯ 0 ¯ 2
¯ A ¯ ¯B ¯ − ¯ A ¯ ¯B ¯
S 30 = i =0
I in
These transformations expressed in matrix format become
S 00 1 cos 2θ sin 2θ 0 S0
    
1
0
 S  1  cos 2θ
 1 =  cos2 2θ 2 sin 4θ 0   S1 
1
 
 S 0  2  sin 2θ sin2 2θ 0   S2
2 sin 4θ

2
S 30 0 0 0 0 S3
which reveals the Mueller matrix for a linear polarizer.
The Mueller matrix for a half wave plate is worked out below. The Mueller
matrix for a quarter wave plate is deferred to problem P6.15
Derivation: Mueller Matrix for a Half Wave Plate
We know that all of the light transmits through the wave plate. This immediately
gives
S 00 = S 0
The wave plate does nothing to unpolarized light. On the other hand, the polarized
portion of the light is influenced by the wave plate as follows (see (6.39)):
A0
· ¸ · ¸· ¸ · ¸
cos 2θ sin 2θ A A cos 2θ + B sin 2θ
= =
B0 sin 2θ − cos 2θ B A sin 2θ − B cos 2θ
As usual, θ is the angle of the fast axis relative to the horizontal. (As expected,
¯ 0 ¯ 2 ¯ 0 ¯2
¯ A ¯ + ¯B ¯ = |A|2 +|B |2 ; the intensity of the light is unaltered.) Using (6.59) we get
¯ 0 ¯ 2 ¯ 0 ¯2
¯ A ¯ − ¯B ¯ |A cos 2θ + B sin 2θ|2 − |A sin 2θ − B cos 2θ|2
S 10 = =
I in Ii n
¡ 2
|A| − |B |2 cos 4θ + (A ∗ B + AB ∗ ) sin 4θ
¢
= = S 1 cos 4θ + S 2 sin 4θ
Ii n
¯ 0 ¯ 2 ¯ 0 ¯2
¯ A ¯ − ¯B ¯ |A cos 2θ + B sin 2θ|2 − |A sin 2θ − B cos 2θ|2
S 10 = =
I in Ii n
¡ 2
|A| − |B |2 cos 4θ + (A ∗ B + AB ∗ ) sin 4θ
¢
= = S 1 cos 4θ + S 2 sin 4θ
Ii n
A 0∗ B 0 + A 0 B 0∗
S 20 =
Ii n
(A ∗ cos 2θ + B ∗ sin 2θ) (A sin 2θ − B cos θ)
=
Ii n
(A cos 2θ + B sin 2θ) (A ∗ sin 2θ − B ∗ cos θ)
+
Ii n
|A|2 − |B |2 AB ∗ + A ∗ B
= sin 4θ − cos 4θ = S 1 sin 4θ − S 2 cos 4θ
Ii n Ii n
A 0∗ B 0 − A 0 B 0∗
S 30 = i
Ii n
(A ∗ cos 2θ + B ∗ sin 2θ) (A sin 2θ − B cos θ)
=i
Ii n
(A cos 2θ + B sin 2θ) (A ∗ sin 2θ − B ∗ cos θ)
−i
Ii n
A ∗ B − AB ∗
= −i = −S 3
Ii n
These transformations expressed in matrix format become
S 00 1 0 0 0 S0
    
 S0   0 cos 4θ sin 4θ   S1
0   
 1 = 
 S0   0 sin 4θ − cos 4θ 0   S2 
2
S 30 0 0 0 −1 S3
which reveals the Mueller matrix for a half wave plate.

Exercises
Exercises for 6.2 Jones Vectors for Representing Polarization

¢∗
Show that A x̂ + B e i δ ŷ · A x̂ + B e i δ ŷ = 1, as defined in connection
¡ ¢ ¡
P6.1
with (6.5).
P6.2 Prove that if 0 < δ < π, the helicity is left-handed, and if π < δ < 2π the
helicity is right-handed.
HINT: Write the relevant real field associated with (6.5)
E (z, t ) = |E eff | x̂A cos kz − ωt + φ + ŷB cos kz − ωt + φ + δ

£ ¡ ¢ ¡ ¢¤
where φ is the phase of E eff . Freeze time at, say, t = φ/ω. Determine the
field at z = 0 and at z = λ/4 (a quarter cycle), say. If E (0, t ) × E (λ/4, t )
points in the direction of k, then the helicity matches that of a wood
screw.
L6.3 Determine how much right-handed circularly polarized light (λvac =

633 nm) is delayed (or advanced) with respect to left-handed circularly
polarized light as it goes through approximately 3 cm of Karo syrup
(the neck of the bottle). This phenomenon is called optical activity.
Because of a definite-handedness to the molecules in the syrup, right-
and left-handed polarized light experience slightly different refractive
indices. (video)
Karo Light
Corn Syrup
Screen
Polarizer Polarizer
Laser
Figure 6.11 Lab schematic for L 6.3
HINT: Linearly polarized light contains equal amounts of right and left
circularly polarized light. Consider
eiφ
· ¸ · ¸
1 1 1
+
2 i 2 −i
where φ is the phase delay of the right circular polarization. Show that
this can be written as
cos φ/2
· ¸
eiδ
sin φ/2
Exercises 167
The overall phase δ is unimportant. Compare this with

cos α
· ¸
sin α
where α is the angle of linearly polarized light (see table 6.1).
Exercises for 6.3 Elliptically Polarized Light
P6.4 For the following cases, what is the orientation of thepmajor axis, and
p = B = 1/ 2; δ = 0 Case II:
what is thepellipticity of the light? Case I: A
A = B = 1/ 2; δ = π/2; Case III: A = B = 1/ 2; δ = π/4.
Exercises for 6.4 Linear Polarizers and Jones Matrices
P6.5 (a) Suppose that linearly polarized light is oriented at an angle α with
respect to the horizontal axis (x-axis) (see table 6.1). What fraction of
the original intensity gets through a vertically oriented polarizer?
(b) If the original light is right-circularly polarized, what fraction of the
original intensity gets through the same polarizer?
Exercises for 6.5 Jones Matrix for Polarizers at Arbitrary Angles
P6.6 Horizontally polarized light (α = 0) is sent through two polarizers, the

first oriented at θ1 = 45◦ and the second at θ2 = 90◦ . What fraction of
the original intensity emerges? What is the fraction if the ordering of
the polarizers is reversed?
P6.7 (a) Suppose that linearly polarized light is oriented at an angle α with
respect to the horizontal or x-axis. What fraction of the original inten-
sity emerges from a polarizer oriented with its transmission at angle θ
from the x-axis?
Answer: cos2 (θ − α); compare with P6.5.
(b) If the original light is right circularly polarized, what fraction of the
original intensity emerges from the same polarizer?
P6.8 Derive (6.12), (6.13), and (6.14).

HINT: Analyze the Jones vector just as you would analyze light in the
laboratory. Put a polarizer in the beam and observe the intensity of the
light as a function of polarizer angle. Compute the intensity via (6.23).
Then find the polarizer angle (call it α) that gives a maximum (or a
minimum) of intensity. The angle then corresponds to an axis of the
ellipse inscribing the E-field as it spirals. When taking the arctangent,
remember that it is defined only over a range of π. You can add π for
another valid result (which corresponds to the second ellipse axis).
Exercises for 6.6 Jones Matrices for Wave Plates
L6.9 Create a source of unknown elliptical polarization by reflecting a lin-

early polarized laser beam (with both s and p-components) from a
metal mirror with a large incident angle (i.e. θi ≥ 80◦ ). Use a quarter-
wave plate and a polarizer to determine the Jones vector of the reflected
beam. Find the ellipticity, the helicity (right or left handed), and the
orientation of the major axis. (video)
o
Silver Mirror
~80 angle of incidence
Polarizer
Laser 1/4 wave

Polarizer plate
o
set at 45
Screen
Figure 6.12 Lab schematic for L 6.9
HINT: A polarizer alone can reveal the direction of the major and minor
axes and the ellipticity, but it does not reveal the helicity. Use a quarter-
wave plate (oriented at a special angle θ) to convert the unknown
elliptically polarized light into linearly polarized light. A subsequent
polarizer can then extinguish the light, from which you can determine
the Jones vector of the light coming through the wave plate. This must
equal the original (unknown) Jones vector (6.11) operated on by the
wave plate (6.37). As you solve the matrix equation, it is helpful to note
that the inverse of (6.37) is its own complex conjugate.
P6.10 What is the minimum thickness (called zero-order thickness) of a

quartz plate made to operate as a quarter-wave plate for λvac = 500 nm?
The indices of refraction are n fast = 1.54424 and n slow = 1.55335.
Exercises for 6.7 Polarization Effects of Reflection and Transmission
P6.11 Light is linearly polarized at α = 45◦ with a Jones vector according to

table 6.1. The light is reflected from a vertical silver mirror with angle
of incidence θi = 80◦ , as described in (P3.15). Find the Jones vector
80
representation for the polarization of the reflected light.
s NOTE: The answer may be somewhat different than the result mea-
sured in L 6.9. For one thing, we have not considered that a silver mirror
p
inevitably has a thin oxide layer.
Figure 6.13 Geometry for P6.11 P6.12 Calculate the angle θ to cut the glass in a Fresnel rhomb such that after
the two internal reflections there is a phase difference of π/2 between
Exercises 169
the two polarization states. The rhomb then acts as a quarter wave
plate.
Fresnel
HINT: You need to find the phase difference between (3.40) and (3.41). Rhomb
Set the difference equal to π/4 for each bounce. The equation you get
does not have a clean analytic solution, but you can plot it to find a
numerical solution. Side
View
= 50◦ and θ ∼
Answer: There are two angles that work: θ ∼ = 53◦ .
Figure 6.14 Fresnel Rhomb geom-

etry for P6.12
Exercises for 6.A Ellipsometry
P6.13 Derive (6.49) and (6.51), often used for ellipsometry measurements.
1−cos 2θ 1+cos 2θ
HINT: Using sin2 θ = 2 and cos2 θ = 2 , first show
³ ´. ¯ ¯2 . 2
r p r s∗ + r s r p∗ |r s |2 tan α ¯r p ¯ |r s | − tan2 α
I ∝ 1− ¯ ¯ . sin 2θ + ¯ ¯ . cos 2θ
¯r p ¯2 |r s |2 + tan2 α ¯r p ¯2 |r s |2 + tan2 α
Exercises for 6.B Partially Polarized Light
P6.14 (a) One way to construct a right-circular polarizer is using a quarter

wave plate with fast axis at 45◦ , followed by a linear polarizer oriented
vertically, and finally a quarter wave plate with fast axis at −45◦ . Calcu-
late the Jones matrix for this system.
· ¸
1 1 i
Answer: 2
−i 1
(b) Check that the device leaves right-circularly polarized light unal-
tered while killing left-circularly polarized light.
P6.15 Derive the Mueller matrix for a quarter wave plate.

Answer:
1 0 0 0
 
1
 0 cos2 2θ 2 sin 4θ − sin 2θ 
1
 
 0
2 sin 4θ sin2 2θ cos 2θ 
0 sin 2θ − cos 2θ 0
Chapter 7
Superposition of Quasi-Parallel
Plane Waves
To this point, we have typically only considered individual plane-wave fields

which have uniform intensity throughout space and time. Some optical fields
can be well-approximated by a plane wave, but most have a more complicated
structure. Nevertheless, it turns out that any field (e.g. pulses or focused beams),
regardless of how complicated, can be described by a superposition of many
plane wave fields. In this chapter, we develop the techniques for superimposing
plane waves.
We begin our analysis with a discrete sum of plane wave fields and show how
to calculate the intensity in this case. We will introduce the concept of group
velocity, which describes the motion of interference ‘ripples’ resulting when
multiple plane waves are superimposed. Group velocity is distinct from phase
velocity that we encountered previously. As we saw in chapter 2, the real part of
refractive index in certain situations can be less than one, indicating superluminal
wave crest propagation (i.e. greater than c)! In this case, the group velocity is
usually less than c. Since group velocity tracks the speed of the interference
ripples, regions of light intensity tend to advance with the group velocity rather
than the phase velocity.
Beginning in section 7.3, we extend our analysis of wave superposition to
waveforms composed of continua of plane waves rather than from discrete sums.
The analysis is based on Fourier theory (see section 0.4 for a review), which
in essence is a tool for keeping track of the plane waves that make up a given
waveform E (r1 , t ). Since it is easiest to deal with plane waves, we will learn how
to decompose arbitrary waveforms into plane waves for purposes of determining
effects such as propagation in a material (with a frequency-dependent index).
Conversely, we will also learn how to reassemble plane waves into a final pulse at
the end of propagation.
Different frequency components of the waveform experience different phase
velocities, causing the waveform to undergo distortion as it propagates, a phe-
nomenon called dispersion. We shall see that the group velocity tracks the move-
171
172 Chapter 7 Superposition of Quasi-Parallel Plane Waves
ment of the center of the wave packet. For narrowband packets (i.e. packets
comprised of a narrow range of frequencies and hence long duration), the packet
tends to maintain its shape (with some spreading) while propagating at the group
velocity. On the other hand, broadband pulses (i.e. packets comprised of many
frequencies and possibly of short duration) tend to distort severely while prop-
agating in materials. Nevertheless, the group velocity tracks the center of the
pulse.
It turns out that group velocity can become superluminal when significant
absorption and/or amplification of the light pulse is involved. This is no cause
for alarm (nor is it cause for an abundance of gee-wiz papers on the subject).
Absorption and amplification can cause a pulse to appear to move unexpectedly
fast through a reshaping effect. Group velocity, or rather its inverse group delay,
takes this into account, which makes it remarkably general. In such a scenario,
Sir Isaac Newton (16431727, En- energy can be lost from the back of a pulse or perhaps added to an already-present
glish) was born in Lincolnshire, England
three months after the death of his fa-
forward portion of a pulse such that the average pulse position appears to advance
ther who was a farmer. Newton spent abruptly. When all energy is accounted for (both the energy in the medium and in
much of his childhood with his ma-
the light pulse), however, nothing advances faster than the universal speed limit
ternal grandmother, after his mother
remarried. (Newton did not like his step- c. Appendix 7.B gives a good look under the hood at how a medium exchanges
father.) In his teenage years, Newton's energy with a pulse to produce these eye-catching effects.
mother tried to persuade him to take up
farming, but his love for education won
out. He became the top-ranked student
and was admitted into Trinity College, 7.1 Intensity of Superimposed Plane Waves
Cambridge at age 18. Newton was in-
uenced by the works of Descartes,
Copernicus, Galileo, and Kepler. Upon
We can construct arbitrary waveforms by adding together many plane waves with
graduation four years later, the univer- different propagation directions, amplitudes, phases, frequencies and polariza-
sity closed for two years because of a
tions. Consider the following discrete sum of plane waves:
plague. Newton's return to farm life co-
incided with a remarkable period when
E j e i (k j ·r−ω j t )
X
he rst developed ideas on calculus, E(r, t ) = (7.1)
gravitation, and optics. Newton later j
returned to Cambridge where he spent
his extraordinarily prolic career and
The corresponding magnetic field according to (2.56) is
became the rst scientist to be knighted.
In optics, Newton advanced the ray the-
ory of light and image formation. He
X kj × Ej
B j e i (k j ·r−ω j t ) = e i (k j ·r−ω j t )
X
B(r, t ) = (7.2)
showed that `white' light is comprised of
j j ωj
many colors and that the amount of re-
fraction depends on color. He built the
rst reecting telescope, which avoids As usual, the (time- and space-independent) individual field components E j
chromatic aberration. Newton advo-
cated against the wave theory of light in
contain both amplitude and phase information for each plane wave.
favor of his `corpuscular' theory. (Imag- The Poynting vector (2.52) associated with the fields (7.1) and (7.2) is
ining that by this Newton foresaw the
quantized nature of light energy gives Re {B (r, t )}
too much credit!) (Wikipedia) S(r, t ) = Re{E (r, t )} ×
µ0
1 n o ³ n o´ (7.3)
Re E j e i (k j ·r−ω j t ) × km × Re Em e i (km ·r−ωm t )
X
=
j ,m ωm µ0
where we have assumed that the km vectors are real. (Recall the conspiracy that
only the real parts of the fields are relevant – crucial when multiplying.) The above
expression is cumbersome because of the many cross terms that arise when
7.1 Intensity of Superimposed Plane Waves 173
the two summations are multiplied. We need some simplifying assumptions

before we can make any real progress on this expression. For example, we can
time-average the rapid fluctuations in the expression that vary on the scale of
optical frequencies. Additionally, it is common to encounter the situation where
all plane-wave components travel roughly parallel to each other, which will be a
big help in simplifying (7.3).
Intensity for Quasi Parallel-Traveling Light
For simplicity, we assume that all vectors k j are real. If the wave vectors are
complex, the result is essentially the same, but, as in (2.62), the field amplitudes E j
correspond to local amplitudes (adjusted for absorption or amplification during
prior propagation). We apply the BAC-CAB rule (P0.3) to (7.3) and obtain
X 1 h ³ n ¡ ¢o n o´
S(r, t ) = km Re E j e i k j ·r−ω j t · Re Em e i (km ·r−ωm t )
j ,m ωm µ0 (7.4)
n o³ n o ¡ í ¢
− Re Em e i (km ·r−ωm t ) Re E j e i k j ·r−ω j t
· km
The last term in (7.4) can be dismissed if all k-vectors are approximately parallel to
each other, in which case all of the km are essentially perpendicular to each of the
E j . We will make this rather stringent assumption and kill the last line in (7.4). The
magnitude of the Poynting vector then becomes (with the help of (0.30))
 ¡ ¢ ¡ ¢
i k ·r−ω j t
X km  E j e j + E∗j e −i k j ·r−ω j t 
S(r, t ) =
j ,m ωm µ0 2
 
( )
Em e i (km ·r−ωm t ) + E∗m e −i (km ·r−ωm t )
·
2 (parallel k-vectors)
X km n
E j · Em e i k j +km ·r− ω j +ωm t + E∗j · E∗m e −i k j +km ·r− ω j +ωm t
£¡ ¢ ¡ ¢ ¤ £¡ ¢ ¡ ¢ ¤
=
j ,m 4ωm µ0
¢ ¤o
+ E j · E∗m e i k j −km ·r− ω j −ωm t + E∗j · Em e −i k j −km ·r− ω j −ωm t
£¡ ¢ ¡ ¢ ¤ £¡ ¢ ¡
(7.5)
The terms involving (ω j + ωm )t oscillate rapidly and time-average to zero. By
comparison, the terms involving (ω j − ωm )t oscillate slowly (especially when the
ω j are all in the neighborhood of the ωm ) or not at all when j = m. We retain the
slower fluctuations and discard the rapid oscillations. For purposes of computing
the intensity (as opposed to determining phase changes with propagation) we can
approximate the index as a constant, and write k m /(ωm µ0 ) ≈ n²0 c. (We seldom
measure intensity inside of materials anyway.) With these simplifications, (7.5)
becomes
∗ i k j −km ·r− ω j −ωm t
£¡ ¢ ¡ ¢ ¤ £¡ ¢ ¤
n²0 c X E j · Em e + E∗j · Em e −i k j −km ·r−(ωn −ωm )t
〈S(r, t )〉osc =
2 j ,m 2
( )
n²0 c X ¡
i k j ·r−ω j t
¢ X
∗ −i (km ·r−ωm t )
= Re Ej e · Em e
2 j m
n²0 c ©
Re E (r, t ) · E∗ (r, t ) .
ª
=
2
(parallel k-vectors) (7.6)
The final expression in (7.6) is already manifestly real so there is no need

to apply the operation Re {}. The time-averaged intensity for light composed of
parallel wave vectors is then well-approximated by
n²0 c
(valid for parallel or antiparallel I (r, t ) = E(r, t ) · E∗ (r, t ) (7.7)
2
k-vectors and constant n)
In a surprising turn of events, it is important that E(r, t ) in (7.7) be written as the
entire complex expression for the electric field rather than just the real part. Then
(7.7) automatically time-averages over rapid oscillations in such a way that I (r, t )
retains a slowly varying time dependence. This expression is reminiscent of (2.62),
but it should be kept in mind that we previously considered only a single plane
wave (perhaps with two distinct polarization components).
If some of the k-vectors point in an anti-parallel direction, we can still use
(7.7) with negative signs entered explicitly for those components. This brings up
a distinction between irradiance S and intensity I . For example, 〈S〉 is zero for
standing waves because there is no net flow of energy, whereas (7.7) still gives a
result. Intensity specifies whether atoms locally experience an oscillating electric
field without regard for whether there is a net flow of energy carried by a light field.
At extreme intensities, however, when the influence of the magnetic field becomes
comparable to that of the electric field, the distinction between propagating and
standing fields becomes important to the behavior of charged particles in that
field.
The assumption that all vectors k j are parallel is not as serious as it might
seem at first. For example, the output of a Michelson interferometer (studied
in chapter 8) is the superposition of two fields, each composed of a range of
frequencies with parallel k j ’s. We can relax the restriction of parallel k j ’s slightly
and apply (7.7) also to plane waves with nearly parallel k j ’s such as occurs in a
Young’s two-slit diffraction experiment (studied in chapter 8). In such diffraction
problems, (7.7) is viewed as an approximation valid to the extent that the vectors
k j are close to parallel. For the remainder of the chapter we will assume that the
k-vectors for all frequency components in our waveform are essentially parallel.
7.2 Group vs. Phase Velocity: Sum of Two Plane Waves

To begin our study of interference, consider just two plane waves with equal
amplitudes given by
E1 = E0 e i (k1 ·r−ω1 t ) and E2 = E0 e i (k2 ·r−ω2 t ) (7.8)
As we previously studied (see P1.9), the velocities of the wave crests for these two
Figure 7.1 Animation showing su-
waves are
perposition of two plane waves
(electric fields) with different fre-
v p1 = ω1 /k 1 and v p2 = ω2 /k 2 (7.9)
quencies and traveling at different These are known as the phase velocities of the individual plane waves.
speeds.
7.2 Group vs. Phase Velocity: Sum of Two Plane Waves 175
Next consider a composite wave created from the superposition of the above
two plane waves:
E(r, t ) = E0 e i (k1 ·r−ω1 t ) + E0 e i (k2 ·r−ω2 t ) (7.10)
The two plane waves interfere, producing regions of higher and lower intensity
that move in time. Remarkably, these intensity peaks can propagate at speeds
quite different from either of the phase velocities in (7.9). The intensity (7.7) for
the field (7.10) is computed as follows:
n²0 c h ih i
I (r, t ) = E0 · E∗0 e i (k1 ·r−ω1 t ) + e i (k2 ·r−ω2 t ) e −i (k1 ·r−ω1 t ) + e −i (k2 ·r−ω2 t )
2
n²0 c h i
= E0 · E∗0 2 + e i [(k2 −k1 )·r−(ω2 −ω1 )t ] + e −i [(k2 −k1 )·r−(ω2 −ω1 )t ]
2 (7.11)
∗
= n²0 cE0 · E0 [1 + cos [(k2 − k1 ) · r − (ω2 − ω1 ) t ]]
= n²0 cE0 · E∗0 [1 + cos (∆k · r − ∆ωt )]
where
∆k ≡ k2 − k1
(7.12)
∆ω ≡ ω2 − ω1
The darker line in Fig. 7.2 shows the intensity computed with (7.11). Keep in
mind that this intensity is averaged over rapid oscillations. For comparison, the
lighter line shows the Poynting flux with the rapid oscillations retained, according
Intensity
to (7.5). It is left as an exercise (see P7.3) to show that the rapid-oscillation peaks
in Fig. 7.2 move with a phase velocity derived from the average k and average ω
of the two plane waves:
ave{ω}
vp ≡ (7.13)
ave{k}
An examination of the cosine argument in (7.11) reveals that the time-averaged
curve in Fig. 7.2 (solid) travels with speed Position
∆ω Figure 7.2 Intensity of two inter-

vg ≡ (7.14) fering plane waves. The solid line
∆k
shows intensity averaged over
This is known as the group velocity. Essentially, v g may be thought of as the rapid oscillations.
velocity for the envelope that encloses the rapid oscillations.
In general, v g and v p are not the same. This means that as the waveform
propagates, the rapid oscillations move within the larger modulation pattern, for
example, continually disappearing at the front and reappearing at the back of
each modulation. The group velocity is identified with the propagation of overall
waveforms. The presence of field energy in a waveform is clearly tied more to v g
than to v p .
Example 7.1
Determine the phase velocity and group velocity for the superposition of two plane
waves in a plasma (see P2.7).
Solution: The index of refraction is given by

q
n plasma (ω) = 1 − ω2p /ω2 < 1 (assuming ω > ωp ) (7.15)
The phase velocity for each frequency is computed as
ω1 + ω1 ∼ c
vp = = (7.16)
n plasma (ω1 )ω1 /c + n plasma (ω2 )ω2 /c n plasma (ω)
For convenience, we have taken ω1 and ω2 to lie very close to each other. Since
n plasma < 1, both of these velocities exceed c. However, the group velocity is
∆ω ∼ d ω d k −1 d ωn plasma (ω) −1
· ¸ · ¸
vg = = = = = n plasma (ω) c (7.17)
∆k dk dω dω c
which is clearly less than c. The derivation of the final expression in (7.17) from
the previous one is left as an exercise.
Example 7.1 illustrates that in an environment where the index of refraction is

real (i.e. no net exchange of energy with the medium), the group velocity does not
John William Strutt (3rd Baron exceed c, even when the phase velocity does. The ‘fast-moving’ phase velocity v p
Rayleigh) (18421919, British) was
born in Langford Grove, Essex, England results merely from an interplay between the field and the plasma. In a similar
and was frequently ill in his youth. He sense, the intersection of an ocean wave with the shoreline can also exceed c, if
entered the University of Cambridge in
1861 and graduated four years later as
different points on the wave front happen to strike the shore nearly simultane-
senior wrangler in mathematics. He mar- ously. The point of intersection between the wave and the shoreline does not
ried in 1871 and became the father of
three sons. In 1873, Strutt inherited the
constitute an actual object under motion. Similarly, wave crests of individual
Barony of Rayleigh (and the title Lord plane waves do not necessarily constitute actual objects that are moving. In short,
Rayleigh) from his father who died that
v p is not the relevant speed at which events up stream influence events down
year. In 1879 Strutt succeeded James
Clerk Maxwell as the Cavendish Profes- stream.
sor of Physics at Cambridge. Rayleigh Individual plane waves have infinite length and infinite duration. They do not
studied a wide variety of subjects. He
is credited with the discovery of argon.
exist in isolation except in our imagination. All real waveforms are comprised of a
He studied how atoms scatter light range of frequency components, and so interference always happens. Energy is
(Rayleigh scattering) and explained why
the sky is blue. He developed the notion
associated with regions of constructive interference between those waves. Group
of group velocity and used it to under- velocity v g tracks the presence of field energy, whether that energy propagates or
stand the propagation of sound. He
is extracted from a medium. Although sometimes v g can exceed c (i.e. when ab-
won the Nobel prize in physics in 1904.
(Wikipedia) sorption or amplification is involved), energy is never transported faster than the
universal speed limit c. An examination of energy flow is given in Appendix 7.B.
7.3 Frequency Spectrum of Light

A waveform constructed from a discrete sum (as in the previous two sections)
must eventually repeat over and over (i.e. it is periodic). To create a waveform that
does not repeat (e.g. a single laser pulse or, technically speaking, any waveform
that exists in the physical world since no light source repeats forever) we must
replace the discrete sum (7.1) with an integral that combines a continuum of
plane waves. Such a waveform at a point r can be expressed as
Z∞
1
E(r, t ) = p E (r, ω) e −i ωt d ω (7.18)
2π
−∞
7.3 Frequency Spectrum of Light 177
The function E (r, ω), called the spectrum, has units of field per frequency. Essen-
tially, it gives the amplitude and phase of each plane wave that makes up the over-
all waveform. It includes any spatially dependent factors such as exp {i k (ω) · r}.
We distinguish the spectrum E (r, ω) from the wholly separate function E(r, t ) by
its argument (i.e. ω instead of t ). (Sorry for using E for both functions, but this
is standard notation.) The operation (7.18) is called an inverse Fourier transform
as outlined in section 0.4; actually, it would be a good idea to review section 0.4
p
thoroughly. Now. Why haven’t you turned to section 0.4 yet? The factor 1/ 2π is
introduced to match our Fourier-transform convention. Regardless of what the
function is called, please notice that (7.18) merely sums together a range of plane
waves in much the same way that our earlier discrete summation (7.1) does.
If we already have/know a waveform E(r, t ), one might wonder what plane
waves should be added together in order to construct it. Equation (7.18) can be
inverted, which remarkably has a very similar form:
Z∞
1
E (r, ω) = p E (r, t ) e i ωt d t (7.19)
2π
−∞
This operation is called the Fourier transform. It is used to generate the spectrum
E (r, ω) from the field E(r, t ) in much the same way that (7.18) is used to generate
the field E(r, t ) from the spectrum E (r, ω).
Although only the real part of E(r, t ) is physically relevant, we can continue
our habit of working with the complex field and taking the real part of E (r, t ) at
our leisure.1 In fact, we will find it advantageous to work with the complex field
instead of only the real part. We will not run into trouble as long as we remember
never to discard the imaginary part of E (r, ω), only the imaginary part of E (r, t ).
The intensity formula (7.7) remains useful for continuous superpositions of
plane waves (i.e. a field defined by the inverse Fourier transform (7.18)):
n²0 c
I (r, t ) ≡ E(r, t ) · E∗ (r, t ) (7.20)
2
Remember, this formula specifically requires the fields to be in complex for-
mat, and it takes care of the time-average over rapid oscillations automatically.2
Moreover, the above expression for I (r, t ) assumes that all relevant k-vectors are
essentially parallel.
Similarly, we will define the power spectrum produced from E (r, ω), which we
write as
n²0 c
I (r, ω) ≡ E (r, ω) · E∗ (r, ω) (7.21)
2
1 Since Fourier transforms are linear, one can take the Fourier transform of the real and imaginary
parts of a field separately. Appropriate modifications to E (r, ω) in the frequency domain will not
cause the two parts to become mingled. Upon taking the inverse Fourier transform to obtain E(r, t )
again, the original real part remains purely real, and the original imaginary part remains purely
imaginary.
2 To use this expression there needs to be a sufficient number of oscillations within the waveform
to make the rapid time average meaningful.

The power spectrum I (r, ω) is what one observes when the waveform is sent into
a spectral analyzer or spectrometer. We must apologize again for the potentially
confusing notation (in wide usage): I (r, ω) is not the Fourier transform of I (r, t )!
They are defined exclusively through (7.20) and (7.21).
Parseval’s theorem (see Example 0.7) imposes an interesting connection be-
tween the time-integral of the intensity and the frequency-integral of the power
spectrum:
Z∞ Z∞
I (r, t )d t = I (r, ω) d ω (7.22)
−∞ −∞
With the above formalities out of the way, we will illustrate the use of Fourier
transforms through some examples.
Example 7.2
Find E (r, ω) associated with the field
2 2T 2
e −i ω0 t
±
E(r, t ) = E0 (r) e −t (7.23)
Figure 7.3 Real part of electric
field (7.23) with T = 4π/ω0 and The real part of this field is shown in Fig. 7.3 for two different durations T . The
T = 10π/ω0 , where 2π/ω0 is the intensity profile computed by (7.20) is shown in Fig. 7.4 .
period of the carrier frequency.
Solution: The argument r is unimportant to our calculation. It merely specifies
that we are considering the field at the point r. We compute the Fourier transform
as follows:
Z∞
1 2
± 2
E (r, ω) = p E0 (r) e −t 2T e −i ω0 t e i ωt d t
2π
−∞
(7.24)
Z∞
E0 (r) −t 2 /2T 2 +i (ω−ω0 )t
= p e dt
2π
−∞
This integral can be performed with the help of (0.55), and we obtain
T 2 (ω−ω0 )2
E (r, ω) = T E0 (r) e − 2 (7.25)
Notice that E (r, ω) has units of Field multiplied by time, or in other words, field per
frequency.
In general, E (r, ω) is a complex function. E (r, ω) keeps track of the amplitude

and phase of each plane wave needed to compose the waveform E(r, t ). More of-
ten than not, E (r, ω) exhibits a complicated complex phase structure, depending
on the time-shape of E(r, t ).
The spectrum of the field in Example 7.2 is shown in Fig. 7.5. The complex
Figure 7.4 The intensity (7.20) of phase turns out to be boringly uniform for this example; if E0 is real, the imaginary
the fields in Fig. 7.3.
part of the spectrum turns out to be zero for all frequencies. The corresponding
power spectrum (7.21) is plotted in Fig. 7.6. As expected, the waveform includes
7.3 Frequency Spectrum of Light 179
frequencies in the neighborhood of ω0 . A range of frequencies are needed to con-

struct waveforms that turn on and off. The shorter the duration of the waveform,
the more frequency components that are necessary. This trend can be seen for
the two pulse durations T plotted.
Example 7.3
Check Parseval’s theorem for the field and spectrum in Example 7.2.
Solution: The time integration in (7.22) yields
Z∞ Z∞
n²0 c 2
±
T2
I (r, t )d t = E0 (r) · E∗0 (r) e −t dt
2
−∞ −∞
n²0 c p
= E0 (r) · E∗0 (r) T π
2
where we have used (0.55) to perform the integration. This result has units of
energy per area. It is the energy per area absorbed by a detector after the pulse has
concluded. The frequency integration in (7.22) yields
Figure 7.5 Spectral components
Z∞ Z∞ (7.25) of the fields in Fig. 7.3 with
n²0 c 2
(ω−ω0 )2 T = 4π/ω0 and T = 10π/ω0 , where
I (r, ω) d ω = E0 (r) · E∗0 (r) T 2 e −T dω
2 2π/ω0 is the period of the carrier
−∞ −∞
p frequency.
n²0 c π
= E0 (r) · E∗0 (r) T 2
2 T
which is the same answer.
As mentioned previously, the inverse Fourier transform is interpreted as sum-

ming together many plane waves to create a waveform.
Example 7.4
Take the inverse Fourier transform of (7.25) to recover the original waveform (7.23).
Solution: The inverse Fourier transform (7.18) is

Z∞
1
E(r, t ) = p E (r, ω) e −i ωt d ω
2π
−∞
Z∞ T 2 (ω−ω0 )2
T E0 (r)
= p e− 2 e −i ωt d ω (7.26)
2π
−∞
Z∞ T 2 ω0 2
T E0 (r) T 2 ω2
+(T 2 ω0 −i t )ω− Figure 7.6 Power spectrum based
= p e− 2 2 dω
2π on (7.21) for the spectral compo-
−∞
nents shown in Fig. 7.5.
This integral can be performed with the help of (0.55), which gives
2
(T 2 ω0 −i t ) T 2 ω2
π
r 0
T E0 (r) 4(T 2 /2)
− 2
E(r, t ) = p e
2π T 2 /2
2 2T 2
e −i ω0 t
±
= E0 (r) e −t
Since only the real part of the time profile E(r, t ) is physically relevant, you
might be curious about how the Fourier transform of the real part of the field
compares with that of the complex version of the field that we have been using.
Indeed, there are situations where it is more appropriate to use the real version
of the field rather than its complex form. For example, if a waveform includes
multiple propagation directions or if a waveform contains only a few cycles, then
the motivation/interpretation behind (7.20) and the convenience of the complex
format begins to wane.
Example 7.5
Take the Fourier transform of just the real part of waveform (7.23).
Solution: The real part of (7.23) is
E(r, t ) + E∗ (r, t )
Er (r, t ) =
2
−i ω0 t
(7.27)
−t 2 2T 2 E0 (r) e
± + E∗0 (r) e i ω0 t
=e
2
2 /2T 2
If E0 (r) is real, then this field can be written as E0 (r) e −t cos (ω0 t ). The Fourier
transform (7.19) yields (see P0.24)
Figure 7.7 Spectrum based on T 2 (ω+ω0 )2 T 2 (ω−ω0 )2

E0 (r) e − 2 + E∗0 (r) e − 2
(7.28) with T = 10π/ω0 . Compare Er (r, ω) = T (7.28)
2
with the lower curve in Fig. 7.5
The spectrum is shown in Fig. 7.7.
From the above example, you might notice that the transform of the real
part of a field tends to be more cumbersome than the transform of the entire
complex field. For the real field, both positive and negative frequency components
contribute to the overall spectrum.3 Moreover, the Fourier transform of a real
function Er (r, t ) obeys the following symmetry relation:
Er (r, −ω) = E∗r (r, ω) (if Er (r, t ) is real) (7.29)
The spectrum in Fig. 7.7 obeys this symmetry relation, whereas the Fourier trans-
form of the complex field depicted in Fig. 7.5 does not.
3 Essentially, the spectrum of the complex representation of the field can be understood to be
twice the spectrum of the real representation, but plotted only for the positive frequencies.
7.4 Packet Propagation and Group Delay 181
7.4 Packet Propagation and Group Delay

Once we have the spectrum for a waveform (obtained by Fourier transform),
we can apply effects to the individual spectral components. In particular, we
can find how an overall waveform propagates in a uniform medium by taking
advantage of our knowledge of how individual plane waves propagate (as studied
in chapter 2). At any point in the medium, we can perform an inverse Fourier
transform, which recombines spectral components (i.e. plane waves) to reveal
how the overall waveform looks as a function of time. Thus, we will be able to
predict the temporal profile of a waveform at any location given knowledge of
that waveform at another location.4
Let E(r0 , t ) be the temporal profile of a pulse at some point r0 in a medium.
The spectrum of this pulse E(r0 , ω) (found using (7.19)) gives the amplitudes and
phases of the individual plane wave components at the point r0 . We already
know how to propagate individual plane waves through a material (see (2.20)). A
phase shift associated with a displacement ∆r modifies the spectral components
according to
E (r0 + ∆r, ω) = E (r0 , ω) e i k(ω)·∆r (7.30)
The k-vector contains the frequency-dependent information about the material
via k = n(ω)ω/c. (A complex wave vector k may also be used if absorption or
amplification is present.) We take the inverse Fourier transform of E (r0 + ∆r, ω) at
the new position to determine the waveform E (r0 + ∆r, t ):
Z∞
1
E(r0 + ∆r, t ) = p E(r0 + ∆r, ω)e −i ωt d ω
2π
−∞
Z∞
1
= p E(r0 , ω)e i (k(ω)·∆r−ωt ) d ω (7.31)
2π
−∞
Example 7.6 If a waveform at r0 = 0 has the form E (0, t ) = E0 e −t /2T e −i ω0 t ,

2 2
compute the waveform at r = z ẑ if propagation occurs in vacuum in the

z-direction.
Solution: Of course, after traversing ∆r = z ẑ in vacuum, the waveform will

look the same, only arriving a time z/c later. We’ll demonstrate that the
tools described above yield this expected result. The Fourier transform of
the Gaussian pulse is given in (7.25):
T 2 (ω−ω0 )2
E (0, ω) = T E0 e − 2
4 See J. D. Jackson, Classical Electrodynamics, 3rd ed., Sect. 7.8 (New York: John Wiley, 1999).
To find the field downstream we invoke (7.30), assuming k (ω) = k vac (ω) ẑ =
ω
c ẑ, which gives the appropriate phase shift for each plane wave compo-
nent: 2 2 T (ω−ω0 ) ω
E (z, ω) = E (0, ω) e i k(ω)·∆r = T E0 e − 2 ei c z
We compute the final waveform using (7.31) and obtain
Z∞
1 T 2 (ω−ω0 )2 ω
2
− (t −z/c)
E (z, t ) = p E0 Te − 2 e i c z e −i ωt d ω = E0 e 2
2T e −i ω0 (t −z/c)
2π
−∞
(7.32)
Not surprisingly, after traveling a distance z though vacuum, the pulse looks
identical to the original pulse, only delayed by time z/c.
A waveform propagating in a material such as glass can undergo significant

temporal dispersion, as different frequency components experience different
indices of refraction. Each frequency component propagates at its own phase
velocity. The speed of the pulse, however, can be quite different; it propagates
approximately with the group velocity, as will be shown below.
The exponent in (7.30) is called the phase delay for the pulse propagation. It
is often expanded in a Taylor series about the pulse carrier frequency ω0 :
∂k ¯¯ 1 ∂2 k ¯¯
· ¯ ¯ ¸
k · ∆r ∼
= k|ω0 + (ω − ω0 ) + (ω − ω0 ) 2
+ · · · · ∆r (7.33)
∂ω ¯ω0 2 ∂ω2 ¯ω0
The k-vector has a sometimes-complicated frequency dependence through the

functional form of n(ω). If we retain only the first two terms in this expansion
then (7.31) becomes
Z∞ ³h i ´
1 ∂k ¯
¯
i k(ω0 )+ (ω−ω0 ) ·∆r−ωt
E(r0 + ∆r, t ) = p E(r0 , ω)e ∂ω ω0
dω
2π
−∞
h i Z∞ ³ ´
∂k ¯ 1 −i ω t − ∂k ¯
¯ ¯
i k(ω0 )−ω0 ·∆r ·∆r
= e ∂ω ω0
p E (r0 , ω) e ∂ω ω0
dω
2π
−∞
Z∞
1
E (r0 , ω) e −i ω(t −t ) d ω
0 0
= e i [k(ω0 )·∆r−ω0 t ] p (7.34)
2π
−∞
where in the last line we have introduced the definition

∂k ¯¯ ∂Re {k} ¯¯
¯ ¯
0
t ≡ ∼
· ∆r = · ∆r (7.35)
∂ω ¯ω0 ∂ω ¯ω0
and assumed that the imaginary part of k is roughly constant near ω0 so that t 0
is real. Then the integral in (7.34) is recognized as the Fourier transform of the
original pulse with a new time argument:
0
E (r0 + ∆r, t ) = E r0 , t − t 0 e i (k(ω0 )·∆r−ω0 t )
¡ ¢
(7.36)
7.5 Quadratic Dispersion 183
Notice that (7.32) from Example 7.6 agrees with this result, since kvac (ω0 ) · ∆r =
ω0 z/c. The second factor in (7.36) merely gives an overall phase shift due to prop-
agation. The phase shift is dictated by the phase velocity of the carrier frequency
(see (7.9)):
ω0
v p (ω0 ) = (7.37)
k (ω0 )
Otherwise (7.36) is unaltered except for a delay t 0 , the time required for the pulse
to traverse the displacement ∆r.
The function ∂Rek ∂ω · ∆r is known as the group delay function, and in (7.35)
±
it is evaluated only at the carrier frequency ω0 . Traditional group velocity is

obtained by dividing the displacement ∆r by the group delay time t 0 to obtain
∂Re{k(ω)} ¯¯
¯
−1
v g (ω0 ) = (7.38)
∂ω ¯
ω0
Group delay (or group velocity) essentially tracks the center of the packet.
In our derivation we have assumed that the phase delay k(ω)·∆r could be well-
represented by the first two terms of the expansion (7.33). While this assumption
gives results that are often useful, higher-order terms can also play a role. In
section 7.5 we’ll find that the next term in the expansion controls the rate at which
the wave packet spreads as it travels. We should also note that there are times
when the expansion (7.33) fails to converge (when ω0 is near a resonance of the
medium), and the above expansion approach is not valid. We’ll analyze pulse
propagation in these sticky situations in section 7.6.
7.5 Quadratic Dispersion

A light pulse traversing a material in general undergoes dispersion when different
frequency components propagate with different phase velocities. As an example,
consider a short laser pulse traversing an optical component such as a lens or
window, as depicted in Fig. 7.8. The short light pulse can broaden in time5 with
the different frequency components becoming separated (often called stretching
or chirping). If absorption (and surface reflections) can be neglected, then the
amplitude of E(r, ω) does not change – only its phase changes – and the power
spectrum (7.21) remains unaltered.
Consider light traveling in a material such as glass. Let the plane-wave com-
ponents all point in the ẑ-direction. We place r0 at the start of the glass where we
assign z = 0, so that k · ∆r = kz. Let the polarization of the field be the same for all
frequencies.
To find the waveform at the new position z (where the pulse presumably
has just exited the glass), we take the inverse Fourier transform (7.31). However,
before doing this we must specify the function k (ω). In general, with the exact
functional form for the index the inverse Fourier transform can only be performed
numerically. For our present purposes, we again resort to an expansion of the

type (7.33), but this time we will keep one additional term:
k (ω) z ∼
= k 0 z + v g−1 (ω − ω0 ) z + α (ω − ω0 )2 z + · · · (7.39)
where
ω0 n (ω0 )
k 0 ≡ k (ω0 ) = (7.40)
c
∂k ¯ = (ω0 ) + ω0 n (ω0 )
0
¯
n
v g−1 ≡
¯
(7.41)
∂ω ¯ω0 c c
2 ¯¯
1 ∂ k¯ n (ω0 ) ω0 n 00 (ω0 )
0
α≡ = + (7.42)
2 ∂ω2 ¯ ω0 c 2c
As before, we have supposed that the imaginary part of the index is negligible.
Unfortunately, we can’t calculate a general formula for the affect of quadratic
dispersion on an arbitrary initial pulse. However, we can get a general idea for
how quadratic dispersion works by considering the specific example of a Gaussian
pulse.
Example 7.7
25 fs A Gaussian waveform similar to that in Example 7.6 propagates throught a piece of
56 fs
glass with thickness ∆r = z. Compute the waveform exiting the glass.
Solution: Again, the Fourier transform of the Gaussian pulse before propagation
Figure 7.8 A 25 fs pulse traversing is given by (7.25):
T 2 (ω−ω0 )2
an ` = 1 cm piece of BK7 glass. E (0, ω) = T E0 e − 2
With the aid of expansion (7.39), the inverse Fourier transform (7.31) (which yields
the pulse after propagation) becomes
Z∞ T 2 (ω−ω0 )2
1 −1
(ω−ω0 )z+i α(ω−ω0 )2 z −i ωt
E (z, t ) = p E0 Te − 2 e i k0 z+i v g e dω
2π
−∞
(7.43)
Z∞
T E0 e i (k0 z−ω0 t ) −(T 2 /2−i αz )(ω−ω0 )2 i v g−1 (ω−ω0 )z−i (ω−ω0 )t
= p e e dω
2π
−∞
We can avoid considerable clutter if we change variables to ω0 ≡ ω − ω0 . Then the

inverse Fourier transform becomes
Z∞
T E0 e i (k0 z−ω0 t ) T2
(1−i 2αz/T 2 )ω02 −i (t −z/v g )ω0 d ω0
E (z, t ) = p e− 2 (7.44)
2π
−∞
7.6 Generalized Context for Group Delay 185
The above integral can be performed with the aid of (0.55). The result is
2
(
t −z/v g )
T E0 e i (k0 z−ω0 t ) π −
s
T 2
4 2 1−i 2αz/T 2
( )
E (z, t ) = p ¢e
T2
¡
2π 2 1 − i 2αz/T 2
2 (7.45)
i
tan−1 2αz
−
( t −z/v g ) (1+i 2αz/T 2 )
e2 T2 2
³ ´
2T 2 1+ 2αz/T 2
( )
= E0 e i (k0 z−ω0 t ) q ¢2 e
4 ¡
1 + 2αz/T 2
Next, we spruce up the appearance of this rather cumbersome formula as follows:
E0 (t −z/v g )2 (t −z/v g )2 Φ(z)+i (k 1

tan−1 Φ(z)
− −i 0 z−ω0 t )+i 2
E (z, t ) = p e 2T̃ 2 (z) e 2T̃ 2 (z) (7.46)
T̃ (z)/T
where
2α
Φ(z) ≡ z (7.47)
T2
and p
T̃ (z) ≡ T 1 + Φ2 (z) (7.48)
We can immediately make a few observations about (7.46). First, note that
at z = 0 (i.e. zero thickness of glass), (7.46) reduces to the input pulse E (0, t ) =
E0 e −t /2T e −i ω0 t , as it should. Secondly, the peak of the pulse moves at speed v g
2 2
2 2
since the factor e −(t −z/v g ) /2T̃ (z) controls the pulse amplitude, while the other
terms (multiplied by i ) in the exponent of (7.46) merely alter the phase. Also
note that the duration of the pulse increases and its peak intensity decreases as it
travels, since T̃ (z) increases with z. In P7.8 we will find that (7.46) also predicts
that for large z, the field of the spread-out pulse oscillates less rapidly at the begin- Figure 7.9 Animation of a
ning of the pulse than at the end (assuming α > 0). This phenomenon, known as Gaussian-envelope pulse (elec-
pulse chirping, means that red frequencies get ahead of blue frequencies during tric field) undergoing dispersion
propagation since the red frequencies experience a lower index of refraction. during transit.
While Example 7.7 is worked out for the specific case of a Gaussian pulse,
the results are qualitatively similar for all pulses. The exact details vary with
pulse shape, but all short pulses eventually broaden and chirp as they propagate
through a dispersive medium such as glass. Higher-order terms in the expansion
(7.33) that were neglected cause additional spreading, chirping, and other defor-
mations to the pulses as they propagates. The influence of each order becomes
progressively more cumbersome to study analytically. In that case, it is easier to
perform the inverse Fourier transform numerically; there is no need to resort to
the expansion of k (ω) if the integration is done numerically.
7.6 Generalized Context for Group Delay

The expansion of k (ω) in (7.33) is inconvenient if the frequency content (band- Figure 7.10 Real and imaginary
width) of a waveform encompasses a substantial portion of a resonance structure. parts of the refractive index for an
In this case, it becomes necessary to retain a large number of terms in (7.33) to absorptive medium.
Initial Pulse Final Pulse

Detector Detector
Figure 7.11 Transit time defined as the difference between arrival time at two points.
describe accurately the phase delay k (ω) · ∆r. Moreover, if the bandwidth of the
waveform is wider than the spectral resonance of the medium, the series alto-
gether fails to converge. These difficulties have led to the traditional viewpoint
that group velocity loses meaning for broadband waveforms near a resonance. In
this section, we study a broader context for group velocity (or rather its inverse,
group delay d k/d ω), which is always valid, even for broadband pulses where the
expansion (7.33) utterly fails. The analysis avoids the expansion and so is not
restricted to a narrowband context.
We are interested in the arrival time of a waveform (or pulse) to a point, say,
where a detector is located. The definition of the arrival time of pulse energy
need only involve the Poynting flux (or the intensity), since it alone is responsible
for energy transport. To deal with arbitrary broadband pulses, the arrival time
should avoid presupposing a specific pulse shape, since the pulse may evolve
in complicated ways during propagation. For example, the pulse peak or the
midpoint on the rising edge of a pulse are poor indicators of arrival time if the
pulse contains multiple peaks or a long and non-uniform rise time.
For the reasons given, we use a time expectation integral (or time ‘center-of-
mass’) to describe the arrival time of a pulse:
R∞
t I (r, t ) d t
−∞
〈t 〉r ≡ (7.49)
R∞
I (r, t ) d t
−∞
For simplification, we have assumed that the light travels in a uniform direction
by using intensity rather than the Poynting vector.
Consider a pulse as it travels from point r0 to point r = r0 + ∆r in a homoge-
neous medium. The difference in arrival times at the two points is
∆t ≡ 〈t 〉r − 〈t 〉r0 (7.50)
The pulse shape can evolve in complicated ways between the two points, spread-
ing with different portions being absorbed (or amplified) during transit as de-
picted in Fig. 7.11. Nevertheless, (7.50) renders an unambiguous time interval
between the passage of the pulse center at each point.
7.6 Generalized Context for Group Delay 187
This difference in arrival time can be shown to consist of two terms (see
P7.11):6
∆t = ∆tG (r) + ∆t R (r0 ) (7.51)
The first term, called the net group delay, dominates if the field waveform is
initially symmetric in time (e.g. an unchirped Gaussian). It amounts to a spectral
average of the group delay function taken with respect to the spectral content of
the pulse arriving at the final point r = r0 + ∆r: Before Propagation
R∞ ³
∂Rek
´
I (r, ω) ∂ω · ∆r d ω
−∞
∆tG (r) = (7.52)
R∞
I (r, ω) d ω
−∞
where I (r, ω) is given in (7.21). The two curves in Fig. 7.12 show I (r0 , ω) (before
propagation) and I (r, ω) (after propagation) for an initially Gaussian pulse. As
After Propagation
seen in (7.52), the pulse travel time depends on the spectral shape of the pulse at
the end of propagation.
Note the close resemblance between the formulas (7.49) and (7.52). Both are
expectation integrals. The former is executed as a ‘center-of-mass’ integral on Figure 7.12 Normalized power
time; the latter is executed in the frequency domain on ∂Rek · ∆r/∂ω, the group spectrum of a broadband pulse
delay function (7.38). The group delay at every frequency present in the pulse before and after propagation
through an absorbing medium
influences the result. If the pulse has a narrow bandwidth in the neighborhood
with the complex index shown in
of ω0 , the integral reduces to ∂Rek/∂ω|ω0 · ∆r, in agreement with (7.38) (see P7.9).
Fig. 7.10. The absorption line eats
The net group delay depends only on the spectral content of the pulse, indepen- a hole in the spectrum.
dent of its temporal organization (i.e., the phase of E (r, ω) has no influence). Only
the real part of the k-vector plays a direct role in (7.52).
The second term in (7.51) is the reshaping delay ∆t R . It represents a delay
that arises solely from a reshaping of the spectral amplitude. Often this term is
negligible. The term takes into account how the pulse time center-of-mass shifts
as portions of the spectrum are removed (or added), as illustrated in Fig. 7.13. It
is computed at r0 before propagation takes place:7
∆t R (r0 ) = 〈t 〉r0 ¯altered − 〈t 〉r0

¯
(7.53)
Here 〈t 〉r0 represents the usual arrival time of the pulse at the initial point r0 ,
according to (7.49). The intensity at this point is associated with a field E (r0 , t )
whose spectrum is E (r0 , ω). On the other hand,
¯
〈t 〉 r0 altered is the arrival time of
¯
a pulse with modified spectrum E (r0 , ω) e −Imk·∆r . Notice that E (r0 , ω) e −Imk·∆r is Figure 7.13 The center of a
still evaluated at the initial point r0 . Only the spectral amplitude (not the phase) chirped pulse can shift owing
is modified, according to what is anticipated to be lost (or gained) during the trip. to the reshaping effect when spec-
trum is removed.
In contrast to the net group delay, the reshaping delay is sensitive to how a pulse
6 M. Ware, S. A. Glasgow, and J. Peatross, “The Role of Group Velocity in Tracking Field Energy in
Linear Dielectrics,” Opt. Express 9, 506-518 (2001).

7 The reshaping delay can instead be computed after propagation takes place, in which case the
net group delay should be computed with the initial rather than final spectrum.
is organized. The reshaping delay is negligible if the pulse is initially symmetric

(in amplitude and phase) before propagation. The reshaping delay also goes to
zero in the narrowband limit, and the total delay reduces to the net group delay.
Example 7.8
Find the time required for a Gaussian pulse (7.23) to traverse a slab of absorption
material (neglecting possible surface reflections). Let the material response be
described by the Lorentz model described in section 2.2 with the carrier frequency
of the pulse ω0 , coinciding with the material resonance frequency. Let the slab
have thickness ∆r = cγ−1 /10 and absorption strength ω2p = 10γ.
Solution: The spectrum of the initially Gaussian pulse is given by (7.25), and its
power spectrum is8
2 2
I (r0 , ω) ∝ e −T (ω−ω0 )
After propagating from r0 to r = r0 + ∆r , the power spectrum becomes
2
(ω−ω0 )2 −2 κ(ω)ω
c ∆r
I (r, ω) ∝ e −T e
The net group delay is then
R∞ ³
∂(ωn/c)
´ R∞ 2
(ω−ω0 )2 −2 κω
³
∂n
´
Figure 7.14 Animation compar- I (r, ω) dω e −T e c ∆r n + ω ∂ω dω
∂ω ∆r
−∞ −∞
ing narrowband vs. broadband ∆tG (r) = ∆r =
R∞ c R∞ κω
c ∆r
2 (ω−ω )2
Gaussian pulses traversing an I (r, ω)d ω e −T 0 e −2 dω
absorbing slab (green stripe) on −∞ −∞
resonance. Note the logarithmic The index of refraction n + i κ is given by (2.39) (see also (2.27) and (2.29)). Since
scale. See Example 7.8. the expressions for n and κ are complicated, the integration in the above formula
must be performed numerically.
p
The result when T = T1 = 10γ−1 / 2 (narrowband) is
∆tG = −5.1/γ = −51∆r /c = −0.72T1

p
and the result when T = T2 = γ−1 / 2 (broadband) is
∆tG = 0.67/γ = 6.7∆r /c = 0.95T2
The reshaping delay 7.53 in both cases is negligible.
The narrowband pulse (with duration T1 ) in Example 7.8 traverses the ab-
sorbing medium superluminally (i.e. faster than c). The negative transit time
means that the ‘center-of-mass’ of the exiting pulse emerges even before the
‘center-of-mass’ of the entering pulse reaches the medium! On the other hand,
the broadband pulse (with the shorter duration T2 ) has a large positive delay time,
indicating that the exiting pulse emerges subluminally.
8 In general, one should write ω̄ to distinguish the carrier frequency of the pulse from the
0
resonance frequency of the material ω0 ; in practice, these are often different.
7.A Pulse Chirping in a Grating Pair 189
Figure 7.14 shows the intensity profiles for these two pulses as they traverse
the absorption slab, calculated with the aid of (7.31). By eye, one can see how
the centers of the two pulses are either advanced or delayed as they go through
the absorption medium. In both cases, the pulse that emerges is well within
the envelope of the original pulse propagated forward at c. In the case of the
broadband pulse, the absorption peak eats a hole in the center of the spectrum
as shown in Fig. 7.12, causing the emerging pulse to be distorted in time. The
analysis in this section predicts the center of pulses, whereas to see the shape of Figure 7.15 Delay as a function of
pulses one needs to calculate (7.31). pulse duration.
The results for the two pulse durations in Example 7.8 indicate a trend. Su-
perluminal behavior only occurs for long boring pulses. In the case of a single
absorption resonance, this comes with a severe cost of attenuation. Figure 7.15
shows the delay time as a function of pulse duration. As the injected pulse be-
comes more sharply defined in time, the superluminal behavior does not persist.
Sharply defined waveforms (i.e. broadband) cannot propagate superluminally
precisely because much of their bandwidth lies away from the frequencies with
superluminal group delays.
We should mention that superluminal propagation cannot persist for indefi-
nite distances since the medium eventually removes the superluminal spectral
components through absorption (or else adds subluminal spectral components Figure 7.16 Narrowband pulse
in the case of amplification). This limits the amount that a pulse center can be traversing an absorbing medium.
advanced—on the scale of the pulse’s own duration.
As we saw for the absorption situation the exiting pulse is tiny and resides
well within the original envelope of the pulse propagated forward at speed c,
as depicted in Fig. 7.16. Without the absorbing material in place, the signal
would be detectable just as early. This statement is also true for any spectral
behavior of a medium, including amplifying media. you can use the Lorentz
model (2.40) to describe an amplifying medium with a negative oscillator strength
f . Figure 7.17 shows narrowband and broadband pulse traversing an amplifying
medium. In this case, superluminal behavior occurs for spectra near by but not
on an amplifying resonance. If the pulse is too broadband, its spectrum will be
amplified, which adds slower components to the overall group delay.
Figure 7.17 Animation compar-
ing narrowband vs. broadband
Appendix 7.A Pulse Chirping in a Grating Pair Gaussian pulses traversing an am-
plifying slab (green stripe) slightly
Grating pairs can be used to introduce large amounts of dispersion into a light off resonance.
pulse. Gratings are especially useful for amplification of ultrashort laser pulses,
where laser pulses are first stretched in time before amplification (to prevent
damage to the amplifier) and then compressed back to short duration just before
the experiment (called chirped pulse amplification). Diffraction from a grating
causes each k-vector to travel at a different angle. A second grating parallel to the
first can realign all of the k-vectors to be parallel to each other. Since laser beams
are not infinitely wide, the light is typically sent through the grating pair twice
to undo the tendency of the different frequency components becoming laterally
separated. In the present analysis, we will consider an infinitely wide plane wave
pulse incident upon grating. The scenario is depicted in Fig. 7.19: A short plane
wave pulse strikes the grating at an angle, and a spreading pulse emerges.
Consider a plane-wave pulse that ricochets between a pair of parallel grating
surfaces. Although different k-vectors point with different angles, they are all
straightened out upon diffracting from the second grating. For simplicity, we will
consider a pulse just before the first bounce and just after the second bounce,
even though we are interested in the dispersion that takes place between the
gratings. Therefore, we can consider all k-vectors as being parallel with each
other.
First Second
Grating Grating Consider the a plane wave incident on a grating at an incident angle θi with
respect to the grating normal (aligned with the x-axis in our coordinate system)
as depicted in Fig. 7.18. The plane wave diffracts from the first grating at an angle
θr (also referenced from the grating normal). This angle is governed by the grating
diffraction formula9 µ ¶
−1 2πc
θr (ω) = sin − sin θi (7.54)
ωd
where d is the grating groove spacing. By examining the geometry of the figure,
we see that the reflected k-vector is given by k = x̂ cos θr + ŷ sin θr ω/c.
¡ ¢
Figure 7.18 Direction of k-vector Suppose we know the pulse at a point r0 on the first grating. Next we choose a
between parallel gratings (top point r0 + ∆r on the second grating where we will determine the outgoing pulse.
view). Grating rulings run in and Since we are considering an infinitely wide plane-wave pulse, it doesn’t matter
out of the page.
where we choose that point as long as it lies on the surface of the second grating.
The waveform will be the same everywhere along the surface of the second gratin,
only its arrival time will trivially differ. For convenience, we might as well take the
second point to be r0 + ∆r = r0 + L x̂ as shown in Fig. 7.18.
The phase delay needed for (7.30) becomes
Lω
k (ω) · ∆r = cos θr (7.55)
c
We will express this as a Taylor-series expansion similar to (7.39) so that we can
perform the inverse Fourier transform analytically. We will approximate (7.55) as
k (ω) · ∆r ≈ k 0 L + v g−1 (ω − ω0 ) L + α (ω − ω0 )2 L + · · · (7.56)
so that we can take advantage of formula (7.46). To calculate the terms in this
expansion we will need the derivative of θr :
d θr
µ ¶ µ ¶
1 2πc 1 2πc
=q ¢2 − ω2 d = p − 2
dω ¡ 2πc
1 − sin2 θr ω d
1− ωd − sin θi
(7.57)
2πc sin θi + sin θr
=− 2 =−
ω d cos θr ω cos θr
The derivatives necessary for the Taylor’s series expansion are

9 This formula is equivalent to d sin θ + d sin θ = λ with λ = 2πc/ω.
i r
7.B Causality and Exchange of Energy with the Medium 191
dk L d θr
µ ¶
· ∆r = cos θr − ω sin θr
dω c dω
L sin θi + sin θr
µ ¶
= cos θr + sin θr (7.58)
c cos θr
L 1 + sin θr sin θi
µ ¶
=
c cos θr
and
d 2k L sin θr (1 + sin θr sin θi ) d θr

µ ¶
· ∆r = sin θ i +
d ω2 c cos2 θr dω
L sin θi + sin θr sin θi + sin θr
µ ¶µ ¶
= − (7.59)
c cos2 θr ω cos θr
L (sin θi + sin θr )2
=−
ωc cos3 θr
The coefficients in (7.56) then are
∆r ω0
k 0 ≡ k|ω0 · = (7.60)
L c
∆r 1 + sin θr sin θi ¯¯
¯ ¯
d k ¯¯
v g−1 ≡ · = (7.61)
d ω ¯ω0 L c cos θr ¯
ω0
1 d 2 k ¯¯ ∆r (sin θi + sin θr )2 ¯¯
¯ ¯
α≡ · =− (7.62)
2 d ω2 ¯ω0 L 2cω cos3 θr ¯ω0
In the case of a Gaussian pulse, we can employ (7.46), where L takes the place of
z, and k 0 , v g−1 and α are defined by (7.60) – (7.62). The duration of the pulse is Figure 7.19 Animation showing a
controlled by (7.62) and the spacing between the gratings L. short plane-wave pulse diffracting
from a grating positioned along
the left edge of the frame.
Appendix 7.B Causality and Exchange of Energy with the
Medium
The group delay function indicates the average arrival of field energy to a point.
Since this is only part of the whole energy story, there is no problem when it
becomes superluminal. The overly rapid appearance of electromagnetic energy at
one point and its simultaneous disappearance at another point merely indicates
an exchange of energy between the electric field and the medium.10
We should not be dazzled by the magician who invites the audience to look
only at the field energy while energy transfers into and out of the ‘unwatched’
domain of the medium. Extra field energy seems to appear ‘prematurely’ down-
stream only if there is already non-zero field energy downstream to stimulate a
10 M. Ware, S. A. Glasgow, and J. Peatross, “Energy Transport in Linear Dielectrics,” Opt. Express 9,
519-532 (2001).
transfer of energy from the medium. The actual transport of energy is strictly
bounded by c; superluminal propagation of a signal front is impossible.
In accordance with Poynting’s theorem (2.51), the total energy density stored
in an electromagnetic field and in a medium is given by
u(r, t ) = u field (r, t ) + u med (r, t ) + u (r, −∞) (7.63)
where the time-dependent accumulation of energy transferred into the medium

from the field (ignoring possible free current Jfree ) is
Zt
¡ 0 ¢ ∂P r, t 0
¡ ¢
u med (r, t ) = E r, t · dt0 (7.64)
∂t 0
−∞
The expression (7.63) for the energy density includes all (relevant) forms of energy,
including a non-zero integration constant u (r, −∞) corresponding to energy
stored in the medium before the arrival of any pulse (important in the case of an
amplifying medium). u field (r, t ) and u med (r, t ) are both zero before the arrival of
the pulse (i.e. at t = −∞). In addition, u field (r, t ), given by (2.53), returns to zero
after the pulse has passed (i.e. at t = +∞).
As u med increases, the energy in the medium increases. Conversely, as u med
decreases, the medium surrenders energy to the electromagnetic field. While it is
possible for u med to become negative, the combination u med + u (−∞) (i.e. the net
energy in the medium) can never go negative since a material cannot surrender
more energy than it has to begin with.
Poynting’s theorem (2.51) has the form of a continuity equation which when
integrated spatially over a small volume V yields
∂
I Z
S · da = − u dV (7.65)
∂t
A V
where the left-hand side has been transformed into an surface integral (via the
divergence theorem (0.11)) representing the power leaving the volume. Let the
volume be small enough to take S to be uniform throughout V .
We can define an energy transport velocity (directed along S) as the effective
speed at which all of the energy density would need to travel in order to achieve
the Poynting flux:
S
vE ≡ (7.66)
u
Note that this ratio of the Poynting flux to the energy density has units of velocity.
When the total energy density u is used in computing (7.66), the energy transport
velocity has a fictitious nature; it is not the actual velocity of the total energy
(since part is stationary), but rather the effective velocity necessary to achieve
the same energy transport that the electromagnetic flux alone delivers. If we
reduce the denominator to the subset of the energy that can move, namely u field ,
the Cauchy-Schwartz inequality (i.e. α2 + β2 ≥ 2αβ) ensures an energy transport
velocity v E remains strictly bounded by the speed of light in vacuum c. The total
energy density u is at least as great as the field energy density u field . Hence, this
strict luminality is maintained.
Centroid of Energy
Consider a weighted average of the energy transport velocity:

vE u d 3 r S d 3r
R R
〈vE 〉 ≡ R 3
= R (7.67)
ud r u d 3r
where we have substituted from (7.66).
Integration by parts leads to
R ∂u 3
r∇ · S d 3 r
R
r d r
〈vE 〉 = − R 3
= R ∂t 3 (7.68)
ud r ud r
where we have assumed that the volume for the integration encloses all energy in
the system and that the field near the edges of this volume is zero. Since we have
included all energy, Poynting’s theorem (2.51) can be written with no source terms
(i.e. ∇ · S + ∂u/∂t = 0). This means that the total energy in the system is conserved
and is given by the integral in the denominator of (7.68). This allows the derivative
to be brought out in front of the entire expression giving
∂ 〈r〉 ru d 3 r
R
〈vE 〉 = where 〈r〉 ≡ R (7.69)
∂t u d 3r
The latter expression represents the ‘center-of-mass’ or centroid of the total en-
ergy in the system, which is guaranteed to evolve strictly luminally since vE is
everywhere luminal.11
It is enlightening to consider u med within a frequency-domain context. In an

isotropic medium, the polarization for an individual plane wave can be written in
terms of the linear susceptibility defined in (2.16):
P (r, ω) = ²0 χ (r, ω) E (r, ω) (7.70)
We can use this to express u med in terms of the electric field and material suscepti-
bility.
11 Although (7.69) guarantees that the centroid of the total energy moves strictly luminally, there is
no such limitation on the centroid of field energy alone. The steps leading to (7.69) are not possible
if u field is used in place of u. Explicitly, that is
∂ ru field d 3 r
¿ À R
S
6=
∂t u field d 3 r
R
u field
As was pointed out, the left-hand side is strictly luminal. However, the right-hand side can easily
exceed c as the medium exchanges energy with the field. In an amplifying medium, for example, the
rapid appearance of a pulse downstream can occur when the leading portion of a pulse stimulates
energy already present in the medium to convert to the form of field energy. Group velocity is
related to this method of accounting, which is why it also can become superluminal.
Expressing u med in terms of the power spectrum
The field E(r, t ) can be expressed as an inverse Fourier transform (7.18). Similarly,
the polarization P can be written as12
Z∞ Z∞
1 ∂P(r, t ) −i
P(r, t ) = p P (r, ω) e −i ωt d ω ⇒ =p ωP (r, ω) e −i ωt d ω (7.71)
2π ∂t 2π
−∞ −∞
The energy density in the medium (7.64) can then be written as
Z∞ Z∞ Z∞
  
 p1
0 0 −i ² 0
E r, ω0 e −i ω t d ω0 · p ωχ (r, ω) E (r, ω) e −i ωt d ω d t 0
¡ ¢ 0
u med (r, ∞) =
2π 2π
−∞ −∞ −∞
(7.72)
where we have incorporated (7.70) and evaluated u med after the pulse is over at
t = ∞. We may change the order of integration and write
Z∞ Z∞ Z∞
¢ 1 0 0
u med (r, ∞) = −i ²0 d ωωχ (r, ω) E (r, ω) · d ω0 E r, ω0 e −i (ω+ω )t d t 0
¡
2π
−∞ −∞ −∞
(7.73)
The final integral is a delta function a delta function similar to (0.54), which allows
the middle integral also to be performed. The expression for u med then reduces to
Z∞
u med (r, ∞) = −i ²0 ωχ (r, ω) E (r, ω) ·E (r, −ω) d ω (7.74)
−∞
In this derivation, we take E(r, t ) and P(r, t ) to be real functions, so we can employ
the symmetry (7.29) along with
P∗ (r, ω) = P (r, −ω) and χ∗ (r, ω) = χ (r, −ω) .
Then we obtain
Z∞
u med (r, ∞) = ²0 ωImχ (r, ω) E (r, ω) · E∗ (r, ω) d ω (7.75)
−∞
The expression (7.75) describes the net energy density transfered to a point
in the medium after all action has finished (i.e. at t = ∞). It involves the power
spectrum of the pulse. We can modify this formula in an intuitive way so that it
describes the transfer of energy density to the medium for any time during the
pulse.
Since the medium is unable to anticipate the spectrum of the entire pulse
before experiencing it, the material responds to the pulse according to the history
of the field up to each instant. In particular, the material has to be prepared for
12 We assume that the real forms of the fields in the time domain are used for the sake of this
multiplication.
the possibility of an abrupt cessation of the pulse at any moment, in which case
all exchange of energy with the medium immediately ceases. In this extreme sce-
nario, there is no possibility for the medium to recover from previously incorrect
attenuation or amplification, so it must have gotten it right already.
If the pulse were in fact to abruptly terminate at a given instant, it would
not be necessary to integrate the inverse Fourier transform (7.19) beyond the
termination time t after which all contributions are zero. Causality requires that
the medium be indifferent to whether a pulse actually terminates if that possibility
lies in the future. Therefore, (7.75) can apply for any time t (not just for t = ∞)
if the spectrum (7.19) is evaluated just for that portion of the field previously
experienced by the medium (up to time t ).
The following is then an exact representation for the energy density (7.64)
transferred to the medium:
Z∞
u med (r, t ) = ²0 ωImχ (r, ω) Et (r, ω) · E∗t (r, ω) d ω (7.76)
−∞
where
Zt
1
E r, t 0 e i ωt d t 0
0
E t (r, ω) ≡ p
¡ ¢
(7.77)
2π
−∞
This time dependence enters only through Et (r, ω) · E∗t (r, ω), known as the instan-
taneous power spectrum.
The expression (7.76) gives physical insight into the manner in which causal
dielectric materials exchange energy with different parts of an electromagnetic
pulse. Since the function E t (ω) is the Fourier transform of the pulse truncated
at the current time t and set to zero thereafter, it can include many frequency
components that are not present in the pulse taken in its entirety. This explains
why the medium can respond differently to the front of a pulse compared to the
back. Even though absorption or amplification resonances may lie outside of
the spectral envelope of a pulse taken in its entirety, the instantaneous spectrum Figure 7.20 Real and imaginary
on a portion of the pulse can momentarily lap onto or off of resonances in the parts of the refractive index for an
medium. amplifying medium.
In view of (7.76) and (7.77) it is straightforward to predict when the electro-
magnetic energy of a pulse will exhibit superluminal or subluminal behavior. In
section 7.5, we saw that this behavior is controlled by the group velocity function.
However, with (7.76) and (7.77), it is not necessary to examine the group velocity
directly, but only the imaginary part of the susceptibility χ (r, ω).
If the entire pulse passing through point r has a spectrum in the neighborhood
of an amplifying resonance, but not on the resonance, superluminal behavior
can result. The instantaneous spectrum during the front portion of the pulse is
generally wider and can therefore lap onto the nearby gain peak. The medium
accordingly amplifies this perceived spectrum, and the front of the pulse grows.
The energy is then returned to the medium from the latter portion of the pulse
as the instantaneous spectrum narrows and withdraws from the gain peak. The
effect is not only consistent with the principle of causality, it is a direct and general
consequence of causality as demonstrated by (7.76) and (7.77). p
As an illustration, consider the broadband waveform with T2 = γ−1 / 2 de-
scribed in Example 7.8. Consider an amplifying medium with index shown in
Fig. 7.20 with the amplifying resonance (negative oscillator strength) set on the
frequency ω0 = ω̄0 + 2γ, where ω̄0 is the carrier frequency. Thus, the resonance
structure is centered a modest distance above the carrier frequency, and there is
only minor spectral overlap between the pulse and the resonance structure.
Superluminal behavior can occur in amplifying materials when the forward
edge of a narrow-band pulse receives extra amplification. Fig. 7.21 shows how the
early portion of a pulse has a wide instantaneous spectrum computed by (7.77)
that can lap onto the amplifying resonance. As the wings grow and access the
neighboring resonance, the pulse extracts more energy from the medium. As the
wings diminish, the pulse surrenders much of that energy back to the medium,
which shifts the center of the pulse forward.
In this appendix we have indirectly proven that a sharply defined signal edge
cannot propagate faster than c. If a signal edge begins abruptly at time t 0 , the
Figure 7.21 Animation of a nar- instantaneous spectrum E t (ω) clearly remains identically zero until that time. In
rowband pulse traversing an am- other words, no energy may be exchanged with the medium until the field energy
plifying medium off resonance. from the pulse arrives. Since, as was pointed out in connection with (7.66), the
The black dot shows the move-
Cauchy-Schwartz inequality prevents the field energy from traveling faster than c,
ment of the center of all energy.
at no point in the medium can a signal front exceed c.
The red line inside the medium
shows the energy held in that
medium, which cannot go nega-
tive. The lower figure shows the
Appendix 7.C Kramers-Kronig Relations
instantaneous spectrum of the
In the late 1920s, of Ralph Kronig and Hendrik Kramers independently discovered
pulse at the front of the medium
relative to the narrow amplifying a remarkable relationship between the real and imaginary parts of a material’s
resonance. susceptibility χ (ω). Recall that the susceptibility as defined in (2.16) relates the
polarization of a material to the field that stimulates the medium:
P (ω) = ²0 χ (ω) E (ω) (7.78)
They made an argument based on causality (i.e. effect cannot precede cause),
which allows one to obtain the real part of χ (ω) from the imaginary part of χ (ω),
if it is known for all ω. Similarly, one can obtain the imaginary part of χ (ω) from
the real part of χ (ω). We develop the Kramers-Kronig formulas below.13
We can replace E (ω) in (7.78) with the Fourier transform of E (t ) in accordance
with (7.19). In addition, we take the inverse Fourier transform (7.19) of both sides
of (7.78) and obtain
Z∞ Z∞
 
²0 1
E t 0 e i ωt d t 0  e −i ωt d ω
0
χ (ω)  p
¡ ¢
P (t ) = p (7.79)
2π 2π
−∞ −∞
Also B. Y.-K. Hu, “Kramers-Kronig in two lines,” Am. J. Phys. 57, 821 (1989).
7.C Kramers-Kronig Relations 197
Next we interchange the order of integration to get
Z∞
 ∞ 
²0
Z
E t  χ (ω) e −i ω(t −t ) d ω d t 0
¡ 0¢ 0
P (t ) = (7.80)
2π
−∞ −∞
Now for the causality argument: The polarization of the medium P (t ) cannot
depend on the field E t 0 at future times t 0 > t . Therefore the expression in square
¡ ¢
brackets must be identically zero unless t − t 0 > 0. This places a restriction on the
functional form of χ (ω) as we shall see.
The causality argument comes explicitly into play when we employ the fol-
lowing integral formula:14
Z∞ 0 0
−i ω(t −t 0 ) 0 1 e−iω (t−t ) 0
e = sign{t − t } dω (7.81)
iπ ω − ω0
−∞
+1 (t > t 0 )
½
Apparently, we require the positive sign since sign{t − t0 } ≡ .
−1 (t < t 0 )
Upon substitution of (7.81) into (7.80) and after changing the order of integra-
tion within the square brackets we obtain
Z∞
 ∞
Z∞
 
²0 1 χ (ω)
Z
d ω e −i ω (t −t ) d ω0  d t 0
¡ 0¢ 0 0
P (t ) = E t   (7.82)
2π iπ ω − ω0
−∞ −∞ −∞
For (7.80) and (7.82) to be the same, we require
Z∞
χ ω0
¡ ¢
1
χ (ω) = d ω0 (7.83)
iπ ω0 − ω
−∞
or
Z∞
Reχ ω0 + i Imχ ω0
¡ ¢ ¡ ¢
1
Reχ (ω) + i Imχ (ω) = d ω0 (7.84)
iπ ω0 − ω
−∞
Finally, equating separately the real and imaginary parts of the above equation
yields
Z∞ Z∞
Imχ ω0 Reχ ω0
¡ ¢ ¡ ¢
1 0 1
Reχ (ω) = dω and Imχ (ω) = − d ω0 (7.85)
π ω0 − ω π ω0 − ω
−∞ −∞
14 This integral, which is a specific instance of Cauchy’s theorem, is tricky because it involves two
diverging pieces, to either side of the singularity ω = ω0 . The divergences have opposite sign so that
they cancel. The integration must approach the singularity in the same manner from either side, in
which case the result is called the principal value. In practical terms, if the integral is performed
numerically, the sampling of points should straddle the singularity symmetrically; other sampling
schemes can change the result dramatically, which is incorrect.
These are known as the Kramers-Kronig relations on real and imaginary parts of
χ.15 If the real part of χ is known at all frequencies, we can use the Kramers-Kronig
relations to generate the imaginary part, and visa versa. We see that the real and
imaginary parts of χ cannot be chosen independently, if we are to respect the
principle of causality.
Example 7.9
Show that the expression in square brackets of (7.80) is zero when t 0 > t , if χ (ω)
satisfies the Kramers-Kronig relations (7.85).
Solution: The expression may be written as
Z∞ Z∞ Z∞
0 0 0
χ (ω) e −i ω(t −t ) d ω = Reχ (ω) e −i ω(t −t ) d ω + i Imχ (ω) e −i ω(t −t ) d ω
−∞ −∞ −∞
Z∞ Z∞ Z∞
 
Reχ ω0
¡ ¢
t −t 0 − 1
0
= Reχ (ω) e −i ω( )d ω + i d ω0  e −i ω(t −t ) d ω
π ω0 − ω
−∞ −∞ −∞
Z∞ Z∞ Z∞ −i ω(t −t 0 )
 
t −t 0 1 e
Reχ (ω) e −i ω( )d ω + Reχ ω0  d ω d ω0
¡
¢
=
iπ ω0 − ω
−∞ −∞ −∞
(7.86)
where we have invoked the Kramers-Kronig relation for Imχ (ω) (7.85) and inter-
changed the order of integration in the final expression. Since we are specifically
considering future times t 0 > t , we have by (7.81)
Z∞ 0
1 e −i ω(t −t ) 0 0
d ω = −e −i ω (t −t )
iπ ω0 − ω
−∞
Hence
Z∞ Z∞ Z∞
−i ω(t −t 0 ) −i ω(t −t 0 ) 0 0
χ (ω) e d ω = Reχ (ω) e dω − Reχ ω0 e −i ω (t −t ) d ω0
¡ ¢
−∞ −∞ −∞
=0
(7.87)
Finally, it is worth noting that the Kramers-Kronig relations also apply to the
15 As with (7.81), the principal value of the integral must be calculated. If the integral is performed
numerically, the sampling of points should straddle the singularity symmetrically. Separately, the
integral on each side of ω0 = ω diverges, but with opposite sign.
7.C Kramers-Kronig Relations 199
real and imaginary parts of the index of refraction (subtract one). 16
Z∞ Z∞
κ ω0 n ω0 − 1
¡ ¢ ¡ ¢
1 0 1
n (ω) − 1 = dω and κ (ω) = − d ω0 (7.88)
π ω0 − ω π ω0 − ω
−∞ −∞
One can use the Kramers-Kronig relations to find the real part of the index from
a measurement of absorption, if the measurement is done over a broad enough
range of the spectrum. This is the most useful form of the Kramers-Kronig rela-
tions.
It is sometimes convenient to multiply the numerator and denominator inside
the integrands of (7.88) by ω0 + ω. Then noting that n is an even function and
κ is an odd function allows us to dismiss either ω0 or ω in the numerator and
integrate17 over positive frequencies only:
Z∞ Z∞ ¡ 0 ¢
ω0 κ ω0 n ω −1
¡ ¢
2 0 2ω
n (ω) − 1 = dω and κ (ω) = − d ω0 (7.89)
π ω02 − ω2 π ω02 − ω2
0 0
16 This follows from Cauchy’s theorem since the index (subtract one) is the square root of χ (ω).
The Kramers-Kronig relations for χ (ω) guarantee that χ (ω) has no poles in the upper half complex
plane, when ω is considered (for mathematical purposes) to be a complex variable. Taking the
square root does not introduce poles into the upper half plane.
17 The integrals (7.88) and (7.89) diverge to either side of ω0 = ω, but with opposite sign. Again,
the principal value of the integral is required, which means a numeric grid should straddle the
singularity symetrically.
Exercises
Exercises for 7.1 Intensity of Superimposed Plane Waves
P7.1 (a) Consider two counter-propagating fields described by x̂E 1 e i (kz−ωt )

and x̂E 2 e i (−kz−ωt ) where E 1 and E 2 are both real. Show that their sum
can be written as
x̂E tot (z) e i (Φ(z)−ωt )
where s
E2 2 E2
µ ¶
E tot (z) = E 1 1 − + 4 cos2 kz
E1 E1
and
(1 − E 2 /E 1 )
· ¸
−1
Φ (z) = tan tan kz
(1 + E 2 /E 1 )
Outside the range − π2 ≤ kz ≤ π
2 the pattern repeats.
(b) Suppose that two counter-propagating laser fields have separate
intensities, I 1 and I 2 = I 1 /100. The ratio of the fields is then E 2 /E 1 =
1/10. In the standing interference pattern that results, what is the ratio
of the peak intensity to the minimum intensity? Are you surprised how
high this is?
P7.2 Equation (7.7) implies that there is no interference between fields that
are polarized along orthogonal dimensions. That is, the intensity of
E(r, t ) = x̂E 0 e i [(k ẑ)·r−ωt ] + ŷE 0 e i [(k x̂)·r−ωt ]
according to (7.7) is uniform throughout space. Of course (7.7) does not

apply since the k-vectors are not parallel. Show that the time-average
of S (r, t ) according to (7.4) exhibits interference in the distribution of
net energy flow.
Exercises for 7.2 Group vs. Phase Velocity: Sum of Two Plane Waves
P7.3 Show that (7.10) can be written as

ω2 +ω1
∆k ∆ω
³ ´ µ ¶
k2 +k1
i ·r− t
E(r, t ) = 2E0 e 2 2
cos ·r− t
2 2
From this show that the speed of the rapid-oscillation intensity peaks
in Fig. 7.2 is v p0 = ω̄/k̄ where
(k 1 + k 2 ) (ω1 + ω2 )
k̄ ≡ and ω̄ ≡
2 2
P7.4 Confirm the right-hand side of (7.17).

Exercises 201
Exercises for 7.3 Frequency Spectrum of Light
P7.5 The continuous field of a very narrowband continuous laser may be

approximated as a pure plane wave: E(r, t ) = E0 e i (k0 z−ω0 t ) . Suppose the
wave encounters a shutter at the plane z = 0.
(a) Compute the power spectrum of the light before the shutter. HINT:
The answer is proportional to the square of a delta function centered
on ω0 (see (0.54)).
(b) Compute the power spectrum after the shutter if it is opened during
the interval −T /2 ≤ t ≤ T /2. Plot the result. Are you surprised that the
shutter appears to create extra frequency components?
HINT: Write your answer in terms of the sinc function defined by
sincα ≡ sin α/α.
P7.6 (a) Determine the Full-Width-at-Half-Maximum of the intensity (i.e.

the width of I (r, t ) represented by ∆t FWHM ) and of the power spectrum
(i.e. the width of I (r, ω) represented by ∆ωFWHM ) for the Gaussian pulse
defined in (7.25).
HINT: Both answers are in terms of T .
(b) Give an uncertainty principle for the product of ∆t FWHM and ∆ωFWHM .
Exercises for 7.5 Quadratic Dispersion
P7.7 The intensity of a Gaussian laser pulse has a FWHM duration TFWHM =
25 fs with carrier frequency ω0 corresponding to λvac = 800 nm. The
pulse goes through a lens of thickness ` = 1 cm (laser quality glass type
BK7) with index of refraction given approximately by
ω
n (ω) ∼
= 1.4948 + 0.016
ω0
What is the full-width-at-half-maximum of the intensity for the emerg-

ing pulse?
HINT: For the input pulse we have
TFWHM
T= p
2 ln 2
(see P7.6).
P7.8 If the pulse defined in (7.46) travels through the material for a very long
distance z such that T̃ (z) → T Φ (z) and tan−1 Φ (z) → π/2, show that
the instantaneous frequency of the pulse is
t − 2z/v g
ω0 +
4αz
COMMENT: As the wave travels, the earlier part of the pulse oscillates
more slowly than the later part. This is called chirp, and it means that
the red frequencies get ahead of the blue ones since they experience a
lower index.
Exercises for 7.6 Generalized Context for Group Delay
P7.9 When the spectrum is narrow compared to features in a resonance

(such as in Fig. 7.10), the reshaping delay (7.53) tends to zero and can
be ignored. Show that when the spectrum is narrow the net group delay
(7.52) reduces to
∂Rek
¯
lim ∆tG (r) = · ∆r¯¯
¯
T →∞ ∂ω ω̄
P7.10 When the spectrum is very broad the reshaping delay (7.53) also tends
to zero and can be ignored. Show that when the spectrum is extremely
broad, the net group delay reduces to
∆r
lim ∆tG (r) =
T →0 c
assuming k and ∆r are parallel. This implies that a sharply defined
signal cannot travel faster than c.
HINT: The real index of refraction n goes to unity far from resonance,
and the imaginary part κ goes to zero.
P7.11 Work through the derivation of (7.51).

HINT: This somewhat lengthy derivation can be found in the reference
in the footnote near (7.51).
Exercises for 7.A Pulse Chirping in a Grating Pair
P7.12 A Gaussian pulse with T = 20 fs is incident with θi = 20◦ on a grating

pair with groove separation d = 1.67 µm. What grating separation L
will lead to a pulse duration of T = 100 ps? Assume two passes through
the grating pair for a total effective separation of 2L. Take the pulse
carrier frequency to corresponds to λ0 = 800 nm.
Chapter 8
Coherence Theory
Coherence theory is the study of correlations that exist between different parts of
a light field. In temporal coherence theory, we focus on the correlation between
the fields at different times, E(r, t ) and E(r, t + τ). In spatial coherence theory,
we focus on the correlations between fields at different spatial locations, E(r, t )
and E(r + ∆r, t ). Because light oscillations are too fast to resolve directly, we
usually need to study optical coherence using interference techniques. In these
techniques, light from different times or places in the light field are brought
together at a detection point. If the two fields have a high degree of coherence,
they consistently interfere either constructively or destructively at the detection
point. If the two fields are not coherent, the interference at the detection point
rapidly fluctuates between constructive and destructive interference, so that a
time-averaged signal does not show interference.
You are probably already familiar with two instruments that measure coher-
ence: the Michelson interferometer, which measures temporal coherence, and
Young’s two-slit interferometer, which measures spatial coherence. Your pre-
liminary understanding of these instruments was probably gained in terms of
single-frequency plane waves, which are perfectly coherent for all separations in
time and space. In this chapter, we build on that foundation and derive descrip-
tions that are appropriate when light with imperfect coherence is sent through
these instruments. We also discuss a practical application known as Fourier spec-
troscopy (Section 8.4) which allows us to measure the spectrum of light using a
Michelson interferometer rather than a grating spectrometer.
Beam
Splitter
8.1 Michelson Interferometer
A Michelson interferometer employs a 50:50 beamsplitter to divide an initial
beam into two identical beams and then delays one beam with respect to the
other before bringing them back together (see Fig. 8.1). Depending on the relative
Detector
path difference d (roundtrip by our convention) between the two arms of the
system, the light can interfere constructively or destructively in the direction of Figure 8.1 Michelson interferome-
the detector. The relative path difference d introduces a relative time delay τ, ter.
203
204 Chapter 8 Coherence Theory
defined by τ ≡ d /c.
If the input light is a plane-wave, the net field at the detector consists of the
field coming from one arm of the interferometer E0 e i (kz−ωt ) added to the field
coming from the other arm E0 e i (kz−ω(t −τ)) . These two fields are identical except
for the delay τ. The intensity seen at the detector as a function of path difference
is computed to be
c²0 h i h i∗
I tot (τ) = E0 e i (kz−ωt ) + E0 e i (kz−ω(t −τ)) · E0 e i (kz−ωt ) + E0 e i (kz−ω(t −τ))
2
Figure 8.2 The intensity seen at c²0 £
2E0 · E∗0 + 2E0 · E∗0 cos(ωτ)
¤
=
the detector of a Michelson in- 2
terferometer with a plane-wave = 2I 0 [1 + cos(ωτ)]
input. Because the plane wave is (Plane Wave Input) (8.1)
infinitely coherent, the output os- c²0 ∗
where I 0 ≡ 2 E0 · E0 is the intensity from one beam alone (when the other arm of
cillates forever in both directions.
the interferometer is blocked). This formula is probably familiar. It describes how
Energy is conserved, so when the
intensity at the detector is zero,
the intensity at the detector oscillates between zero and four times the intensity
all of the input light is being sent of the beam from one arm when the other is blocked,1 as plotted in Fig. 8.2.
back on the input arm of the inter- When light containing a continuous band of frequencies is sent through the
ferometer. interferometer, (8.1) no longer holds. Instead of repeating indefinitely, the oscilla-
tions in the intensity at the detector become less pronounced as τ increases. The
concept of temporal coherence describes how fast fringe visibility diminishes as
delay is introduced in an arm of the Michelson interferometer. The less coherent
the light source, the faster the fringes die out as τ is increased. To model this
behavior, we need to expand our analysis beyond (8.1).
Consider an arbitrary waveform E(t ) (comprised of many frequency compo-
nents) that has traveled through the first arm of a Michelson interferometer to
arrive at the detector in Fig. 8.1. Again, E(t ) is the value of the field at the detector
when the second arm is blocked. The beam that travels through the second arm
of the interferometer is identical, but delayed by the round-trip delay τ: E (t − τ).
The total field at the detector is the sum of these two fields:
Etot (t , τ) = E (t ) + E (t − τ) (8.2)
The total intensity I tot at the detector is found using (7.21) with n = 1:
c²0
I tot (t , τ) = Etot (t , τ) · E∗tot (t , τ)
2
c²0 £
E(t ) · E∗ (t ) + E(t ) · E∗ (t − τ) + E(t − τ) · E∗ (t ) + E(t − τ) · E∗ (t − τ)
¤
=
2
c²0 £
= I (t ) + I (t − τ) + E(t ) · E∗ (t − τ) + E(t − τ) · E∗ (t )
¤
2
= I (t ) + I (t − τ) + c²0 Re E(t ) · E∗ (t − τ)
© ª
(8.3)
The function I (t ) corresponds to the intensity of one of the beams arriving at the
detector while the opposite path of the interferometer is blocked.
1 Keep in mind that if a 50:50 beam splitter is used, then the intensity arriving to the detector
from one arm alone (with other arm blocked) is one fourth of the original beam, since the light
meets the beam splitter twice.
8.1 Michelson Interferometer 205
For now we treat E (t ) as a pulse with a finite duration and energy to simplify
the math. Later we illustrate how to adapt this analysis for continuous light
sources. In (8.3) we have retained the t dependence of I tot (t , τ) in addition to the
dependence on the path delay τ. This allows for pulses with arbitrary duration
and shape. The rapid oscillations of the light are automatically averaged away in
I (t ) since we used (7.21), but the slowly varying envelope of the pulse is retained.
For a pulsed source, the physical signal from a Michelson interferometer is
proportional to the total amount of pulse energy arriving at the detector as a
function of τ.2 This physical signal, which we’ll denote by Sig(τ), is proportional
to the total energy per area, or fluence, accumulated at the detector:
Z∞
Sig(τ) ∝ Itot (t, τ) dt (8.4)
−∞
The proportionality constant will depend on the area of the beam, as well as the
units with which the detector reports Sig(τ) (volts, etc.). We can manipulate the Albert Abraham Michelson (1852
fluence integral in (8.4) into a more useful form that will make the coherence 1931, United States) was born in
Poland, but he immigrated to the
properties more evident. US with his parents and grew up in
the rough mining towns of California
and Nevada where his father was a
Manipulation of the fluence integral merchant. Michelson attended high

school in San Fransisco. He entered
the US Naval Academy in 1869 (with
Inserting (8.3) in the fluence integral, we have
intervention from US President Grant
after Michelson pleaded his case on
Z∞ Z∞ Z∞ Z∞ the grounds near the White House).
∗
I tot (t , τ) d t = I (t )d t + I (t − τ) d t + c²0 Re E (t ) · E (t − τ) d t (8.5) After two years at sea, Michelson re-
−∞ −∞ −∞ −∞ turned to the Naval Academy to teach

physics and mathematics for several
3 years. Michelson was fascinated by the
The first two integrals on the right-hand side of (8.5) are equal, and give the
problem of determining the speed of
fluence E from one arm of the interferometer when the other arm is blocked: light, and developed successive exper-
iments to measure it more accurately.
Z∞ Z∞
He is probably most famous for his ex-
E≡ I (t )d t = I (t − τ) d t (8.6) periment conducted at Case School of
−∞ −∞ Applied Science in Cleveland with Ed-
ward Morley to detect the motion of
The final integral in (8.5) remains unchanged if we take a Fourier transform fol- the earth through the ether. Michelson
later was a professor at the University of
lowed by an inverse Fourier transform:
Chicago and then at Caltech. In 1907,
he became the rst American to win the
Z∞ Z∞ Z∞ Z∞
 
1 1 Nobel prize, for his contributions to op-
E(t ) · E∗ (t − τ) d t = p d ωe −i ωτ  p d τe i ωτ E (t ) · E∗ (t − τ) d t  tics. Michelson married late in life and
2π 2π was the father of four. (Wikipedia)
−∞ −∞ −∞ −∞
(8.7)
The reason for this procedure is so that we can take advantage of the autocor-
relation theorem described in P0.27. With it, the expression in square brackets
2 For sub-nanosecond laser pulses, a detector automatically integrates the entire energy of the
pulse since a detector cannot keep up with temporal variations on such a rapid time scale. For
longer pulses, it may be necessary to force the integration.
3 Note that the second integral is insensitive to τ since a change of variables t 0 = t − τ converts it
into the first integral.

p p
simplifies to 2πE (ω) · E∗ (ω) = 2π2I (ω) /c²0 . Then with the aid of (8.6) and (8.7),
the overall fluence (8.5) becomes
Z∞ Z∞
 
1
I tot (t , τ) d t = 2E 1 + Re I (ω)e −i ωτ d ω (8.8)
E
−∞ −∞
With (8.8), we can rewrite the physical signal (8.4) in the more useful form
Sig(τ) ∝ 2E 1 + Re γ (τ)
£ © ª¤
(8.9)
where the dependence on the path delay τ is entirely contained in the degree of
coherence function γ (τ):4
R∞
I (ω) e −i ωτ d ω
−∞
γ (τ) ≡ (8.10)
R∞
I (ω) d ω
−∞
The denominator of (8.10) was rewritten with the help of Parseval’s theorem
R∞ R∞
E ≡ −∞ I (t )d t = −∞ I (ω) d ω. Remarkably, the signal out of the Michelson inter-
ferometer does not depend on the phase of E (ω). It depends only on the amount
of light associated with each frequency through I (ω) ≡ ²20 c E (ω) · E∗ (ω).
Alternate derivation of (8.9)
We could have derived (8.9) using another strategy, which may seem more intuitive
than the approach above. Equation (8.1) gives the intensity at the detector when a
single plane wave of frequency ω goes through the interferometer. Now suppose
that a waveform composed of many frequencies is sent through the interferometer.
The intensity associated with each frequency acts independently, obeying (8.1)
individually.
The total energy (per area) accumulated at the detector is then a linear superposi-
tion of the spectral intensities of all frequencies present:
Z∞ Z∞
I tot (ω, τ) d ω = 2I (ω) [1 + cos (ωτ)] d ω (8.11)
−∞ −∞
While this procedure may seem obvious, the fact that we can do it is remarkable!
Remember that it is usually the fields that we must add together before finding the
intensity of the resulting superposition. The formula (8.11) with its superposition
of intensities relies on the fact that the different frequencies inside the interferom-
eter when time-averaged (over all time) do not interfere. Certainly, the fields at
different frequencies do interfere (or beat in time). However, they constructively
interfere as often as they destructively interfere, and in a time-averaged picture it
is as though the individual frequency components transmit independently. Again,
4 M. Born and E. Wolf, Principles of Optics, 7th ed., p. 570 (Cambridge University Press, 1999).
8.1 Michelson Interferometer 207
in writing (8.11) we considered the light to be pulsed rather than continuous so

that the integrals converge.
We can manipulate (8.11) as follows:
 R∞ 
Z∞ Z∞ I (ω) cos (ωτ) d ω
 
−∞
I tot (ω, τ) d ω = 2 I (ω) d ω 
 
1 + (8.12)

R∞ 
−∞ −∞ I (ω) d ω
−∞
This is the same as (8.8) since we can replace cos(ωτ) with Re e −i ωτ , and we can
© ª
apply Parseval’s theorem (8.6) to the other integrals. Thus, the above arguments
lead to (8.9) and (8.10).
Example 8.1
Compute the output signal when a Gaussian pulse with spectrum (7.25) is sent
into a Michelson interferometer.
Solution: The power spectrum of the pulse is5

²0 c 2 2
I (r, ω) = E0 · E∗0 T 2 e −T (ω−ω0 )
2
where T is the pulse duration, not to be confused with τ, the delay of the interfer-
ometer arm. As shown in Example 7.3, we also have
Z∞ 0
²0 c p
-1
I (r, ω) d ω = E0 · E∗0 T π 0 1
2
−∞
The degree of coherence (8.10) is then Figure 8.3 The output or signal
Z∞ from a Michelson interferometer
T 2
(ω−ω0 )2 −i ωτ for light with a Gaussian spec-
γ (τ) = p e −T e dω
π trum.
−∞
Z∞ 2 ω −i τ 2
π (2T )
r
T −T 2 ω2 +(2T 2 ω0 −i τ)ω−T 2 ω20 T 0
−T 2 ω20
=p e dω = p e 4T 2
π π T2
−∞
τ2
e −i ω0 τ
−
=e 4T 2
Formula (0.55) was used to complete the integration. According to (8.9), the signal
at the detector is then
2
· ¸
− τ2
Sig(τ) ∝ 2E 1 + Re γ (τ) = 2E 1 + e cos (ω0 τ)
£ © ª¤
4T
Figure 8.3 shows this signal for a given T . As delay is added (or subtracted), the
output signal oscillates. Eventually enough delay is introduced such that the
very short pulses no longer interfere (arriving sequentially), and the output signal
becomes steady.
5 Technically, the output intensity is one fourth this, but our calculation of the degree of coher-
ence is insensitive to amplitude.

8.2 Coherence Time and Fringe Visibility

The degree of coherence function γ (τ) describes the oscillations in intensity
at the detector as the mirror in one arm of the interferometer is moved. The
real part of γ (τ) is analogous to cos(ωτ) in (8.1). However, for large delays τ,
the oscillations tend to die off as different frequencies get out of sync—some
interfere constructively, while others interfere destructively. Narrowband light is
temporally more coherent than broadband light because there is less opportunity
for frequencies to get out of sync. Still, for large path differences, the oscillations
eventually die off, and the time-averaged intensity at the detector then remains
steady as the mirror is moved further.
The coherence time τc is the amount of delay necessary to cause γ(τ) to quit
oscillating (i.e. its amplitude approaches zero). This definition is not very precise,
since the oscillations do not usually have an abrupt end, but instead slowly die off
as τ increases. A useful (although arbitrary) analytic definition for the coherence
time is
Z∞ Z∞
¯2 ¯2
τc ≡ γ (τ) d τ = 2 ¯γ (τ)¯ d τ
¯ ¯
¯ ¯ (8.13)
−∞ 0
The coherence length is the distance that light travels in this time:
`c ≡ cτc (8.14)
Another useful concept is fringe visibility. The fringe visibility is defined in

the following way: £ ¤ £ ¤
max Sig(τ) − min Sig(τ)
V (τ) ≡ £ ¤ £ ¤ (8.15)
max Sig(τ) + min Sig(τ)
£ ¤
where max Sig(τ) refers to the detector signal when the mirror is positioned
such that the amount of throughput to the detector is a local maximum, and
£ ¤
min Sig(τ) refers to the detector signal when the mirror is positioned such that
the amount of throughput to the detector is a local minimum. The minimum
and the maximum don’t occur at exactly the same τ, but for optical frequencies
the difference in τ is only about half an optical period. As the mirror moves
a large distance from the equal-path-length position, the oscillations in Sig(τ)
become less pronounced as the max and min tend to the same value, and the
fringe visibility goes to zero when γ (τ) = 0. It is left as an exercise (see P8.1) to
show that the fringe visibility can be written as6
V (τ) = ¯γ (τ)¯
¯ ¯
(8.16)
Note that the fringe visibility depends only upon the frequency content of the
light without regard to whether the frequency components are organized into a
short pulse or a longer time pattern.
8.3 Temporal Coherence of Continuous Sources 209
Example 8.2
Find the fringe visibility and the coherence time for the Gaussian pulse studied in
Example 8.1.
Solution: By (8.16), the fringe visibility is

2
− τ
V (τ) = ¯γ (τ)¯ = e 4T 2 .
¯ ¯
This is shown as the dashed line in Fig. 8.4. As expected, the fringe visibility dies
off as delay τ gets farther from the origin, the point where the interferometer arms
are equidistant. From (8.13) the coherence time is
Z∞ Z∞ Figure 8.4 Re γ(τ) (solid) and
© ª
2
− τ2 p
τc = ¯γ (τ)¯2 d τ = d τ = 2πT
¯ ¯
e 2T |γ(τ)| (dashed) for a light pulse
−∞ −∞ with a Gaussian spectrum as in
which is the delay necessary to cause the fringes to substantially diminish. examples 8.1 and 8.2.
8.3 Temporal Coherence of Continuous Sources

Consider a continuous light source such as starlight or a continuous wave (CW)
R∞
laser. The integral −∞ I (t )d t diverges for such a source, since it is on forever (or
at least for a very long time) and emits infinite (or very much) energy. However,
note that the integrals on both sides of (8.5) diverge in the same way. We can
renormalize (8.5) in this case by replacing the integrals on each side with the
average value of the intensity:
ZT /2
1
I ave ≡ 〈I (t )〉t = I (t )d t (continuous source) (8.17)
T
−T /2
The duration T must be large enough to average over any fluctuations that are
present in the light source. The average in (8.17) should not be used on a pulsed
light source since the result would depend on the duration T of the temporal
window.
For a continuous light source, the signal at the detector (8.9) becomes
£ ¤
Sig(τ) ∝ 2 〈I (t )〉t 1 + Reγ (τ) (continuous source) (8.18)
Although technically the integrals used in (8.10) to compute γ (τ) also diverge
in the case of continuous light, the numerator and the denominator diverge in
the same way. Therefore, we may renormalize I (ω) in any way we like to deal
with this problem. Both the numerator and denominator of (8.10) contain I (ω),
so regardless of how large I (ω) is or what units the measurement gives (volts
or whatever), we can just plug the instrument reading directly into (8.10). The
units in the numerator and denominator cancel so that γ (τ) always remains
dimensionless. Once we have the degree of coherence function γ(τ), we can
calculate the coherence time and fringe visibility just as we did for pulsed sources.
8.4 Fourier Spectroscopy

As we saw in (8.8), the signal output from a Michelson interferometer for a pulsed
input may be written as
Z∞
Sig (τ) ∝ 2E + 2Re I (ω)e −i ωτ d ω (8.19)
−∞
Typically, the signal comes in the form of a voltage or a current from a sensor.
However, the signal can easily be normalized to the beam fluence. In particular,
for large τ the fringe visibility goes to zero (i.e. γ (τ) = 0), and the normalized
signal must approach
Z ∞
lim Sig (τ) = 2E = 2 I (t )d t (8.20)
τ→∞ −∞
We will assume that this normalization has taken place and write (8.19) as an
equality.
Given our measurement of Sig(τ), we would like to find the power spectrum
I (ω). Unfortunately, I (ω) is buried within an integral in (8.19). However, since the
integral looks like an inverse Fourier transform of I (ω), we will be able to extract
the desired spectrum after some manipulation. This procedure for extracting I (ω)
from an interferometric measurement is known as Fourier spectroscopy.7
Extracting I (ω)
We first take the Fourier transform of (8.19):8

Z∞
 
 
−i ωτ
F Sig (τ) = F {2E } + F 2Re dω
© ª
I (ω) e (8.21)
 
−∞
The left-hand side is known since it is the measured data, and a computer can be
employed to take the Fourier transform of it. The first term on the right-hand side
is the Fourier transform of a constant:
Z∞ p
1
F {2E } = 2E p e i ωτ d τ = 2E 2πδ (ω) (8.22)
2π
−∞
Notice that (8.22) is zero everywhere except where ω = 0, where a spike occurs.
This represents the DC component of F Sig (τ) .
© ª
The second term of (8.21) can be written as

Z∞
 ∞
Z∞
  
  Z 
−i ωτ −i ωτ i ωτ
F 2Re I (ω) e dω = F I (ω) e dω + I (ω) e d ω
   
−∞ −∞ −∞
7 J. Peatross and S. Bergeson, “Fourier Spectroscopy of Ultrashort Laser Pulses,” Am. J. Phys. 74,
842-845 (2006).
8 This is weird since normally we take Fourier transforms on fields rather than expressions
involving intensity!
8.5 Young’s Two-Slit Setup and Spatial Coherence 211
Carrying out the Fourier transform then gives
Z∞ Z∞ Z∞ Z∞
   
1 0 1 0
p I (ω0 )e −i ω τ d ω0  e i ωτ d τ + p I (ω0 )e i ω τ d ω0  e i ωτ d τ
2π 2π
−∞ −∞ −∞ −∞
which we rearrange to
 ∞
Z∞ Z∞ Z∞
    
p Z
0 1 −i (ω0 −ω)τ 0 0 1 −i (ω0 +ω)τ 0
2π  I (ω )  e d τ d ω + I (ω )  e d τ d ω 
2π 2π
−∞ −∞ −∞ −∞
From (0.52) we note that the terms in parentheses are delta functions, so we have
 ∞
Z∞

p Z
2π  I (ω0 )δ ω0 − ω d ω0 + I (ω0 )δ ω0 + ω d ω0 
¡ ¢ ¡ ¢
−∞ −∞
The remaining frequency integrals can then be easily performed to obtain our final
form:
Z∞
 
  p
F 2Re I (ω) e −i ωτ d ω = 2π [I (ω) + I (−ω)] (8.23)
 
−∞
With (8.22) and (8.23) we can write (8.21) as
F Sig (τ)
© ª
p = 2E 0 δ (ω) + I (ω) + I (−ω) (8.24)
2π
The Fourier transform of the measured signal is seen to contain three terms, one
of which is the power spectrum I (ω) that we are after. Fortunately, when graphed
as a function of ω (shown in Fig. 8.5), the three terms on the right-hand side
typically do not overlap. As a reminder, the measured signal as a function of τ
looks something like that in Fig. 8.3. The oscillation frequency of the fringes lies
in the neighborhood of ω0 . The procedure to obtain I (ω) is (1) Record Sig (τ); (2)
if desired, normalize by its value at large τ; (3) take its Fourier transform; and (4) A graphical depiction of
Figure 8.5±p
F {Sig(τ)} 2π .
extract the curve at positive frequencies.
8.5 Young’s Two-Slit Setup and Spatial Coherence

In close analogy with the Michelson interferometer, which is able to investigate
temporal coherence, a Young’s two-slit setup can be used to investigate spatial co-
herence of quasi-monochromatic light. Thomas Young, who lived nearly a century
before Michelson, used his two-slit setup for the first conclusive demonstration
that light propagates as a wave. The Young’s double-slit setup and the Michelson
interferometer have in common that two beams of light travel different paths
and then interfere. In the Michelson interferometer, one path is delayed with
respect to the other so that temporal effects can be studied. In the Young’s two-slit
setup, two laterally separate points of the same wave are compared as they are
sent through two slits.
Fringe Pattern
Point Source
Figure 8.6 A point source produces coherent (locked phases) light. When this light
which traverses two slits and arrives at a screen it produces a fringe pattern.
Depending on the coherence of the light entering each slit, the fringe pattern
observed can exhibit good or poor visibility. Just as the Michelson interferometer
is sensitive to the spectral content of light, the Young’s two-slit setup is sensitive
to the spatial extent of the light source illuminating the two slits. For example, if
light from a distant star (restricted by a filter to a narrow spectral range) is used to
illuminate a double-slit setup, the resulting interference pattern appearing on a
subsequent screen shows good or poor fringe visibility depending on the angular
width of the star. Michelson was the first to use this type of setup to measure the
angular width of stars.
Light emerging from a single ideal point source has wave fronts that are
spatially uniform in a lateral sense (see Fig. 8.6). Such wave fronts are said to be
spatially coherent, even if the temporal coherence is not perfect (i.e. if a range
of frequencies is present). When spatially coherent light illuminates a Young’s
two-slit setup, fringes of maximum visibility are seen at a distant screen, meaning
the fringes vary between a maximum intensity and zero.
Consider a Young’s two-slit setup illuminated by a single point source. We
represent the fields on a subsequent screen that transmit through each slit, re-
spectively, as E0 e i (kd1 −ωt ) and E0 e i (kd2 −ωt ) . We have assumed that the slits are
equidistant from the point source and that the two fields at the screen are identi-
cal other than for their phases. In close analogy with (8.1), the resulting intensity
pattern on a far-away screen is
£ ¡ ¢¤
I tot (h) = 2I 0 [1 + cos (kd 2 − kd 1 )] = 2I 0 1 + cos kh y/D (8.25)
Notice the close similarity between this expression and the output from a Michel-
son interferometer for a plane wave (8.1). We will consider h (the separation of
the slits) to be the counterpart of τ (the delay introduced by moving a mirror in
the Michelson interferometer). To obtain the final expression in (8.25) we made
8.5 Young’s Two-Slit Setup and Spatial Coherence
Extended Source 213
Fringe Pattern
Figure 8.7 Light from an extended source is only partially coherent. Fringes are still
possible, but they exhibit less contrast.
use of the following Taylor expansions:

s ¡ ¢2 " ¡ ¢2 #
y − h/2 y − h/2
q
¢2
∼
¡ ¢ ¡
d1 y = y − h/2 + D 2 = D 1 + = D 1+ + · · · (8.26)
D2 2D 2
and
s ¡ ¢2 " ¡ ¢2 #
y + h/2 y + h/2
q
¢2
∼
¡ ¢ ¡
d2 y = y + h/2 + D2 =D 1+ = D 1+ + · · · (8.27)
D2 2D 2
These approximations are valid as long as D À y and D À h.

We next consider how to modify (8.25) so that it applies to the case when
the two slits are illuminated by a collection of point sources distributed over a
finite lateral extent. This situation is depicted in Fig. 8.7 and it leads to partial
spatial coherence if the phase of each point emitter fluctuates randomly.9 When
a Young’s two-slit setup is illuminated by an extended random source, the wave
fronts at the two slits are less correlated. This makes the fringes move around on
the screen rapidly and partially ‘wash out’ when time averaged, meaning worse
fringe visibility.
To simplify our analysis, we restrict the distribution of point sources to vary
only in the y 0 dimension.10 We assume that the light is quasi-monochromatic so
that its frequency is approximately ω with a phase that fluctuates randomly over
time intervals much longer than the period of oscillation 2π/ω.11
The light emerging from the j th point at y 0j travels by means of two very
narrow slits to a point y on a screen. Let E1 (y 0j ) and E2 (y 0j ) be the fields on the
9 A laser beam does not fit the definition of a light source with randomly varying spatial phase.
Instead, in this section we consider a source such as the surface of a star (filtered to a narrow
frequency range). See appendix 8.B for more discussion.
10 The results can be generalized to a two-dimensional source.
11 Random phase fluctuations necessarily imply some frequency bandwidth, however small.
Hence the need to specify quasi-monochromatic light.

screen at y, both originating from the point y 0j , but traveling respectively through
the two different slits. We assume that these fields have the same polarization,
and we will suppress the vectorial nature of the fields. For simplicity, we assume
the two fields have the same (real) amplitude at the screen E 0 (y 0j ). Thus, we write
the two fields as
n h i o
0 0
0 0 i k r 1 (y j )+d 1 (y) −ωt +φ(y j )
E 1 (y j ) = E 0 (y j )e (8.28)
and n h i o
i k r 2 (y 0j )+d 2 (y) −ωt +φ(y 0j )
E 2 (y 0j ) = E 0 (y 0j )e (8.29)
We have explicitly included an arbitrary phase φ(y 0j ), which we will take to be
different for each point source.
We now set about finding the cumulative field at y arising from the many
points indexed by the subscript j . The total field on the screen at point y is
Xh i
E tot (h) = E 1 (y 0j ) + E 2 (y 0j ) (8.30)
j
Obviously, in addition to h, the total field depends on y, R, D, and k as well as on

the phase φ(y 0j ) at each point. Nevertheless, in the end we will mainly emphasize
the dependence on the slit separation h. The intensity associated with (8.30) is
²0 c
I tot (h) = |E tot (h)|2
2 " #· ¸∗
²0 c X 0 0
X 0 0
Thomas Young (17731829, English) = E 1 (y j ) + E 2 (y j ) E 1 (y m ) + E 2 (y m )
was born in Milverton, Somerset, and
2 j m
was the oldest of ten children. By age ²0 c Xh i
fourteen, he had become procient at a = E 1 (y 0j )E 1∗ (y m
0
) + E 2 (y 0j )E 2∗ (y m
0
) + E 1 (y 0j )E 2∗ (y m
0
) + E 2 (y 0j )E 1∗ (y m
0
)
dozen dierent languages. As a young
2 j ,m
adult, he studied medicine and then
²0 c X ¯¯
· ³ ´ ³ ´
¯¯
0 ¯¯ 0 ¯
¯ i k r 1 (y 0 )−r 1 (y m0 ) i k r 2 (y 0j )−r 2 (y m
0
)
went to Göttingen, Germany where he = ¯E 0 (y j )¯ E 0 (y m ) e j
+e
earned a doctoral degree in physics. In 2 j ,m
1801, he was appointed professor of ½ ³ ´ ¾¸ ³ ´
natural philosophy at the Royal Insti- i k r 1 (y 0j )−r 2 (y m
0
) i k (d 1 (y)−d 2 (y)) i φ(y 0j )−φ(y m
0
)
tute, but he also maintained an active
+2Re e e e
medical practice on the side. He con-
tributed to a wide variety of elds and
(8.31)
helped to decipher ancient Egyptian hi- At this juncture we make a critical assumption: that the phase of the emission
eroglyphs, including the Rosetta Stone.
φ(y 0j ) varies in time independently at every point on the source. This is sometimes
He published descriptions of the heart
and arteries as well as how the eye ac- called the stochastic assumption, and it is appropriate for the emission from ther-
commodates to see at dierent depths mal sources such as starlight, a glowing filament (filtered to a narrow frequency
and how the eye perceives color. In en-
gineering elds, Young is well known
range), or spontaneous emission from an excited gas or plasma. However, it is
his analysis of stresses and strains in not appropriate for coherent sources like lasers.
elastic media. Young's double-slit exper-
iment gave convincing evidence of the
A wonderful simplification happens to (8.31) when the phase difference
wave nature of light, overturning New- φ(y 0j )−φ(y m0
) varies randomly. If j 6= m, then exp{i (φ(y 0j ) − φ(y m
0
))} time-averages
ton's corpusculor theory. Regarding this,
Thomas Young traded ideas with Au-
to zero. On the other hand, if j = m, then the factor reduces to e 0 = 1. Formally,
gustin Fresnel through correspondence. this is written
(Wikipedia)
¿ ³ ´À ½
i φ(y 0j )−φ(y m
0
) 1 if j = m,
e = δ j ,m ≡ (random phase assumption) (8.32)
t 0 if j 6= m.
8.5 Young’s Two-Slit Setup and Spatial Coherence 215
where δ j ,m is known as the Kronecker delta function. The time-averaged intensity

under the stochastic assumption (8.32) then reduces to
( ³ ´ )
0 0
0 0 0 i k r 1 (y j )−r 2 (y j ) i k (d 1 (y)−d 2 (y))
X X X
〈I (h)〉 = I (y ) + I (y ) + 2Re
tot t j j I (y )e j e
j j j
(8.33)
We may use (8.26) to simplify d 1 (y) − d 2 (y) = ∼ h y/D. Very similarly, we may also
write r 1 (y 0j )−r 2 (y 0j ) ∼
= h y 0j /R. The only thing left to do is to put (8.33) into a slightly
more familiar form:
" #
0
〈I tot (h)〉t = 2 I (y j ) 1 + Re γ (h)
X £ © ª¤
(random phase assumption) (8.34)
j
We have introduced
kh y 0
kh y j
e −i I (y 0j )e −i
P
D R
j
γ (h) ≡ (8.35)
I (y 0j )
P
j
which is known as the degree of coherence. It controls the fringe pattern seen at
the screen.
We can generalize (8.34) so that it applies to the case of a continuous distribu-
tion of light as opposed to a collection of discrete point sources. In Appendix 8.A
we show how summations in (8.34) and (8.35) become integrals over the source
intensity distribution, and we write
〈I net (h)〉t = 2 〈I oneslit 〉t 1 + Re γ(h)

£ © ª¤
(random phase assumption) (8.36)
where
kh y R∞ kh y 0
e −i D I (y 0 )e −i R d y0
−∞
γ (h) ≡ (8.37)
R∞
I (y 0 )d y 0
−∞
Here I (y 0 ) has units of intensity per length.

¡ ¢
The factor exp −i kh y/D defines the positions of the periodic fringes on
the screen. The remainder of (8.37) controls the depth of the fringes as the slit
separation h is varied. When the slit separation h increases, the amplitude of γ (h)
tends to diminish until the intensity at the screen becomes uniform. When the
kh y 0
two slits have very small separation (such that e −i R ∼ = 1) then we have ¯γ (h)¯ = 1
¯ ¯
and very good fringe visibility results. γ (h) dictates the degree of spatial coherence
in much the same way that γ (τ) dictates the degree of temporal coherence. Notice
the close similarity between (8.37) and (8.10).
As the slit separation h increases, the fringe visibility
V (h) = ¯γ (h)¯
¯ ¯
(8.38)
diminishes, eventually approaching zero (see (8.16)). In analogy to the temporal

case (see (8.13)), we can define a slit separation sufficiently large to make the
fringes at the screen ‘wash out’:
Z∞
¯2
h c ≡ 2 ¯γ (h)¯ d h
¯
(8.39)
0
Appendix 8.A Spatial Coherence for a Continuous Source

In this appendix we examine the spatial coherence of light from a continuous spa-
tial distribution (as opposed to a collection of discrete point sources) and justify
(8.37) and (8.38). We begin by replacing the summations in (8.31) with integrals
over a continuous emission source. We make the following replacements:
Z∞ Z∞
E 1 (y 0j ) → E 1 (y 0 )d y 0 0
E 1 (y 00 )d y 00
X X
and E 1 (y m ) →
j m
−∞ −∞
(8.40)
Z∞ Z∞
E 2 (y 0j ) → E 2 (y 0 )d y 0 0
E 2 (y 00 )d y 00
X X
and E 2 (y m ) →
j m
−∞ −∞
Rather than deal with a time average of randomly varying phases, we will instead
work with a linear superposition of all conceivable phase factors. That is, we will
write the phase φ(y 0 ) as K y 0 , where K is a parameter with units of inverse length,
which we allow to take on all possible real values with uniform likelihood. The
way we modify (8.32) for the continuous case is then
¿ h iÀ Z∞
i φ(y 0j )−φ(y m
0
) 1 0 00
e = δ j ,m → e i K (y −y ) d K = δ(y 00 − y 0 ) (8.41)
t 2π
−∞
With the replacements in (8.40) and (8.41), (8.31) becomes

Z∞ Z∞
²0 c ¯h 0 00 0 00
d y E (y 0 )¯
0¯
d y 00 ¯E (y 00 )¯ e i k (r 1 (y )−r 1 (y )) + e i k (r 2 (y )−r 2 (y ))
¯ ¯ ¯
I tot (h) =
2
−∞ −∞
n 0 00
oi
+2Re e i k (d1 (y)−d2 (y)) e i k (r 1 (y )−r 2 (y )) δ(y 00 − y 0 )
(8.42)
∼
Again, consistent with (8.26), we may write d 1 (y) − d 2 (y) = h y/D and r 1 (y 0 ) −
r 2 (y 0 ) ∼
= h y 0 /R, and (8.42) reduces to
Z∞ Z∞
 
kh y kh y 0
 
I tot (h) = 2 I (y 0 )d y 0 + 2Re e −i D I (y 0 )e −i R d y 0 (8.43)
 
−∞ −∞
where
1 ¯2
I (y 0 ) ≡ ²0 c ¯E (y 0 )¯
¯
(8.44)
2
8.B Van Cittert-Zernike Theorem 217
For I tot to have normal units of intensity, I (y 0 ) must have units of intensity per
length of source, implying that E (y 0 ) has units of field per square root of length.
R∞
Hence, −∞ I (y 0 )d y 0 is the intensity at the screen caused by the entire extended
source when only one slit is open. We see that (8.43) is equivalent to (8.36) and
(8.37).
Appendix 8.B Van Cittert-Zernike Theorem

In this appendix we avoid making the assumption of randomly varying phase.
This would be the case when the source of light is, for example, a laser. In place of
(8.42) we have
¯ ∞ ¯2 ¯ ∞ ¯2
k y 02 kh y 0 k y 02 kh y 0
¯Z · ¸ ¯ ¯Z · ¸ ¯
0 ¯ i φ( y 0 )+i 2R −i 2R 0¯ 0 ¯ i φ(y 0 )+i 2R i 2R 0¯
¯ ¯ ¯ ¯ ¯ ¯ ¯ ¯
I tot (h) ∝ ¯
¯ ¯E (y ) e e dy ¯ +¯ ¯ ¯ E (y ) e e dy ¯
−∞ −∞
¯ ¯ ¯ ¯
 ∞  ∞ ∗
k y 02 kh y 0 k y 02 kh y 0
kh y 
Z · ¸   Z ·¯ ¡ ¢¯ ¸
0 0
¯E y 0 ¯ e i φ(y )+i 2R e −i 2R d y 0 ¯E y 0 ¯ e i φ( y )+i 2R e i 2R d y 0

+ 2Ree i D
¯ ¡ ¢¯
  
−∞ −∞
(8.45)
where we have employed (8.26) and (8.27) and similar expressions involving R
and y 0 .
The first term on the right-hand side of (8.45) is the intensity on the screen
when the lower slit is covered. The second term is the intensity on the screen
when the upper slit is covered. The last term is the interference term, which
modifies the sum of the individual intensities when both slits are uncovered.
Notice the occurrence of Fourier transforms (over position) on the quantities
inside of the square brackets. Later, when we study diffraction theory, we will
recognize these transforms as determining the strength of fields impinging on
the individual slits. This corresponds to a major difference between a spatially
coherent source and a random-phase source. With the random-phase source, the
slits are always illuminated with the same strength regardless of the separation.
However, with a coherent source, ‘beaming’ can occur such that the strength as
well as phase of the field at each slit depends on the slit separation.
A beautiful simplification occurs when the phase of the emitted light has the
following distribution:
k y 02
φ(y 0 ) = − (8.46) (converging spherical wave)
2R
Equation (8.46) is not as arbitrary as it may first appear. This particular phase
is an approximation to a concave spherical wave front converging to the center
between the two slits. This type of wave front is created when a plane wave passes
through a lens. With the special phase (8.46), the intensity (8.45) reduces to
¯ ∞ ¯2 ¯ ∞ ¯2
¯Z ¯ ¯Z ¯
kh y 0 kh y 0
I tot (h) ∝ ¯¯ ¯E (y 0 )¯ e −i 2R d y 0 ¯¯ + ¯¯ ¯E y 0 ¯ e i 2R d y 0 ¯¯
¯ ¯ ¯ ¯ ¯ ¯ ¡ ¢¯ ¯
¯ ¯ ¯ ¯
−∞ −∞
(converging spherical wave)  ∞ 2 (8.47)
kh y 0
kh y
Z ¯ 
i D ¯E (y 0 )¯ e −i 2R d y 0
¯
+ 2Ree
 
−∞
There is a close resemblance between the expression in the first term

¯ ∞ ¯
¯Z ¯
kh y 0
¯ ¯E (y 0 )¯ e −i 2R d y 0 ¯
¯ ¯ ¯ ¯
¯ ¯ (8.48)
¯ ¯
−∞
and the magnitude of the degree of coherence V = ¯γ (h/2)¯ from (8.37). Again,
¯ ¯
this corresponds to the field that goes through the upper slit, when it is positioned
at h/2, and which impinges on the screen. Let this field be denoted by |E 1 (h/2)|.
The field strength when the single slit is positioned at h compared to that when it
is positioned at zero is
¯ ∞ ¯
¯ R ¯ kh y 0 ¯
¯E (y 0 )¯ e −i R d y 0 ¯
¯
¯ ¯ ¯¯
¯ E 1 (h) ¯ ¯−∞ ¯
(converging spherical wave ¯=¯ (8.49)
¯ ¯
R∞ ¯
¯ E (0) ¯ ¯ ¯
1
¯
¯E (y 0 )¯ d y 0
¯
assumption) ¯
¯ −∞
¯
¯
This looks very much like ¯γ (h)¯ of (8.37) except that the magnitude of the field
¯ ¯
appears in (8.49), whereas the intensity appears in (8.37).

This may seem rather contrived, but at least it is cute, and it is known as the
van Cittert-Zernike theorem.12 It says that the spatial coherence of an extended
source with randomly varying phase drops off with lateral slit separation in the
same way that the field pattern at the focus of a converging spherical wave would
drop off, whose field amplitude distribution is the same as the original intensity
distribution.
Exercises 219
Exercises
Exercises for 8.2 Coherence Time and Fringe Visibility
P8.1 (a) Verify that (8.16) gives the fringe visibility.

HINT: Write γ = ¯γ¯ e i φ and assume that the oscillations ¯in¯ γ that give
¯ ¯
rise to fringes are due entirely to changes in φ and that ¯γ¯ is a slowly
varying function in comparison to the oscillations.
(b) What is the coherence time τc of the light in P8.4?
P8.2 (a) Show that the fringe visibility of a Gaussian spectral distribution
(see Example 8.2) goes from 1 to e −π/2 = 0.21 as the round-trip path in
one arm of the instrument is extended by a coherence length.
(b) Find the FWHM bandwidth in wavelength ∆λFWHM in terms of the
coherence length `c and the center wavelength λ0 .
HINT: First determine ∆ωFWHM , defined to be the width of I (ω) at half
of its peak. To convert to a wavelength difference, use ω = 2πc λ ⇒
∼ 2πc
∆ωFWHM = − 2 ∆λFWHM . You can ignore the minus sign; it simply
λ0
means that wavelength decreases as frequency increases.
Exercises for 8.3 Temporal Coherence of Continuous Sources
P8.3 Show that Re{γ(τ)} defined in (8.10) reduces to cos (ω0 τ) in the case of
a plane wave E (t ) = E 0 e i (k0 z−ω0 t ) being sent through a Michelson inter-
ferometer. In other words, the output intensity from the interferometer
reduces to
I = 2I 0 [1 + cos (ω0 τ)]
as you already expect.
HINT: Don’t be afraid of delta functions. After integration, the left-over
delta functions cancel.
P8.4 Light emerging from a dense hot gas has a collisionally broadened
power spectrum described by the Lorentzian function
I (ω0 )
I (ω) = ´2
ω−ω0
³
1+ ∆ωFWHM /2
The light is sent into a Michelson interferometer. Make a graph of the

average power arriving to the detector as a function of τ.
HINT: See (0.56).
P8.5 Consider the light source described in P8.4

(a) Regardless of how the phase of E (ω) is organized, the oscillation
of the energy arriving to the detector as a function of τ is the same.
The spectral phase of the light in P8.4 is randomly organized. Describe
qualitatively how the light probably looks as a function of time.
(b) Now suppose that the phase of the light is somehow neatly orga-
nized such that ω
i E (ω0 ) e i c z
E (ω) =
i + ∆ωω−ω 0
FWHM /2
Perform the inverse Fourier transform on the field and find how the
intensity of the light looks a function of time.
HINT:
Z∞
e −i ax −2i πe i aβ
½
if a>0 ¡ ¢
dx = Imβ > 0
x +β 0 if a<0
−∞
The constants I (ω0 ), and ∆ωFWHM will appear in the answer.
Exercises for 8.4 Fourier Spectroscopy
L8.6 (a) Use a scanning Michelson interferometer to measure the wave-

length of the ultrashort laser pulses from a mode-locked Ti:sapphire
oscillator.13
(b) Measure the coherence length of the source by observing the dis-
tance over which the visibility diminishes. From your measurement,
Beam what is the bandwidth ∆λFWHM of the source, assuming the Gaussian
Splitter profile in the previous problem? See P8.2.
(c) Use a computer to perform a fast Fourier transform (FFT) of the
signal output. For the positive frequencies, plot the laser spectrum as a
function of λ and compare with the results of (a) and (b).
Detector (d) How do the results change if the ultrashort pulses are first stretched
in time by traversing a thick piece of glass?
Figure 8.8
Exercises for 8.5 Young’s Two-Slit Setup and Spatial Coherence
P8.7 (a) A point source with wavelength λ = 500 nm illuminates two parallel
slits separated by h = 1.0 mm. If the screen is D = 2 m away, what is
the separation between the diffraction peaks on the screen? Make a
sketch.
13 J. Peatross and S. Bergeson, “Fourier Spectroscopy of Ultrashort Laser Pulses,” Am. J. Phys. 74,
842-845 (2006).
Exercises 221
(b) A thin piece of glass with thickness d = 0.01 mm and index n = 1.5 is
placed in front of one of the slits. By how many fringes does the pattern
at the screen move?
HINT: Add ∆φ to k (d 2 − d 1 ) in (8.25) , where ∆φ ≡ φ2 −φ1 is the relative
phase between the two paths. Compare the phase of the light when
traversing the glass versus traversing an empty region of the same
thickness.
L8.8 (a) Carefully measure the separation of a double slit in the lab (h ∼
0.1 mm separation) by shining a HeNe laser (λ = 633 nm) through it
and measuring the diffraction peak separations on a distant wall (say,
2 m from the slits).
HINT: For better accuracy, measure across several fringes and divide.
Double slit
Single slit separation h
Diffuser
width a Filter
Laser
CCD
Camera
Rotating diffuser
to create phase
variation
Figure 8.9
(b) Create an extended light source with a HeNe laser using a time-
varying diffuser followed by an adjustable single slit. (The diffuser
must rotate rapidly to create random time variation of the phase at
each point as would occur automatically for a natural source such
as a star.) Place the double slit at a distance of R ≈ 100 cm after the
first slit. (Take note of the exact value of R, as you will need it for the
next problem.) Use a lens to image the diffraction pattern that would
have appeared on a far-away screen into a video camera. Observe
the visibility of the fringes. Adjust the width of the source with the
single slit until the visibility of the fringes disappears. After making the
source wide enough to cause the fringe pattern to degrade, measure
the single slit width a by shining a HeNe laser through it and observing
the diffraction pattern on the distant wall. (video)
HINT: As we will study later, a single slit of width a produces an inten-
sity pattern on a screen a distance L away described by
³ πa ´
I (x) = I peak sinc2 x
λL
sin α sin α
where sinc (α) ≡ α and lim α = 1.
α→0
NOTE: It would have been nicer to vary the separation of the two slits
to determine the width of a fixed source. However, because it is hard to
make an adjustable double slit, we varied the size of the source until
the spatial coherence of the light matched the slit separation.
P8.9 (a) Compute h c for a uniform intensity distribution of width a using

(8.39).
(b) Use this formula to check that your measurements in L8.8 agree
with spatial coherence theory.
HINT: In your experiment h c is the double slit separation. Use your
measured R and h to calculate what the width of the single slit (i.e.
a) should have been when the fringes disappeared and compare this
calculation to your direct measurement of a.
Solution: (This is only a partial solution)
a/2
y0

y −i kh
a/2 y a/2 y0
e −i kh D  e
R
y0 y
h ³ í
d y0 e −i kh D e −i kh R d y0
R R
I 0 exp −i kh R + D

−i kh
R
−a/2 −a/2 −a/2
γ (h) = = =
a/2 a a
I0d y 0
R
−a/2
a/2 −a/2
 
−i kh
y
e
R − e −i kh R y
 = e −i kh D sinc kha
= e −i kh D
−2i kh a/2
R
2R
Note that
Z∞
sin2 αx π
dx =
(αx)2 2α
0
R29 T or F: In our notation (widely used), I (t ) is the Fourier transform of

I (ω).
R30 T or F: The integral of I (t ) over all t equals the integral of I (ω) over all
ω.
R31 T or F: The phase velocity of light (the speed of an individual frequency

component of the field) never exceeds the speed of light c.
R32 T or F: The group velocity of light in a homogeneous material can

exceed c if absorption or amplification takes place.
R33 T or F: The group velocity of light never exceeds the phase velocity.
R34 T or F: A Michelson interferometer can be used to measure the spectral

intensity of light I (ω).
R35 T or F: A Michelson interferometer can be used to measure the duration

of a short laser pulse and thereby characterize its chirp.
R36 T or F: A Michelson interferometer can be used to measure the wave-

length of light.
R37 T or F: A Michelson interferometer can be used to measure the phase

of E (ω).
R38 T or F: The Fourier transform (or inverse Fourier transform if you prefer)
of I (ω) is proportional to the degree of temporal coherence.
R39 T or F: A Michelson interferometer is ideal for measuring the spatial

coherence of light.
R40 T or F: The Young’s two-slit setup is ideal for measuring the temporal
coherence of light.
223
R41 T or F: Vertically polarized light illuminates a Young’s double-slit setup

and fringes are seen on a distant screen with good visibility. A half wave
plate is placed in front of one of the slits so that the polarization for that
slit becomes horizontally polarized. Here’s the statement: The fringes
at the screen will shift position but maintain their good visibility.
Problems
Horizontal Vertical
Polarizer Polarizer R42 (a) Horizontally polarized light enters a system and first travels through
a horizontal and then a vertical polarizer in series. What is the Jones
vector of the transmitted field?
(b) Now a polarizer at 45◦ is inserted between the two polarizers in the
system described in (a). What is the Jones vector of the transmitted
field? How does the final intensity compare to initial intensity?
(c) Now a quarter wave plate with a fast-axis angle at 45◦ is inserted
between the two polarizers (instead of the polarizer of part (b)). What
Figure 8.10 is the Jones vector of the transmitted field? How does the final intensity
compare to initial intensity?
R43 (a) Find the Jones matrix for half wave plate with its fast axis making an
arbitrary angle θ with the x-axis.
HINT: Project an arbitrary polarization with E x and E y onto the fast
and slow axes of the wave plate. Shift the slow axis phase by π, and then
project the field components back onto the horizontal and vertical axes.
The answer is
cos2 θ − sin2 θ 2 sin θ cos θ
· ¸
2 sin θ cos θ sin2 θ − cos2 θ
(b) We desire to create a variable attenuator for a polarized laser beam

y-axis
using a half wave plate and a polarizer aligned to the initial polarization
Fast axis of the beam (see figure). The fast axis of the half wave plate is initially
Transmission Axis aligned in the direction of polarization and then rotated through an
x-axis
angle θ. What is the ratio of the intensity exiting the polarizer to the
incoming intensity as a function of θ?
R44 (a) What is the spectral content (i.e., I (ω)) of a square laser pulse
E 0 e −i ω0 t , |t | ≤ τ/2
½
Figure 8.11 Polarizing Elements E (t ) =
0 , |t | > τ/2
Make a sketch of I (ω), indicating the location of the first zeros.

(b) What is the temporal shape (i.e., I (t )) of a light pulse with frequency
content
E 0 , |ω − ω0 | ≤ ∆ω/2
½
E (ω) =
0 , |ω − ω0 | > ∆ω/2
225
where in this case E 0 has units of E-field per frequency. Make a sketch
of I (t ), indicating the location of the first zeros.
(c) If E (ω) is known (any arbitrary function, not the same as above), and
the light goes through a material of thickness ` and index of refraction
n (ω), how would you find the form of the pulse E (t ) after passing
through the material? Please set up the integral.
R45 (a) Prove Parseval’s theorem:

Z∞ Z∞
2
|E (ω)| d ω = |E (t )|2 d t .
−∞ −∞
HINT:
Z∞
1
e i ω(t −t ) d ω
0
0
δ t −t =
¡ ¢
2π
−∞
(b) Explain the physical relevance of Parseval’s theorem to light pulses.

Suppose that you have a detector that measures the total energy in
a pulse of light, say 1 mJ directed onto an area of 1 mm2 . Next you
measure the spectrum of light and find it to have a width of ∆λ =
50 nm, centered at λ0 = 800 nm. Assume that the light has a Gaussian
frequency profile
¡ ω−ω0 ¢2
I (ω) = I (ω0 )e − δω
Use as an approximate value δω ∼

= 2πc
λ2
∆λ. Find a value and correct
units for I (ω0 ).
HINT:
Z∞
π B 2 /4A+C
r
−Ax 2 +B x+C
e dx = e Re {A} > 0
A
−∞
R46 Continuous light entering a Michelson interferometer has a spectrum

described by
I 0 , |ω − ω0 | ≤ ∆ω/2
½
I (ω) =
0 , |ω − ω0 | > ∆ω/2
The Michelson interferometer uses a 50:50 beam splitter. The emerging
light has intensity 〈I det (t , τ)〉t = 2 〈I (t )〉t 1 + Reγ (τ) , where degree of
£ ¤
coherence is
Z∞ , Z∞
γ(τ) = I (ω) e −i ωτ d ω I (ω)d ω
−∞ −∞
Find the fringe visibility V ≡ (I max − I min )/(I max + I min ) as a function of τ
(i.e. the round-trip delay due to moving one of the mirrors).
Extended Source
Fringe Pattern
R47 Light emerging from a point travels by means of two very narrow slits
to a point y on a screen. The intensity at the screen arising from a point
source at position y 0 is found to be
y y0
½ · µ ¶¸¾
¡ 0 ¢ 0
I screen y , h = 2I (y ) 1 + cos kh +
D R
where an approximation has restricted us to small angles.
(a) Now, suppose that I (y 0 ) characterizes emission from a wider source
with randomly varying phase across its width. Write down an expres-
sion (in integral form) for the resulting intensity at the screen:
Z∞
I screen y 0 , h d y 0
¡ ¢
I screen (h) ≡
−∞
(b) Assume that the source has an emission distribution with the form
02 02
I (y 0 ) = I 0 /∆y 0 e −y /∆y . What is the function γ(h) where the intensity
¡ ¢
p
is written I screen (h) = 2 πI 0 1 + Reγ(h) ?
£ ¤
HINT:
Z∞
π B 2 /4A+C
r
−Ax 2 +B x+C
e dx = e Re {A} > 0.
A
−∞
(c) As h varies, the intensity at a point on the screen y oscillates. As h

grows wider, the amplitude of oscillations decreases. How wide must
the slit separation h become (in terms of R, k, and ∆y 0 ) to reduce the
visibility to
I max − I min 1
V≡ =
I max + I min 3
Selected Answers
R42: (b) 1/4, (c) 1/2.

R45: (b) 3.8 × 10−16 J/ cm2 · s−1 .
¡ ¢
Chapter 9
Light as Rays
So far in our study of optics, we have described light in terms of waves, which sat-
isfy Maxwell’s equations. However, as you are probably aware, in many situations
light can be thought of as rays pointing along the direction of wave propagation. A
ray picture is useful when one is interested in the macroscopic flow of light energy,
but rays fail to reveal fine details, in particular wave and diffraction phenomena.
For example, simple ray theory suggests that a lens can focus light down to a point.
However, if a beam of light were concentrated onto a true point, the intensity
would be infinite! Nevertheless, ray theory is useful for predicting where a focus
occurs. It is also useful for describing imaging properties of optical systems (e.g.
lenses and mirrors).
Beginning in section 9.3 we study the details of ray theory and the imaging
properties of optical systems. First, however, we examine the justification for ray
theory starting from Maxwell’s equations. In the short-wavelength limit, Maxwell’s
equations give rise to the eikonal equation, which governs the direction of rays
in a medium with an index of refraction that varies with position. The German
word ‘eikonal’ comes from the Greek ‘²ικων’ from which the modern word ‘icon’
derives. The eikonal equation therefore has a descriptive title since it controls the
formation of images. Although we will not use the eikonal equation extensively,
we will show how it embodies the underlying justification for ray theory. As will be
apparent in its derivation, the eikonal equation relies on an approximation that
the features of interest in the light distribution are large relative to the wavelength
of the light.
The eikonal equation describes the direction of ray propagation, even in com-
plicated situations such as desert mirages where air is heated near the ground and
has a different index than the air farther from the ground. Rays of light from the
sky that initially are directed toward the ground can be bent such that they travel
parallel to or even up from the ground, owing to the inhomogeneous refractive
index. The eikonal equation can also be used to deduce Fermat’s principle, which
in short says that light travels from point A to point B following a path that takes
the minimum time. This principle can be used, for example, to‘derive’ Snell’s law.
Of course Fermat asserted his principle more than a century before Maxwell’s
227
228 Chapter 9 Light as Rays
equations were known, but it is nice to give justification retroactively to Fermat’s

principle using the modern perspective.
In this chapter, we will analyze the propagation of rays through optical systems
composed of lenses and/or curved mirrors in the context of paraxial ray theory.
The paraxial approximation restricts rays to travel nearly parallel to the axis of
such systems. We consider the effects of three basic optical elements acting on
paraxial rays: 1) Unobstructed propagation through a distance d in a uniform
medium; a ray may move farther away from (or closer to) the optical axis, as it
travels. 2) Reflection from a curved spherical mirror, which changes a ray’s angle
with respect to the optical axis. 3) Transmission through a spherical interface
between two materials with differing refractive indices. The effects of each of
these basic elements on a ray of light can be represented as a 2 × 2 matrix, which
can be multiplied together to construct more complex imaging systems (such as
a lens or a series of lenses and curved mirrors).
We will study image formation in the context of the paraxial approximation,
which in the case of a curved mirror or a thin lens gives rise to the familiar formula
1 1 1
= + (9.1)
f do di
Even a complicated multi-element optical system obeys (9.1) if d o and d i are

measured from principal planes rather than the single plane of, for example, a
thin lens.
Paraxial ray theory can also be used to study the stability of laser cavities. The
formalism predicts whether a ray, after many round trips in the cavity, remains
near the optical axis (trapped and therefore stable) or if it drifts endlessly away
from the axis of the cavity on successive round trips.
In appendix 9.A we address deviations from the paraxial ray theory known
as aberrations. We also comment on ray-tracing techniques, used for designing
optical systems that minimize such aberrations.
9.1 The Eikonal Equation

In this chapter, we consider light to consist of only a single frequency ω. The wave
equation (2.13) for a medium with a real index of refraction in this case may be
written as
n 2 (r) ω2
∇2 E(r, t ) + E (r, t ) = 0 (9.2)
c2
where we have already performed the time differentiation on the assumed time
dependence e −i ωt . Although in chapter 2 we considered solutions to the wave
equation in a homogeneous material, the wave equation remains valid when the
index of refraction varies throughout space (i.e. if n (r) is an arbitrary function
of r). In this case, the usual plane-wave solutions no longer satisfy the wave
equation.
9.1 The Eikonal Equation 229
As a trial solution for (9.2), we take
E(r, t ) = E0 (r) e i [kvac R(r)−ωt ] (9.3)
where
ω 2π
k vac = = (9.4)
c λvac
Here R (r) is a real scalar function (which depends on position) having the dimen-
sion of length. By taking R (r) to be real, we do not take into account possible
absorption or amplification in the medium. Even though the trial solution (9.3)
looks somewhat like a plane wave,1 the function R (r) accommodates wave fronts
that can be curved or distorted as depicted in Fig. 9.1. At any given instant t , the
phase of the curved surfaces described by R (r) = const ant can be interpreted
as wave fronts of the solution. The wave fronts travel in the direction for which
R (r) varies the fastest. This direction is is aligned with ∇R (r), which lies in the
direction perpendicular to surfaces of constant phase.
The substitution of the trial solution (9.3) into the wave equation (9.2) gives
1 h i
2
∇2 E0 (r) e i kvac R(r) + n 2 (r) E0 (r) e i kvac R(r) = 0 (9.5)
k vac Figure 9.1 Wave fronts (i.e. sur-
faces of constant phase given by
where we have divided each term by e −i ωt . R(r)) distributed throughout space
in the presence of a spatially inho-
mogeneous refractive index. The
Computing the Laplacian in (9.5) gradient of R gives the direction of
travel for a wavefront.
The gradient of the x component of the field is
h i
∇ E 0x (r) e i kvac R(r) = [∇E 0x (r)] e i kvac R(r) + i k vac E 0x (r) [∇R (r)] e i kvac R(r)
The Laplacian of the x component is

h i ©
∇ · ∇ E 0x (r) e i kvac R(r) = ∇2 E 0x (r) − k vac
2
E 0x (r) [∇R (r)] · [∇R (r)]
+i k vac E 0x (r) ∇ R (r) + 2i k vac [∇E 0x (r)] · [∇R (r)] e i kvac R(r)
£ 2 ¤ ª
Upon combining the result for each vector component of E0 (r), the required spatial
derivative can be written as
h i ¡
∇2 E0 (r) e i kvac R(r) = ∇2 E0 (r) − k vac
2
E0 (r) [∇R (r)] · [∇R (r)] + i k vac E0 (r) ∇2 R (r)
£ ¤
© £ ¤
+2i k vac x̂ [∇E 0x (r)] · [∇R (r)] + ŷ ∇E 0 y (r) · [∇R (r)]
+ ẑ [∇E 0z (r)] · [∇R (r)]}) e i kvac R(r)
1 If the index is spatially independent (i.e. n (r) → n), then (9.3) reduces to the usual plane-wave
solution of the wave equation. In this case, we have R (r) = k · r/k vac and the field amplitude
becomes constant (i.e. E0 (r) → E0 ).
After performing the Laplacian and after some rearranging, (9.5) becomes
∇2 E0 (r) i 2i
∇R(r) · ∇R(r) − n 2 (r) E0 (r) = ∇2 R (r) +
£ ¤
2
+ x̂∇E 0x (r) · ∇R (r)
k vac k vac k vac
2i £ £ ¤ ¤
+ ŷ ∇E 0 y (r) · ∇R (r) + ẑ∇E 0z (r) · ∇R (r)
k vac
(9.6)
Don’t be afraid; at this point we are ready to make an important approxima-
tion. We take the limit of a very short wavelength (i.e. 1/k vac = λvac /2π → 0), and
the entire right-hand side of (9.6) vanishes. (Thank goodness!) With it we lose
the effects of diffraction. We also lose surface reflections at abrupt index changes
unless specifically considered. This approximation works best in situations where
only macroscopic features are of concern.
Our wave equation has been simplified to
[∇R (r)] · [∇R (r)] = n 2 (r) (9.7)
Written another way, this equation is
∇R (r) = n (r) ŝ (r) (9.8)
where ŝ is a unit vector pointing in the direction ∇R (r), the direction normal to
wave front surfaces. Equation (9.8) is called the eikonal equation.2
y
Example 9.1
h
Suppose that a region of air above the desert
¡ ¢ on a phot day has an index of refraction
¡ ¢
that varies with height y according to n y = n 0 1 + y 2 /h 2 . Verify that R x, y =
h/2 n 0 x ± y 2 /2h is a solution to the eikonal equation. (See problem P9.1 for a more
¡ ¢
h/4 general solution.)

x
Solution: The gradient of our trial solution gives
Figure 9.2 Depiction of possible ¡ ¢ ¡ ¢
light ray paths in a region with ∇R x, y = n 0 x̂ ± ŷ y/h
varying index.
Substituting this into (9.7) gives
∇R · ∇R = n 0 x̂ ± ŷ y/h · n 0 x̂ ± ŷ y/h = n 02 1 + y 2 /h 2 = n 2 y
¡ ¢ ¡ ¢ ¡ ¢ ¡ ¢
which confirms that it is a solution. The direction of light propagation is

¡ ¢
¡ ¢ ∇R n 0 x̂ ± ŷ y/h x̂ ± ŷ y/h
ŝ y ≡ = p = p
|∇R| n 0 1 + y 2 /h 2 1 + y2
Computed at various heights, the direction for rays turns out to be
x̂ ± ŷ x̂ ± ŷ/2 x̂ ± ŷ/4
ŝ (h) = p ŝ (h/2) = p ŝ (h/4) = p
2 5/4 17/16
9.2 Fermat’s Principle 231
These are represented in Fig. 9.2. In a desert mirage, light from the sky can appear
to come from a lower position. We can determine a path for the rays by setting
d y/d x equal to the slope of ŝ:
dy y
= ⇒ y = y 0 e ±(x−x0 )/h
dx h
Under the assumption of an infinitely short wavelength, the Poynting vector

is directed along ŝ as demonstrated in P9.2. In other words, the direction of
ŝ specifies the direction of energy flow. The unit vector ŝ at each location in
space points perpendicular to the wave fronts and indicates the direction that the
waves travel as seen in Fig. 9.1. We refer to a collection of vectors ŝ distributed
throughout space as rays.
In retrospect, we might have jumped straight to (9.8) without going through
the above derivation. After all, we know that each part of a wave front advances
in the direction of its gradient ∇R (r) (i.e. in the direction that R (r) varies most
rapidly). We also know that each part of a wave front defined by R (r) = constant
travels at speed c/n (r). The slower a given part of the wave front advances, the
more rapidly R (r) changes with position r and the closer the contours of constant
phase. It follows that ∇R (r) must be proportional to n (r) since ∇R (r) denotes the
rate of change in R (r).
9.2 Fermat’s Principle

Pierre de Fermat (16011665, French)
As we have seen, the eikonal equation (9.8) governs the path that rays follow as was born in Beaumont-de-Lomagne,
France to a wealthy merchant family.
they traverse a region of space, where the index varies with position. Another way
He attended the University of Toulouse
of deducing the correct path of rays is via Fermat’s principle.3 Fermat’s principle before moving to Bordeaux in the late
1620s where Fermat distinguished him-
says that if a ray happens to travel through both points A and B, it will follow a
self as a mathematician. Fermat was
path between them that takes the least time. procient in many languages and went
on to obtain a law degree in 1631 from
the University of Orleans. He continued
his study of mathematics as a hobby
Derivation of Fermat’s Principle from the Eikonal Equation throughout his life. He corresponded
with a number of notable mathemati-
4
We begin by taking the curl of (9.8) to obtain cians, and through his letters made
notable contributions to analytic ge-
∇ × [n (r) ŝ (r)] = ∇ × [∇R (r)] = 0 (9.9) ometry, probability theory, and number
theory. He was often quite secretive
about the methods used to obtain his
This can be integrated over an open surface of area A to give
results. Mathematicians suspect that
Z I Fermat didn't actually prove his famous
∇ × [n (r) ŝ (r)] d a = n (r) ŝ (r) · d ` = 0 (9.10) last theorem, which was not able to be
veried until the 1990's. Fermat was
A C
the rst to assert that the path taken
by a beam of light is the one that can
where we have applied Stokes’ theorem (0.12) to convert the area integral into a be traveled in the least amount of time.
path integral around the perimeter contour C . (Wikipedia)
4 The curl of a gradient is identically zero for any function.
Equation (9.10) states that the integration of nŝ · d ` around a closed loop is always
zero. If we consider a closed loop comprised of a path from point A to point B and
then a different path from point back to point A again, the integrals for the two
legs always cancel, even while holding one path fixed while varying the other. This
means
ZB
nŝ · d ` is independent of path from A to B. (9.11)
A
Now consider a path from A to B that is parallel to ŝ, as depicted in Fig. 9.3. In
this case, the cosine in the dot product is always one. If we choose some other
path that connects A and B, the cosine associated with the dot product is less than
one at most points along that path, whereas the result of the integral is the same.
Therefore, if we artificially remove the dot product from the integral (i.e. exclude
the cosine factor), the result of the integral will exceed the true value unless the
path chosen follows the direction of ŝ (i.e. the path that corresponds to the one
that light rays actually follow).
In mathematical form, this argument can be expressed as
A ZB
B 
Z 
nŝ · d ` = min nd ` (9.12)
 
A A
The integral on the right is called the optical path length (OP L) between points A
and B:
ZB
B
B OP L| A ≡ nd ` (9.13)
A
Figure 9.3 A ray of light leaving The conclusion is that the true path that light follows between two points (i.e.
point A arriving at B. the one that stays parallel to ŝ) is the one with the shortest optical path length.
The index n may vary with position and therefore can be different for each of the
incremental distances d `.
Fermat’s principle is usually stated in terms of the time it takes light to travel
between points. The travel time ∆t depends not only on the path taken by the
light but also on the velocity of the light v (r), which varies spatially with the
refractive index:
ZB ZB
B d` d` OP L|BA
∆t | A = = = (9.14)
v(r) c/n(r) c
A A
To find the correct path for the light ray that leaves point A and crosses point
B, we need only minimize the optical path length between the two points. Mini-
mizing the optical path length is equivalent to minimizing the time of travel since
it differs from the time of travel only by the constant c. The optical path length
is not the actual distance that the light travels; it is proportional to the number
of wavelengths that fit into that distance (see (2.24)). Thus, as the wavelength
shortens due to a higher index of refraction, the optical path length increases.
The correct ray traveling from A to B does not necessarily follow a straight line
but can follow a complicated curve according to how the index varies.
9.2 Fermat’s Principle 233
An imaging situation occurs when many paths from point A to point B have
the same optical path length. An example of this occurs when a lens causes an
image to form. In this case all rays leaving point A (on an object) and traveling
A B
through the system to point B (on the image) experience equal optical path
lengths. This situation is depicted in Fig. 9.4. Note that while the rays traveling
through the center of the lens have a shorter geometric path length, they travel
through more material so that the optical path length is the same for all rays. Figure 9.4 Rays of light leaving
To summarize Fermat’s principle, of the many rays that might emanate from point A with the same optical path
a point A, the ray that crosses a second point B is the one that follows the shortest length to B.
optical path length. If many rays tie for having the shortest optical path, we
say that an image of point A forms at point B. It should be noted that Fermat’s
principle, as we have written it, does not work for anisotropic media such as
crystals where n depends on the direction of a ray as well as on its location (see
P9.4).
Example 9.2
Use Fermat’s principle to derive Snell’s law.
Solution: Consider the many rays of light that leave point A seen in Fig. 9.5. Only
one of the rays passes through point B. Within each medium we expect the light to
travel in a straight line since the index is uniform. However, at the boundary we
must allow for bending since the index changes.
The optical path length between points A and B may be written
q q
OP L = n i x i2 + y i2 + n t x t2 + y t2 (9.15)
We need to minimize this optical path length to find the correct one according to
Fermat’s principle. B
Since points A and B are fixed, we may regard x i and x t as constants. The distances
y i and y t are not constants although the combination
y tot = y i + y t (9.16)
is constant. Thus, we may rewrite (9.15) as A

q q
¢2
OP L y i = n i x i2 + y i2 + n t x t2 + y tot − y i
¡ ¢ ¡
(9.17)
Figure 9.5 Rays of light leaving
where everything on the right-hand side is constant except for y i . point A; not all of them will tra-
We now minimize the optical path length by taking the derivative and setting it verse point B.
equal to zero:
¡ ¢
d (OP L) yi − y tot − y i
= ni q + nt q ¢2 = 0 (9.18)
d yi x i2 + y i2 x t2 + y tot − y i
¡
Notice that
yi yt
sin θi = q and sin θt = q (9.19)
x i2 + y i2 x t2 + y t2
When these are substituted into (9.18) we obtain
n i sin θi = n t sin θt (9.20)
which is the familiar Snell’s law.
Example 9.3
Use Fermat’s principle to derive the equation of curvature for a reflective surface
that causes all rays leaving one point to image to another. Do the calculation in
two dimensions rather than in three.5
Solution: We adopt the convention that the origin is half way between the points,
which are separated by a distance 2a, as shown in Fig. 9.6. If the points are to
image to each other, Fermat’s principle requires that the total path length be a
constant; call it b. By inspection of the figure, we that path (which reflects once)
from one point to the other is
q q
(x + a)2 + y 2 + (x − a)2 + y 2 = b (9.21)
To get (9.21) into a more recognizable form, we isolate the first square root and
square both sides of the equation, which gives
q
(x + a)2 + y 2 = b 2 + (x − a)2 + y 2 − 2b (x − a)2 + y 2
After squaring the two binomial terms, some nice cancelations occur, and we get
Figure 9.6
q
4ax − b 2 = −2b (x − a)2 + y 2
which we square again to obtain
16a 2 x 2 − 4ab 2 x + b 4 = 4b 2 x 2 − 2ax + a 2 + y 2

¡ ¢
After some cancellations and regrouping this becomes
16a 2 − 4b 2 x 2 − 4b 2 y 2 = 4a 2 b 2 − b 4
¡ ¢
Finally, we divide both sides by the term on the right to obtain the (hopefully)
familiar form of an ellipse
x2 y2
³ 2´ + ³ 2 ´ =1 (9.22)
b b 2
4 4 − a
5 This configuration is used to direct flash lamp energy into a laser amplifier rod. One ‘point’ in
Fig. 9.6 represents the end of an amplifier rod while the other represents the end of a thin flash-lamp
tube.
9.3 Paraxial Rays and ABCD Matrices 235
9.3 Paraxial Rays and ABCD Matrices

We now turn our attention to the effects of curved mirrors and lenses on rays of
light. Keep in mind that when describing light as a collection of rays rather than
as waves, the results can only describe features that are macroscopic compared to
a wavelength. The rays of light at each location in space describe approximately
the direction of travel of the wave fronts at that location. Since the wavelength of
visible light is extraordinarily small compared to the macroscopic features that we
perceive in our day-to-day world, the ray approximation is often a very good one.
This is the reason that ray optics was developed long before light was understood
as a wave.
We consider ray theory within the paraxial approximation, meaning that
we restrict our attention to rays that are near and almost parallel to an optical
axis of a system, say the z-axis. It is within this approximation that the familiar
imaging properties of lenses occur. An image occurs when all rays from a point
on an object converge to a corresponding point on what is referred to as the
image. To the extent that the paraxial approximation is violated, the clarity of
an image can suffer, and we say that there are aberrations present. The field of
optical engineering is often concerned with the minimization aberrations in cases
where the paraxial approximation is not strictly followed. This is done so that, for
example, a camera can take pictures of objects that occupy a fairly wide angular
field of view, where rays violate the paraxial approximation. Optical systems are
typically engineered using the science of ray tracing, which is described briefly in
section 9.A.
As we develop paraxial ray theory, we should remember that rays impinging
on devices such as lenses or curved mirrors should strike the optical component
at near normal incidence. To quantify this statement, the paraxial approximation
is valid to the extent that
sin θ ∼
=θ (9.23)
is a good approximation, and similarly
tan θ ∼
=θ (9.24)
Here, the angle θ (in radians) represents the angle that a particular ray makes
with respect to the optical axis. There is an important mathematical reason for
this approximation. The sine is a nonlinear function, but at small angles it is
approximately linear and can be represented by its argument. It is this linearity
that is crucial to the process of forming images. The linearity also greatly simplifies
the formulation since it reduces the problem to linear algebra. Conveniently, we
will be able to keep track of imaging effects with a 2×2 matrix formalism.
Consider a ray propagating in the y–z plane where the optical axis is in the z- Figure 9.7 The behavior of a ray as
direction. Let us specify a ray at position z 1 by two coordinates: the displacement light traverses a distance d .
from the axis y 1 and the orientation angle θ1 (see Fig. 9.7). If the index is uniform
everywhere, the ray travels along a straight path. It is straightforward to predict the
coordinates of the same ray down stream, say at z 2 . First, since the ray continues
in the same direction, we have
θ2 = θ1 (9.25)
By referring to Fig. 9.7 we can write y 2 in terms of y 1 and θ1 :
y 2 = y 1 + d tan θ1 (9.26)
where d ≡ z 2 − z 1 . Equation (9.26) is nonlinear in θ1 . However, in the paraxial

approximation (9.24) becomes linear, which after all is the point of the approxi-
mation. In this approximation the expression for y 2 simplifies to
y 2 = y 1 + θ1 d (9.27)
Equations (9.25) and (9.27) describe a linear transformation which in matrix

notation can be consolidated into the form
· ¸ · ¸· ¸
y2 1 d y1
ABCD matrix for propagation = (9.28)
θ2 0 1 θ1
through a distance d
Here, the vectors in this equation specify the essential information about the ray
before and after traversing the distance d , and the matrix describes the effect of
traversing the distance. This type of matrix is called an ABCD matrix;,6 sometimes
physicists are not very inventive with names.
Example 9.4
Let the distance d be subdivided into two distances, a and b, such that d = a +
b. Show that an application of the ABCD matrix for distance a followed by an
application of the ABCD matrix for b renders same result as an application of the
ABCD matrix for distance d .
Solution: Individually, the effects of propagation through a and through b are

· ¸ · ¸· ¸ · ¸ · ¸· ¸
y mid 1 a y1 y2 1 b y mid
= and = (9.29)
θmid 0 1 θ1 θ2 0 1 θmid
where the subscript “mid” refers to the ray in the middle position after traversing
the distance a. If we combine the equations, we get
· ¸ · ¸· ¸· ¸
y2 1 b 1 a y1
= (9.30)
θ2 0 1 0 1 θ1
which is in agreement with (9.28) since the ABCD matrix for the entire displace-
ment is · ¸ · ¸· ¸ · ¸
A B 1 b 1 a 1 a +b
= = (9.31)
C D 0 1 0 1 0 1
6 P. W. Milonni and J. H. Eberly, Lasers, Sect. 14.2 (New York: Wiley, 1988).
9.4 Reflection and Refraction at Curved Surfaces 237
9.4 Reflection and Refraction at Curved Surfaces

We next consider the effect of reflection from a spherical surface as depicted in
Fig. 9.8. We consider only the act of reflection without considering propagation
before or after the reflection takes place. Thus, the incident and reflected rays
in the figure are symbolic only of the direction of propagation before and after
reflection; they do not indicate any amount of travel. We immediately write
y2 = y1 (9.32)
since the ray has no chance to go anywhere.

We adopt the widely used convention that, upon reflection, the positive z-
direction is reoriented so that we consider the rays still to travel in the positive
z sense. An easy way to remember this is that the positive z direction is always
taken to be down stream of where the light is headed. Notice that in Fig. 9.8, the
reflected ray approaches the z-axis. In this case θ2 is a negative angle (as opposed
to θ1 which is drawn as a positive angle) and is equal to
θ2 = − (θ1 + 2θi ) (9.33)
where θi is the angle of incidence with respect to the normal to the spherical
mirror surface. By the law of reflection, the incident and reflected ray both occur
at an angle θi referenced to the surface normal. The surface normal points towards
the center of curvature of the mirror surface, which we assume is on the z-axis a
distance R away. By convention, the radius of curvature R is a positive number
if the mirror surface is concave and a negative number if the mirror surface is
convex.
Elimination of θi from (9.33) in favor of θ1 and y 1
By inspection of Fig. 9.8 we can write

y1
= sin φ ∼
=φ (9.34)
R
where we have applied the paraxial approximation (9.23). (The angles in Fig. 9.8
are exaggerated. In fact, when φ is small enough for (9.34) to hold, we may also
neglect the small distance δ.) By inspection of the geometry, we also have
φ = θ1 + θi (9.35)
Figure 9.8 A ray depicted in the
and when this is combined with (9.34), we get
act of reflection from a spherical
y1 surface.
θi = − θ1 (9.36)
R
With this we are able to put (9.33) into a useful linear form:
2
θ2 = − y 1 + θ1 (9.37)
R
Equations (9.32) and (9.37) describe a linear transformation that can be con-
cisely formulated as
· ¸ · ¸· ¸
y2 1 0 y1
ABCD matrix for a curved mirror = (9.38)
θ2 −2/R 1 θ1
The ABCD matrix in this transformation describes the act of reflection from a
concave mirror with radius of curvature R. The radius R is negative when the
mirror is convex.
The final basic element that we shall consider is a spherical interface between
two materials with indices n i and n t (see Fig. 9.9). This has an effect similar to
that of the curved mirror, which changes the direction of a ray without altering
its distance y 1 from the optical axis. Please note that here the radius of curvature
is considered to be positive for a convex surface (opposite convention from that
of the mirror). In this way, if the lower index is on the left, a positive radius R for
either the interface or the mirror tends to deflect rays towards the axis. Again, we
are interested only in the act of transmission without any travel before or after
the interface. As before, (9.32) applies (i.e. y 2 = y 1 ).
At the interface, the rays obey Snell’s obeys, which in the paraxial approxima-
tions is written
n i θi = n t θt (9.39)
The angles θi and θt are referenced from the surface normal, as seen in Fig. 9.9.
Substituting θ1 , θ2 and y 1 into Snell’s Law
Figure 9.9 A ray depicted in the By inspection of Fig. 9.9, we have

act of transmission at a curved
material interface. θ i = θ1 + φ (9.40)
and
θt = θ2 + φ (9.41)
where φ is the angle that the surface normal makes with the z-axis. As before (see
(9.34)), within the paraxial approximation we may write
φ∼
= y 1 /R
When this is used in (9.40) and (9.41), which are substituted into (9.39), Snell’s law
becomes
ni y 1 ni
µ ¶
θ2 = −1 + θ1 (9.42)
nt R nt
The compact matrix form of (9.32) and (9.42) is written

· ¸ · ¸· ¸
y2 1 0 y1
ABCD matrix for a curved = (9.43)
θ2 (n i /n t − 1) /R n i /n t θ1
interface
9.5 ABCD Matrices for Combined Optical Elements 239
9.5 ABCD Matrices for Combined Optical Elements

To summarize the previous two sections, we have developed ABCD matrices for Distance within a material, ex-
three basic elements: 1) propagation through a region of uniform index (9.28), cluding interfaces
2) reflection from a curved mirror (9.38), and 3) transmission through a curved ·
1 d
¸
interface between regions with different indices (9.43). All other ABCD matrices 0 1
that we will use are composites of these three. For example, one can construct the
Window, starting and stopping in
ABCD matrix for a lens by using two matrices like those in (9.43) to represent the air
entering and exiting surfaces of the lens. A distance matrix (9.28) can be inserted
· ¸
1 d /n
to account for the thickness of the lens. It is left as an exercise to derive the ABCD 0 1
matrix for a thick lens (seeP9.6). Thin lens or Mirror

· ¸
1± 0
−1 f 1
Example 9.5 ³ ´
Thin Lens: 1f = (n − 1) R1 − R1
Derive the ABCD matrix for a thin lens, where the thickness between the two lens 1 2
surfaces is ignored. (See P 9.6 for the more general case of a thick lens.) Mirror: 1f = R2
Thick lens
Solution: A thin lens is depicted in Fig. 9.10. R 1 is the radius of curvature for the ³ ´
1+ d n 1 −1 d
 
R1 n
first surface (which is positive if convex as drawn), and R 2 is the radius of curvature  µ ¶ ³ ´ ³

´
(1−n) R1 − R1 + R dR 2− n 1 −n 1− d 1 −1

for the second surface (which is negative as drawn). For either surface, the radius 1 2 1 2 R2 n
of curvature is considered to be positive if the surface is convex from the perspective

of rays that encounter it. Table 9.1 Summary of ABCD matri-
We take the index outside of the lens to be unity while that of the lens material to ces for common optical elements.
be n. We apply the ABCD matrix (9.43) in sequence, once for entering the lens and
once for exiting:
· ¸ · ¸· ¸
A B 1 0 ¡ 11 0
= 1 1
¢ 1
C D R2 (n − 1) n R1 n −1 n
" # (9.44) ABCD matrix for a thin lens
1³ 0
= ´
− (n − 1) R11 − R12 1
The matrix for the first interface is written on the right, where it operates first on
an incoming ray vector. In this case, n i = 1 and n t = n. The matrix for the second
surface is written on the left so that it operates afterwards. For the second surface,
n i = n and n t = 1.
Notice the close similarity between the ABCD matrix for a thin lens (9.44) and
the ABCD matrix for a curved mirror (9.38). The ABCD matrix for either the thin
lens or the mirror can be written as
· ¸ · ¸
A B 1 0
= (9.45)
C D −1/ f 1 Figure 9.10 Thin lens.
where in the case of the thin lens the focal length is given by the lens maker’s
formula µ ¶
1 1 1
= (n − 1) − (focal length of thin lens) (9.46)
f R1 R2
and in the case of a curved mirror, the focal length is
f = R/2 (focal length for a curved mirror) (9.47)
Table 9.1 is a summary of ABCD matrices of common optical elements.
Example 9.6
Derive the ABCD matrix for a window with thickness d and index n.
Solution: We can again take advantage of the ABCD matrix for a curved interface
(9.43), only in this problem we will let R 1 = ∞ and R 2 = ∞ to provide flat surfaces.
We take the index outside of the window to be unity and the index inside the
window to be n. We use the ABCD matrix (9.43) twice, once for each interface,
sandwiching matrix (9.31), which endows the window with thickness:
· ¸ · ¸· ¸· ¸
A B 1 0 1 d 1 0
= 1
C D 0 n 0 1 0 n
· ¸ (9.48)
1 d /n
= (window)
0 1
As far as rays are concerned, a window is effectively shorter to traverse than free
space.7 Fig. 9.11 illustrates why this is the case. The displacement of the exiting ray
is not as great as it would have been without the window. The window impedes
Figure 9.11 Window.
the rate at which the ray can move away from or toward the optical axis.
Example 9.7
h i
y
h i
y
Find ray θ22 that results when θ11 propagates through a distance a, reflects from
a mirror of radius R, and then propagates through a distance b. See Fig. 9.12.
Solution: The final ray in terms of the initial one is computed as follows:
· ¸ · ¸· ¸· ¸· ¸
y2 1 b 1 0 1 a y1
=
θ2 0 1 −2/R 1 0 1 θ1
· ¸· ¸
1 − 2b/R a + b − 2ab/R y1
= (9.49)
−2/R 1 − 2a/R θ1
(1 − 2b/R) y 1 + (a + b − 2ab/R) θ1
· ¸
=
(−2/R) y 1 + (1 − 2a/R) θ1
As always, the ordering of the matrices is important. The first effect that the ray
experiences is represented
h i by the matrix on the right, which is in the position that
y1
first operates on θ1 .
Figure 9.12 A ray that travels

through a distance a, reflects from
7 In contrast, the optical path length OPL is effectively longer than free space by the factor n.
a mirror, and then travels through
a distance b.
9.6 Image Formation 241
We have derived our basic ABCD matrices for rays traveling in the y–z plane,
as suggested in Figs. 9.7–9.12. This may have given the impression that it is
necessary to work within a plane that contains the optical axis (i.e. the z-axis
in our case). However, within the paraxial approximation, the ABCD matrices
are valid for rays that become displaced simultaneously in both the x and y
dimensions during propagating along z.
As we demonstrate below, the behavior of rays functions independently in
the x and y dimensions. h y Ifi desired, one can write a ray vector for each dimen-
£x¤
sion, namely θx and θ y . Moreover, the identical matrices, for example any
in table 9.1, are used for either dimension. Figs. 9.7–9.12 therefore represent
projections of rays onto the y–z plane. To complete the story, one can imagine
corresponding figures representing the projection of the rays onto the x–z plane.
Independence of Rays in the x and y Dimensions
Imagine a ray contained within a plane that is parallel to the y–z plane but for
which x > 0. One might be concerned that when the ray meets, for example, a
spherically concave mirror, the radius of curvature in the perspective of the y–z
dimension might be different for x > 0 than for x = 0 (at the center of the mirror).
This concern is actually quite legitimate and is the source of what is known as Galileo di Vincenzo Bonaiuti de'
spherical aberration. Nevertheless, in the paraxial approximation the intersection Galilei (15641642, Italian) was born
with the curved mirror of all planes that are parallel to the optical axis gives the in Pisa, Italy, the son of a musician.
Galileo enrolled in the University of Pisa
same curve.
with the intent to study medicine but
To see why this is so, consider the curvature of the mirror in Fig. 9.8. As we soon became diverted into mathematics.
He served three years as chair of math-
move away from the mirror center (in the x or y-dimension or some combination
ematics in Pisa beginning in 1589 and
thereof), the mirror surface deviates to the left by the amount then moved to the University of Padua
where he taught geometry, mechanics,
δ = R − R cos φ (9.50) and astronomy for two decades. While
Galileo did not invent the telescope, he
∼ 2
. φ = 1 − φ /2. And since in this approxi-
considerably improved the design. With
In the paraxial approximation, we have cos
it he discovered four moons of Jupiter
mation we may also write φ ∼
p
= x 2 + y 2 R, (9.50) becomes and was the rst to observe sunspots
and mountains and valleys on the Moon.
Galileo also was the rst to document
x2 y2
δ∼
= + (9.51) the phases of Venus, similar to the
2R 2R phases of the moon. He used these
observations to argue in favor of the
In the paraxial approximation, we see that the curve of the mirror is parabolic, and Copernican model of the solar system,
therefore separable between the x and y dimensions. That is, the curvature in but this conicted with the prevailing
the x-dimension (i.e. ∂δ/∂x = x/R) is independent of y, and the curvature in the views of the Catholic Church at the
time, and he was placed under house
y-dimension (i.e. ∂δ/∂y = y/R) is independent of x. A similar argument can be arrest and forbidden to publish of any
made for a spherical interface between two media. of his works. While under house arrest,
he wrote much on kinematics and other
principles of physics and is considered to
be the father of modern physics. Galileo
attempted to measure the speed of
light by observing an assistant uncover
9.6 Image Formation a lantern on a distant hill in response
to a light signal. He concluded that
light is really fast if not instantaneous.
Consider Example 9.7 where a ray travels a distance a, reflects from a curved
(Wikipedia)
mirror, and then travels a distance b. From (9.49), the ABCD matrix for the overall
process is
· ¸ · ¸
A B 1 − b/ f a + b − ab/ f
= (9.52)
C D −1/ f 1 − a/ f
where by (9.47) we have replaced 2/R with 1/ f . Because of the similarity between
the behavior of a curved mirror and a thin lens, the above expression can also
represent a ray traveling a distance a, traversing a thin lens with focal length f ,
and then traveling a distance b. The only difference is that, in the case the thin
lens, f is given by lens maker’s formula (9.46).
As is well known, it is possible to form an image with either a curved mirror
or a lens. Suppose that the initial ray is one of many rays that leaves a particular
point on an object positioned a = d o before the mirror (or lens). In order for an
image to occur at d i = b, it is essential that all rays leaving the particular point on
the object converge to a corresponding point on the image. That is, we want rays
leaving the point y 1 on the object (which may take on a range of angles θ1 ) all to
converge to a single point y 2 at the image. In the following equation we need y 2
to be independent of θ1 :
Ay 1 + B θ1
· ¸ · ¸· ¸ · ¸
y2 A B y1
= = (9.53)
θ2 C D θ1 C y 1 + Dθ1
The condition for image formation is therefore
Image B =0 (condition for image formation) (9.54)

Object
When this condition is applied to (9.52), we obtain
do di 1 1 1
Figure 9.13 Image formation by a
do + di − =0⇒ = + (9.55)
f f do di
thin lens.
which is the familiar imaging formula (9.1). When the object is infinitely far away
(i.e. d o → ∞), the image appears at d i → f . This gives a physical interpretation
to the focal length f , as we have been calling it. Please note that d o and d i can
each be either positive (real as depicted in Fig. 9.13) or negative (virtual meaning
a screen cannot be inserted to display the image).
The magnification of the image is found by comparing the size of y 2 to y 1 .
From (9.52)–(9.55), the magnification is found to be
y2 di di
M≡ = A = 1− = − (9.56)
y1 f do
The negative sign indicates that for positive distances d o and d i the image is
inverted.
In the above discussion, we have examined image formation by a thin lens
or a curved mirror. Of course, images can also be formed by thick lenses or by
more complex composite optical systems (e.g. a system of lenses and spaces).
The ABCD matrices for the elements in a composite system are simply multiplied
together (the first element that rays encounter appearing on the right) to obtain an
9.6 Image Formation 243
overall ABCD matrix. The principles for image formation with an arbitrary ABCD
matrix are the same as those for a thin lens or curved mirror. As before, consider
propagation a distance d o from an object to the optical element followed by
propagation a distance d i to an image. The ABCD matrix for the overall operation
is
· ¸· ¸· ¸ · ¸
1 di A B 1 do A + d iC d o A + B + d o d iC + d i D
=
0 1 C D 0 1 C d oC + D
· 0 (9.57)
A B0
¸
=
C 0 D0
An image occurs according to (9.54) when B 0 = 0, or
d o A + B + d o d iC + d i D = 0, (9.58) general condition for image

formation
with magnification
M = A + d iC (9.59)
For a complex lens system, the matrix elements A, B , C and D can be complicated
expressions. There is a convenient way to simplify the analysis, which is discussed
in the next section.
Example 9.8
Beginning students are often taught to draw ray diagrams such as the one in Fig.
9.14, which shows a real image formed by a thin lens. Several key rays aid in a
graphic prediction of the location and size of the image. Use ABCD-matrix analysis
to describe the effect of the lens on the three rays drawn.
object
A
C
image
Figure 9.14 Formation of a real image by a thin lens.
Solution: Ray A is parallel to the axis with height y 1 before traversing the lens. Just
after the lens, ray A is described by
· ¸ · ¸· ¸ · ¸
y2 1 0 y1 y1
= =
θ2 −1/ f 1 0 −y 1 / f
· ¸· ¸ · ¸
1 f y1 0
which crosses the axis at the focus d = f , since 0 1 −y 1 / f
=
−y 1 / f
.
Meanwhile, ray B traverses the lens just where it crosses the axis. The lens
does nothing to this ray:
µ ¶ µ ¶µ ¶ µ ¶
y2 1 0 0 0
= =
θ2 −1/ f 1 −y 1 /d o −y 1 /d o
Ray B is un-deflected.
Finally, ray C, which goes through the point d = − f before the lens, becomes
parallel to the axis following the lens:
· ¸ · ¸· ¸ · ¸
y2 1 0 −M y 1 −M y 1
= =
θ2 −1/ f 1 −M y 1 / f 0
Note that starting from the left focus, we have just before the lens
· ¸· ¸ · ¸
1 f 0 −M y 1
=
0 1 −M y 1 / f −M y 1 / f
9.7 Principal Planes for Complex Optical Systems

For every ABCD matrix representing an optical system, there exist two principal
planes,8 located (in our convention) a distance p 1 before entering the system and
a distance p 2 after exiting the system. When the matrices corresponding to these
(appropriately chosen) distances are appended to the original ABCD matrix of
the system, the overall matrix simplifies to one that looks identical to the matrix
for a simple thin lens (9.45).
With knowledge of the positions of the principal planes, one can treat the
complicated imaging system in the same way that one treats a simple thin lens.
That is, we can simply use the common formulas (9.55) and (9.56). The only
difference is that d o is the distance from the object to the first principal plane and
d i is the distance from the second principal plane to the image. In the case of an
actual thin lens, both principal planes are at p 1 = p 2 = 0. For a composite system,
p 1 and p 2 can be either positive or negative.
First Second
Principal Principal
We assert that for any optical system,9 p 1 and p 2 can always be selected such
Plane Plane that we can write
· ¸· ¸· ¸ · ¸
Figure 9.15 A multi-element sys- 1 p2 A B 1 p1 A + p 2C p 1 A + B + p 1 p 2C + p 2 D
=
tem represented as an ABCD ma- 0 1 C D 0 1 C p 1C + D
trix for which principal planes · ¸
1 0
always exist. =
−1/ f eff 1
(9.60)
The final matrix is that of a simple thin lens, and it takes the place of the composite
system including the distances to the principal planes.
8 R. Guenther, Modern Optics, p. 186 (New York: Wiley, 1990).

9 The starting and ending refractive index must be the same.
9.8 Stability of Laser Cavities 245
Determination of p 1 and p 2 and Justification of (9.60)
Our task is to find the values of p 1 and p 2 that make (9.60) true. We can straight-
away make the definition
f eff ≡ −1/C (9.61)
We can also solve for p 1 and p 2 by setting the diagonal elements of the matrix to 1.
Explicitly, we get
1−D
p 1C + D = 1 ⇒ p 1 = (9.62)
C
and
1− A
A + p 2C = 1 ⇒ p 2 = (9.63)
C
It remains to be shown that the upper right element in (9.60) (i.e. p 1 A + B +
p 1 p 2C + p 2 D) automatically goes to zero for our choices of p 1 and p 2 . This may
seem unlikely at first, but watch what happens!
When (9.62) and (9.63) are substituted into the upper right matrix element of (9.60)
we get
1−D 1−D 1− A 1− A
p 1 A + B + p 1 p 2C + p 2 D = A +B + C+ D
C C C C
1
= [1 − AD + BC ] (9.64)
C
µ ¯ ¯¶
1 ¯ A B ¯
= 1−¯¯ ¯
C C D ¯
This vanishes (as desired) if the determinant of the original ABCD matrix equals
one. Fortunately, this is always the case as long as we begin and end in the same
index of refraction: ¯ ¯
¯ A B ¯
¯ C D ¯=1 (9.65)
¯ ¯
Notice that the determinants of all of the matrices in table 9.1 are one. Moreover,
ABCD matrices constructed of these will also have determinants equal to one.10
9.8 Stability of Laser Cavities

The ABCD matrix formulation provides a powerful tool to analyze the stability of a
laser cavity.11 The basic elements of a laser cavity include an amplifying medium
and mirrors to provide feedback. Presumably, at least one of the end mirrors is
partially transmitting so that energy is continuously extracted from the cavity.
Here, we dispense with the amplifying medium and concentrate our attention on
the optics providing the feedback.
10 The determinant of (9.43) is not one since it starts and ends with different indices of refraction.
However, when this matrix is used in succession to form a lens, the resulting matrix has determinant
equal to one.
11 P. W. Milonni and J. H. Eberly, Lasers, Sect. 14.3 (New York: Wiley, 1988).
As might be expected, the mirrors must be carefully aligned or successive

reflections might cause rays to ‘walk’ continuously away from the optical axis,
(a) so that they eventually leave the cavity out the side. If a simple cavity is formed
with two flat mirrors that are perfectly aligned parallel to each other, one might
suppose that the mirrors would provide ideal feedback. However, all rays except
for those that are perfectly aligned to the mirror surface normals would eventually
wander out of the side of the cavity as illustrated in Fig. 9.16a. Such a cavity is said
(b) to be unstable. We would like to do a better job of trapping the light in the cavity.
To improve the situation, a cavity can be constructed with concave end mir-
rors to help confine the beams within the cavity. Even so, one must choose
carefully the curvature of the mirrors and their separation L. If this is not done
(c) correctly, the curved mirrors can ‘overcompensate’ for the tendency of the rays
to wander out of the cavity and thus aggravate the problem. Such an unstable
scenario is depicted in Fig. 9.16b.
Figure 9.16c depicts a cavity made with curved mirrors where the separation
(d) L is chosen appropriately to make the cavity stable. Although a ray, as it makes
successive bounces, can strike the end mirrors at a variety of points, the curvature
of the mirrors keeps the ‘trajectories’ contained within a narrow region so that
they cannot escape out the sides of the cavity.
There are many ways to make a stable laser cavity. For example, a stable cavity
Figure 9.16 (a) A ray bouncing can be made using a lens between two flat end mirrors as shown in Fig. 9.16d. Any
between two parallel flat mirrors.
combination of lenses (perhaps more than one) and curved mirrors can be used
(b) A ray bouncing between two
to create stable cavity configurations. Ring cavities can also be made to be stable
curved mirrors in an unstable
configuration. (c) A ray bouncing where in no place do the rays retro-reflect from a mirror but circulate through
between two curved mirrors in a a series of elements like cars going around a racetrack. The ABCD matrix for a
stable configuration. (d) Stable round trip in the cavity will be useful for this analysis.
cavity utilizing a lens and two flat
end mirrors.
Example 9.9
Find the round-trip ABCD matrix for the cavities shown in Figs. 9.16c and 9.16d.
Solution: The round-trip ABCD matrix for the cavity shown in Fig. 9.16c is
· ¸ · ¸· ¸· ¸· ¸
A B 1 L 1 0 1 L 1 0
= (9.66)
C D 0 1 −2/R 2 1 0 1 −2/R 1 1
where we have begun the round trip just after a reflection from the first mirror.
The round-trip ABCD matrix for the cavity shown in Fig. 9.16d is
· ¸ · ¸· ¸· ¸· ¸
A B 1 2L 1 1 0 1 2L 2 1 0
= (9.67)
C D 0 1 −1/ f 1 0 1 −1/ f 1
where we have begun the round trip just after a transmission through the lens
moving to the right. It is somewhat arbitrary where a round trip begins. The
multiplication on the above matrices will need to be carried out to do problems
P9.15 and P9.16.
9.A Aberrations and Ray Tracing 247
To determine whether a given configuration of a cavity is stable, we need to

know what a ray does after making many round trips in the cavity. To find the
effect of propagation through many round trips, we multiply the round-trip ABCD
matrix together N times, where N is the number of round trips that we wish to
consider. We can then examine what happens to an arbitrary ray after making N
round trips in the cavity as follows:
· ¸ · ¸N · ¸
y N +1 A B y1
= (9.68)
θN +1 C D θ1
At this point you might be concerned that taking an ABCD matrix to the N th
power can be a lot of work. (It is already a significant work just to compute the
ABCD matrix for a single round trip.) In addition, we are interested in letting N
be very large, perhaps even infinity. You can relax because we have a neat trick to
accomplish this daunting task.
By Sylvester’s theorem in appendix 0.3, we have
¸N
· · ¸
A B 1
= (9.69)
where
1
cos θ =
(A + D) . (9.70)
2
This is valid as long as the determinant of the ABCD matrix is one. As noted
earlier (see (9.65)), we are in luck! The determinant is one any time a ray begins
and stops in the same refractive index, which by definition is guaranteed for any
round trip. We therefore can employ Sylvester’s theorem for any N that we might
choose, including very large integers.
We would like the elements of (9.69) to remain finite as N becomes very large.
If this is the case, then we know that a ray remains trapped within the cavity
and stays reasonably close to the optical axis. Since N only appears within the
argument of a sine function, which is always bounded between −1 and 1 for
real arguments, it might seem that the elements of (9.69) always remain finite
as N approaches infinity. However, it turns out that θ can become imaginary
depending on the outcome of (9.70), in which case the sine becomes a hyperbolic
sine, which can ‘blow up’ as N becomes large. In the end, the condition for cavity
stability is that a real θ must exist for (9.70), or in other words we need
1
−1 < (A + D) < 1 (condition for a stable cavity) (9.71)
2
It is left as an exercise to apply this condition to (9.66) and (9.67) to find the
necessary relationships between the various element curvatures and spacing in
order to achieve cavity stability.
Appendix 9.A Aberrations and Ray Tracing

(a)
The paraxial approximation places serious limitations on the performance of
optical systems (see (9.23) and (9.24)). To stay within the approximation, all rays
traveling in the system should travel very close to the optic axis with very shallow
angles with respect to the optical axis. To the extent that this is not the case, the
collection of rays associated with a single point on an object may not converge to
a single point on the associated image. The resulting distortion or “blurring” of
the image is known as aberration.
(b)
Common experience with photographic and video equipment suggests that
it is possible to image scenes that have a relatively wide angular extent (many
tens of degrees), in apparent serious violation of the paraxial approximation.
The paraxial approximation is indeed violated in these devices, so they must be
designed using more complicated analysis techniques than those we have learned
in this chapter. The most common approach is to use a computationally intensive
Figure 9.17 (a) Paraxial theory pre- procedure called ray tracing in which sin θ and tan θ are rendered exactly. The
dicts that the light imaged from nonlinearity of these functions precludes the possibility of obtaining analytic
a point source will converge to
solutions describing the imaging performance of such optical systems.
a point (i.e. have spherical wave
fronts coming to the image point). The typical procedure is to start with a collection of rays from a test point such
(b) The image of a point source as shown in Fig. 9.18. Each ray is individually traced through the system using
made by a real lens with aberra- the exact representation of geometric surfaces as well as the exact representation
tions is an extended and blurred of Snell’s law. On close analysis, the rays typically do not converge to a distinct
patch of light and the converg- imaging point. Rather, the rays can be ‘blurred’ out over a range of points where
ing wavefronts are only quasi- the image is supposed to occur. Depending on the angular distribution of the
spherical.
rays as well as on the elements in the setup, the spread of rays around the image
point can be large or small. The engineer who designs the system must determine
whether the amount of aberration is acceptable, given the various constraints of
the device.
To minimize aberrations below typical tolerance levels, several lenses can
be used together. If properly chosen, the lenses (some positive, some negative)
separated by specific distances, can result in remarkably low aberration levels
over certain ranges of operation for the device. Ray tracing is best done with
commercial software designed for this purpose (e.g. Zemax or other professional
products). Such software packages are able to develop and optimize designs for
specific applications. A nice feature is that the user can specify that the design
Figure 9.18 Ray tracing through a should employ only standard optical components available from known optics
simple lens. companies. In any case, it is typical to specify that all lenses in the system should
have spherical surfaces since these are much less expensive to manufacture. We
mention briefly a few types of aberrations that you may encounter. Multiple
aberrations can often be observed in a single lens.
Chromatic aberration arises from the fact that the index of refraction for glass
varies with the wavelength of light. Since the focal length of a lens depends on
the index of refraction (see, for example, Eq. (9.46)), the focal length of a lens
varies with the wavelength of light. Chromatic aberration can be compensated
for by using a pair of lenses made from two types of glass as shown in Fig. 9.19
9.A Aberrations and Ray Tracing 249
(the pair is usually cemented together to form a “doublet” lens). The lens with the
shortest focal length is made of the glass whose index has the lesser dependence
on wavelength. By properly choosing the prescription of the two lenses, you
can exactly compensate for chromatic aberration at two wavelengths and do a
good job for a wide range of others. Achromatic doublets can also be designed to
minimize spherical aberration (see below), so they are often a good choice when
you need a high quality lens.
Monochromatic aberrations arise from the shape of the lens rather than the
variation of n with wavelength. Before the advent computers facilitated the
widespread use of ray tracing, these aberrations had to be analyzed primarily
with analytic techniques. The analytic results derived previously in this chapter
were based on first order approximations (e.g. sin θ ≈ θ). This analysis predicts
that a lens can image a point source to an exact image point, which predicts
spherically converging wavefronts at the image point as shown in Fig. 9.17(a). You
can increase the accuracy of the theory for non-paraxial rays by retaining second-
order correction terms in the analysis. With these second-order terms included,
the wave fronts converging towards an image point are mostly spherical, but have
second-order aberration terms added in (shown conceptually in Fig. 9.17(b)).
There are five aberration terms in this second-order analysis, and these represent
a convenient basis for discussing aberration.
The first aberration term is known as spherical aberration. This type of aber- high dispersion
low dispersion
ration results from the fact that rays traveling through a spherical lens at large glass glass
radii experience a different focal length than those traveling near the axis. For a
Figure 9.19 Chromatic aberration
converging lens, this causes wide-radius rays to focus before the near-axis rays
causes lenses to have different
as shown in Fig. 9.20. This problem can be helped by orienting lenses so that
focal lengths for different wave-
the face with the least curvature is pointed towards the side where the light rays lengths. It can be corrected using
have the largest angle. This procedure splits the bending of rays more evenly an achromatic doublet lens.
between the front and back surface of the lens. As mentioned above, you can also
cement two lenses made from different types of glass together so that spherical
aberrations from one lens are corrected by the other.
The aberration term referred to as astigmatism occurs when an off-axis object
point is imaged to an off-axis image point. In this case a spherical lens has a
different focal length in the horizontal and vertical dimensions. For a focusing
lens this causes the two dimensions to focus at different distances, producing a
vertical line at one image plane and a horizontal line at another. A lens can also be
inherently astigmatic even when viewed on axis if it is football shaped rather than
spherical. In this case, the astigmatic aberration can be corrected by inserting a
cylindrical lens at the correct orientation (this is a common correction needed in
eyeglasses).
A third aberration term is referred to as coma. This is observed when off-axis
points are imaged and produces a comet shaped tail with its head at the point
predicted by paraxial theory. (The term ‘coma’ refers to the atmosphere of a
comet, which is how the aberration got its name.) This aberration is distinct from
astigmatism, which is also observed for off-axis points, since coma is observed Figure 9.20 Spherical aberration
in a plano-convex lens.
c
b
a
c
a
Image on screen
Figure 9.21 Illustration of coma. Rays traveling through the center of the lens are im-
aged to point a as predicted by paraxial theory. Rays that travel through the lens at
radius ρ b in the plane of the figure are imaged to point b. Rays that travel through the
lens at radius ρ b , but outside the plane of the figure are imaged to other points on the
circle (in the image plane) containing point b. Rays at that travel through the lens at
other radii on the lens (e.g. ρ c ) also form circles in the image plane with radius propor-
tional to ρ 2 with the center offset from point a a distance proportional to ρ 2 . When
light from each of these circles combines on the screen it produces an imaged point
Undistorted
with a “comet tail.”
even when all of the rays are in one plane (see Fig. 9.21). You have probably seen
coma if you’ve ever played with a magnifying glass in the sun—just tilt the lens
slightly and you see a comet-like image rather than a point.
The curvature of the field aberration term arises from the fact that spherical
Barrel Distortion lenses image spherical surfaces to another spherical surface, rather than imaging
a plane to a plane. This is not so bad for your eyeball, which has a curved screen,
but for things like cameras and movie projectors we would like to image to a flat
screen. When a flat screen is used and the curvature of the field aberration is
present, the image will be focus well near the center, but become progressively
out of focus as you move to the edge of the screen (i.e. the flat screen is farther
from the curved image surface as you move from the center).
The final aberration term is referred to as distortion. This aberration occurs
Pincushion Distortion
when the magnification of a lens depends on the distance from the center of
Figure 9.22 Distortion occurs the screen. If magnification decreases as the distance from the center increases,
when magnification is not con- then ‘barrel’ distortion is observed. When magnification increases with distance,
stant across an extended image.
‘pincushion’ distortion is observed (see Fig. 9.22).
All lenses will exhibit some combination of the aberrations listed above (i.e.
chromatic aberration plus the five second-order aberration terms). In addition to
the five named monochromatic aberrations, there are many other higher order
aberrations that also have to be considered. Aberrations can be corrected to a high
degree with multiple-element systems (designed using ray-tracing techniques)
composed of lenses and irises to eliminate off-axis light. For example, a camera
lens with a focal length of 50 mm, one of the simplest lenses in photography, is
typically composed of about six individual elements. However, optical systems
never completely eliminate all aberration, so designing a system always involves
some degree of compromise in choosing which aberrations to minimize and
which ones you can live with.
Exercises 251
Exercises
Exercises for 9.1 The Eikonal Equation
P9.1 Consider the index described in Example 9.1. The solution given in
the example corresponds to rays that asymptotically approach y = 0. A
more general solution is given by
p
µ q ¶
∇R = n 0 x̂ 1 + α ± ŷ y 2 /h 2 − α 1 + α > 0 and y 2 /h 2 − α > 0
¡ ¢
This corresponds to rays that either hit the ground or return toward the
sky without reaching the ground, depending on the sign of α.
(a) Verify that ∇R satisfies the eikonal equation and determine the
¡ ¢
function R x, y .
ξ
³ p ´
HINT: d ξ ξ2 − α = 2 ξ2 − α − α2 ln ξ + ξ2 − α (ξ − α > 0).
R p p
p ³ ´
(b) Verify that the light path is given by y = h α cosh x−x p 0 when
h 1+α
p ¯
¯ x−x
¯
α > 0 and is given by y = h |α| sinh ¯ p 0
¯ when α < 0. Consider only
¯
h 1+α
the region y > 0 (i.e. above ground). Notice that these solutions can
make rays that travel either to the right or to the left.
HINT: cosh2 ξ − sinh2 ξ = 1 ddξ cosh ξ = sinh ξ ddξ sinh ξ = cosh ξ.
(c) Make a sketch of these two solution classes in the case of α = ±4.
P9.2 Prove that under the approximation of very short wavelength, the
Poynting vector is directed along ∇R (r) or ŝ.
Solution: (partial)
First, from Faraday’s law (1.36) we have
i ³ ´
B(r, t ) = ∇ × E0 (r)e i (kvac R(r)−ωt )
ω
Applying the identity ∇ × aψ = ψ(∇ × a) + ∇ψ × a to this equation, we obtain:

¡ ¢
i ³ i (kvac R(r)−ωt ) ´
B(r, t ) = e [∇ × E0 (r)] + i k vac e i (kvac R(r)−ωt ) [∇R(r) × E0 (r)]
ω
i λvac i [kvac R(r)−ωt ] 1
= e [∇ × E0 (r)] − e i [kvac R(r)−ωt ] [∇R (r) × E0 (r)]
2πc c
The first term vanishes in the limit of very short wavelength, and we have:
1
B(r, t ) → − [∇R (r)] × E0 (r) e i [kvac R(r)−ωt ] . (9.72)
c
Next, from Gauss’s law (1.34) and the constitutive relation (2.16) we have
h¡ i
∇ · 1 + χ(r) E0 (r)e i (kvac R(r)−ωt ) = 0
¢
Applying the identity ∇ · (aψ) = a · ∇ψ + ψ∇ · a to this expression yields:
e i (kvac R(r)−ωt ) ∇ · 1 + χ (r) E0 (r) + i k vac e i (kvac R(r)−ωt ) 1 + χ(r) [∇R (r) · E0 (r)] = 0
£¡ ¢ ¤ ¡ ¢
Canceling the common exponential term, using k vac = 2π/λvac , and some algebra then gives
∇ · 1 + χ(r) E0 (r)
£¡ ¢ ¤
−i λvac + ∇R(r) · E0 (r) = 0
2π 1 + χ(r)
¡ ¢
In the limit of very short wavelength, this becomes
∇R(r) · E0 (r) → 0 (9.73)
Finally, compute the time average of the Poynting vector
1
S= Re {E(r, t )} × Re {B(r, t )}
µ0
1 £
E (r, t ) + E∗ (r, t ) × B(r, t ) + B∗ (r, t )
¤ £ ¤
=
4µ0
You will need to employ expressions (9.72) and (9.73), as well as the BAC-CAB rule (see P0.3).
Exercises for 9.2 Fermat’s Principle
P9.3 Use Fermat’s Principle to derive the law of reflection (3.6) for a reflective
surface.
A
B
HINT: Do not consider light that goes directly from A to B; require a
single bounce.
P9.4 Show that Fermat’s Principle fails to give the correct path for an extraor-
dinary ray entering a uniaxial crystal whose optic axis is perpendicular
to the surface.
HINT: With the index given by (5.29), show that Fermat’s principle
Figure 9.23 leads to an answer that neither agrees with the direction of the k-vector
(5.32) nor with the direction of the Poynting vector (5.40).
Exercises for 9.4 Reflection and Refraction at Curved Surfaces
P9.5 Derive the ABCD matrix that takes a ray on a round trip through a
simple laser cavity consisting of a flat mirror and a concave mirror of
radius R separated by a distance L. HINT: Start at the flat mirror. Use
the matrix in (9.28) to travel a distance L. Use the matrix in (9.38) to
represent reflection from the curved mirror. Then use the matrix in
(9.28) to return to the flat mirror. The matrix for reflection from the flat
mirror is the identity matrix (i.e. R flat → ∞).
Exercises 253
P9.6 Derive the ABCD matrix for a thick lens made of material n 2 sur-
rounded by a liquid of index n 1 . Let the lens have curvatures R 1 and R 2
and thickness d .
Answer:
 ³
n1
´
n

1 + Rd −1 d n1
· ¸
A B 1 ´ n2 ³2
= ³
n2
´³ ³
n n
´
n1
´ 
C D − n − 1 R − R + R dR 2 − n1 − n2
1 1
1 − Rd n2 − 1
1 1 2 1 2 2 1 2
Exercises for 9.6 Image Formation C
P9.7 (a) Show that the ABCD matrix for a thick lens (see P9.6) reduces to that A
of a thin lens (9.45) when the thickness goes to zero. Take the index
image object
outside of the lens to be n 1 = 1. B
(b) Find the ABCD matrix for a thick window (thickness d ). Take the
index outside of the window to be n 1 = 1. HINT: A window is a thick
lens with infinite radii of curvature. Figure 9.24 Formation of a virtual
image by a thin lens.
P9.8 An object is placed in front of a concave mirror. Find the location of
the image d i and magnification M when d o = R, d o = R/2, d o = R/4,
and d o = −R/2 (virtual object). Make a diagram for each situation,
depicting rays traveling from a single off-axis point on the object to A
a corresponding point on the image. You may want to emphasize
especially the ray that initially travels parallel to the axis and the ray
C
that initially travels in a direction intersecting the axis at the focal point
R/2. object image B
P9.9 Perform an analysis similar to example 9.8 for the virtual image formed
by the positive lens in Fig. 9.24.
Figure 9.25 Formation of a virtual
P9.10 Perform an analysis similar to example 9.8 for the virtual image formed image by a thin lens with negative
by the negative lens in Fig. 9.25. focal length.
Exercises for 9.7 Principal Planes for Complex Optical Systems

unknown
P9.11 A complicated lens element is represented by an ABCD matrix. An element
object placed a distance d 1 before the unknown element causes an
image to appear a distance d 2 after the unknown element.
Suppose that when d 1 = `, we find that d 2 = 2`. Also, suppose that
when d 1 = 2`, we find that d 2 = 3`/2 with magnification −1/2. What is
Figure 9.26
the ABCD matrix for the unknown element?
HINT: Use the conditions for an image (9.58) and (9.59). If the index
of refraction is the same before and after, then (9.65) applies. HINT:
First find linear expressions for A, B , and C in terms of D. Then put the
results into (9.65).
Principal Principal
Plane Plane
P9.12 (a) Consider a lens with thickness d = 5 cm, R 1 = 5 cm, R 2 = −10 cm,
n = 1.5. Compute the ABCD matrix of the lens. HINT: See P9.6.
(b) Where are the principal planes located and what is the effective
focal length f eff for this system?
L9.13 Deduce the positions of the principal planes and the effective focal
Figure 9.27
length of a compound lens system. Reference the positions of the
principal planes to the outside ends of the metal hardware that encloses
the lens assembly. (video)
HINT: Obtain three sets of distances to the object and image planes
and place the data into (9.58) to create three distinct equations for the
unknowns A, B, C, and D. Find A, B, and C in terms of D and place the
results into (9.65) to obtain the values for A, B, C, and D. The effective
Figure 9.28 focal length and principal planes can then be found through (9.61)–
(9.63).
P9.14 Use a computer program to calculate the ABCD matrix for the com-
pound system shown in Fig. 9.29, known as the “Tessar lens.” The
details of this lens are as follows (all distances are in the same units,
and only the magnitude of curvatures are given—you decide the sign):
Convex-convex lens 1 (thickness 0.357, R 1 = 1.628, R 2 = 27.57, n =
1.6116) is separated by 0.189 from concave-concave lens 2 (thickness
0.081, R 1 = 3.457, R 2 = 1.582, n = 1.6053), which is separated by 0.325
from plano-concave lens 3 (thickness 0.217, R 1 = ∞, R 2 = 1.920, n =
1.5123), which is directly followed by convex-convex lens 4 (thickness
0.396, R 1 = 1.920, R 2 = 2.400, n = 1.6116).
1 2 3 4 HINT: You can reduce the number of matrices you need to multiply by
using the “thick lens” matrix.
Figure 9.29
Exercises for 9.8 Stability of Laser Cavities
P9.15 (a) Show that the cavity depicted in Fig. 9.16c is stable if
L L
µ ¶µ ¶
0 < 1− 1− <1
R1 R2
(b) The two concave mirrors have radii R 1 = 60 cm and R 2 = 100 cm.
Over what range of mirror separation L is it possible to form a stable
laser cavity?
HINT: There are two different stable ranges with an unstable range
between them.
P9.16 Find the stable ranges for L 1 = L 2 = L for the laser cavity depicted in
Fig. 9.16d with focal length f = 50 cm.
Exercises 255
L9.17 Experimentally determine the stability range of a HeNe laser with ad-
justable end mirrors. Check that this agrees reasonably well with theory.
Figure 9.30
Can you think of reasons for any discrepancy? (video)
Chapter 10
Diffraction
In the 1600’s, Christian Huygens developed a wave description for light. Unfor-
tunately, his ideas were largely overlooked at the time because Sir Isaac Newton
promoted a competing theory. Newton proposed that light should be thought
of as many tiny bullets, or corpuscles, as he called them. Newton’s ideas pre-
vailed for more than a century, perhaps because he was right on so many other
things, until 1807 when Thomas Young performed his famous two-slit experiment,
conclusively demonstrating the wave nature of light. Even then, Young’s conclu-
sions were accepted only gradually by others, a notable exception being a young
Frenchman named Augustin Fresnel. The two formed a close friendship through
correspondence, and it was Fresnel that followed up on Young’s conclusions and
dedicated his life to a study of light.
Fresnel’s skill as a mathematician allowed him to transform physical intuition
into powerful and concise ideas. Perhaps Fresnel’s greatest accomplishment was Christiaan Huygens (16291695,
the adaptation of Huygens’ principle of wavelet superposition into a mathematical Dutch) was born in The Hague, Nether-
lands. His father was friends with the
formula. Ironically, he used Newton’s calculus to achieve this. Huygens’ principle mathematician René Descartes, which
asserts that a wave front can be thought of as many wavelets, which propagate and probably inuenced his upbringing. Huy-
gens studied law and mathematics at
interfere to form new wave fronts. This is illustrated in Fig. 10.1. The phenomenon the University of Leiden, which preceded
of diffraction is then understood as the spilling of wavelets around obstructions a very productive career as a scientist
and mathematician. During mid career,
in the path of light.
Huygens held a position in the French
After formulating Huygens’ principle as a diffraction integral, Fresnel made Academy of Sciences in Paris for 15
an approximation to his own formula, called the Fresnel approximation, for the years, but spent the majority of his life
in The Hague. Huygens was the rst
sake of making the integration easier to perform. As far as approximations go, to advocate the wave theory of light.
the Fresnel approximation is surprisingly accurate in describing the light field He was able to explain birefringence in
terms of his wave theory together with
in the region down stream from an aperture. The diffraction pattern can evolve a refractive index that varied with direc-
in complicated ways as the distance from an aperture increases. At distances far tion. Huygens constructed a telescope
with which he discovered Saturn's moon
down stream from an aperture, the diffraction pattern acquires a final form that
Titan. He also made the rst detailed
no longer evolves, other than to grow in proportion to distance. This far-field observations of the Orion nebula. Huy-
limit is often of interest, and it turns out that the Fresnel diffraction formula can gens made signicant advancements in
clock-making technology and wrote a
be simplified further in this case. The far-field limit of the Fresnel diffraction book on probability theory. Huygens
formula is called the Fraunhofer approximation. was one of the earliest science-ction
writers and speculated that life exists on
From the modern perspective, Fresnel’s diffraction formula needs justifica- other planets in his book Cosmotheoros.
(Wikipedia)
257
258 Chapter 10 Diffraction
tion starting from Maxwell’s equation. The diffraction formula is based on scalar
diffraction theory, which ignores polarization effects. In some situations, ignor-
ing polarization is benign, but in other situations ignoring polarization effects
produces significant errors. These issues as well as the approximations leading to
scalar diffraction theory are discussed in section 10.2.
10.1 Huygens’ Principle as Formulated by Fresnel

In this section we discuss the calculus of summing up the contributions from the
many wavelets originating in an aperture illuminated by a light field. Each point
in the aperture is thought of as a source of a spherical wavelet.1 In our modern
notation, such a spherical wave can be written as proportional to e i kR /R, where R
is the distance from the source. As a spherical wave propagates, its strength falls
off in proportion to the distance traveled and the phase is related to the distance
propagated, similar to the phase of a plane wave.
You should be aware that a spherical wave of the form e i kR /R (even if some
sort of polarization is attached) is a poor solution to Maxwell’s equations (see
P10.2). It utterly fails near R = 0. However, if R is much larger than a wavelength,
this spherical wave starts to approximate actual solutions to Maxwell’s equations.
It is within this regime that the diffraction formula derived here is successful. It
Figure 10.1 Wave fronts depicted should be noted that by choosing k, we consider only a single wavelength of light
as a series of Huygens’ wavelets. (i.e. one frequency).
Consider an aperture or opening in an opaque screen in the plane z = 0. Let
the aperture be illuminated with a light field distribution E (x 0 , y 0 , z = 0) within
the aperture. Then for a point (x, y, z) lying somewhere after the aperture (z > 0)
the net field is given by adding together wavelets emitted from each point in the
aperture.
Each spherical wavelet takes on the strength and phase of the field at the
point where it originates. Mathematically, this summation takes the form
i e i kR
Ï
E (x, y, z) = − E (x 0 , y 0 , 0) d x 0d y 0 (10.1)
λ R
aperture
where q
R= (x − x 0 )2 + (y − y 0 )2 + z 2 (10.2)
Figure 10.2 is the radius of each wavelet as it individually intersects the point (x, y, z). The
factor −i /λ in front of the integral in (10.1) ensures the right phase and field
strength (not to mention units). Justification for this factor is given in section
10.3 and in appendix 10.A. To summarize, (10.1) tells us how to compute the field
1 For simplicity, we use the term ‘spherical wave’ in this book to refer to waves of the type
imagined by Huygens (i.e. of the form e i kR /R). There is a different family of waves based on
spherical harmonics that are also sometimes referred to as spherical waves. These waves have
angular as well as radial dependence, and they are solutions to Maxwell’s equations. See J. D.
Jackson, Classical Electrodynamics, 3rd ed., pp. 429–432 (New York: John Wiley, 1999).
10.1 Huygens’ Principle as Formulated by Fresnel 259
down stream given knowledge of the field in an aperture. The field at each point
(x 0 , y 0 ) in the aperture, which may vary with strength and phase, is treated as the
source for a spherical wave. The integral in (10.1) sums the contributions for all
of these wavelets.
Example 10.1
Find the on-axis2 (i.e. x, y = 0) intensity following a circular aperture of diameter `
illuminated by a uniform plane wave.
Solution: The diffraction integral (10.1) takes the form

p 02 02 2
i ¢ e i k x +y +z
Ï
0 0
d x0d y 0
¡
E (0, 0, z) = − E x , y ,0 p
λ x 02 + y 02 + z 2
aperture
Figure 10.3 Circular aperture illu-

The circular hole encourages a change to cylindrical coordinates: x 0 = ρ 0 cos φ0 and
minated by a plane wave.
y 0 = ρ 0 sin φ0 ; d x 0 d y 0 → ρ 0 d ρ 0 d φ0 . In this case, the limits of integration define of
the geometry of the aperture, and the integration is accomplished as follows:
Z2π Z`/2 i k pρ 02 +z 2
i E0 0 e
E (0, 0, z) = − dφ ρ0 d ρ0
λ ρ 02 + z 2
p
0 0
p 02 2 ¯`/2 µ p
i E0 e i k ρ +z ¯¯
¶
2 2
=− 2π ¯ = −E 0 e i k (`/2) +z − e i kz
λ ik ¯
0
The on axis intensity becomes

µ p ¶µ p ¶
2 2 2 2
I (0, 0, z) ∝E (0, 0, z) E ∗ (0, 0, z)=|E 0 |2 e i k (`/2) +z − e i kz e −i k (`/2) +z − e −i kz
· µ q ¶¸
= 2 |E 0 |2 1 − cos k (`/2)2 + z 2 − kz
(10.3)
See problem P10.6 for a graph of this function.
When an aperture has a complicated shape, it may be convenient to break up

the diffraction integral (10.1) into several pieces. You are probably already used
to doing this sort of piecewise approach to integration in other settings. It seems
hardly worth giving a name to this technique, but it is called Babinet’s principle;
perhaps in Babinet’s day people were not as comfortable with calculus.
As an example of how to use Babinet’s principle, suppose that we have an
Figure 10.4 Aperture comprised of
aperture that consists of a circular obstruction within a square opening as de-
the region between a circle and a
picted in Fig. 10.4. Thus, the light transmits through the region between the circle square.
and the square. One can evaluate the overall diffraction pattern by first evaluating
the diffraction integral for the entire square (ignoring the circular block) and then
subtracting the diffraction integral for a circular opening having the shape of the
2 An analytical solution is not possible off axis.
block. This removes the unwanted part of the previous integration and yields the
overall result. It is important to add and subtract the integrals (i.e. fields), not
their squares (i.e. intensity).
As trivial as Babinet’s principle may seem to you, it may not be obvious at
first that Babinet’s principle also applies to an infinitely wide plane wave that
is interrupted by finite obstructions. In this case, one computes the diffraction
of the blocked portions of the field as though these portions were openings in a
mask. This result is then subtracted from the plane wave (no integration needed
for the plane), as depicted in Fig. 10.5.
Mask When Fresnel first presented his diffraction formula to the French Academy
Block of Sciences, a certain judge of scientific papers named Siméon Poisson noticed
that Fresnel’s formula predicted that there should be light in the center of the
geometric shadow behind a circular obstruction. This seemed so absurd to
Figure 10.5 A block in a plane Poisson that he initially disbelieved the theory, until the spot was shortly thereafter
wave giving rise to diffraction in experimentally confirmed, much to Poisson’s chagrin. Needless to say, Fresnel’s
the geometric shadow. paper was then awarded first prize, and this spot appearing behind circular blocks
has since been known as Poisson’s spot.
Example 10.2
Find the on-axis (i.e. x, y = 0) intensity behind a circular block of diameter ` placed
in a uniform plane wave.
Solution:
³ Fromp Example ´ 10.1, the on-axis field behind a circular aperture is
i kz i k (`/2)2 +z 2
E0 e − e . Babinet’s principle says to subtract this result from a plane
wave to obtain the field behind the circular block. The situation is depicted in
Fig. 10.5 (side view). The on-axis field is then
µ p ¶ p
2 2 2 2
E (0, 0, z) = E 0 e i kz − E 0 e i kz − e i k (`/2) +z = E 0 e i k (`/2) +z
The on axis intensity becomes

p p
(`/2)2 +z 2 −i k (`/2)2 +z 2
I (0, 0, z) ∝ E (0, 0, z) E ∗ (0, 0, z) = |E 0 |2 e i k e = |E 0 |2
This result says that, in the exact center of the shadow behind a circular obstruction,
the intensity is the same as the illuminating plane wave for all distance z. A spot of
light in the center forms right away; no wonder Poisson was astonished!
10.2 Scalar Diffraction Theory

In this section we provide the background motivation for Huygen’s principle and
Fresnel’s formulation of it. Consider a light field with a single frequency ω. The
light field can be represented by E (r) e −i ωt , and the time derivative in the wave
equation (2.13) can be easily performed. It reduces to
∇2 E (r) + k 2 E (r) = 0 (10.4)

10.2 Scalar Diffraction Theory 261
where k ≡ nω/c is the magnitude of the usual wave vector (see also (9.2)). Equa-
tion (10.4) is called the Helmholtz equation. Again, it is merely the wave equation
written for the case of a single frequency, where the trivial time dependence has
been removed. To obtain the full wave solution, just append the factor e −i ωt to
the solution of the Helmholtz equation E (r).
At this point we take an egregious step: We ignore the vectorial nature of E(r)
and write (10.4) using only the magnitude E (r). When using scalar diffraction
theory, we must keep in mind that it is based on this serious step. Under the
scalar approximation, the vector Helmholtz equation (10.4) becomes the scalar
Helmholtz equation:
∇2 E (r) + k 2 E (r) = 0 (10.5)
This equation of course is consistent with (10.4) in the case of a plane wave.
However, we are interested in spherical waves of the form E (r ) = E 0 r 0 e i kr /r . It
turns out that such spherical waves are exact solutions to the scalar Helmholtz
equation (10.5). The proof is left as an exercise (see P10.3). Nevertheless, spherical Francois Jean Dominique Arago
waves of this form only approximately satisfy the vector Helmholtz equation (10.4). (1786-1853, French) was born in Cata-
lan France, where his father was the
We can get away with this sleight of hand if the radius r is large compared to a Treasurer of the Mint. As a teenager,
wavelength (i.e., kr À 1) and if we restrict r to a narrow range perpendicular to Arago was sent to a municipal college
in Perpignan where he developed a
the polarization. deep interest in mathematics. In 1803,
he entered the Ecole Polytechnique in
Paris, where he purportedly was dis-
Significance of the Scalar Wave Approximation appointed that he was not presented
with new knowledge at a higher rate.
He associated with famous French
The solution of the scalar Helmholtz equation is not completely unassociated with
mathematicians Siméon Poisson and
the solution to the vector Helmholtz equation. In fact, if E scalar (r) obeys the scalar Pierre-Simmon Laplace. He later worked
Helmholtz equation (10.5), then with Jean-Baptiste Biot to measure
the meridian arch to determine the
E (r) = r × ∇E scalar (r) (10.6) exact length of the meter. This work
took him to the Balearic Islands, Spain,
where he was imprisoned as a spy, be-
obeys the vector Helmholtz equation (10.4).
ing suspected because of lighting res
Consider a spherical wave, which is a solution to the scalar Helmholtz equation: atop a mountain as part of his survey-
ing eorts. After a heroic prison escape
E scalar (r) = E 0 r 0 e i kr /r (10.7) and a subsequent string of misfortunes,

he eventually made it back to France
where he took a strong interest in op-
Remarkably, when this expression is placed into (10.6) the result is zero. Although
tics and the wave theory of light. Arago
zero is in fact a solution to the vector Helmholtz equation, it is not very interesting. and Fresnel established a fruitful collab-
A more interesting solution to the scalar Helmholtz equation is oration that extended for many years.
It was Arago who demonstrated Pois-
i e i kr son's spot (sometimes called Arago's
µ ¶
E scalar (r) = r 0 E 0 1 − cos θ (10.8) spot). Arago also invented the rst po-
kr r larizing lter. In later life, he served a
brief stint as the French prime minister.
which is one of an infinite number of unique ‘spherical’ solutions that exist. Notice (Wikipedia)
that in the limit of large r , this expression looks similar to (10.7), aside from the
factor cos θ. The vector form of this field according to (10.6) is
i e i kr
µ ¶
E (r) = −φ̂r 0 E 0 1 − sin θ (10.9)
kr r
This field looks approximately like the scalar spherical wave solution (10.7) in the
limit of large r if the angle is chosen to lie near θ ∼
= π/2 (spherical coordinates).
Since our use of the scalar Helmholtz equation is in connection with this spherical
wave under these conditions, the results are close to those obtained from the
vector Helmholtz equation.
Fresnel developed his diffraction formula (10.1) a half century before Maxwell
assembled the equations of electromagnetic theory. In 1887, Gustav Kirchhoff
demonstrated that Fresnel’s diffraction formula satisfies the scalar Helmholtz
equation. In doing this he clearly showed the approximations implicit in the
theory, and made a slight revision to the formula:
i ¢ e i kR 1 + cos (R, ẑ)

Ï · ¸
0 0
d x 0d y 0
¡ ¢ ¡
E x, y, z = − E x ,y ,z = 0 (10.10)
λ R 2
aperture
Figure 10.6 The factor in square brackets, Kirchhoff’s revision, is known as the obliquity factor.
Here, cos(R, ẑ) indicates the cosine of the angle between R and ẑ. Notice that this
factor is approximately equal to one when the point (x, y, z) is chosen to be in
the forward direction; we usually study diffraction under this circumstance. On
the other hand, the obliquity factor equals zero for fields traveling in the reverse
direction (i.e. in the −ẑ direction). This fixes a problem with Fresnel’s version of
the formula (10.1) based on Huygens’ wavelets, which suggested that light could
as easily diffract in the reverse direction as in the forward direction
In honor of Kirchhoff’s work, (10.10) is referred to as the Fresnel-Kirchhoff
diffraction formula. The details of Kirchhoff’s more rigorous derivation, including
how the factor −i /λ naturally arises, are given in Appendix 10.A. Since the Fresnel-
Kirchhoff formula can be understood as a superposition of spherical waves, it is
not surprising that it satisfies the scalar Helmholtz equation (10.5).
10.3 Fresnel Approximation

Although the Fresnel-Kirchhoff integral looks innocent enough, it is actually quite
difficult to evaluate analytically. It is problematic even if the field E x 0 , y 0 , z = 0 is
¡ ¢
constant across the aperture and if the obliquity factor (1 + cos (r, ẑ))/2 is approxi-
mated as one (i.e. forward direction).
Fresnel introduced an approximation3 to his diffraction formula that makes
the integration somewhat easier to perform. The approximation is analogous
to the paraxial approximation made for rays in chapter 9. Thus, the Fresnel
approximation requires the avoidance of large angles with respect to the z-axis
Besides letting the obliquity factor be one, Fresnel approximated R by the
distance z in the denominator of (10.10) . Then the denominator can be brought
out in front of the integral since it no longer depends on x 0 and y 0 . This is valid to
the extent that we restrict ourselves to small angles:
R∼
=z (denominator only; Fresnel approximation) (10.11)
3 J. W. Goodman, Introduction to Fourier Optics, Sect. 4-1 (New York: McGraw-Hill, 1968).
10.3 Fresnel Approximation 263
The above approximation is wholly inappropriate in the exponent of (10.10) since

small changes in R can result in dramatic variations in the periodic function e i kR .
To approximate R in the exponent, we must proceed with caution. To this end
we expand (10.2) under the assumption z 2 À (x − x 0 )2 + (y − y 0 )2 . Again, this is
consistent with the idea of restricting ourselves to relatively small angles. The
expansion of (10.2) is written as
s ¢2 ¢2 ¡ ¢2
(x − x 0 )2 + y − y 0 ∼
" #
x − x0 + y − y 0
¡ ¡
R = z 1+ = z 1+ +···
z2 2z 2
(exponent; Fresnel approximation) (10.12)
Substitution of (10.11) and (10.12) into the Fresnel diffraction formula (10.1)
yields
k
ie i kz i 2z
e (x 2 +y 2 ) Ï
¢ k 02 02 k 0 0
E x, y, z ∼ E x 0 , y 0 , 0 e i 2z (x +y ) e −i z (xx +y y ) d x 0 d y 0
¡ ¢ ¡
=−
λz
aperture
(Fresnel approximation) (10.13)
This is Fresnel’s approximation to his diffraction integral formula. It may look a bit
messier than before, but in terms of being able to make progress on integration
we are better off than previously. Notice that the integral can be interpreted as a
¢ k 02 02
two-dimensional Fourier transform on E x 0 , y 0 , 0 e i 2z (x +y ) .
¡
Example 10.3
Compute the Fresnel diffraction field following a rectangular aperture (dimensions
∆x by ∆y) illuminated by a uniform plane wave.
Solution: According to (10.13), the field down stream is
∆x/2 ∆y/2
e i kz i k (x 2 +y 2 )
Z Z
k 02 ky
0 i 2z x −i kx
z x
0 k 02 y0
d y 0 e i 2z y e −i
¡ ¢
E x, y, z = −i E 0 e 2z dx e e z
λz
−∆x/2 −∆y/2
Unfortunately, the integration in the preceding example must be performed

numerically. This is often the case for diffraction integrals in the Fresnel approx-
imation. Figure 10.7 shows the result of such an integration for a rectangular
aperture with a height twice its width.
Figure 10.7 Field amplitude fol-
Paraxial Wave Equation lowing a rectangular aperture
computed in the Fresnel approxi-
If we assume that the light coming through the aperture is highly directional, such mation.
that it propagates mainly in the z-direction, we are motivated to write the field
as E (x, y, z) = Ẽ (x, y, z)e i kz . Upon substitution of this into the scalar Helmholtz
equation (10.5), we arrive at
∂2 Ẽ ∂2 Ẽ ∂Ẽ ∂2 Ẽ
+ + 2i k + =0 (10.14)
∂x 2 ∂y 2 ∂z ∂z 2
2
At this point we make the paraxial wave approximation,4 which is |2k ∂∂zẼ | À | ∂∂zẼ2 |.
That is, we assume that the amplitude of the field varies slowly in the z-direction
such that the wave looks much like a plane wave. We permit the amplitude to
change as the wave propagates in the z-direction as long as it does so on a scale
much longer than a wavelength. This leads to the paraxial wave equation:
∂2 ∂2 ∂
µ ¶
+ + 2i k Ẽ (x, y, z) ∼
=0 (paraxial wave equation) (10.15)
∂x 2 ∂y 2 ∂z
It turns out that the Fresnel approximation (10.13) is an exact solution to the
paraxial wave equation. As demonstrated in problem P10.5, (10.15) is satisfied by
Ï∞ h i
i 2
i k (x−x 0 ) +( y−y 0 )
2
Ẽ (x, y, z) ∼
=− Ẽ (x 0 , y 0 , 0)e 2z d x0d y 0 (10.16)
λz
−∞
When the factor e i kz is appended, this field is identical to (10.13).
Joseph von Fraunhofer (17871826, 10.4 Fraunhofer Approximation

German) was born in Straubing, Bavaria.
He was orphaned at age 11, whereupon An additional approximation to the diffraction integral was made famous by
he was apprenticed to a glassmaker.
The workshop collapsed, trapping him
Joseph von Fraunhofer. The Fraunhofer approximation is the limiting case of
in the rubble. The Prince of Bavaria the Fresnel approximation when the field is observed at a distance far after the
directed the rescue eorts and thereafter
aperture (called the far field). A diffraction pattern continuously evolves along the
took an interest in Fraunhofer's educa-
tion. The prince required the glassmaker z-direction, as described by the Fresnel approximation. Eventually it evolves into
to allow young Joseph time to study, a final diffraction pattern that maintains itself as it continues to propogate (al-
and he naturally took an interest in
optics. Fraunhofer later worked at the
though it increases its size in proportion to distance). It is this far-away diffraction
Optical Institute at Benediktbeuern, pattern that is obtained from the Fraunhofer approximation. Since the Fresnel
where he learned techniques for making
the nest optical glass in his day. Fraun-
approximation requires the angles to be small (i.e. the paraxial approximation),
hofer developed numerous glass recipes so does the Fraunhofer approximation.
and was expert at creating optical de-
In many textbooks, the Fraunhofer approximation is presented first because
vices. Fraunhofer was the inventor of
the spectroscope, making it possible to the formula is easier to use. However, since it is a special case of the Fresnel
do quantitative spectroscopy. Using his approximation, it logically should be discussed afterwards as we are doing here.
spectroscope, Fraunhofer was the rst
to observe and document hundreds of To obtain the diffraction pattern very far after the aperture, we make the following
absorption lines in the sun's spectrum. approximation:5
He also noticed that these varied for k 02 02
dierent stars, thus establishing the eld e i 2z (x +y ) ∼
=1 (far field) (10.17)
of stellar spectroscopy. He was also the
inventor of the diraction grating. In The validity of this approximation depends on a comparison of the size of the
1822, he was granted an honorary doc- aperture to the distance z where the diffraction pattern is observed. We need
torate from the University of Erlangen.
Fraunhofer passed away at age 39, per-
haps due to heavy-metal poisoning from
k¡ ¢2
zÀ aperture radius (condition for far field) (10.18)
glass blowing. (Wikipedia) 2
4 P. W. Milonni and J. H. Eberly, Laser, Sect. 14.4 (New York: Wiley, 1988).
5 J. W. Goodman, Introduction to Fourier Optics, p. 61 (New York: McGraw-Hill, 1968).
10.5 Diffraction with Cylindrical Symmetry 265
By removing the factor (10.17) from (10.13), we obtain the Fraunhofer diffrac-
tion formula:
k 2 2
i e i kz e i 2z (x +y )
Ï
k 0 0
∼ E x 0 , y 0 , 0 e −i z (xx +y y ) d x 0 d y 0
¡ ¢ ¡ ¢
E x, y, z = − (10.19) (Fraunhofer approximation)
λz
aperture
k 02 02
Obviously, the removal of e i 2z (x +y ) from the integrand improves our chances of
being able to perform the integration. Notice that the integral can now be inter-
preted as a two-dimensional Fourier transform on the aperture field E x 0 , y 0 , 0 .
¡ ¢
Once we are in the Fraunhofer regime, a change in z is not very interesting

since it appears in the combination x/z or y/z inside the integral. At a larger
distance z, the same diffraction pattern is obtained with a proportionately larger
value of x or y. The Fraunhofer diffraction pattern thus preserves itself indefinitely
as the field propagates. It grows in size as the distance z increases, but the angular
size defined by x/z or y/z remains the same.
Example 10.4
Compute the Fraunhofer diffraction pattern following a rectangular aperture (di-
mensions ∆x by ∆y) illuminated by a uniform plane wave.

∆x/2 ∆y/2
e i kz i k (x 2 +y 2 )
Z Z
ky
0 −i kx
z x
0 y0
d y 0 e −i
¡ ¢
E x, y, z = −i E 0 e 2z dx e z
λz
−∆x/2 −∆y/2
It is left as an exercise (see P10.8) to perform the integration and compute the
Figure 10.8 Fraunhofer diffraction
intensity. The result turns out to be
pattern (field amplitude) gener-
∆x 2 ∆y 2 2 π∆x 2 π∆y
ated by a uniformly illuminated
µ ¶ µ ¶
¡ ¢
I x, y, z = I 0 sinc x sinc y (10.20) rectangular aperture with a height
λ2 z 2 λz λz
twice the width.
where sincξ ≡ sin ξ/ξ. Note that lim sincξ = 1.
ξ→0
10.5 Diffraction with Cylindrical Symmetry

Sometimes the field transmitted by an aperture is cylindrically symmetric. In this
case, the field at the aperture can be written as
E (x 0 , y 0 , z = 0) = E (ρ 0 , z = 0) (10.21)
where ρ ≡ x 2 + y 2 . Under cylindrical symmetry, the two-dimensional integra-

p
tion over x 0 and y 0 in (10.13) or (10.19) can be reduced to a single-dimensional

integral over a cylindrical coordinate ρ 0 . With the coordinate transformation
x ≡ ρ cos φ y ≡ ρ sin φ x 0 ≡ ρ 0 cos φ0 y 0 ≡ ρ 0 sin φ0 (10.22)

the Fresnel diffraction integral (10.13) becomes

z = 25/k
kρ 2 Z2π
i e i kz e i ¢ kρ02
Z
2z k
ρ 0 d ρ 0 E ρ 0 , 0 e i 2z e −i z (ρρ cos φcos φ +ρρ sin φsin φ )
0 0 0 0
E ρ, z = − d φ0
¡ ¢ ¡
λz
0 aperture
(10.23)
100/k
Notice that in the exponent of (10.23) we can write
ρ 0 ρ cos φ0 cos φ + sin φ0 sin φ = ρ 0 ρ cos φ0 − φ

¡ ¢ ¡ ¢
(10.24)
500/k
With this simplification, the diffraction formula (10.23) can be written as
z = 75/k
kρ 2 Z2π
i e i kz e i ¢ kρ02 kρρ 0
Z
2z
d φ0 e −i z cos(φ−φ )
0
E ρ, z = − ρ d ρ E ρ 0 , 0 e i 2z
0 0
¡ ¢ ¡
(10.25)
λz
aperture 0
We are able to perform the integration over φ0 with the help of the formula (0.57):
Z2π
kρρ 0
µ ¶
kρρ 0
−i z cos(φ−φ0 ) 0
e d φ = 2πJ 0 (10.26)
500/k z
0
z = 200/k J 0 is called the zero-order Bessel function. Equation (10.25) then reduces to
kρ 2
2πi e i kz e i kρρ 0
µ ¶
¡ 0 ¢ i kρ02
Z
2z
0 0
E ρ, z = − ρ d ρ E ρ , 0 e 2z J 0
¡ ¢
λz z
aperture
(Fresnel approximation with cylindrical symmetry) (10.27)
¢ kρ02
The integral in (10.27) is called a Hankel transform on E ρ 0 , 0 e i 2z .
¡
500/k In the case of the Fraunhofer approximation, the diffraction ³ integral becomes
kρ 02
´
a Hankel transform on just the field E ρ , z = 0 since exp i 2z goes to one.
¡ 0 ¢
z = 1000/k
Under cylindrical symmetry, the Fraunhofer approximation is
kρ 2
2πi e i kz e i kρρ 0
Z µ ¶
2z
0 0
¡ 0 ¢
E ρ, z = − ρ d ρ E ρ , 0 J0
¡ ¢
λz z
aperture
(Fraunhofer approximation with cylindrical symmetry) (10.28)
Just as fast Fourier transform algorithms aid in the numerical evaluation of diffrac-
500/k tion integrals in Cartesian coordinates, fast Hankel transforms also exist and can
be used with cylindrically symmetric diffraction integrals.
Figure 10.9 Field amplitude fol-
lowing a circular aperture com-
puted in the Fresnel approxima- Example 10.5
tion. Compute the Fresnel and Fraunhofer diffraction patterns following a circular
aperture (diameter `) illuminated by a uniform plane wave.

kρ 2 Z`/2
2πe i kz e i kρ 02 kρρ 0
µ ¶
2z
E ρ, z = −i E 0 ρ 0 d ρ 0 e i 2z J 0
¡ ¢
λz z
0
10.A Fresnel-Kirchhoff Diffraction Formula 267
Unfortunately, this Fresnel integral must be performed numerically. The result

of the calculation for a uniform field illuminating a circular aperture is shown in
Fig. 10.9.
On the other hand, the field in the Fraunhofer limit (10.28) is
kρ 2 Z`/2
2πe i kz e i kρρ 0
µ ¶
2z
E ρ, z = −i E 0 ρ0d ρ0 J0
¡ ¢
λz z
0
which can be integrated analytically. It is left as an exercise to perform the integra- !

tion and to show that the intensity of the Fraunhofer pattern is
µ 2 ¶2 · ¢ ¸2
π`
¡
J 1 k`ρ/2z
I ρ, z = I 0
¡ ¢
2 ¡ ¢ (10.29)
4λz k`ρ/2z
2J (ξ) Figure 10.10 Fraunhofer diffrac-

The function 1ξ (sometimes called the jinc function) looks similar to the sinc
function (see Example 10.4) except that its first zero is at ξ = 1.22π rather than at π. tion pattern (field amplitude) gen-
2J (ξ) erated for a uniformly illuminated
Note that lim 1ξ = 1.
ξ→0 circular aperture.
Appendix 10.A Fresnel-Kirchhoff Diffraction Formula

To begin the derivation of the Fresnel-Kirchhoff diffraction formula,6 we employ
Green’s theorem (proven in appendix 10.B):
∂V ∂U
I · ¸ Z
U ∇2V − V ∇2U d v
£ ¤
U −V da = (10.30)
∂n ∂n
S V
The notation ∂/∂n implies a derivative in the direction normal to the surface. We
choose the following functions:
V ≡ e i kr /r
(10.31)
U ≡ E (r)
where E (r) is assumed to satisfy the scalar Helmholtz equation, (10.5). When
these functions are used in Green’s theorem (10.30), we obtain
I " # Z " #
∂ e i kr e i kr ∂E 2e
i kr
e i kr 2
E − da = E∇ − ∇ E dv (10.32)
∂n r r ∂n r r
S V
The right-hand side of this equation vanishes7 since we have
e i kr e i kr 2 e i kr e i kr 2
E ∇2 − ∇ E = −k 2 E + k E =0 (10.33)
r r r r
6 See J. W. Goodman, Introduction to Fourier Optics, Sect. 3-3 (New York: McGraw-Hill, 1968).
7 We exclude the point r = 0; see P0.4 and P0.5.
where we have taken advantage of the fact that E (r) and e i kr /r both satisfy (10.5).
This is exactly the reason for our judicious choices of the functions V and U since
with them we were able to make half of (10.30) disappear. We are left with
I " #
∂ e i kr e i kr ∂E
E − da = 0 (10.34)
∂n r r ∂n
S
Now consider a volume between a small sphere of radius ² at the origin and an
outer surface of whatever shape. The total surface that encloses the volume is
comprised of two parts (i.e. S = S 1 + S 2 as depicted in Fig. 10.11).
When we apply (10.34) to the surface in Fig. 10.11, we have
I " # I " #
∂ e i kr e i kr ∂E ∂ e i kr e i kr ∂E
E − da = − E − da (10.35)
∂n r r ∂n ∂n r r ∂n
S2 S1
Our motivation for choosing this geometry with multiple surfaces is that eventu-
ally we want to find the field at the origin (inside the little sphere) from knowledge
of the field on the outside surface. To this end, we assume that ² is so small that
E (r) is approximately the same everywhere on the surface S 1 . Then the integral
over S 1 becomes
Figure 10.11 A two-part surface
enclosing volume V . I " # Z2π Zπ " Ã ! #
∂ e i kr e i kr ∂E ∂ e i kr ∂r e i kr ∂E ∂r
µ ¶
E − d a = lim dφ E − r 2 sin φd φ
∂n r r ∂n r =²→0 ∂r r ∂n r ∂r ∂n
S1 0 0
(10.36)
where we have used spherical coordinates. Notice that we have employed the
chain rule to execute the normal derivative ∂/∂n. Since r always points opposite
to the direction of the surface normal n̂, the normal derivative ∂r /∂n is always
equal to −1.8 We can now perform the integration in (10.36) as well as take the
limit as ² → 0 to obtain
I " # " Ã ! ¶#
∂ e i kr e i kr ∂E 2 e i kr e i kr 2e
i kr µ
∂E
lim E − d a = −4π lim r − 2 + i k E −r
²→0 ∂n r r ∂n ²→0 r r r ∂r
S1 r =²
∂E
·³ ´ µ ¶ ¸
= −4π lim −e i k² + i k²e i k² E − e i k² ²
²→0 ∂r r =²
= 4πE (0)
(10.37)
With the aid of (10.37), Green’s theorem applied to our specific geometry
reduces to I " i kr #
1 e ∂E ∂ e i kr
E (0) = −E da (10.38)
4π r ∂n ∂n r
S2
8 From the definition of the normal derivative we have ∂r /∂n ≡ ∇r · n̂ = −n̂ · n̂ = −1.
10.A Fresnel-Kirchhoff Diffraction Formula 269
If we know E everywhere on the outer surface S 2 , this equation allows us to predict

the field E (0) at the origin. Of course we are free to choose any coordinate system
in order to find the field anywhere inside the surface S 2 , by moving the origin.
Now let us choose a specific surface S 2 . Consider an infinite mask with a finite
aperture connected to a hemisphere of infinite radius R → ∞. In the end, we
will suppose that light that enters through the mask and propagates to our origin
(among other points). In our present coordinate system, the vectors r and n̂ point
opposite to the incoming light.
We must evaluate (10.38) on the surface depicted in the figure. For the portion
of S 2 which is on the hemisphere, the integrand tends to zero as R becomes large. mask
To argue this, it is necessary to recognize the fact that at large distances the field
takes on a form proportional to e i kR /R so that the two terms in the integrand
cancel. On the mask, we assume, as did Kirchhoff, that both ∂E /∂n and E are aperture
zero.9 Thus, we are left with only the integration over the open aperture:
origin
Ï " i kr #
1 e ∂E ∂ e i kr
E (0) = −E da (10.39)
4π r ∂n ∂n r
aperture
We have essentially arrived at the result that we are seeking. The field coming Figure 10.12 Surface S 2 depicted
through the aperture is integrated to find the field at the origin, which is located as a mask and a large hemisphere.
beyond the aperture. Let us manipulate the formula a little further. The second
term in the integral of (10.39) can be rewritten as follows:
Ã !
∂ e i kr ∂ e i kr ∂r ik i ke i kr
µ ¶
1 i kr
= = − 2 e cos (r, n̂) → cos (r, n̂) (10.40)
∂n r ∂r r ∂n r r r Àλ r
where ∂r /∂n = cos (r, n̂) indicates the cosine of the angle between r and n̂. We
have also assumed that the distance r is much larger than a wavelength in order
to drop a term. Next, we assume that the field illuminating the aperture can be
written as E ∼
= Ẽ x, y e i kz . This represents a plane-wave field traveling through
¡ ¢
the aperture from left to right. Then, we have
∂E ∂E ∂z
= i k Ẽ x, y e i kz (−1) = −i kE
¡ ¢
= (10.41)
∂n ∂z ∂n
Substituting (10.40) and (10.41) into (10.39) yields
i e i kr
· ¸
1 + cos (r, n̂)
Ï
E (0) = − E da (10.42)
λ r 2
aperture
Finally, we wish to rearrange our coordinate system to that depicted in Fig. 10.2.
In our derivation, it was less cumbersome to place the origin at a point after the
9 Later Sommerfeld noticed that these two assumptions actually contradict each other, and he
revised Kirchhoff’s work to be more accurate. In practice this revision makes only a tiny difference
as light spills onto the back of the aperture, over a length scale of a wavelength. We will ignore
this effect and go with Kirchhoff’s (slightly flawed) assumption. For further discussion see J. W.
Goodman, Introduction to Fourier Optics, Sect. 3-4 (New York: McGraw-Hill, 1968).
aperture. Now that we have completed our mathematics, it is convenient to make

a change of coordinate system and move the origin to the plane of the aperture as
in Fig. 10.2. Then, we can obtain the field at a point lying somewhere after the
aperture by computing
i ¢ e i kR 1 + cos (r, ẑ)

Ï · ¸
E x 0, y 0, z = 0 d x 0d y 0
¡ ¢ ¡
E x, y, z = d = − (10.43)
λ R 2
aperture
where q
¢2
(x − x 0 )2 + y − y 0 + d 2
¡
R= (10.44)
Equation (10.10) is the same as (10.42) after applying a coordinate transformation.
It is called the Fresnel-Kirchhoff diffraction formula and it agrees with (10.1)
except for the obliquity factor [1 + cos (r, ẑ)]/2.
Appendix 10.B Green’s Theorem

To derive Green’s theorem, we begin with the divergence theorem (see (0.11)):
I Z
f · n̂ d a = ∇ · f d v (10.45)
S V
The unit vector n̂ always points normal to the surface of volume V over which
the integral is taken. Let the vector function f be U ∇V , where U and V are both
analytical functions of the position coordinate r. Then (10.45) becomes
I Z
(U ∇V ) · n̂ d a = ∇ · (U ∇V ) d v (10.46)
S V
We recognize ∇V · n̂ as the directional derivative of V , directed along the surface

normal n̂. This is often represented in shorthand notation as
∂V
∇V · n̂ ≡ (10.47)
∂n
The argument of the integral on the right-hand side of (10.46) can be expanded
with the chain rule:
∇ · (U ∇V ) = ∇U · ∇V +U ∇2V (10.48)
With these substitutions, (10.46) becomes
∂V
I Z
∇U · ∇V +U ∇2V d v
£ ¤
U da = (10.49)
∂n
S V
Actually, so far we haven’t done much. Equation (10.49) is nothing more than the
divergence theorem applied to the vector function U ∇V . Similarly, we can apply
the divergence theorem to an alternative vector function given by the reverse
10.B Green’s Theorem 271
combination V ∇U . Thus, we can write an equation similar to (10.49) where U

and V are interchanged:
∂U
I Z
∇V · ∇U + V ∇2U d v
£ ¤
V da = (10.50)
∂n
S V
We subtract (10.50) from (10.49), and this leads to (10.30) known as Green’s theo-
rem.
Exercises
Exercises for 10.1 Huygens’ Principle as Formulated by Fresnel
P10.1 Huygens’ principle is often used to describe diffraction through a slits,

but it can be also used to describe refraction. Use a drawing program
or a ruler and compass to produce a picture similar to Fig. 10.13, which
shows that the graphical prediction of refracted angle from the Huy-
gens’ principle. Verify that the Huygens picture matches the numerical
prediction from Snell’s Law for an incident angle of your choice. Use
n i = 1 and n t = 2.
HINT: Draw the wavefronts hitting the interface at an angle and treat
each point where the wavefronts strike the interface as the source of
circular waves propagating into the n = 2 material. The wavelength of
Figure 10.13
the circular waves must be exactly half the wavelength of the incident
light since λ = λvac /n. Use at least four point sources and connect the
matching wavefronts by drawing tangent lines as in the figure.
P10.2 (a) Show that the function
A
f (r ) = cos (kr − ωt )
r
is a solution to the wave equation in spherical coordinates with only
radial dependence,
1 ∂ 2 ∂f 1 ∂2 f
µ ¶
r =
r 2 ∂r ∂r v 2 ∂t 2
Determine what v is, in terms of k and ω.

(b) If the electric field were a scalar field, we might be done there.
However, it’s a vector field, and moreover it must satisfy Maxwell’s
equations. We know from experience that it’s generally transverse, and
since it’s traveling radially let’s make a guess that it’s oscillating in the φ̂
direction:
A
E(r ) = cos (kr − ωt ) φ̂
r
Show that this choice for E is not consistent with Maxwell’s equations.
In particular: (i) show that it does satisfy Gauss’s Law (1.1); (ii) compute
the curl of E use Faraday’s Law (1.3) to deduce B; (iii) Show that this B
does satisfy Gauss’s Law for magnetism (1.2); (iv) but this B it does not
satisfy Ampere’s law (1.4).
(c) A somewhat more complicated ‘spherical’ wave
A sin φ
· ¸
1
E(r, φ) = cos (kr − ωt ) − sin (kr − ωt ) φ̂
r kr
Exercises 273
does satisfy Maxwell’s equations. Describe how this wave behaves as

a function of r and φ. What conditions need to be satisfied for this
equation to reduce to the spherical wave formula used in the diffraction
formulas?
Exercises for 10.2 Scalar Diffraction Theory
P10.3 Show that E (r ) = E 0 r 0 e i kr /r is a solution to the scalar Helmholtz equa-

tion (10.5).
HINT:
1 ∂2 ¡ ¢ ∂ ∂ψ ∂2 ψ
µ ¶
2 1 1
∇ ψ= r ψ + sin θ +
r ∂r 2 r 2 sin θ ∂θ ∂θ r 2 sin2 θ ∂φ2
P10.4 Learn by heart the derivation of the Fresnel-Kirchhoff diffraction for-

mula (outlined in Appendix 10.A). Indicate the percentage of how well
you understand the derivation. If you write 100% percent, it means
that you can reproduce the derivation after closing your notes.
P10.5 Check that (10.16) is the solution to the paraxial wave equation (10.15).
Exercises for 10.4 Fraunhofer Approximation
P10.6 (a) Repeat Example 10.1 to find the on-axis intensity after a circular
aperture in both the Fresnel and Fraunhofer approximations. (HINT:
Use (10.27) and (10.28) to obtain the fields ρ = 0.) Also make suitable
approximations directly to (10.3) to obtain the same answers.
(b) Check how well the Fresnel and Fraunhofer approximations work
by graphing the three curves (i.e. (10.3) and the curves obtained in part
(a)) on a single plot as a function of z. Take ` = 10 µm and λ = 500 nm.
To see the result better, use a log scale on the z-axis.
4
3 Fresnel
Fraunhofer
2
1 Fresnel-Kirchoff
Figure 10.14 “The Fraunhofer Ap-
proximation” by Sterling Cornaby
0
-3 -2 -1 0
10 10 10 10
z (mm)
Figure 10.15
L10.7 (a) Why does the on-axis intensity behind a circular opening fluctuate
(see Example 10.1) whereas the on-axis intensity behind a circular
obstruction remains constant (see Example 10.2)?
(b) Create a collimated laser beam several centimeters wide. Observe
the on-axis intensity on a movable screen (e.g. a hand-held card) be-
hind a small circular aperture and behind a small circular obstruction
placed in the beam. (video)
(c) In the case of the circular aperture, measure the distance to several
on-axis minima and check that it agrees with prediction. (See problem
P10.6.)
Laser
Figure 10.16
P10.8 Calculate the Fraunhofer diffraction field and intensity patterns for a
rectangular aperture (dimensions ∆x by ∆y) illuminated by a plane
wave E 0 . In other words, derive (10.20).
P10.9 A single narrow slit has a mask placed over it so the aperture function
is not a square pulse but rather a cosine: E (x 0 , y 0 , 0) = E 0 cos(x 0 /L) for
−L/2 < x 0 < L/2 and E (x 0 , y 0 , 0) = 0 otherwise. Calculate the far-field
(Fraunhofer) diffraction pattern. Make a plot of intensity as a function
of xkL/2z; qualitatively compare the pattern to that of a regular single
slit.
Exercises for 10.5 Diffraction with Cylindrical Symmetry
P10.10 Calculate the Fraunhofer diffraction intensity pattern (10.29) for a cir-
cular aperture (diameter `) illuminated by a plane wave E 0 .
Chapter 11
Diffraction Applications
In this chapter, we consider a number of practical examples of diffraction. We first

discuss diffraction theory in systems involving lenses. The Fraunhofer diffraction
pattern discussed in section 10.4, applicable in the far-field limit, is imaged to the
focus of a lens when the lens is placed in the stream of light. This has important
implications for the resolution of instruments such as telescopes or the human
eye.
The array theorem, which applies to Fraunhofer limit, is introduced in sec-
tion 11.3. This theorem is a powerful mathematical tool that enables one to deal
conveniently with diffraction from an array of identical apertures. One of the
important uses of the array theorem is in determining Fraunhofer diffraction
from a grating, since a diffraction grating can be thought of as an array of narrow
slit apertures. In section 11.5, we study the workings of a diffraction spectrometer.
To find the resolution limitations, one combines the diffraction properties of
gratings with the Fourier properties of lenses.
Finally, we consider a Gaussian laser beam to understand its focusing and
diffraction properties. The information presented here comes up remarkably
often in research activity. We often think of lasers as collimated beams of light
that propagate indefinitely without expanding. However, the laws of diffraction
require that every finite beam eventually grow in width. The rate at which a laser
beam diffracts depends on its beam waist size. Because laser beams usually have
narrow divergence angles and therefore obey the paraxial approximation, we can
calculate their behavior via the Fresnel approximation discussed in section 10.3.
Appendix 11.A discusses the ABCD law for Gaussian beams, which is a method
of computing the effects of optical elements represented by ABCD matrices on
Gaussian laser beams.
11.1 Fraunhofer Diffraction Through a Lens

The Fraunhofer limit corresponds to the ultimate amount of diffraction that
light in an optical system experiences. As has been previously discussed, the
Fraunhofer approximation applies to diffraction when the propagation distance
275
276 Chapter 11 Diffraction Applications
from an aperture is sufficiently large (see (10.18) and (10.19)). Mathematically, it

is obtained via a two-dimensional Fourier transform. The intensity of the far-field
diffraction pattern is
¯ ¯2
¯ ¯
¢ 1 1
Ï
¡ ¯ ¡ 0 0 ¢ −i k x x 0 + y y 0
¡ ¢
0 0¯
¯
I x, y, z = c²0 ¯
¯ E x , y ,0 e z z dx dy ¯ (11.1)
2 ¯ λz ¯
aperture
Notice that the dependence of the diffraction on x, y, and z comes only

through the combinations θx ∼ = x/z and θ y ∼ = y/z. Therefore, the diffraction
pattern in the Fraunhofer limit is governed by the two angles θx and θ y , and
the pattern preserves itself indefinitely. As the light continues to propagate, the
pattern increases in size at a rate proportional to distance traveled so that the
angular width is preserved. The situation is depicted in Fig. 11.1.
Recall that in order to use the Fraunhofer diffraction formula we need to
¢2
satisfy z À π aperture radius /λ (see (10.18)). As an example, if an aperture
¡
with a 1 cm radius (not necessarily circular) is used with visible light, the light
must travel more than a kilometer in order to reach the Fraunhofer limit. It
may therefore seem unlikely to reach the Fraunhofer limit in a typical optical
Figure 11.1 Diffraction in the far system, especially if the aperture or beam size is relatively large. Nevertheless,
field. spectrometers, which typically utilize diffraction gratings many centimeters wide,
depend on achieving the Fraunhofer limit within the confines of a manageable
instrument box. This is accomplished using imaging techniques. The Fraunhofer
limit is also important to the performance of other optical instruments that use
lenses (e.g. a telescope).
Consider a lens with focal length f placed in the path of light following an
aperture (see Fig. 11.2). Let the lens be placed an arbitrary distance L after the
aperture. The lens produces an image of the Fraunhofer pattern at a new location
d i following the lens according to the imaging formula (see (9.55))
1 1 1
= + . (11.2)
f − (z − L) d i
Keep in mind that the lens interrupts the light before the Fraunhofer pattern
has a chance to form. This means that the Fraunhofer diffraction pattern may
Image of the pattern that

would have appeared at
infinity
Figure 11.2 Imaging of the Fraunhofer diffraction pattern to the focus of a lens.
11.1 Fraunhofer Diffraction Through a Lens 277
be thought of as a virtual object a distance z − L to the right of the lens. Since

the Fraunhofer diffraction pattern occurs at very large distances (i.e. z → ∞) the
image of the Fraunhofer pattern appears at the focus of the lens:
di ∼
= f. (11.3)
Thus, a lens makes it very convenient to observe the Fraunhofer diffraction pat-
tern even from relatively large apertures. It is not necessary to let the light propa-
gate for kilometers. We need only observe the pattern at the focus of the lens as
shown in Fig. 11.2. Notice that the spacing L between the aperture and the lens is
unimportant to this conclusion.
Even though we know that the Fraunhofer diffraction pattern occurs at the
focus of a lens, the question remains as to the size of the image. To find the answer,
let us examine the magnification (9.56), which is given by
di
M =− (11.4)
− (z − L)
Taking the limit of very large z and employing (11.3), the magnification becomes
f
M→ (11.5)
z
This is a remarkable result. When the lens is inserted, the size of the diffraction
pattern decreases by the ratio of the lens focal length f to the original distance
z to a far-away screen. Since in the Fraunhofer regime the diffraction pattern is
proportional to distance (i.e. si ze ∝ z), the image at the focus of the lens scales
in proportion to the focal length (i.e. si ze ∝ f ). This means that the angular
width of the pattern is preserved! With the lens in place, we can rewrite (11.1)
straightaway as
¯ ¯2
¯ ¯
¢ 1 ¯ 1
Ï
¡ 0 0 ¢ −i k (xx 0 +y y 0 ) 0 0 ¯
∼
¡
I x, y, L + f = c²0 ¯¯ E x , y ,0 e f d x d y ¯¯ (11.6)
2 ¯λf ¯
aperture
which describes the intensity distribution pattern at the focus of the lens.
Although (11.6) correctly describes the intensity, we cannot easily write the
electric field since the imaging techniques that we have used do not render the
phase information. To obtain an expression for the field, it will be necessary to
employ the Fresnel diffraction formula. In addition, we need to know how a lens
adjusts the phase fronts of the light passing through it.
Phase Front Alteration by a Lens
Consider a monochromatic light field that goes through a thin lens with focal
length f . In traversing the lens, the wavefront undergoes a phase shift that varies
across the lens. We will reference the phase shift to that experienced by the light
that goes through the center of the lens. In the Fig. 11.3, R 1 is a positive radius
of curvature, and R 2 is a negative radius of curvature, according to our previous

convention. We take the distances `1 and `2 , as drawn, to be positive.
The light passing through the off-axis portion of the lens experiences less material
than the light passing through the center. The difference in optical path length is
(n − 1) (`1 + `2 ) (see discussion connected with (9.14)). This means that the phase
of the field passing through the off-axis portion of the lens relative to the phase of
the field passing through the center is
∆φ = −k (n − 1) (`1 + `2 ) . (11.7)
The negative sign indicates a phase advance (i.e. same sign as −ωt ). Since the
Figure 11.3 A thin lens, which
off-axis light travels through less material, the phase of the wave front gets ahead
modifies the phase of a field pass-
of the light traveling through the center of the lens. In (11.7), k represents the wave
ing through.
number in vacuum (i.e. 2π/λvac ); since `1 and `2 correspond to distances outside
of the lens material.
We can find expressions for `1 and `2 from the equations describing the spherical
surfaces of the lens:
(R 1 − `1 )2 + x 2 + y 2 = R 12
(11.8)
(R 2 + `2 )2 + x 2 + y 2 = R 22
In the Fresnel approximation, which takes place in the paraxial limit, it is appro-
priate to neglect the terms `21 and `22 in comparison with the other terms present.
Within this approximation, equations (11.8) become
x2 + y 2 x2 + y 2
`1 ∼
= and `2 ∼
=− (11.9)
2R 1 2R 2
Substitution into (11.7) yields
x2 + y 2
¶¡ ¢
k ¡ 2
µ
1 1
∆φ = −k (n − 1) x + y2
¢
− =− (11.10)
R1 R2 2 2f
where the focal length of a thin lens f has been introduced according to the lens-
maker’s formula (9.46).
In summary, the light traversing a lens experiences a relative phase shift given by
¢ −i k x 2 +y 2 )
E x, y, z after lens = E x, y, z before lens e 2 f (
¡ ¢ ¡
(11.11)
Equation (11.11) introduces a wave-front curvature to the field. For example, if a

plane wave (i.e. a uniform field E 0 ) passes through the lens, the field emerges with
a spherical-like wave front converging towards the focus of the lens.
Figure 11.4 The phase fronts of a
plane wave are bent as they pass
through a lens. We compute the diffraction pattern after the lens in three steps, as illustrated
in Fig. 11.5. First, we use the Fresnel diffraction formula to compute the field
arriving at the lens. Second, we adjust the phase front of the light passing through
the lens according to (11.11). Third, we use the field exiting the lens as the input
for a second Fresnel diffraction integral to find the field at the lens focus. The
result gives an intensity pattern in agreement with (11.6). It also provides the full
expression for the field, including its phase.
11.2 Resolution of a Telescope 279
Starting from the known field E x 0 , y 0 , 0 at the aperture, we compute the field
¡ ¢
incident on the lens using the Fresnel approximation:

k 002
+y 002 )
e i kL e i 2L (x
Ï
k 02
+y 02 ) −i kL (x 00 x 0 +y 00 y 0 )
E (x 00 , y 00 , L) = −i E (x 0 , y 0 , 0)e i 2L (x e d x 0d y 0
λL
(11.12)
(The double primes keep track of distinct variables in sequential diffraction
integrals.) As mentioned, the field gains a phase factor according to (11.11) upon
transmitting through the lens. Finally, we use the Fresnel diffraction formula a
second time to propagate the distance f from the back of the thin lens:
i 2kf (x 2 +y 2 )
eik f e
Ï h
−i k x 002 +y 002 )
E (x 00 , y 00 , L)e 2 f (
¡ ¢ i
E x, y, L + f = −i
λf
i 2kf (x 002 +y 002 ) −i kf (xx 00 +y y 00 )
×e e d x 00 d y 00 (11.13)
As you can probably appreciate, the injection of (11.12) into (11.13) makes a
rather long formula involving four dimensions of integration. Nevertheless, two
of the integrals can be performed in advance of choosing the aperture (i.e. those
over x 00 and y 00 ). This is accomplished with the help of the integral formula (0.55)
(even though in this instance the real part of a is zero). After this cumbersome
work, (11.13) reduces to
Figure 11.5 Diffraction from an

kL 2 2
i 2kf (x 2 +y 2 ) −i 2 f 2 (x +y ) aperture viewed at the focus of a
e i k (L+ f ) e e
Ï k
−i ( xx 0 +y y 0 )d x0d y 0
E (x 0 , y 0 , 0)e
¡ ¢
E x, y, L + f = −i f
lens.
λf
(11.14)
Notice that at least the integration portion of this formula looks exactly like
the Fraunhofer diffraction formula! This happened even though in the preceding
discussion we did not at any time specifically make the Fraunhofer approximation.
The result (11.14) implies the intensity distribution (11.6) as anticipated. However,
the phase of the field is also revealed in (11.14).
In general, the field caries a wave front curvature as it passes through the
focal plane of the lens. In the special case L = f , the diffraction formula takes a
particularly simple form:
e 2i k f
Ï
−i kf (xx 0 +y y 0 )
E (x 0 , y 0 , L + f )¯L= f = −i E (x 0 , y 0 , 0)e d x 0d y 0
¯
(11.15)
λf
When the lens is placed at this special distance following the aperture, the Fraun-
hofer diffraction pattern viewed at the focus of the lens carries a flat wave front.
11.2 Resolution of a Telescope

In the previous section we learned that the Fraunhofer diffraction pattern appears
at the focus of a lens. This has important implications for telescopes and other
optical instruments. In essence, any optical instrument incorporates an aperture,

limiting the light that enters. If nothing else, the diameter of a lens itself acts
effectively as an aperture. The pupil of the human eye is an aperture that induces
a Fraunhofer diffraction pattern to occur at the retina. Cameras have irises which
aperture the light, again causing a Fraunhofer diffraction pattern to occur at the
image plane.
Of course, the focus of the lens is just where one needs to look in order to
see images of distant objects. The Fraunhofer pattern, which occurs at the focus,
represents the ultimate amount of diffraction caused by an aperture. This has the
effect of blurring out features in the image and limiting resolution. This illustrates
why it is impossible to focus light to a true point.
Figure 11.6 To resolve distinct im- Suppose you point a telescope at two distant stars. An image of each star is
ages at the focus of a lens, the an-
formed in the focal plane of the lens. The angular separation between the two
gular separation must exceed the
images (referenced from the lens) is the same as the angular separation between
width of the Fraunhofer diffraction
patterns. the stars.1 This is depicted in Fig. 11.6.
A resolution problem occurs when the Fraunhofer diffraction causes the
image of each star to blur by more than the angular separation between them.
In this case the two images cannot be resolved because they ‘bleed’ into one
another.
1
The Fraunhofer diffraction pattern from a circular aperture was computed
previously (see (10.29)). At the focus of a lens, this pattern becomes
µ 2 ¶2 · ¢ ¸2
π`
¡
J 1 k`ρ/2 f
0 I ρ, f = I 0
¡ ¢
2 ¡ ¢ (11.16)
4λ f k`ρ/2 f
where f , the focal length of the lens, takes the place of z in the diffraction formula.
-1 The parameter ` is its diameter of the lens. This intensity pattern contains the
first order Bessel function J 1 , which behaves somewhat like a sine wave as seen in
1 Fig. 11.7. The main differences are that the zero crossings are not exactly periodic
and the function slowly diminishes with larger arguments. The first zero crossing
(after x = 0) occurs at 1.22π.
The intensity pattern described by (11.16) contains the factor 2J 1 (ξ)/ξ, where
ξ represents the combination k`ρ/2 f . As noticed in Fig. 11.7, J 1 (ξ) goes to zero
at ξ = 0. Thus, we have a zero-divided-by-zero situation when evaluating 2J 1 (ξ)/ξ
at the origin. This is similar to the sinc function (i.e. sin (ξ)/ξ), which approaches
one at the origin. In fact, 2J 1 (ξ)/ξ is sometimes called the jinc function because it
0 also approaches one at the origin. The square of the jinc is shown in Fig. 11.7b.
0 2 3 4
This curve is proportional to the intensity described in (11.16). This pattern is
sometimes called an Airy pattern after Sir George Biddell Airy (English, 1801–1892)
1 In the thin-lens approximation, the ray from either star that traverses the center of the lens (i.e.
Figure 11.7 (a) First-order Bessel
function. (b) Square of the Jinc y = 0) maintains its angle:
function. ·
0
¸ ·
1 0
¸·
0
¸ ·
0
¸
= =
θ2 −1/ f 1 θ1 θ1
11.2 Resolution of a Telescope 281
who first described the pattern. As can be seen in Fig. 11.7b, the intensity quickly
drops at larger radii.
We now return to the question of whether the images of two nearby stars
as depicted in Fig. 11.6 can be distinguished. Since the peak in Fig. 11.7b is the
dominant feature in the diffraction pattern, we will say that the two stars are
resolved if the angle between them is enough to keep their respective diffraction
peaks from seriously overlapping. We will adopt the criterion suggested by Lord
Rayleigh that the peaks are distinguishable if the peak of one pattern is no closer
than the first zero to the other peak. This situation is shown in Fig. 11.8.
The angle that corresponds to this separation of diffraction patterns is found
by setting the argument of (11.16) equal to 1.22π, the location of the first zero:
k`ρ
= 1.22π (11.17)
2f
With a little rearranging we have
ρ 1.22λ
θmin ∼
= = (11.18)
f `
Here we have associated the ratio ρ/ f (i.e. the radius of the diffraction pattern
compared to the distance from the lens) with an angle θmin . The Rayleigh criterion
requires that the diffraction patterns be separated by at least this angle before we
say that they are resolved. Figure 11.8 The Rayleigh criterion
θmin depends on the diameter of the lens ` as well as on the wavelength of the for a circular aperture.
light. Since the angle between the images and the angle between the objects is
the same, θmin tells the minimum angle between objects that can be resolved with
a given instrument. This analysis assumes that the light from the two objects is
incoherent, meaning the intensities in the image plane add; interferences between
the two fields fluctuate rapidly in time and average away.
Example 11.1
What minimum telescope diameter is required to distinguish a Jupiter-like planet
(orbital radius 8 × 108 km) from its star if they are 10 light-years away?
Solution: From (11.18) and assuming 500 nm light, we need
1.22λ 1.22(500 × 10−9 m) 9.5 × 1015 m

`> = × = 0.07m
θmin (8 × 1011 m)/(10ly) ly
This seems like a piece of cake; a telescope with a diameter bigger than 7cm will do
the trick. However, the vastly unequal brightness of the star and the planet is the
real technical challenge. The faint diffraction rings in the star’s diffraction pattern
completely swamp the faint signal from the planet.
11.3 The Array Theorem

In this section we develop the array theorem, which is used for calculating the
Fraunhofer diffraction from an array of N identical apertures. We will be using
the theorem to compute diffraction from a grating, which may be thought of as a
mask with many closely spaced identical slits. However, the array theorem can be
applied to apertures with any shape, as suggested by Fig. 11.9.
Consider N apertures in a mask, each with the identical field distribution
described by E aperture (x 0 , y 0 , 0). Outside of the aperture, we suppose E aperture (x 0 , y 0 , 0)
is zero, so in a diffraction integral we won’t need to worry about the limits of
integration; we can just integrate over the whole mask. Each identical aperture has
a unique location on the mask. Let the location of the n th aperture be designated
by the coordinates x n0 , y n0 . The field associated with the n th aperture is then
¡ ¢
E aperture (x 0 − x n0 , y 0 − y n0 , 0), where the offset in the arguments shifts the location of
Figure 11.9 Array of identical aper- the aperture. The field comprising all of the identical apertures is
tures. N
E x 0, y 0, 0 = E aperture (x 0 − x n0 , y 0 − y n0 , 0)
¡ ¢ X
(11.19)
n=1
We next compute the Fraunhofer diffraction pattern for the above field. Upon
inserting (11.19) into the Fraunhofer diffraction formula (10.19) we obtain
k 2 +y 2 ∞ Z∞
e i kz e i 2z (x ) X
N Z k 0 0
0
d y 0 E aperture x 0 − x n0 , y 0 − y n0 , 0 e −i z (xx +y y )
¡ ¢ ¡ ¢
E x, y, z = −i dx
λz n=1
−∞ −∞
(11.20)
where we have taken the summation out in front of the integral. We have also
integrated over the entire (infinitely wide) mask since E aperture is nonzero only
inside each aperture.
Even without yet choosing the shape of the identical apertures, we can make
some progress on (11.20) with the change of variables x 00 ≡ x 0 −x n0 and y 00 ≡ y 0 − y n0 :
k 2
+y 2 ) X Z∞ Z∞
e i kz e i 2z (x N
00
d y 00 E aperture x 00 , y 00 , 0
¡ ¢ ¡ ¢
E x, y, z = −i dx
λz n=1
−∞ −∞
k 00
+x n0 )+y ( y 00 +y n0 )]
× e −i z [x (x
(11.21)
Next we pull the factor exp {−i kz (xx n0 + y y n0 )} out in front of the integral to arrive
at our final result:
" #
N
−i kz (xx n0 +y y n0 )
¡ ¢ X
E x, y, z = e
n=1
Z∞ Z∞
 
k 2
+y 2 )
e i kz e i 2z (x k 0 0
d x0 d y 0 E aperture x 0 , y 0 , 0 e −i (xx +y y ) 
¡ ¢
× −i z
λz
−∞ −∞
(11.22)
11.3 The Array Theorem 283
For the sake of elegance, we have traded back x 0 for x 00 and y 0 for y 00 as the
variables of integration. Equation (11.22) is known as the array theorem.2 Note
that the second factor in brackets is exactly the Fraunhofer diffraction pattern
from a single aperture centered on x 0 = 0 and y 0 = 0. When more than one
identical aperture is present, we only need to evaluate the Fraunhofer diffraction
formula for a single aperture. Then, the single-aperture result is multiplied by the
summation in front, which entirely contains the information about the placement
of the (many) identical apertures.
Example 11.2
Calculate the Fraunhofer diffraction pattern for two identical circular apertures
with diameter ` whose centers are separated by a spacing h.
Solution: As computed previously, the single-slit Fraunhofer diffraction pattern

from a circular aperture (see (10.29)) is
µ 2 ¶2 · ¢ ¸2
π`
¡
J 1 k`ρ/2z
I ρ, z = I 0
¡ ¢
2 ¡ ¢
4λz k`ρ/2z
From the array theorem (11.22), the intensity of the overall diffraction pattern is
¯2 ¢ ¸2 Figure 11.10 Fraunhofer diffrac-
¯ µ 2 ¶2 ·
π`
¡
2
¢ ¯¯ X J 1 k`ρ/2z tion pattern from two identical
−i kz (xx n0 +y y n0 ) ¯
¡ ¯
I x, y, z = ¯ e ¯ × I0 2 ¡ ¢
circular holes separated by twice
¯n=1 ¯ 4λz k`ρ/2z
their diameters.
Let y 10 = y 20 = 0. To create the separation h, let x 10 = −h/2 and x 20 = h/2. Then
2 ³ ´ ³ ´ µ
khx
¶
k 0 0 −i k − hx −i k hx
e −i z (xxn +y y n ) = e z 2 + e z 2 = 2 cos
X
n=1 2z
The overall pattern then becomes

¶2 · ¢ ¸2
π`2
¡
J 1 k`ρ/2z 2 khx
µ µ ¶
¡ ¢
I x, y, z = I 0 2 ¡ ¢ cos
2λz k`ρ/2z 2z
This pattern can be seen in Fig. 11.10.

2 A somewhat abstract alternative route to the array theorem recognizes that the field for
each aperture can

¢ be written as a 2-D convolution (see P0.26) between the aperture function
E aperture x 0 , y 0 , 0 and delta functions specifying the aperture location:
¡
Z∞ Z∞
E aperture x 0 − x n0 , y 0 − x n0 , 0 = 0
d y 0 δ x 00 − x n0 δ y 00 − x n0 E aperture x 0 − x 00 , y 0 − y 00 , 0
¡ ¢ ¡ ¢ ¡ ¢ ¡ ¢
dx
−∞ −∞
The integral in (11.20) therefore may be viewed as a 2-D Fourier transform of a convolution, where
kx/z and k y/z play the role of spatial frequencies. The convolution theorem (see P0.26) indicates
that this is the same as the product of Fourier transforms. The 2-D Fourier transform for the delta
function (times 2π) is
Z∞ Z∞
k 00 00 k 0 0
d x 00 d y 00 δ x 00 − x n0 δ y 00 − y n0 e −i z (xx +y y ) = e −i z (xxn +y y n )
¡ ¢ ¡ ¢
−∞ −∞
The array theorem (11.22) exhibits this factor. It multiplies the single-slit Fraunhofer diffraction
integral, which is the Fourier transform of the other function.
11.4 Diffraction Grating

In this section we will use the array theorem to calculate the diffraction from
a grating comprised of an array of equally spaced identical slits. An array of
uniformly spaced slits is called a transmission grating (see Fig. 11.11). Reflection
gratings are similar, being composed of an array of narrow rectangular mirrors
that behave similarly to the slits.
The Fraunhofer diffraction pattern from a single rectangular aperture was
previously calculated (see Example 10.4 and problem P10.8):
∆x∆ye i kz i k (x 2 +y 2 ) π∆x π∆y

µ ¶ µ ¶
¡ ¢
E aperture x, y, z = −i E 0 e 2z sinc x sinc y (11.23)
λz λz λz
The only part of (11.22) that remains to be evaluated is the summation out in
front. Let the apertures be positioned at
N +1
µ ¶
0
Figure 11.11 Transmission grating. xn = n − h, y n0 = 0 (11.24)
2
where N is the total number of slits. Then the summation in the array theorem,
(11.22), becomes
N k 0 0 khx
N
¡ N +1 ¢ X khx
e −i z (xxn +y y n ) = e i e −i n
X
z 2 z (11.25)
n=1 n=1
This summation is recognized as a geometric sum, which can be performed using

formula (0.65).
Equation (11.25) then simplifies to
khx
N
N
X −i kz (xx n0 +y y n0 ) i kz
¡ N +1 ¢
xh −i khx e −i z −1
e =e 2 e z
−i khx
n=1 e ³− 1 ´
z
(11.26)
e −i
khx
2z
N
− ei
khx
2z
N sin N khx
2z
= khx khx
= ³ ´
e −i 2z − ei 2z sin khx
2z
By combining (11.23) and (11.26) we obtain the full Fraunhofer diffraction pattern
for a diffraction grating. The expression for the field is
³ ´
khx "
¢ sin N 2z
¶#
∆x∆ye i kz i k (x 2 +y 2 ) π∆x π∆y
µ ¶ µ
¡
E x, y, z = ´ −i E 0 e 2z sinc x sinc y
λz λz λz
³
sin khx
2z
(11.27)
Now let us suppose that the slits are really tall (parallel to the y-dimension)
such that ∆y À λ. If the slits are infinitely tall, the final sinc function in Eq. (11.27)
can be approximated as one. 3 The intensity pattern in the horizontal direction
3 This is mostly the right idea, but is still a bit of a fake. In fact, the field often does not have a
uniform phase along the entire slit in the y-dimension, so our use of the function sinc π∆y/λz y
£¡ ¢ ¤
was inappropriate to begin with. The energy in a real spectrometer is usually spread out in a diffuse
pattern in the y-dimension. However, its form in y is of little relevance; the spectral information is
carried in the x-dimension only.
11.5 Spectrometers 285
can then be written in terms of the peak intensity of the diffraction pattern on the
screen: ³ ´
¶ sin2 N πhx
π∆x
µ
λz
I (x) = I peak sinc2 x (11.28)
λz
³ ´
N 2 sin πhx
2
λz
N=2
sin N α
Note that lim = N so we have placed N 2 in the denominator when intro-
α→0 sin α
ducing our definition of I peak , which represents the intensity on the screen at
x = 0. In principle, the intensity I peak is a function of y and depends on the exact
details of how the slits are illuminated as a function of y, but this is usually not of
interest as long as we stay with a given value of y as we scan along x.
It is left as an exercise to study the functional form of (11.28), especially how N=5
the number of slits N influences the behavior. The case of N = 2 describes
the diffraction pattern for a Young’s double slit experiment. We now have a
description of the Young’s two-slit pattern in the case that the slits have finite
openings of width ∆x rather than infinitely narrow ones.
A final note: You may wonder why we are interested in Fraunhofer diffraction
from a grating. The reason is that we are actually interested in separating different
wavelengths by observing their distinct diffraction patterns separated in space. In N = 10
order to achieve good spatial separation between light of different wavelengths,
it is necessary to allow the light to propagate a far distance. Optimal separation
(the maximum possible) occurs therefore in the Fraunhofer regime.
11.5 Spectrometers
The formula (11.28) can be exploited to make wavelength measurements. This N = 100
forms the basis of a diffraction grating spectrometer. A spectrometer has relatively
poor resolving power compared to a Fabry-Perot interferometer. Nevertheless, a
spectrometer is not hampered by the serious limitation imposed by free spectral
range. A spectrometer is able to measure a wide range of wavelengths simulta-
neously. The Fabry-Perot interferometer and the grating spectrometer in this
sense are complementary, the one being able to make very precise measurements -4 -2 0 2 4
within a narrow wavelength range and the other being able to characterize wide
ranges of wavelengths simultaneously.
To appreciate how a spectrometer works, consider Fraunhofer diffraction Figure 11.12 Diffraction through
from a grating, as described by (11.28). The structure of the diffraction pattern various numbers of slits, each
with ∆x = h/2 (slit widths half
has various peaks. For example, Fig. 11.12a shows the diffraction peaks from a
the separation). The dotted line
Young’s double slit (i.e. N = 2). The diffraction pattern is comprised of the typical shows the single slit diffraction
Young’s double-slit
³ pattern
´ multiplied
³ ´ by³ the´diffraction pattern of a single slit. pattern. (a) Diffraction from a
(Note that sin2 2 πhx
λz /4sin
2 πhx
λz = cos
2 πhx
λz .)
double slit. (b) Diffraction from 5
As the number of slits N is increased, the peaks seen in the Young’s double-slit slits. (c) Diffraction from 10 slits.
(d) Diffraction from 100 slits.
pattern tend to sharpen with additional smaller peaks appearing in between.
Figure 11.12b shows the case for N = 5. The more significant peaks occur when
sin(πhx/λz) in the denominator of (11.28) goes to zero. Keep in mind that the
numerator goes to zero at the same places, creating a zero-over-zero situation, so

the peaks are not infinitely tall.
With larger values of N , the peaks can become extremely sharp, and the small
secondary peaks in between are smaller in comparison. Fig. 11.12c shows the
case of N = 10 and Fig. 11.12d, shows the case of N = 100.
When very many slits are used, the diffraction pattern becomes very useful for
measuring spectra of light, since the position of the diffraction peaks depends on
wavelength (except for the center peak at x = 0). If light of different wavelengths
is simultaneously present, then the diffraction peaks associated with different
wavelengths appear in different locations. It helps to have very many slits involved
(i.e. large N ) so that the diffraction peaks are sharply defined. Then closely spaced
wavelengths can be more easily distinguished.
Consider the inset in Fig. 11.12d, which gives a close-up view of the first-order
Figure 11.13 Animation showing diffraction peak for N = 100. The location of this peak on a distant screen varies
diffraction through a number of with the wavelength of the light. How much must the wavelength change to cause
slits. the peak to move by half of its ‘width’ as marked in the inset of Fig. 11.12d? We
will say that this is the minimum separation of wavelengths that still allows the
two peaks to be distinguished.
Finding the Minimum Distinguishable Wavelength Separation
As mentioned, the main diffraction peaks occur when the denominator of (11.28)
goes to zero, i.e.
πhx
= mπ (11.29)
λz
The numerator of (11.28) goes to zero at these same locations (i.e. N πhx/λz =
N mπ), so the peaks remain finite. If two nearby wavelengths λ1 and λ2 are sent
through the grating simultaneously, their m th peaks are located at
mzλ1 mzλ2
x1 = and x 2 = (11.30)
h h
These are spatially separated by
mz
∆x λ ≡ x 2 − x 1 = ∆λ (11.31)
h
where ∆λ ≡ λ2 − λ1 .
Meanwhile, we can find the spatial width of, say, the first peak by considering the
change in x 1 that causes the sine in the numerator of (11.28) to reach the nearby
zero (see inset in Fig. 11.12d). This condition implies
πh x 1 + ∆x peak
¡ ¢
N = N mπ + π (11.32)
λ1 z
We will say that two peaks, associated with λ1 and λ2 , are barely distinguishable
when ∆x λ = ∆x peak . We also substitute from (11.30) to rewrite (11.32) as
πh (mzλ1 /h + mz∆λ/h) λ
N = N mπ + π ⇒ ∆λ = (11.33)
λ1 z Nm
Here we have dropped the subscript on the wavelength in the spirit of λ1 ≈ λ2 ≈ λ.
11.6 Diffraction of a Gaussian Field Profile 287
As we did for the Fabry-Perot interferometer, we can define the resolving

power of the diffraction grating as
λ
RP ≡ = mN (11.34)
∆λ
The resolving power is proportional to the number of slits illuminated on the
diffraction grating. The resolving power also improves for higher diffraction
orders m.
Example 11.3
What is the resolving power with m = 1 of a 2-cm-wide grating with 500 slits per
millimeter, and how wide is the 1st-order diffraction peak for 500-nm light after
1-m focusing?
Solution: From (11.34) the resolving power is
500
RP = mN = 2 cm = 104
0.1 cm
and the minimum distinguishable wavelength separation is
∆λ = λ/RP = 500 nm/104 = 0.05 nm
From (11.31), with z → f , we have
mf 1m
∆x = ∆λ = 0.05nm = 25 µm
h 2 × 10−6 m
11.6 Diffraction of a Gaussian Field Profile

Consider a Gaussian field profile described, at the plane z = 0, with the functional
form 02 02 x +y
−
w2
E (x 0 , y 0 , 0) = E 0 e 0 (11.35)
where w 0 , called the beam waist, specifies the radius of Gaussian profile. It is
depicted in Fig. 11.14. To better appreciate the meaning of w 0 , consider the
intensity of the above field distribution:
02 2
I x 0 , y 0 , 0 = I 0 e −2ρ /w 0
¡ ¢
(11.36)
z-axis
where ρ 02 ≡ x 02 + y 02 . In (11.36) we see that w 0 indicates the radius at which the
intensity reduces by the factor e −2 = 0.135.
We would like to know how this field evolves when it propagates forward from
the plane z = 0. We compute the field downstream using the Fresnel approxima-
tion (10.13): Figure 11.14 Diffraction of a Gaus-
sian field profile.
k 2 +y 2 ∞ Z∞
e i kz e i 2z (x ) Z h 02 02 2
i k 02 02 k 0 0
d x0 d y 0 E 0 e −(x +y )/w 0 e i 2z (x +y ) e −i z (xx +y y )
¡ ¢
E x, y, z = −i
λz
−∞ −∞
(11.37)
The Gaussian profile itself limits the dimension of the emission region, so there is
no problem in integrating to infinity. Equation (11.37) can be rewritten as
∞ ∞
µ ¶ µ ¶
k 2 +y 2
E 0 e i kz e i 2z (x ) Z k ky
− 1
−i 2z x 02 −i kx
z x
0 Z − 1 k
+i 2z y 02 −i z y 0
0 w2 0 w2
¡ ¢
E x, y, z = −i dx e 0 dy e 0
λz
−∞ −∞
(11.38)
The integrals over x 0 and y 0 have the identical form and can be done individually
with the help of the integral formula (0.55). The algebra is cumbersome, but the
integral in the x 0 dimension becomes
1  ³ ´2 
Z∞ −i kx
µ ¶ 
2
− 1
w2
−i k 02
x −i kx
x 0
π z
d x 0e
2z z
0 =  exp  
k
1
 ³ ´
− i 2z 1 k
−∞ w 02 4 w 2 − i 2z
0
 1  
2
2
π −kx
= ´  exp  ³
   
³ ´
k 2z 2z
−i 2z 1 + i kw 2 2z kw 2 − i
0 0
 1
2 
h i
2 2z
 λz   −kx kw 02
+ i 
= r  exp 
 
· ¸
³ ´2 i tan−1 2z  ³ ´ 2 
 2z kw 2
 2z
1 + kw 2 e 0 2z 1 + kw 2
0 0
(11.39)
A similar expression results from the integration on y 0 .
When (11.39) and the equivalent expression for the y-dimension are used in
(11.38), the result is
(xÃ2 +y 2 )!
µ ¶
1 k
− 2 +i 2z
k 2 2 w2
¡ ¢ e i kz e i 2z (x +y ) 1+ 2z
kw 2
0
−i tan−1 2z
kw 2
E x, y, z = E 0 r ³ ´2 e
0 e 0 (11.40)
2z
1 + kw 2
0
This rather complicated-looking expression for the field distribution is in fact very
useful and can be directly interpreted, as discussed in the next section.
Gaussian Field in Cylindrical Coordinates
A Gaussian field profile is one of few diffraction problems that can be handled con-
veniently in either the Cartesian (as above) or cylindrical coordinate. In cylindrical
11.7 Gaussian Laser Beams 289
coordinates, the Fresnel diffraction integral (10.27) is

kρ 2 Z∞
2πi e i kz e i kρ 02 kρρ 0
µ ¶
2z 02 2
E ρ, z = − ρ 0 d ρ 0 E 0 e −ρ /w 0 e i 2z J 0
¡ ¢
λz z
0
We can use the integral formula (0.59) to obtain

kρ 2
µ ¶
z
− " #
1 −i k
kρ 2 4
i kz i w2 2z
2πe e 2z e 0
E ρ, z = −i E 0
¡ ¢
λz
· ¸
1 k
2 w 02
− i 2z
µ ¶
ρ2 1 k
− +i 2z
kρ 2
!2
w2
Ã
e i kz e i 2z 0 2z
2z 1+ −i tan−1
kw 2 kw 2
= E0 s µ ¶2 e
0 e 0
2z
1+ kw 02
which is identical to (11.40).
11.7 Gaussian Laser Beams

The cumbersome Gaussian-field expression (11.40) can be cleaned up through
the judicious introduction of new quantities:
w 0 − ρ22 i kz+i 2R(z)

kρ 2
−i tan−1 zz
E ρ, z = E 0
¡ ¢
e w (z) e 0 (11.41)
w (z)
where
ρ2 ≡ x 2 + y 2, (11.42)
q
w (z) ≡ w 0 1 + z 2 /z 02 , (11.43)
R (z) ≡ z + z 02 /z, (11.44)
2
kw 0
z0 ≡ (11.45)
2
This formula describes the lowest-order Gaussian mode, the most common laser
beam profile. (Please be aware that some lasers are multimode and exhibit more
complicated structures.)
It turns out that (11.41) works equally well for negative values of z. The
expression can therefore be used to describe the field of a simple laser beam
everywhere (before and after it goes through a focus). In fact, the expression
works also near z = 0!4 At z = 0 the diffracted field (11.41) returns the exact
4 There is good reason for this since the Fresnel diffraction integral is an exact solution to the
paraxial wave equation (10.15). The beam (11.41) therefore satisfies the paraxial wave equation for
all z.
expression for the original field profile (11.35) (see P11.11). In short, (11.41) may
be used with impunity as long as the divergence angle of the beam is not too wide.
To begin our interpretation of (11.41), consider the intensity profile I ∝ E ∗ E
as depicted in Fig. 11.15:
w 2 − 2ρ2 I0 2ρ 2
− 2
I ρ, z = I 0 2 0 e w 2 (z) =
¡ ¢
e w (z) (11.46)
w (z) 1 + z 2 /z 02
By inspection, we see that w (z) gives the radius of the beam anywhere along
z. At z = 0, the beam waist, w (z = 0) reduces to w 0 , as expected. The parameter
z 0 , known as the Rayleigh range, specifies the distance along the axis from z = 0
to the point where the intensity decreases by a factor of 2. Note that w 0 and z 0
are not independent of each other but are connected through the wavelength
Figure 11.15 A Gaussian laser field according to (11.45). There is a tradeoff: a small beam waist means a short depth
profile in the vicinity of its beam of focus. That is, a small w 0 means a small Rayleigh range z 0 .
waist. We next consider the phase terms that appear in the field expression (11.41).
The phase term i kz + i kρ 2 /2R (z) describes the phase of curved wave fronts,
where R (z) is the radius of curvature of the wave front at z. At z = 0, the radius of
curvature is infinite (see (11.44)), meaning that the wave front is flat at the laser
beam waist. In contrast, at very large values of z we have R (z) ∼ = z (see (11.44)).
kρ 2 ∼ p 2
In this case, we may write these phase terms as kz + 2R(z) = k z + ρ 2 . This
describes a spherical wave front emanating from the origin out to point ρ, z . The
¡ ¢
Fresnel approximation (same as the paraxial approximation) represents spherical

wave fronts with the former parabolic approximation. As a reminder, to restore the
temporal dependence of the field, we append e −i ωt to the solution, as discussed
in connection with (10.4).
The phase −i tan−1 z/z 0 is perhaps a bit more mysterious. It is called the Gouy
shift and is actually present for any light that goes through a focus, not just laser
beams. The Gouy shift is not overly dramatic since the expression tan−1 z/z 0
ranges from −π/2 (at z = −∞) to π/2 (at z = +∞). Nevertheless, when light goes
through a focus, it experiences an overall phase shift of π.
Example 11.4
Write the beam waist w 0 in terms of the f-number, defined to be the ratio of z to
the diameter of the beam diameter 2w(z) far from the beam waist.
Solution: Far away from the beam waist (i.e. z >> z 0 ) the laser beam expands
along a cone. That is, its diameter increases in proportion to distance.
q
w (z) = w 0 1 + z 2 /z 02 → w 0 z/z 0
The cone angle is parameterized by the f-number, the ratio of the cone height to
its base:
z z z0
f # ≡ lim = =
z→±∞ 2w (z) 2w 0 z/z 0 2w 0
Figure 11.16
11.A ABCD Law for Gaussian Beams 291
Substitution of (11.45) into this expressions yields
2λ f #
w0 = (11.47)
π
Equation (11.47) gives a convenient way to predict the size of a laser focus. One
calculate the f-number by dividing the diameter of the beam at a lens by the
distance to the focus. However, in practice you may be very surprised at how badly
a beam focuses compared to the theoretical prediction (due to aberrations, etc.).
It is always good practice to directly measure your focus if its size is important to
an experiment.
Appendix 11.A ABCD Law for Gaussian Beams

In this section we discuss and justify the ABCD law for Gaussian beams. The
law enables one to predict the parameters of a Gaussian beam that exits from an
optical system, given the parameters of an input Gaussian beam. To make the
prediction, one needs only the ABCD matrix for the optical system, taken as a
whole. The system may be arbitrarily complex with many optical components.
At first, it may seem unlikely that such a prediction should be possible since
ABCD matrices were introduced to describe the propagation of rays. On the other
hand, Gaussian beams are governed by the laws of diffraction. As an example of
this dichotomy, consider a collimated Gaussian beam that traverses a converging
lens. By ray theory, one expects the Gaussian beam to focus near the focal point
of the lens. However, a collimated beam by definition is already in the act of going
through focus. In the absence of the lens, there is a tendency for the beam to grow
via diffraction, especially if the beam waist is small. This tendency competes with
the focusing effect of the lens, and a new beam waist can occur at a wide range of
locations, depending on the exact outcome of this competition.
A Gaussian beam is characterized by its Rayleigh range z 0 . From this, the
beam waist radius w 0 may be extracted via (11.45), assuming the wavelength is
known. Suppose that a Gaussian beam encounters an optical system at position
z, referenced to the position of the beam’s waist as shown in Fig. 11.17. The beam
exiting from the system, in general, has a new Rayleigh range z 00 . The waist of the
new beam also occurs at a different location. Let z 0 denote the location of the exit
of the optical system, referenced to the location of the waist of the new beam. If
the exiting beam diverges as in Fig. 11.17, then it emerges from a virtual beam
waist located before the exit point of the system. In this case, z 0 is taken to be
positive. On the other hand, if the emerging beam converges to an actual waist,
then z 0 is taken to be negative since the exit point of the system occurs before the
focus.
The ABCD law is embodied in the following relationship:
A (z + i z 0 ) + B
z 0 + i z 00 = (11.48)
C (z + i z 0 ) + D
Figure 11.17 Gaussian laser beam traversing an optical system described by an ABCD
matrix. The dark lines represent the incoming and exiting beams. The gray line repre-
sents where the exiting beam appears to have been.
where A, B , C
p, and D are the matrix elements of the optical system. The imaginary
number i ≡ −1 imbues the law with complex arithmetic. It makes two equations
from one, since the real and imaginary parts of (11.48) must separately be equal.
We now prove the ABCD law. We begin by showing that the law holds for
two specific ABCD matrices. First, consider the matrix for propagation through a
distance d : · ¸ · ¸
A B 1 d
= (11.49)
C D 0 1
We know that simple propagation has minimal effect on a beam. The Rayleigh
range is unchanged, so we expect that the ABCD law should give z 00 = z 0 . The
propagation through a distance d modifies the beam position by z 0 = z + d . We
now check that the ABCD law agrees with these results by inserting (11.49) into
(11.48):
1 (z + i z 0 ) + d
z 0 + i z 00 = = z + d + i z 0 (propagation through distance d) (11.50)
0 (z + i z 0 ) + 1
Thus, the law holds in this case.

Next we consider the ABCD matrix of a thin lens (or a curved mirror):
· ¸ · ¸
A B 1 0
= (11.51)
C D −1/ f 1
A beam that traverses a thin lens undergoes the phase shift −kρ 2 /2 f , according
to (11.11). This modifies the original phase of the wave front kρ 2 /2R (z), seen in
(11.41). The phase of the exiting beam is therefore
kρ 2 kρ 2 kρ 2
= − (11.52)
2R (z 0 ) 2R (z) 2 f
where we do not keep track of unimportant overall phases such as kz or kz 0 . With

(11.44) this relationship reduces to
1 1 1 1 1 1
= − ⇒ 2 0
= 2
− (11.53)
R (z 0 ) R (z) f 0 0
z + z 0 /z z + z 0 /z f
11.A ABCD Law for Gaussian Beams 293
In addition to this relationship, the local radius of the beam given by (11.43)
cannot change while traversing the ‘thin’ lens. Therefore,
2
Ã !
z0 z2
µ ¶
¡ 0¢ 0
w z = w (z) ⇒ z 0 1 + 2 = z 0 1 + 2 (11.54)
z 00 z0
On the other hand, the ABCD law for the thin lens gives
1 (z + i z 0 ) + 0
z 0 + i z 00 = ¡ ¢ (traversing a thin lens with focal length f )
− 1/ f (z + i z 0 ) + 1
(11.55)
It is left as an exercise (see P11.14) to show that (11.55) is consistent with (11.53)
and (11.54).
So far we have shown that the ABCD law works for two specific examples,
namely propagation through a distance d and transmission through a thin lens
with focal length f . From these elements we can derive more complicated sys-
tems. However, the ABCD matrix for a thick lens cannot be constructed from just
these two elements. However, we can construct the matrix for a thick lens if we
sandwich a thick window (as opposed to empty space) between two thin lenses.
The proof that the matrix for a thick window obeys the ABCD law is left as an
exercise (see P11.17). With these relatively few elements, essentially any optical
system can be constructed, provided that the beam propagation begins and ends
in the same index of refraction.
To complete our proof of the general ABCD law, we need only show that when
it is applied to the compound element
· ¸ · ¸· ¸ · ¸
A B A2 B2 A1 B1 A 2 A 1 + B 2C 1
A2B1 + B2D 1
= =
C D C2 D2 C1 D1 C 2 A 1 + D 2C 1
C2B1 + D 2D 1
(11.56)
it gives the same answer as when the law is applied sequentially, first on
· ¸
A1 B1
C1 D1
and then on · ¸
A2 B2
C2 D2
Explicitly, we have
A 2 z 0 + i z 00 + B 2
¡ ¢
00 00
z + i z0 =
C 2 z 0 + i z 00 + D 2
¡ ¢
h i
A 2 CA11(z+i
(z+i z 0 )+B 1
z 0 )+D 1 + B 2
= h i
C 2 CA11(z+i
(z+i z 0 )+B 1
z 0 )+D 1 + D 2
A 2 [A 1 (z + i z 0 ) + B 1 ] + B 2 [C 1 (z + i z 0 ) + D 1 ] (11.57)
=
C 2 [A 1 (z + i z 0 ) + B 1 ] + D 2 [C 1 (z + i z 0 ) + D 1 ]
(A 2 A 1 + B 2C 1 ) (z + i z 0 ) + (A 2 B 1 + B 2 D 1 )
=
(C 2 A 1 + D 2C 1 ) (z + i z 0 ) + (C 2 B 1 + D 2 D 1 )
A (z + i z 0 ) + B
=
C (z + i z 0 ) + D
Thus, we can construct any ABCD matrix that we wish from matrices that are
known to obey the ABCD law. The resulting matrix also obeys the ABCD law.
Exercises 295
Exercises
Exercises for 11.1 Fraunhofer Diffraction Through a Lens
P11.1 Fill in the steps leading to (11.14) from (11.13). Show that the intensity
distribution (11.6) is consistent with (11.14).
L11.2 Set up a collimated ‘plane wave’ in the laboratory using a HeNe laser
(λ = 633 nm) and appropriate lenses.
(a) Choose a rectangular aperture (∆x by ∆y) and place it in the plane
wave. Observe the Fraunhofer diffraction on a very far away screen (i.e.,
¢2
where z À k2 aperture radius is satisfied). Check that the location of
¡
the ‘zeros’ agrees with (10.20).

(b) Place a lens in the beam after the aperture. Use a CCD camera
to observe the Fraunhofer diffraction profile at the focus of the lens.
Check that the location of the ‘zeros’ agrees with (10.20), replacing z
with f .
(c) Repeat parts (a) and (b) using a circular aperture with diameter `.
Check the position of the first ‘zero’. (video)
CCD
Camera
Filters
Screen
Laser
Far-away Removable
mirror mirror Aperture
Figure 11.18
Exercises for 11.2 Resolution of a Telescope
P11.3 On the night of April 18, 1775, a signal was sent from the Old North
Church steeple to Paul Revere, who was 1.8 miles away: “One if by
land, two if by sea.” If in the dark, Paul’s pupils had 4 mm diameters,
what is the minimum possible separation between the two lanterns
that would allow him to correctly interpret the signal? Assume that the
predominant wavelength of the lanterns was 580 nm.
HINT: In the eye, the index of refraction is about 1.33 so the wavelength
is shorter. This leads to a smaller diffraction pattern on the retina.
However, in accordance with Snell’s law, two rays separated by an angle
580 nm outside of the eye are separated by an angle θ/1.33 inside the
eye. The two rays then hit on the retina closer together. As far as
resolution is concerned, the two effects exactly compensate.
L11.4 Simulate two stars with laser beams (λ = 633 nm). Align them nearly
parallel with a small lateral displacement. Send the beams down a long
corridor until diffraction causes both beams to grow into one another
so that it is no longer apparent that they are from two distinct sources.
Use a lens to image the two sources onto a CCD camera. The camera
should be placed close to the focal plane of the lens. Use a variable iris
near the lens to create different pupil openings.
Laser
CCD
Laser Camera
Filter
Pupil
Figure 11.19
Experimentally determine the pupil diameter that just allows you to

resolve the two sources according to the Rayleigh criterion. Check your
measurement against theoretical prediction. (video)
HINT: The angular separation between the two sources is obtained by
dividing propagation distance into the lateral separation of the beams.
Exercises for 11.3 The Array Theorem
P11.5 Find the diffraction pattern created by an array of nine circles, each
with radius a, which are centered at the following (x 0 , y 0 ) coordinates:
(−b, b), (0, b), (b, b), (−b, 0), (0, 0), (b, 0), (−b, −b), (0, −b), (b, −b) (a is
less than b). Make a plot of the result for the situation where (in some
choice of units) a = 1, b = 5a, and k/d = 1. View the plot at different
“zoom levels” to see the finer detail.
P11.6 (a) A plane wave is incident on a screen of N 2 uniformly spaced identi-

cal rectangular apertures of dimension ∆x by ∆y (see Fig. 11.20). Their
positions are described by x n = h n − N2+1 and y m = s m − N2+1 . Find
¡ ¢ ¡ ¢
the far-field (Fraunhofer) pattern of the light transmitted by the grid.

(b) You look at a distant sodium street lamp (somewhat monochro-
matic) through a curtain made from a fine mesh fabric with crossed
Figure 11.20 threads. Make a sketch of what you expect to see (how the lamp will
look to you).
HINT: Remember that the lens of your eye causes the Fraunhofer
diffraction of the mesh to appear at the retina.
Exercises 297
Exercises for 11.4 Diffraction Grating
P11.7 Consider Fraunhofer diffraction from a grating of N slits having widths

∆x and equal separations h. Make plots (label relevant points and
scaling) of the intensity pattern for N = 1, N = 2, N = 5, and N =
1000 in the case where h = 2∆x, ∆x = 5 µm, and λ = 500 nm. Let the
Fraunhofer diffraction be observed at the focus of a lens with focal
length f = 100 cm. Do you expect I peak to be the same value for all of
these cases?
P11.8 For the case of N = 1000 in P11.7, you wish to position a narrow slit at
the focus of the lens so that it transmits only the first-order diffraction
¡ ¢
peak (i.e. at khx/ 2 f = ±π). (a) How wide should the slit be if it is to
be half the separation between the first intensity zeros to either side of
the peak?
(b) What small change in wavelength (away from λ = 500 nm) will
cause the intensity peak to shift by the width of the slit found in part
(a)?
Exercises for 11.5 Spectrometers
L11.9 (a) Use a HeNe laser to determine the period h of a reflective grating.
(b) Give an estimate of the blaze angle φ on the grating. HINT: Assume
that the blaze angle is optimized for first-order diffraction of the HeNe
laser (for one side) at normal incidence. The blaze angle enables a
mirror-like reflection of the diffracted light on each groove. (video)
(c) You have two mirrors of focal length 75 cm and the reflective grating Figure 11.21
in the lab. You also have two very narrow adjustable slits and the ability
to ‘tune’ the angle of the grating. Sketch how to use these items to make
a monochromator (scans through one wavelength at a time). If the
beam that hits the grating is 5 cm wide, what do you expect the ultimate
resolving power of the monochromator to be in the wavelength range
of 500 nm? Do not worry about aberration such as astigmatism from
using the mirrors off axis.
Light out
Slit
Grating
Slit
Light in
Figure 11.22
L11.10 Study the Jarrell Ash monochromator. Use a tungsten lamp as a source
and observe how the instrument works by taking the entire top off.
Do not breathe or touch when you do this. In the dark, trace the light
inside of the instrument with a white plastic card and observe what
happens when you change the wavelength setting. Place the top back
on when you are done. (video)
(a) Predict the best theoretical resolving power that this instrument can
do assuming 1200 lines per millimeter.
(b) What should the width ∆x of the entrance and exit slits be to obtain
this resolving power? Assume λ = 500 nm.
HINT: Set ∆x to be the distance between the peak and the first zero of
the diffraction pattern at the exit slit for monochromatic light.
Exercises for 11.7 Gaussian Laser Beams
P11.11 (a) Confirm that (11.41) reduces to (11.35) when z = 0.

(b) Take the limit z À z 0 to find the field far from the laser focus.
P11.12 Use the Fraunhofer integral formula (either (10.19) or (10.28)) to deter-
mine the far-field pattern of a Gaussian laser focus (11.35).
HINT: The answer should agree with P11.11 part (b).
L11.13 Consider the following setup where a diverging laser beam is collimated
using an uncoated lens. A double reflection from both surfaces of the
lens (known as a ghost) comes out in the forward direction, focusing
Ghost Beam after a short distance. Use a CCD camera to study this focused beam.
The collimated beam serves as a reference to reveal the phase of the
focused beam through interference. Because the weak ghost beam
Figure 11.23
Exercises 299
concentrates near its focus, the two beams can have similar intensities
for optimal interference effects. (video)5
Uncoated
Filter Pin Hole Lens
Laser CCD
Camera
Lens
150 cm
Figure 11.24
The ghost beam E 1 ρ, z is described by (11.41), where the origin is at

¡ ¢
the focus. Let the collimated beam be approximated as a plane wave

E 2 e i kz+i φ , where φ is the relative phase between the two beams. The
¯2
net intensity is then I t ρ, z ∝ ¯E 1 ρ, z + E 2 e i kz+i φ ¯ or
¡ ¢ ¯ ¡ ¢
kρ 2 −1 z
· q µ ¶¸
I t ρ, z = I 2 + I 1 ρ, z + 2 I 2 I 1 ρ, z cos −φ
¡ ¢ ¡ ¢ ¡ ¢
− tan
2R (z) z0
where I 1 ρ, z is given by (11.46). We now have a formula that retains

¡ ¢
both R (z) and the Gouy shift tan−1 z/z 0 , which are not present in the
intensity distribution of a single beam (see (11.46)).
(a) Determine the f-number for the ghost beam (see Example 11.4).
Use this measurement to predict a value for w 0 . HINT: You know that
z=0 z = +z0
at the lens, the focusing beam is the same size as the collimated beam.
(b) Measure the actual spot size w 0 at the focus. How does it compare
to the prediction?
HINT: Before measuring the spot size, make a subtle adjustment to
the tilt of the lens. This incidentally causes the phase between the two z = -z0 z = +2z0
beams to vary by small amounts, which you can set to φ = ±π/2. Then
at the focus the cosine term vanishes and the two beams don’t interfere
(i.e. the intensities simply add). This is accomplished if the center of
the interference pattern is as dark as possible either far before or far
after the focus.
z = -2z0 z = +3z0
(c) Observe the effect of the Gouy shift. Since tan−1 z/z 0 varies over a
range of π, you should see that the ring pattern before versus after the
focus inverts (i.e. the bright rings exchange with the dark ones).
(d) Predict the Rayleigh range z 0 and check that the radius of curvature
R (z) ≡ z + z 02 /z agrees with measurement. z = -3z0 z = +4z0
HINT: You should see interference rings similar to those in Fig. 11.25.
The only phase term that varies with ρ is kρ 2 /2R (z). If you count N Figure 11.25
fringes out to a radius ρ, then kρ 2 /2R (z) has varied by 2πN .
5 J. Peatross and M. V. Pack, “Viewing the Mathematical Structure of Gaussian Laser Beams in a
Student Laboratory,” Am. J. Phys. 69, 1169 (2001).

Exercises for 11.A ABCD Law for Gaussian Beams
P11.14 Find the solutions to (11.55) (i.e. find z 0 and z 00 in terms of z and z 0 ).
Show that the results are in agreement with (11.53) and (11.54).
P11.15 Assuming a collimated beam (i.e. z = 0 and beam waist w 0 ), find the
location L = −z 0 and size w 00 of the resulting focus when the beam goes
through a thin lens with focal length f .
L11.16 Place a lens in a HeNe laser beam soon after the exit mirror of the cavity.
Characterize the focus of the resulting laser beam, and compare the
results with the expressions derived in P11.15.
P11.17 Prove the ABCD law for a beam propagating through a thick window of
material with matrix
· ¸ · ¸
A B 1 d /n
=
C D 0 1
Chapter 12
Interferograms and Holography
In chapter 8, we studied a Michelson interferometer in an idealized sense: 1) The

light entering the instrument was considered to be a plane wave. 2) The retro-
reflecting mirrors were considered to be aligned perpendicular to the beams
impinging on them. 3) All reflective surfaces were taken to be perfectly flat. If any
of these conditions are not met, the beam emerging from the interferometer is
likely to exhibit an interference or fringe pattern. A recorded fringe pattern (on a
CCD or photographic film) is called an interferogram. In this chapter, we examine
typical fringe patterns that can be produced in an interferometer. Such patterns
are very useful for testing the prescription and quality of optical components.1
We will also study holography, where an interference pattern (or fringe pat-
tern) is recorded and then later used to diffract light, in much the same way that
gratings diffract light.2 A recorded fringe pattern, when used for this purpose,
is called a hologram. When light diffracts from a hologram, it can mimic the
light field originally used to generate the fringe pattern. This is true even for
complicated fields, recorded when light scatters from arbitrary three-dimensional
objects. When the light field is re-created through diffraction, the resulting image
looks three-dimensional, since the holographic fringes re-construct the original
light field over a wide range of viewing angles.
12.1 Interferograms
Consider the Michelson interferometer seen in Fig. 12.1. Suppose that the beam-
spliter divides the fields evenly, so that the overall output intensity is given by
(8.1):
Figure 12.1 Michelson interferom-
I tot = 2I 0 [1 + cos (ωτ)] (12.1) eter.
As a reminder, τ is the roundtrip delay time of one path relative to the other. This
equation is based on the idealized case, where the amplitude and phase of the two
1 See M. Born and E. Wolf, Principles of Optics, 7th ed., Sect. 7.5.5 (Cambridge: Cambridge
University Press, 1999).

2 In fact, a grating can be considered to be a hologram and holographic techniques are often
employed to produce gratings.
301
302 Chapter 12 Interferograms and Holography
(a) beams are uniform and perfectly aligned to each other following the beamsplitter.
The entire beam ‘blinks’ on and off as the delay path τ is varied.
What happens if one of the retro-reflecting mirrors is misaligned by a small
angle θ? The fringe patterns seen in Fig. 12.2 (a)-(c) are the result. By the law
of reflection, the beam returning from the misaligned mirror deviates from the
‘ideal’ path by an angle 2θ. This puts a relative phase variation of
φ = kx sin (2θx ) + k y sin 2θ y

¡ ¢
(12.2)
(b)
on the misaligned beam.3 Here θx represents the tilt of the mirror in the x-
dimension and θ y represents the amount of tilt in the y-dimension.
When the two plane waves join, the resulting intensity pattern is
I tot = 2I 0 1 + cos φ + ωτ
£ ¡ ¢¤
(12.3)
(c) The phase term φ depends on the local position within the beam through x and
y. Regions of uniform phase, called fringes (in this case individual stripes), have
the same intensity. As the delay τ is varied, the fringes seem to ‘move’ across the
detector. In this case, the fringes appear at one edge of the beam and disappear
at the other.
Another interesting situation arises when the beams in a Michelson interfer-
ometer are diverging. A fringe pattern of concentric circles will be seen at the
(d) detector when the two beam paths are unequal (see Fig. 12.2 (d)). The radius of
curvature for the beam traveling the longer path is increased by the added amount
of delay d = τc. Thus, if beam 1 has radius of curvature R 1 when returning to the
beam splitter, then beam 2 will have radius R 2 = R 1 +d upon return (assuming flat
mirrors). The relative phase (see phase term in (11.41)) between the two beams is
φ = kρ 2 /2R 1 − kρ 2 /2R 2 (12.4)

(e)
and the intensity pattern at the detector is given as before by (12.3).
12.2 Testing Optical Components

A Michelson interferometer is ideal for testing the quality of optical surfaces. If
any of the flat surfaces (including the beam splitter) in the interferometer are
distorted, the fringe pattern readily reveals it. Figure 12.3 shows an example of a
Figure 12.2 Fringe patterns for
a Michelson interferometer: (a)
fringe pattern when one of the mirrors in the interferometer has an arbitrary de-
Horizontally misaligned beams. formity in the surface figure.4 A new fringe stripe occurs for every half wavelength
(b) Vertically misaligned beams. that the surface varies. (The round trip turns a half wavelength into a whole
(c) Both vertically and horizontally wavelength.) This makes it possible to determine the flatness of a surface with
misaligned beams. (d) Diverging very high precision. Of course, in order to test a given surface in an interferometer,
beam with unequal paths. (e) Di- the quality of all other surfaces in the interferometer must first be ensured.
verging beam with unequal paths
3 This ignores an additive constant for fixed z.
and horizontal misalignment.
4 The surface figure is a name for how well a surface contour matches a desired prescription.
12.3 Generating Holograms 303
A typical industry standard for research-grade optics is to specify the surface

flatness to within one tenth of an optical wavelength (633 nm HeNe laser). This (a)
means that the interferometer should reveal no more than one fifth of a fringe
variation across the substrate surface. The fringe pattern tells the technician how
the surface should continue to be polished in order to achieve the desired surface
flatness. Figure 12.3(a) shows the fringe pattern for a surface with significant
variations in the surface figure.
When testing a surface, it is not necessary to remove all tilt from the alignment
before the effects of surface variations become apparent in the fringe pattern.
In fact, it can be helpful to observe the distortions as deflections in a normally (b)
regularly striped fringe pattern. Figure 12.3(b) shows fringes from a distorted
surface when some tilt is left in the interferometer alignment. An important
advantage to leaving some tilt in the beam is that one can better tell the sign of
the phase errors. We can see, for example, in the case of tilt that the two major
distortion regions in Fig. 12.3 have opposite phase; we can tell that one region of
the substrate protrudes while the other dishes in. On the other hand, this is not
clear for an interferogram with no tilt.
Figure 12.3 (a) Fringe pattern aris-
Other types of optical components (besides flat mirrors) can also be tested ing from an arbitrarily distorted
with an interferometer. Figure 12.4 shows how a lens can be tested using a mirror in a perfectly aligned inter-
convex mirror to compensate for the focusing action of the lens. With appropriate ferometer with plane wave beams.
spacing, the lens-mirror combination can act like a flat surface. Distortions in the (b) Fringe pattern from the same
lens figure are revealed in the fringe pattern. In this case, the surfaces of the lens mirror as (a) when the mirror is
are tested together, and variations in optical path length are observed. In order tilted (still plane wave beams).
The distortion due to surface varia-
to record fringes, say with a CCD camera, it is often convenient to image a larger
tion is still easily seen.
beam onto a relatively small active area of the detector. The imaging objective
should be adjusted to produce an image of the test optic on the detector screen.
The diameter of the objective lens needs to accommodate the whole beam.
Optic to
be tested
12.3 Generating Holograms
In the late 1940’s, Dennis Gabor developed the concept of holography, but it wasn’t
until after the invention of the laser that this field really blossomed. Consider
a coherent monochromatic beam of light that is split in half by a beamsplitter,
similar to that in a Michelson interferometer. Let one beam, called the reference
Imaging
beam, proceed directly to a recording film, and let the other beam scatter from Objective
an arbitrary object back towards the same film. The two beams interfere at the Camera
recording film. It is best to split the beam initially into unequal intensities such
that the light scattered from the object has an intensity similar to the reference Figure 12.4 Twyman-Green setup
for testing lenses.
beam at the film.
The purpose of the film is to record the interference pattern. It is important
that the coherence length of the light be much longer than the difference in
path length starting from the beam splitter and ending at the film. In addition,
during exposure to the film, it is important that the whole setup be stable against
vibrations on the scale of a wavelength since this will cause the fringes to wash
Object
Film out. For simplicity, we neglect the vector nature of the electric field, assuming
that the scattering from the object for the most part preserves polarization and
that the angle between the two beams incident on the film is modest (so that the
electric fields of the two beams are close to parallel). To the extent that the light
scattered from the object contains the polarization component orthogonal to that
of the reference beam, it provides a uniform (unwanted) background exposure to
Beamsplitter the film on top of which the fringe pattern is recorded.
Figure 12.5 Exposure of holo- In general terms, we may write the electric field arriving at the film as5
graphic film. E film (r) e −i ωt = E object (r) e −i ωt + E ref (r) e −i ωt (12.5)
Here, the coordinate r indicates locations on the film surface, which may have
arbitrary shape, but often is a plane. The field E object (r), which is scattered from
the object, is in general very complicated. The field E ref (r) may be equally compli-
cated, but typically it is convenient if it has a simple form such as a plane wave,
since this beam must be re-created later in order to view the hologram.
The intensity of the field (12.5) is given by
1 ¯ ¯2
I film (r) = c²0 ¯E object (r) + E ref (r)¯
2
1 h¯ ¯2 i (12.6)
= c²0 ¯E object (r)¯ + |E ref (r)|2 + E ref
∗ ∗
(r) E object (r) + E ref (r) E object (r)
2
For typical photographic film, the exposure of the film is proportional to the
intensity of the light hitting it. This is known as the linear response regime. That
Dennis Gabor (19001979, Hungarian) is, after the film is developed, the transmittance T of the light through the film is
was born in Budapest. As a teenager, proportional to the intensity of the light that exposed it (I film ). However, for low
he fought for Hungary in World War
I. Following the war, he studied at the
exposure levels, or for film specifically designed for holography, the transmission
Technical University of Budapest and of the light through the film can be proportional to the square of the intensity
later at the Technical University of
Berlin. In 1927, Gabor completed his
of the light that exposes the film. Thus, after the film is exposed to the fringe
doctoral dissertation on cathode ray pattern and developed, the film acquires a spatially varying transmission function
tubes and began a long career work-
according to
ing on electron-beam devices such as
2
oscilloscopes, televisions, and electron T (r) ∝ I film (r) (12.7)
microscopes. It was in the context of
èlectron optics' that he invented the
If at a later point in time light of intensity I incident is directed onto the film, it will
concept of holography, which relied on transmit according to I transmitted = T (r)I incident . In this case, the field, as it emerges
the wave nature of electron beams. Ga-
bor did this work while working for a
from the other side of the film, will be
British company, after eeing Germany
when Hitler came to power. Holography
E transmitted (r) = t (r) E incident (r) ∝ I film (r) E incident (r) (12.8)
did not become practical until after the
p
invention of the laser, which provided
where t (r) = T (r).
a bright coherent light source. (Gabor
had attempted to make holograms ear-
lier using a spectral line from a mercury 12.4 Holographic Wavefront Reconstruction
lamp.) In 1964 the rst hologram was
produced. Soon after, holograms be-
came commercially available and were To see a holographic image, we re-illuminate film (previously exposed and devel-
popularized. Gabor accepted a post as oped) with the original reference beam. That is, we send in
professor of applied physics at the Impe-
rial College of London from 1958 until E incident (r) = E ref (r) (12.9)
he retired in 1967. He was awarded the
Nobel prize in physics in 1971 for the
5 See P. W. Milonni and J. H. Eberly, Lasers, Sect. 16.4-16.5 (New York: Wiley, 1988); G. R. Fowles,
invention of holography. (Wikipedia) Introduction to Modern Optics, 2nd ed., Sect. 5.7 (Toronto: Dover, 1975).
12.4 Holographic Wavefront Reconstruction 305
Image
and view the light that is transmitted. According to (12.6) and (12.8), the trans-
Film
mitted field is proportional to
E transmitted (r) ∝ I film (r) E ref (r)

h¯ ¯2 i
= ¯E object (r)¯ + |E ref (r)|2 E ref (r) + |E ref (r)|2 E object (r) + E ref
2 ∗
(r) E object (r)
(12.10)
Although (12.10) looks fairly complicated, each of the three terms has a direct
interpretation. The first term is just the reference beam E ref (r) with an amplitude Observer
modified by the transmission through the film. It is the residual undeflected beam,
Figure 12.6 Holographic recon-
similar to the zero-order diffraction peak for a transmission grating. The second struction of wavefront through
term is interpreted as a reconstruction of the light field originally scattered from diffraction from fringes on film.
the object E object (r). Its amplitude is modified by the intensity of the reference Compare with Fig. 12.5.
beam, but if the reference beam is uniform across the film, this hardly matters.
An observer looking into the film sees a wavefront identical to the one produced
by the original object (superimposed with the other fields in (12.10)). Thus,
the observer sees a virtual image at the location of the original object. Since Reference
Beam Film
the wavefront of the original object has genuinely been recreated, the image
looks ‘three-dimensional’, because the observer is free to view from different
perspectives.
The final term in (12.10) is proportional to the complex conjugate of the
original field from the object. It also contains twice the phase of the reference
beam, which we can overlook if the reference beam is uniform on the film. In
this case, the complex conjugate of the object field actually converges to a real
Point
image of the original object. This image is located on the observer’s side of the Object
film, but it is often of less interest since the image is inside out. An ideal screen for
viewing this real image would be an item shaped identical to the original object,
which of course defeats the purpose of the hologram! To the extent that the film is
not flat or to the extent that the reference beam is not a plane wave, the phase of
2
E ref (r) severely distorts the image. On the other hand, the virtual image previously
described never suffers from this problem.
Example 12.1
Analyze the three field terms in (12.10) for a hologram made from a point object,
as depicted in Fig. 12.7.
Figure 12.7 Exposure to holo-
graphic film by a point source
Solution: Presumably, the point object is illuminated sufficiently brightly so as to
and a reference plane wave. The
make the scattered light have an intensity similar to the reference beam at the film.
holographic fringe pattern for a
Let the reference plane wave strike the film at normal incidence. Then the reference point object and a plane wave ref-
field will have constant amplitude and phase across it; call it E ref . The field from erence beam exposing a flat film is
the point object can be treated as a spherical wave: shown on the right.
E ref L p 2 2
E object ρ = p e i k L +ρ
¡ ¢
(point source example) (12.11)
L2 + ρ2
Here ρ represents the radial distance from the center of the film to some other
Reference Undeflected point on the film. We have taken the amplitude of the object field to match E ref in
beam Film beam the center of the film.
After the film is exposed, developed, and re-illuminated by the reference beam, the
field emerging from the right-hand-side of the film, according to (12.10), becomes
· 2 2
E ref L
¸
E ref L p 2 2
E transmitted ρ ∝ 2 2
e i k L +ρ
¡ ¢
+ E ref E ref + E ref p
L +ρ
2 2
L2 + ρ2
p (12.12)
E ref L 2 2
2
+ E ref e −i k L +ρ
L +ρ
p
2 2
Reference
beam Film We see the three distinct waves that emerge from the holographic film. The first
term in (12.12) represents the plane wave reference beam passing straight through
the film with some variation in amplitude (depicted in Fig. 12.8 (a)). The second
Virtual term in (12.12) has the identical form as the field from the original object (aside
image
from an overall amplitude factor). It describes an outward-expanding spherical
wave, which gives rise to a virtual image at the location of the original point object,
as depicted in Fig. 12.8 (b). The final term in (12.12) corresponds to a converging
spherical wave, which focuses to a point at a distance L from the observer’s side of
the screen (depicted in Fig. 12.8 (c)).
Field associated
with virtual
image
Reference
beam
Real
image
Field associated
with real
image
Film
Figure 12.8 Reference beam in-

cident on previously exposed
holographic film. (a) Part of the
beam goes through. (b) Part of the
beam takes on the field profile of
the original object. undeflected.
(c) Part of the beam converges to a
real image of the original object.
Exercises 307
Exercises
Exercises for 12.1 Interferograms
P12.1 An ideal Michelson interferometer that uses flat mirrors is perfectly

aligned to a wide collimated laser beam. Suppose that one of the mir-
rors is then misaligned by 0.1◦ . What is the spacing between adjacent
fringes on the screen if the wavelength is λ = 633 nm? What would
happen if, instead of tilting one of the mirrors, the angle of the input
beam (before the beamsplitter) changed by 0.1◦ ?
P12.2 An ideal Michelson interferometer uses flat mirrors perfectly aligned

to an expanding beam that diverges from a point 50 cm before the
beamsplitter. Suppose that one mirror is 10 cm away from the beam
splitter, and the other is 11 cm. Suppose also that the center of the
resulting bull’s-eye fringe pattern is dark. If a screen is positioned 10 cm
after the beam splitter, what is the radial distance to the next dark fringe
on the screen if the wavelength is λ = 633 nm?
Exercises for 12.2 Testing Optical Components
L12.3 Set up an interferometer and observe distortions to a mirror substrate

when the setscrew is over tightened.
Exercises for 12.3 Generating Holograms
P12.4 Consider a diffraction grating as a simple hologram. Let the light from
the “object” be a plane wave (object placed at infinity) directed onto
a flat film at angle θ. Let the reference beam strike the film at normal
incidence, and take the wavelength to be λ.
(a) What is the period of the fringes?
(b) Show that when re-illuminated by the reference beam, the three
terms in (12.10) give rise to zero-order and 1st-order diffraction (occur-
ring on each side of zero-order).
P12.5 (a) Show that the phase of the real image in (12.12) may be approxi-
mated as ∆φ = −kρ 2 /2L, aside from a spatially independent overall
phase. Compare with (11.10) and comment.
(b) This hologram is similar to a Fresnel zone plate, used to focus
extreme ultraviolet light or x-rays, for which it is difficult to make a lens.
Graph the field transmission for the hologram as a function of ρ and
superimpose a similar graph for a “best-fit” mask that has regions of
either 100% or 0% transmission. Use λ = 633 nm and L = (5 × 105 − 14 )λ
(this places the point source about a 32 cm before the screen). See
Fig. 12.9.
Consider the holographic pattern produced by the point object de-
scribed in section 12.4.
L12.6 Make a hologram.
Zone Plate Transmittance

1
0.5 Hologram
Transmittance
0
0 0.5 1
Figure 12.9 Field transmission for

a point-source hologram (upper)
and a Fresnel zone plate (middle),
and a plot of both as a function of
radius (bottom).
R48 T or F: The eikonal equation and Fermat’s principle depend on the

assumption that the wavelength is relatively small compared to features
of interest.

assumption that the index of refraction varies only gradually.

assumption that the angles involved must not be too big.

assumption that the polarization is important to the problem.
R52 T or F: Spherical aberration can be important even when the paraxial

approximation works well.
R53 T or F: Chromatic aberration (the fact that refractive index depends on

frequency) is an example of the violation of the paraxial approximation.
R54 T or F: The Fresnel approximation falls within the paraxial approxima-

tion.
R55 T or F: The imaging relation 1/ f = 1/d o + 1/d i relies on the paraxial ray
approximation.
R56 T or F: The spherical waves given by e i kR /R are exact solutions to

Maxwell’s equations.
R57 T or F: Spherical waves can be used to understand diffraction from

apertures that are relatively large compared to λ.
R58 T or F: Fresnel was the first to conceive of spherical waves.
R59 T or F: Spherical waves were accepted by Poisson immediately without

experimental proof.
309
R60 T or F: The array theorem is useful for deriving the Fresnel diffraction
from a grating.
R61 T or F: A diffraction grating with a period h smaller than a wavelength

is ideal for making a spectrometer.
R62 T or F: The blaze on a reflection grating can improve the amount of

energy in a desired order of diffraction.
R63 T or F: The resolving power of a spectrometer used in a particular

diffraction order depends only on the number of lines illuminated (not
wavelength or grating period).
R64 T or F: The central peak of the Fraunhofer diffraction from two nar-
row slits separated by spacing h has the same width as the central
diffraction peak from a single slit with width ∆x = h.
R65 T or F: The central peak of the Fraunhofer diffraction from a circular

aperture of diameter ` has the same width as the central diffraction
peak from a single slit with width ∆x = `.
R66 T or F: The Fraunhofer diffraction pattern appearing at the focus of a

lens varies in angular width, depending on the focal length of the lens
used.
R67 T or F: Fraunhofer diffraction can be viewed as a spatial Fourier trans-

form (or inverse transform if you prefer) on the field at the aperture.
Problems
R68 (a) Derive Snell’s law using Fermat’s principle.

(b) Derive the law of reflection using Fermat’s principle.
R69 (a) Consider a ray of light emitted from an object, which travels a
distance d o before traversing a lens of focal length f and then traveling
a distance d i .
· ¸ · ¸
y2 y1
Write a vector equation relating to . Be sure to simplify
θ2 θ1
image the equation so that only one ABCD matrix is involved.
object · ¸ · ¸
1 0 1 d
HINT: ,
−1/ f 1 0 1
(b) Explain the requirement on the ABCD matrix in part (a) that ensures
Figure 12.10 that an image appears for the distances chosen. From this requirement,
extract a familiar constraint on d o and d i . Also, make a reasonable
definition for magnification M in terms of y 1 and y 2 , then substitute to
find M in terms of d o and d i .
311
(c) A telescope is formed with two thin lenses separated by the sum of
their focal lengths f 1 and f 2 . Rays from a given far-away point all strike
the first lens with essentially the same angle θ1 . Angular magnification
M θ quantifies the telescope’s purpose of enlarging the apparent angle
between points in the field of view.
Give a sensible definition for angular magnification in terms of θ1 and
θ2 . Use ABCD-matrix formulation to derive the angular magnification
of the telescope in terms of f 1 and f 2 . Figure 12.11
· ¸
A B
R70 (a) Show that a system represented by a matrix (beginning
C D
and ending in the same index of refraction) can be made to look like
the matrix for a thin lens if the beginning and ending positions along
the z-axis are referenced from two principal planes, located distances
p 1 and p 2 before and after the system.
¯ ¯
¯ A B ¯
HINT: ¯ ¯ ¯ = 1.
C D ¯
(b) Where are the principal planes located and what is the effective
focal length for two identical thin lenses with focal lengths f that are
separated by a distance d = f (see Fig. 12.12)?
R71 Derive the on-axis intensity (i.e. x, y = 0) of a Gaussian laser beam if Figure 12.12
you know that at z = 0 the electric field of the beam is
ρ 02
− 2
E ρ 0 , z = 0 = E 0 e w0
¡ ¢
Fresnel:
k 2 +y 2
i e i kd e i 2d (x )Ï ¢ k 02 02 k 0 0
E x, y, d ∼ E x 0 , y 0 , 0 e i 2d (x +y ) e −i d (xx +y y ) d x 0 d y 0
¡ ¢ ¡
=−
λd
Z∞
π B 2 +C
r
−Ax 2 +B x+C
e dx = e 4A .
A
−∞
R72 (a) You decide to construct a simple laser cavity with a flat mirror and
another mirror with concave curvature of R = 100 cm. What is the
longest possible stable cavity that you can make?
HINT: Sylvester’s theorem is
¸N
· · ¸
A B 1
=
where cos θ = 12 (A + D).

(b) The amplifier is YLF crystal, which lases at λ = 1054 nm. You decide
to make the cavity 10 cm shorter than the longest possible (i.e. found in
part (a)). What is the value of w 0 , and where is the beam waist located
inside the cavity (the place we assign to z = 0)?
HINT: One can interpret the parameter R (z) as the radius of curvature
of the wave front. For a mode to exist in a laser cavity, the radius of
curvature of each of the end mirrors must match the radius of curvature
of the beam at that location.
w 0 − ρ22 i kz+i kρ2 −i tan−1 zz

E ρ, z = E 0
¡ ¢
e w (z) e 2R(z) e 0
w (z)
ρ2 ≡ x 2 + y 2
q
w (z) ≡ w 0 1 + z 2 /z 02
R (z) ≡ z + z 02 /z
kw 02
z0 ≡
2
R73 (a) Compute the Fraunhofer diffraction intensity pattern for a uni-
formly illuminated circular aperture with diameter `.
HINT:
k 2
+y 2 )
i e i kd e i 2d (x
ZZ
k 0 0
E x, y, d ∼ E x 0 , y 0 , 0 e −i d (xx +y y ) d x 0 d y 0
¡ ¢ ¡ ¢
=−
λd
Z2π
1
e ±i α cos(θ−θ ) d θ 0
0
J 0 (α) =
2π
0
Za
a
J 0 (bx) xd x = J 1 (ab)
b
0
J 1 (1.22π) = 0
2J 1 (x)
lim =1
x→0 x
(b) The first lens of a telescope has a diameter of 30 cm, which is the
only place where light is clipped. You wish to use the telescope to
examine two stars in a binary system. The stars are approximately 25
light-years away. How far apart need the stars be (in the perpendicular
sense) for you to distinguish them in the visible range of λ = 500 nm?
Compare with the radius of Earth’s orbit, 1.5 × 108 km.
313
R74 (a) Derive the Fraunhofer diffraction pattern for the field from a uni-
formly illuminated single slit of width ∆x. (Don’t worry about the
y-dimension.)
(b) Find the Fraunhofer intensity pattern for a grating of N slits of width
∆x positioned on the mask at x n0 = h n − N2+1 so that the spacing
¡ ¢
between all slits is h.

N k 0
e −i d xxn
P
HINT: The array theorem says that the diffraction pattern is
n=1
times the diffraction pattern of a single slit. You will need
N rN −1
rn =r
X
n=1 r −1
(c) Consider Fraunhofer diffraction from the grating in part (b). The
grating is 5.0 cm wide and is uniformly illuminated. For best resolution
in a monochromator with a 50 cm focal length, what should the width
of the exit slit be? Assume a wavelength of λ = 500 nm.
R75 (a) A monochromatic plane wave with intensity I 0 and wavelength λ

is incident on a circular aperture of diameter ` followed by a lens of
focal length f . Write the intensity distribution at a distance f behind
the lens.
(b) You wish to spatially filter the beam such that, when it emerges from
the focus, it varies smoothly without diffraction rings or hard edges. A
pinhole is placed at the focus, which transmits only the central portion
of the Airy pattern (inside of the first zero). Calculate the intensity Figure 12.13
pattern at a distance f after the pinhole using the approximation given
in the hint below.
HINT: A reasonably good approximation of the transmitted field is
2 2
that of a Gaussian E ρ, 0 = E f e −ρ /w 0 , where E f is the magnitude of
¡ ¢
the field at the center of the focus found in part (a), and the width
is w 0 = 2λ f # /π and f # ≡ f /`. The figure below shows how well the
Gaussian approximation fits the actual curve. We have assumed that
the first aperture is a distance f before the lens so that at the focus after
the lens the wave front is flat at the pinhole. To avoid integration, you
may want to use the result of P11.12 or P11.11(b) to get the Fraunhofer
limit of the Gaussian profile. (See figure below.)
Selected Answers Figure 12.14
R72: (a) 100 cm (b) 0.32 mm.

R73: (b) 4.8 × 108 km.
R74: (c) 5 µm.

Chapter 13
Blackbody Radiation
Hot objects glow. In 1860, Kirchhoff proposed that the radiation emitted by hot
objects as a function of frequency is approximately the same for all materials.1
The notion that all materials behave similarly led to the concept of an ideal
blackbody radiator. Most materials have a certain shininess that causes light to
reflect or scatter in addition to being absorbed and reemitted. However, light
that falls upon an ideal blackbody is absorbed perfectly before the possibility of
reemission, hence the name blackbody.
The distribution of frequencies emitted by a blackbody radiator is related
to its temperature. We often consider a blackbody radiator that is in thermal
equilibrium with the surrounding light that is absorbed and reemitted. If it is
not in thermal equilibrium, for example, if more light is emitted than absorbed,
then the object inevitably cools as light escapes to the environment, moving the
system toward thermal equilibrium.
The Sun is a good example of a blackbody radiator. The light emitted from the Gustav Kirchho (18241887, German)
was born in Konigsberg, the son of a
Sun is associated with its surface temperature. Any light that arrives to the Sun lawyer. Kirchho attended the Univer-
from outer space is virtually 100% absorbed, however little light that might be, so sity of Konigsberg. While still a student,
he developed what are now called Kirch-
the name blackbody aptly describes it. Mostly, light escapes to the much colder ho 's law for electrical circuits. During
surrounding space (i.e. it is not in thermal equilibrium), and the temperature of his career, Kirchho was a professor in
Breslau, Heidelberg, and nally Berlin.
the Sun’s surface is maintained by the fusion process within. As another example, Kirchho was one of the rst to study
a glowing tungsten filament in an ordinary light bulb may be reasonably described the spectra emitted by various objects
when heated. Not coincidentally, his
as a blackbody radiator. However, surface reflections make it less than ideal both
colleague in heidelberg was Robert
for absorption and emission. Bunsen, inventor of the Bunsen burner.
Experimentally, a near perfect blackbody radiator can be constructed from Kirchho coined the term `blackbody'
radiation. He demonstrated that an ex-
a hollow object. An example is shown in Fig. 13.1. As the interior of the object cited gas gives o a discrete spectrum,
is heated, the light present inside the internal cavity is in equilibrium with the and that an unexcited gas surrounding
a blackbody emitter produces dark lines
glowing walls. A small hole can be drilled through the wall to observe the radiation in the blackbody spectrum. Together
inside without significantly disturbing the system. The observation hole can be Kirchho and Bunsen discovered cae-
sium and rubidium. Later in his career,
thought of as a perfect blackbody since any light entering the hole from the
Kirchho showed how to derive Fres-
outside is eventually absorbed (before being potentially reemitted), if not on the nel's diraction formula starting from
the wave equation. (Wikipedia)
1 An important exception is atomic vapors, which have relatively few discrete spectral lines.
However, Kirchhoff’s assumption holds quite well for most solids, which are sufficiently complex.
315
316 Chapter 13 Blackbody Radiation
first bounce then on subsequent bounces inside the cavity.

In this chapter, we develop a theoretical understanding of blackbody radiation
and provide some historical perspective. The explanation given by Max Planck
in 1900 marks the birth of quantum mechanics. He postulated the existence of
electromagnetic quanta, which we now call photons. Einstein used Planck’s ideas
to explain the photoelectric effect and to develop the concept of stimulated and
spontaneous emission. Because of his analysis, Einstein can be thought of as the
father of light amplification by stimulated emission of radiation (LASER).
13.1 Stefan-Boltzmann Law

One of the earliest properties deduced about blackbody radiation is known as the
Stefan-Boltzmann law, first suggested by Stefan in 1879 and derived thermody-
Figure 13.1 Blackbody radiator. namically by Boltzmann in 1884.2 This early (somewhat cumbersome) derivation
Thermal light emerges from the is provided in appendix 13.A. 3 The Stefan-Boltzmann law says that the intensity I
small hole in the end.
(including all frequencies) that flows outward from an object’s surface is given by
I = eσT 4 , (13.1)
where σ is called the Stefan-Boltzmann constant and T is the absolute temper-

ature (in Kelvin) of the surface. The value of the Stefan-Boltzmann constant is
σ = 5.6696×10−8 W/m2 ·K4 . The dimensionless parameter e, called the emissivity,
is equal to one for an ideal blackbody surface. However, it takes on smaller values
for actual materials because of surface reflections. For example, the emissivity
of tungsten is approximately e = 0.4. This takes into account surface reflections,
which make it harder for a material to emit light as well as to absorb light.4
As mentioned in the introduction, one can construct an ideal blackbody
radiator from a material with e < 1 by creating an enclosure, or cavity, as depicted
in Fig. 13.2. A small hole in the wall behaves to the outside world like an ideal
blackbody surface. From the perspective of the outside world, the hole’s ‘surface’
has emissivity e = 1. Light within the cavity recirculates until it is eventually
absorbed. The intensity emerging from the hole automatically approaches that of
an ideal blackbody radiator.
It is sometimes useful to express intensity in terms of the energy density of
the light field u field (given by (2.53) in units of energy per volume). The connec-
tion between the intensity emerging from the observation hole in the wall of a
blackbody cavity and the energy density of the thermal light within the cavity is
Figure 13.2 Blackbody radiator
cu field 4σT 4
constructed as a cavity with a I= ⇒ u field = (13.2)
small hole to sample the internal 4 c
light. 2 See P. W. Milonni, The Quantum Vacuum An Introduction to Quantum Electrodynamics, Sect.
1.2 (San Diego: Academic Press, 1994).

3 It is less effort to obtain the Stefan-Boltzmann law using the Planck radiation formula as a
starting point (see P13.3).

4 Emissivity typically has some frequency dependence, so what is presented here is an oversim-
plification.
13.2 Failure of the Equipartition Principle 317
Within the enclosed cavity, light travels at speed c isotropically in all directions. A
factor of 1/2 arrises because only half of the energy travels towards the hole from
within the cavity as opposed to away. The remaining factor of 1/2 occurs because
the light emerging from the hole is directionally distributed over a hemisphere as
opposed to flowing in the direction of the surface normal n̂. The average over the
hemisphere is carried out as follows:
2π π/2 2π π/2
dφ r · n̂ sin θd θ dφ r cos θ sin θd θ
R R R R
0 0 0 0 1
= = (13.3)
2π π/2 2π π/2 2
dφ r sin θd θ dφ r sin θd θ
R R R R
0 0 0 0
Although (13.1) describes the total intensity of the light that leaves a blackbody
surface, it does not describe what frequencies make up the radiation field. This
frequency distribution was not fully described for another two decades, when
Max Planck developed his famous formula. Planck was first to arrive at the correct
formula for the spectrum of blackbody radiation, building on the work of others,
most notably Wien, who came very close. At first, Planck tweaked Wien’s formula
to match newly available experimental data. When he attempted to explain
it, he was forced to introduce the concept of light quanta. Even Planck was
uncomfortable with and perhaps disbelieved the assumption that his formula
implied, but he deserves credit for recognizing and articulating it.
13.2 Failure of the Equipartition Principle

In 1900, Lord Rayleigh attempted to explain the blackbody spectral distribution
(intensity per frequency) as a function of temperature by applying the equipar-
tition theorem to the problem. James Jeans gave a more complete derivation in
1905, which included an overall proportionality constant. They were hopelessly
behind, since Planck nailed the answer in 1900, but their failed (classical) ap-
proach is useful pedagogically, and for that reason it gets more attention than it
deserves. In this section, we also will examine the Rayleigh-Jeans approach to
illustrate the shortcomings of classical concepts. This will help us better appreci-
ate the quantum ideas in the following section. As we will see, the Rayleigh-Jeans
approach actually gets the right answer in the long-wavelength limit. In fairness
to Rayleigh and Jeans, they represented their formula as being useful only for long
wavelengths.
The thermodynamic law of equipartition implies that the energy in a system
on the average is distributed equally among all degrees of freedom in the system.
For example, a system composed of oscillators (say, electrons attached to ‘springs’
representing the response of the material on the walls of a blackbody radiator)
has an energy of k B T /2 for each degree of freedom, where k B = 1.38 × 10−23 J/K
is Boltzmann’s constant. Rayleigh and Jeans supposed that each unique mode
of the electromagnetic field should carry energy k B T just as each mechanical
spring in thermal equilibrium carries energy k B T (k B T /2 as kinetic and k B T /2 as
potential energy). The problem then reduces to that of finding the number of
unique modes for the radiation at each frequency.5 The idea is that requiring each
mode of electromagnetic energy to hold energy k B T should reveal the spectral
shape of blackbody radiation.
Number of Modes in an Electromagnetic Field

q
Each frequency is associated with a specific wave number k = k x2 + k y2 + k z2 . No-
tice that there are many ways (i.e. combinations of k x , k y , and k z ) to come up
with the same wave number k = ω/c (corresponding to a single frequency ν). To
count these ways properly, we can let our experience with Fourier series guide
us. Consider a box with each side of length L. The Fourier theorem (0.42) states
that the total field inside the box (no matter how complicated the distribution)
can always be represented as a superposition of sine (and cosine) waves. The total
field in the box can therefore be written as6
( )
∞ ∞ ∞
i (nk 0 x+mk 0 y+`k 0 z )
X X X
Re E e
n,m,` (13.4)
n=−∞ m=−∞ `=−∞
where each component of the wave number in any of the three dimensions is an
integer times
k 0 = 2π/L (13.5)
Considering a box of size L does not artificially restrict our analysis, since we may
later take the limit L → ∞ so that our box represents the entire universe. Moreover,
L will naturally disappear from our calculation when we later consider the density
of modes.
Figure 13.3 The volume of a thin
spherical shell in n, m, ` space. We can think of a given wave number k as specifying the equation of a sphere in a
coordinate system with axes labeled n, m, and `:
¶2
k
µ
n 2 + m 2 + `2 = (13.6)
k0
The fact that the integers n, m, and ` range over both positive and negative values
automatically takes into account that the field may travel in the forwards or the
backwards direction.
We need to know how many more ways there are to choose n, m, and ` when the
wave number k/k 0 increases to (k + d k)/k 0 . The answer is the difference in the
volume of the two spheres shown in Fig. 13.3:
k2 d k
µ ¶
# modes in (k,k+d k) = 4π 2 (13.7)
k0 k0
This is the number of terms in (13.4) associated with a wave number between k
and k + d k.
5 See O. Svelto, Principles of Lasers, 4th ed., translated by D. C. Hanna, Sect. 2.2.1 (New York:
Plenum Press, 1998).

6 The Fourier expansion 13.4 implies that the field on the right and left of each dimension match
up, which is known as periodic boundary conditions.

13.3 Planck’s Formula 319
According to the Rayleigh-Jeans assumption, each mode should carry on

average equal energy k B T . The energy density associated with a specified range of
wave numbers d k is then k B T /L 3 times the number of modes within that range
(13.7).
The total energy density in the field involving all wave numbers is then7
Z∞ Z∞ 2
k B T 4πk 2 k
u field = 2 × 3 × 3 d k = k B T dk (13.8)
L k0 π2
0 0
where the extra factor of 2 accounts for two independent polarizations, not speci-
fied in (13.4). As anticipated, the dependence on L has disappeared from (13.8)
after substituting from (13.5).
We can immediately see that (13.8) disagrees drastically with the Stefan-
Boltzmann law (13.2), since (13.8) is proportional to temperature rather than
to its fourth power. In addition, the integral in (13.8) is seen to diverge, meaning
that regardless of the temperature, the light carries infinite energy density! This
has since been named the ultraviolet catastrophe since the divergence occurs
on the short wavelength end of the spectrum. This is a clear failure of classical
physics to explain blackbody radiation. Nevertheless, Rayleigh emphasized the
fact that his formula works well for the longer wavelengths.
It is instructive to make the change of variables k = ω/c in the integral to write
Z∞
ω2
u field = k B T dω (13.9)
π2 c 3
0
James Jeans (18771946, English) was
2 2 3
The important factor ω /π c can now be understood to be the number of modes born in Ormskirk, England. He attended
Cambridge University and later taught
per frequency. Then (13.9) is rewritten as there for most of his career. He also
taught at Princeton University for a
Z∞ number of years. One of his major con-
u field = ρ (ω) d ω (13.10) tributions was the development of Jeans

length, the critical radius for interstel-
0 lar clouds, which determines whether
a cloud will collapse to form a star. In
where his later career, Jeans became some-
ω2 what well known to the public for his
ρ Rayleigh-Jeans (ω) = k B T 2 3 (13.11) lay-audience books highlighting scien-
π c tic advances, in particular relativity
describes (incorrectly) the spectral energy density of the radiation field associated and cosmology. (Wikipedia)
with blackbody radiation.
13.3 Planck’s Formula

In the late 1800’s as spectrographic technology improved, experimenters acquired
considerable data on the spectra of blackbody radiation. For the first time, de-
tailed maps of the intensity per frequency associated with blackbody radiation

became available over a fairly wide wavelength range. In keeping with Kirchhoff’s
notion of an ideal blackbody radiator, the results were observed to be indepen-
dent of the material for most solids. The intensity per frequency depended only
on temperature and when integrated over all frequencies agreed with the Stefan-
Boltzmann law (13.1).
In 1896, Wilhelm Wien considered the known physical and mathematical
constraints on the spectrum of blackbody radiation and proposed a spectral
function that seemed to work:8
ħω3 e −ħω/kB T
0 2 4 6 8 10 ρ Wien (ω) = (13.12)
π2 c 3
Figure 13.4 Energy density per

An important feature of (13.12) is that it gives a result proportional to T 4 when
frequency according to Planck, integrated over all frequency ω (i.e. the Steffan-Boltzmann law).
Wien, and Rayleigh-Jeans. Wien’s formula did a fairly good job of fitting the experimental data. However,
in 1900 Lummer and Pringshein, colleagues of Max Planck, reported experimental
data that deviated from the Wien distribution at long wavelengths (infrared).
Planck was privy to this information early on and introduced a modest revision
to Wien’s formula that fit the data beautifully everywhere:
ħω3
ρ Planck (ω) = (13.13)
π2 c 3 e ħω/kB T − 1
£ ¤
where ħ = 1.054 × 10−34 J · s is an experimentally determined constant.9

Figure 13.4 shows the Planck spectral distribution curve together with the
Rayleigh-Jeans curve (13.11) and the Wien curve (13.12). As is apparent, the Wien
distribution does a good job nearly everywhere. However, at long wavelengths
it was off by just enough for the experimentalists to notice that something was
wrong.
At this point, it may seem fair to ask, what did Planck do that was so great?
Wilhelm Wien (18641928, Country) After all, he simply guessed a function that was only a slight modification of
was born in Gaken, Prussia (now Pri-
morsk, Russia). As a teenager, he at- Wien’s distribution. And he knew the ‘answer from the back of the book’, namely
tended schools in Rastenburg and then Lummer’s and Pringshein’s well done experimental results. (At the time, Planck
Heidelberg. He later attended the Uni-
versity of Göttingen and then the Uni-
was unaware of the work by Rayleigh.)
versity of Berlin. In 1886, he received Planck gets well-deserved credit for interpreting the meaning of his new
his Ph.D. after working under Hermann
von Helmholtz where he studied the in-
formula. His interpretation was what he called an “act of desperation.” He did
uence of materials on the color of light. not necessarily believe in the implications of his formula; in fact, he presented
In 1896 Wien developed an empirical
them somewhat apologetically. It was several years later that the young Einstein
formula for the spectral distribution of
blackbody radiation. He collaborated published his paper explaining the photoelectric effect in the context of Planck’s
with Planck, who gave the law a foun- work.
dation in electromagnetic and thermody-
namic theory. Planck later improved the
Planck’s insight was an enormous step toward understanding the quantum
formula, whereupon it became known by nature of light. Nevertheless, it took another three decades to develop a more
his name. However, Wien's formula for
the peak wavelength of the blackbody 8 The constant h had not yet been introduced by Planck. The actual way that Wien wrote his
curve, called Wien's displacement law,
distribution was ρ Wien (ω) = aω3 e −bω/T , where a and b were parameters used to fit the data.
remains valid. In 1898, Wien identied 9 Planck’s constant was first introduced as h = 6.626 × 10−34 J · s, convenient for working with
a positive particle equal in mass to the
hydrogen atom, which was later named
frequency ν, expressed in Hz. It is common to write ħ ≡ h/2π when working with frequency ω,
the proton. Wien received the Nobel expressed in rad/s.
prize in 1911 for his work on heat and
radiation. (Wikipedia)
13.3 Planck’s Formula 321
complete theory of quantum electrodynamics. Students should appreciate that

the very people who developed quantum mechanics were also bothered by its
confrontation with deep-seated intuition. If quantum mechanics bothers you,
you are in good company!
Planck found that he could derive his formula only if he made the following
strange assumption: A given mode of the electromagnetic field is not able to
carry an arbitrary amount of energy (for example, k B T as Rayleigh and Jeans
used, which varies continuously as the temperature varies). Rather, the field
can only carry discrete amounts of energy separated by spacing ħω. Under this
assumption, the probability P n that a mode of the field is excited to the n th level
is proportional to the Boltzmann statistical weighting factor e −nħω/kB T . A review
of the Boltzmann factor is given in Appendix 13.B.
Probable Energy in Each Field Mode
The Boltzmann factor can be normalized by dividing by the sum of all such factors
to obtain the probability of having energy nħω in a particular mode:
Max Planck (18581947, German)
e −nħω/kB T h i was born in Kiel, the sixth child in his
Pn = ∞ = e −nħω/kB T 1 − e −ħω/kB T (13.14) family. His father was a law professor.
P −mħω/k T When Max was about nine years old,
e B
m=0 his family moved to Munich where he
attended gymnasium. A mathematician,
We used (0.66) to accomplish the above sum, which is a geometric series. Herman Muller took an interest in his
schooling and tutored him in mechanics
The expected energy in a particular mode of the field is the sum of each possible and astronomy. Planck was a gifted
energy level (i.e. nħω) times the probability of it occurring: musician, but he decided to pursue a
career in physics. At age 16 he enrolled
∞ h ∞
iX in the University of Munich. By age 22,
−ħω/k B T −nħω/k B T
X
nħωP n = ħω 1 − e ne he had nished his doctoral dissertation
n=0 n=0 and habilitation thesis. He was initially
h
−ħω/k B T
i ∂ ∞ ignored by the academic community and
e −nħω/kB T
X
= ħω 1 − e worked for a time as an unpaid lecturer.
∂ (ħω/k B T ) n=0
(13.15) He became an associate professor of
h
−ħω/1−k B T
i ∂ 1 theoretical physics at the University of
= −ħω 1 − e Kiel and then a few years later took
∂ (ħω/k B T ) 1 − e −ħω/kB T over Kirchho 's post at the University
ħω of Berlin. After nearly twenty years of
= idillic and happy family life, a series
e ħω/kB T − 1
of tragedies hit the Planck household.
We used (0.66) again as well as a clever derivative trick. Planck's rst wife and mother of four,
died. Then his eldest son was killed
in action during World War I. Soon
after, his twin daughters each died
Equation (13.15) provides the expected energy in any of the modes of the radi- giving birth to their rst child. Later
Planck's remaining son from his rst
ation field, as dictated by Planck’s assumption. To obtain the Planck distribution
marriage was executed for participating
(13.13), we replace k B T in the Rayleigh-Jeans formula (13.10) with the correct in a failed attempt to assassinate Hitler.
expected energy (13.15).10 Planck won the Nobel prize in 1918 for
his introduction of energy quanta, but
It is interesting that we are now able to derive the constant in the Stefan- he had serious reservations about the
Boltzmann law (13.2) in terms of Planck’s constant ħ (see P13.3). The Stefan- course that quantum mechanics theory
took. (Wikipedia)
Boltzmann law is obtained by integrating the spectral density function (13.13)

over all frequencies to obtain the total field energy density, which is in thermal
equilibrium with the blackbody radiator:
Z∞
4 π2 k B4 4 4
u field = ρ Planck (ω)d ω = T ≡ σT 4 (13.16)
c 60c 2 ħ3 c
0
Since Planck’s constant was not introduced until a couple decades after the Stefan-
Boltzmann law was developed, one might more appropriately say that the Stefan-
Boltzmann constant pins down Planck’s constant.
Example 13.1
Determine ρ Planck (λ) such that
Z∞ Z∞
u field = ρ Planck (ω) d ω = ρ Planck (λ) d λ
0 0
where ρ Planck (ω) and ρ Planck (λ) represent distinct functions distiguished by their
arguments.
Solution: The change of variables λ ≡ 2πc/ω ⇒ d ω = −2πcd λ/λ2 gives
Z0 ¶ Z∞
ħ (2πc/λ)3 dλ
µ
16ħc
u field = ¤ −2πc 2 = ¤dλ
π2 c 3 e ħ(2πc/λ)/kB T − 1 λ λ5 e 2πħc/λkB T − 1
£ £
∞ 0
By inspection, we get
8πhc
ρ Planck (λ) = (13.17)
λ5 e hc/λkB T − 1
£ ¤
where we have written h ≡ 2πħ. It is interesting to note that the maximum of

ρ Planck (λ) occurring at λmax and the maximum of ρ Planck (ω) occurring at ωmax do not
correspond to a matching wavelength and frequency. That is, λmax 6= 2πc/ωmax ,
because of the nonlinear nature of the variable transformation. (See problem
P13.4.)
13.4 Einstein’s A and B Coefficients

More than a decade after Planck introduced his formula, and after Niels Bohr
had proposed that electrons occupy discrete energy states in atoms, Einstein
reexamined blackbody radiation in terms of Bohr’s new idea. If the material of a
blackbody radiator interacts with a particular mode of the field with frequency ω,
then electrons in the material must make transitions between two energy levels
with energy separation hω. Since the radiation of a blackbody is in thermal equi-
librium with the material, Einstein postulated that the field stimulates electron
13.4 Einstein’s A and B Coefficients 323
transitions between energy levels. In addition, he postulated that some transi-

tions must occur spontaneously. (If the possibility of spontaneous transitions is
not included, then there can be no way for a field mode to receive energy if none
is present to begin with.)
Einstein wrote down rate equations for populations of the two levels N1 and
N2 associated with the transition ħω:11
Ṅ1 = A 21 N2 − B 12 ρ (ω) N1 + B 21 ρ (ω) N2 ,

(13.18)
Ṅ2 = −A 21 N2 + B 12 ρ (ω) N1 − B 21 ρ (ω) N2
The coefficient A 21 is the rate of spontaneous emission from state 2 to state 1,

B 12 ρ (ω) is the rate of stimulated absorption from state 1 to state 2, and B 21 ρ (ω)
is the rate of stimulated emission from state 2 to state 1.
In thermal equilibrium, the rate equations (13.18) are both equal to zero (i.e.,
Ṅ1 = Ṅ2 = 0), since the relative populations of each level must remain constant.
We can then solve for the spectral density ρ (ω) at the given frequency. In this case,
either expression in (13.18) yields
Albert Einstein (18791955, German)
is without a doubt the most famous sci-
A 21
ρ (ω) = N1
(13.19) entist in history. Time Magazine named
him Person of the Century. Born in
N2 B 12 − B 21 Ulm to a (non practicing) Jewish fam-
ily, young Albert was inuenced by a
In thermal equilibrium, the spectral density must match the Planck spectral medical student, Max Talmud, who took
density formula (13.13). In making the comparison, we should first rewrite the meals with his family and enthusiasti-
cally introduced the 10-year-old Albert
ratio N1 /N2 of the populations in the two levels using the Boltzmann probability
to geometry and other topics. Einstein's
factor (see Appendix 13.B): father wanted Albert to be trained as an
electrical engineer, but Albert clashed
N1 e −E 1 /kB T with his teachers in that program and

= = e (E 2 −E 1 )/kB T = e hω/kB T (13.20) withdrew. Einstein then attended school
N2 e −E 2 /kB T in Switzerland, and subsequently en-
tered a mathematics program at the
Then when equating (13.19) to the Planck blackbody spectral density (13.13) we Polytechnic in Zurich. There, Einstein
met his rst wife, Mileva Maric, a fellow
get math student, who he later divorced
A 21 ħω3 before marrying Elsa Lowenthal. Early
= (13.21) on, Einstein could not nd a job as
e ħω/kB T B 12 − B 21 π2 c 3 e ħω/kB T − 1
£ ¤
a professor, and so he worked in the
12 Swiss patent oce until his "Miracle
From this expression we deduce that Year" (1905), when published four ma-
jor papers, including relativity and the
B 12 = B 21 (13.22) photoelectric eect (for which he later

received the Nobel prize). Thereafter,
job oers were never in short supply.
and In 1933, as the Nazi regime came to
3
ħω power, Einstein immigrated from Berlin
A 21 = 2 3 B 21 (13.23)
π c to the US and became a professor at
Princeton University. Einstein is most
We see from (13.22) that the rate of stimulated absorption is the same as the noted for special and general relativity,
for which he became a celebrity scientist
rate of stimulated emission. In addition, if one knows the rate of stimulated
in his own lifetime. Einstein also made
11 See P. W. Milonni, The Quantum Vacuum An Introduction to Quantum Electrodynamics, Sect. huge contributions to statistical and
quantum mechanics. (Wikipedia)
12 We assume that energy levels 1 and 2 are non-degenerate. Some modifications must be made
in the case of degenerate levels, but the procedure is similar.

emission between a pair of states, it follows from (13.23) that one also knows the
rate of spontaneous emission. This is remarkable because to derive A 21 directly,
one needs to use the full theory of quantum electrodynamics (the complete
photon description). However, to obtain B 21 , it is actually only necessary to use a
semiclassical theory, where the light is treated classically and the energy levels in
the material are treated quantum-mechanically using the Schrödinger equation.
In writing the rate equations, (13.18), Einstein predicted the possibility of
creating lasers fifty years in advance of their development. These rate equations
are still valid even if the light is not in thermal equilibrium with the material.
The equations suggest that if the population in the upper state 2 can be made
artificially large, then amplification will result via the stimulated transition. The
rate equations also show that a population inversion (more population in the
upper state than in the lower one) cannot be achieved by ‘pumping’ the material
with the same frequency of light that one hopes to amplify. This is because the
stimulated absorption rate is balanced by the stimulated emission rate. The
material-dependent parameters A 21 and B 12 = B 21 are called the Einstein A and B
coefficients.
Appendix 13.A Thermodynamic Derivation of the Stefan-

Boltzmann Law
In this appendix, we derive the Stefan-Boltzmann law without relying on the
Planck blackbody formula.13 This derivation is included mainly for historical
interest. The derivation relies on the 1st and 2nd laws of thermodynamics.
Consider a container whose walls are all at the same temperature and in
thermal equilibrium with the radiation field inside. Notice that the units of energy
density u field (energy per volume) are equivalent to force per area, or in other
words pressure. It turns out that the radiation exerts a pressure of
P = u field /3 (13.24)
on the walls of the container. This can be derived from the fact that radiation of
energy ∆E imparts a momentum
∆E
∆p = cos θ (13.25)
c
when it is absorbed with incident angle θ on a surface.14 A similar momentum is

imparted when radiation is emitted.
13 See P. W. Milonni, The Quantum Vacuum An Introduction to Quantum Electrodynamics, Sect.

14 The fact that light carries momentum was understood well before the development of the
theory of relativity and the photon description of light.

13.A Thermodynamic Derivation of the Stefan-Boltzmann Law 325
Derivation of (13.24)
Consider a thin layer of space adjacent to a container wall with area A. If the layer
has thickness ∆z, then the volume in the layer is A∆z. Half of the radiation inside
the layer flows toward the wall, where it is absorbed. The total energy in the layer
that will be absorbed is then ∆E = (A∆z)u field /2, which arrives during the interval
∆t = ∆z/(c cos θ), assuming for the moment that all light is directed with angle θ;
we must average the angle of light propagation over a hemisphere.
The pressure on the wall due to absorption (i.e. force or d p/d t per area) is then
2π π/2
∆p 1
dφ sin θ d θ
R R
∆t A Zπ/2
0 0 u field u field
P abs = = cos2 θ sin θ d θ = (13.26)
2π π/2 2 6
dφ sin θ d θ
R R
0
0 0
In equilibrium, an equal amount of radiation is also emitted from the wall. This
gives an additional pressure P emit = P abs , which confirms that the total pressure is
given by (13.24).
Figure 13.5 Field inside a black-
body radiator.
We derive the Stefan-Boltzmann law using the concept of entropy, which is
defined in differential form by the quantity
dQ
dS ≡ (13.27)
T
where d Q is the injection of heat (or energy) into the radiation field in the box
and T is the temperature at which that injection takes place. We would like to
write d Q in terms of u field , V , and T . Then we may invoke the fact that S is a state
variable, which implies
∂2 S ∂2 S
= (13.28)
∂T ∂V ∂V ∂T
This is a mathematical statement of the fact that S is fully defined if the internal
energy, temperature, and volume of a system are specified. That is, S does not
depend on past temperature and volume history; it is dictated by the present
state of the system.
To obtain d Q in the form that we need, we can use the 1st law of thermody-
namics. It states that a change in internal energy dU = d (u fieldV ) can take place
by the injection of heat d Q or by doing work dW = P dV as the volume increases:
d Q = dU + P dV = d (u fieldV ) + P dV
1
= V d u field + u field dV + u field dV (13.29)
3
d u field 4
=V d T + u field dV
dT 3
We have used energy density times volume to obtain the total energy U in the radi-
ation field in the box. We have also used (13.24) to obtain the work accomplished
by pressure as the volume changes.
We can use (13.29) to rewrite (13.27) as
V d u field 4u field
dS = dT + dV (13.30)
T dT 3T
When we differentiate (13.30) with respect to temperature or volume we get
∂S 4u field
=
∂V 3T (13.31)
∂S V d u field
=
∂T T dT
We are now able to evaluate the partial derivatives in (13.28), which give
∂2 S 4 ∂ u field 4 1 ∂u field 4 u field

= = −
∂T ∂V 3 ∂T T 3 T ∂T 3 T2 (13.32)
2
∂ S 1 d u field
=
∂V ∂T T d T
Since by (13.28) these two expressions must be equal, we get a differential
equation relating the internal energy of the system to the temperature:
4 1 ∂u field 4 u field 1 d u field ∂u field 4u field

− = ⇒ = (13.33)
3 T ∂T 3 T 2 T dT ∂T T
The solution to this differential equation is (13.2), where 4σ/c is a constant to be
determined experimentally.
Appendix 13.B Boltzmann Factor

The entropy of an object is defined by
S obj = k B ln n obj (13.34)
which depends on the number of configurations n obj for a given state (defined,
for example, by fixed energy and volume). Now imagine that the object is placed
in contact with a very large thermal reservoir. The ‘object’ could be the electro-
magnetic radiation inside a hollow blackbody apparatus, and the reservoir could
be the walls of the apparatus, capable of holding far more energy than the light
field can hold. The condition for thermal equilibrium between the object and the
reservoir is
∂S obj ∂S res 1
= ≡ (13.35)
∂Uobj ∂Ures T
where temperature has been introduced as a definition, which is consistent with
(13.27).
The total number of configurations for the combined system is N = n obj n res ,
where n obj and n res are the number of configurations available within the object
and the reservoir separately. A thermodynamic principle is that all possible
13.B Boltzmann Factor 327
configurations are equally probable. In thermal equilibrium, the probability for a

given configuration in the object is therefore proportional to
N
P∝ = n res = e S res /kB (13.36)
n obj
where we have invoked (13.34).

Meanwhile, a Taylor’s series expansion of S res yields
¡ eq ¢ ∂S res ¯
¯
∼ eq
¡ ¢
S res (Ures ) = S res Ures + ¯ Ures −Ures + ... (13.37)
∂Ures Ures
¯ eq
Higher order terms are not needed since we assume the reservoir to be very large
so that it is disturbed only slightly by variations in the object. Since the overall
energy of the system is fixed, we may write
eq
Ures −Ures = ∆Ures = −∆Uobj (13.38)
where ∆Uobj is a small change in energy in the object. When (13.35), (13.37), and
(13.38) are introduced into (13.36), the probability for the specific configuration
eq ∆Uob j
1
S (U )−
becomes P ∝ e kB res res kB T , or simply
∆Uob j
−
P ∝e kB T
(13.39)
since the first term in the exponent is constant. ∆Uobj represents an amount
energy added to the object to establish a configuration. In the case of blackbody
radiation, a mode takes on energy ∆Uobj = nħω, where n is the number of energy
quanta in the mode. The probability that a mode carries energy nħω is therefore
− knħω
proportional to e B T
.
Exercises
Exercises for 13.1 Stefan-Boltzmann Law
P13.1 The Sun has a radius of R S = 6.96 × 108 m. What is the total power that
it radiates, given a surface temperature of 5750 K?
P13.2 A 1 cm-radius spherical ball of polished gold hangs suspended inside

an evacuated chamber that is at room temperature 20◦ C. There is no
pathway for thermal conduction to the chamber wall.
(a) If the gold is at a temperature of 100◦ C, what is the initial rate of
temperature loss in ◦ C/s? The emissivity for polished gold is e = 0.02.
The specific heat of gold is 129 J/kg · ◦ C and its density is 19.3 g/cm3 .
HINT: Q = mc∆T and Power = Q/∆t .
(b) What is the initial rate of temperature loss if the ball is coated with
flat black paint, which has emissivity e = 0.95?
HINT: You should consider the energy flowing both ways.
Exercises for 13.3 Planck’s Formula
P13.3 Derive (or try to derive) the Stefan-Boltzmann law by integrating the
(a) Rayleigh-Jeans energy density
Z∞
u field = ρ Rayleigh-Jeans (ω) d ω
0
Please comment.
(b) Wien energy density
Z∞
u field = ρ Wien (ω) d ω
0
Please evaluate σ.
R∞ 6
HINT: x 3 e −ax d x = a4
.
0
(c) Planck energy density
Z∞
u field = ρ Planck (ω) d ω
0
Please evaluate σ. Compare results of (b) and (c).

R∞ 3 x π4
HINT: exaxd−1 = 15a 4.
0
Exercises 329
P13.4 (a) Derive Wien’s displacement law
0.00290 m · K
λmax =
T
which gives the strongest wavelength present in the blackbody spectral
distribution.
HINT: See Example 13.1. You may like to know that the solution to the
transcendental equation (5 − x) e x = 5 is x = 4.965.
(b) What is the strongest wavelength emitted by the Sun, which has a
surface temperature of 5750 K (see P13.1)?
(c) Also find νmax and show that it is not the same as c/λmax . Why would
we be interested mainly in λmax ?
Index
ABCD Law for Gaussian Beams, 291 Brewster’s Angle, 80

ABCD matrices Brewster, David, 80
transmission through a curved broadband, 172
surface, 238
ABCD Matrices for Combined Optical carrier frequency, 182
Elements, 239 Cartesian coordinates, 1
ABCD matrix, 236 causality, 195, 196
aberration, 235 Causality and Exchange of Energy
Aberrations and Ray Tracing, 248 with the Medium, 191
absolute value, 1, 10 centroid, 193
Airy pattern, 280 characteristic matrix, 108
Ampere’s Law, 32 chirped pulse amplification, 189
Ampere’s law, 27 chirping, 185
angle addition formula, 7 Christiaan Huygens, 126
anisotropic, 117 chromatic aberration, 248
aperture, 258 circular polarization, 146
Arago, Francois Jean Dominique, 261 circular polarizer, 160
array theorem, 275 circularly polarized light, 145
Array Theorem, The, 282 Clausius-Mossotti Relation, 63
arrival time, 186 coefficient of finesse, 95
astigmatism, 249 coherence length, 208
coherence time, 208
beam waist, 275, 287, 290 Coherence Time and Fringe Visibility,
Beyond Critical Angle: Tunneling of 208
Evanescent Waves, 96 coma, 249
biaxial, 123 complex angle, 11
Biaxial and Uniaxial Crystals, 123 complex conjugate, 10
Biot, Jean-Baptiste, 30 complex notation, 45, 47
Biot-Savart law, 30 Complex Numbers, 6
birefringence, 117, 122, 126 complex plane, 9
blackbody, 315 complex polar representation, 9
blackbody radiation, 315 concave, 237
Bohr, Niels, 322 conductivity, 70
Boltzmann Factor, 326 constitutive relation, 49
boundary conditions, 75 Constitutive Relation in Crystals, 117
Boundary Conditions For Fields at an continuity equation, 33
Interface, 84 convex, 237
331
332 INDEX
convolution theorem, 24 elliptical polarization, 145, 147, 148

cosine complex representation, 7 Elliptically Polarized Light, 148
Coulomb’s law, 28 ellipticity, 149, 158
critical angle, 81 emissivity, 316
cross product, 2 energy density, 192, 316
curl, 3 Energy Density of Electric Fields, 66
current density, 29 Energy Density of Magnetic Fields, 68
curvature of the field (aberration), 250 energy transport velocity, 192
cylindrical coordinates, 3 Equipartition Principle, Failure of, 317
Euler’s formula, 7
degree of coherence, 206, 208 evanescent waves, 82
degree of polarization, 146, 159, 162 extraordinary, 117
density of modes, 318 extraordinary index, 123, 124
depth of focus, 290
determinant, 12 f-number, 290
dielectric, 45 Fabry, Charles, 98
Diffraction Grating, 284 Fabry-Perot, 98
Diffraction of a Gaussian Field Pro- Fabry-Perot etalon, 100
file, 287 Fabry-Perot interferometer, 100
Diffraction with Cylindrical Symme- Fabry-Perot, Distinguishing Nearby
try, 265 Wavelengths, 101
Dirac delta function, 17 Failure of the Equipartition Principle,
dispersion, 45, 171, 182, 183 317
dispersion relation, 46 far field, 257, 264
in crystals, 121 Faraday’s Law, 31
displacement current, 34 Faraday’s law, 27, 46
Distinguishing Nearby Wavelengths Faraday, Michael, 31
in a Fabry-Perot Instrument, fast axis, 153
101 Fermat’s Principle, 231
distortion, 250 Fermat’s principle, 227
divergence, 3 Fermat, Pierre, 231
divergence theorem, 5 finesse, 104
dot product, 2 finesse, coefficient of, 95
Double-Interface Problem Solved fluence, 205
Using Fresnel Coefficients, focal length, 242
90 Fourier expansion, 14
Fourier integral theorem, 14, 16
eikonal equation, 227, 230 Fourier Spectroscopy, 210
Eikonal Equation, The, 228 Fourier Theory, 13
Einstein’s A and B Coefficients, 322 Fourier transform, 16, 177
Einstein, Albert, 323 Fraunhofer Approximation, 264
electric field, 28 Fraunhofer Diffraction Through a
Electric Field in Crystals, 131 Lens, 275
ellipsometer, 157 Fraunhofer, Joseph, 264
Ellipsometry, 157 free spectral range, 102
INDEX 333
frequency, 46 Huygens’ Elliptical Construct for a

Frequency Spectrum of Light, 176 Uniaxial Crystal, 134
Fresnel Approximation, 262 Huygens’ Principle as Formulated by
Fresnel Coefficient, 77 Fresnel, 258
Fresnel coefficients, 78 Huygens, Christian, 257
Fresnel Coefficients, The, 77 hyperbolic cosine, 7
Fresnel’s equation, 121 hyperbolic sine, 7
Fresnel, Augustin, 77
Fresnel-Kirchhoff Diffraction For- identity matrix, 11
mula, 267 image, 233, 235
fringe, 302 Image Formation, 241
fringe pattern, 301 imaginary number, 7
fringe visibility, 208 imaginary part, 8
fringes, 100 Index of Refraction, 48
frustrated total internal reflection, 96 index of refraction, 45
Index of Refraction of a Conductor,
Gabor, Dennis, 304 54
Galileo, 241 instantaneous power spectrum, 195
Gauss’ Law, 28 intensity, 59
Gauss’ law, 27, 28 Intensity of Superimposed Plane
Gauss’ Law for Magnetic Fields, 29 Waves, 172
Gauss, Friedrich, 29 Interferograms, 301
Gaussian Laser Beams, 289 inverse Fourier transform, 177
Generalized Context for Group Delay, inverse matrix, 11
185 irradiance, 56, 59
Generating Holograms, 303 Irradiance of a Plane Wave, 58
Gouy shift, 290 isotropic, 117
gradient, 3 isotropic medium, 58
grating, 189
Green’s Theorem, 270 Jeans, James Hopwood, 319
group delay, 172, 187 Jones Matrices for Wave Plates, 153
group delay function, 183 Jones matrix, 145
group velocity, 171, 175, 183 Jones Matrix for Polarizers at Arbi-
Group vs. Phase Velocity: Sum of Two trary Angles, 152
Plane Waves, 174 Jones vector, 145, 148
Jones Vectors for Representing Polar-
half-wave plate, 154 ization, 147
Hankel transform, 266 Jones, R. Clark, 147
helicity, 149, 158
Helmholtz equation, 261 Kirchhoff, Gustav, 315
hologram, 301 Kramers-Kronig Relations, 196
Holographic Wavefront Reconstruc- Kronecker delta function, 215
tion, 304 Laplacian, 4
holography, 301 laser, 316, 324
laser beam, 289
334 INDEX
laser cavity, 245 Packet Propagation and Group Delay,

law of reflection, 75 181
lens, 239 paraxial approximation, 228, 235
lens maker’s formula, 239 paraxial ray theory, 228
Linear Algebra, 11 Paraxial Rays and ABCD Matrices,
linear medium, 49 235
linear polarization, 145 paraxial wave equation, 263, 264
Linear Polarizers and Jones Matrices, Parseval’s relation, 18
149 Parseval’s theorem, 178
Linear, Circular, and Elliptical Polar- Partially Polarized Light, 159
ization, 146 pellicle, 99
Lorentz model, 51 phase delay, 182
Lorentz Model of Dielectrics, The, 51 phase velocity, 174, 183
Lorentz, Hendrik, 51 photometry, 60
Lorentz-Lorenz formula, 64 photon, 316
Planck’s Formula, 319
magnetic field, 29 Planck, Max, 321
magnification, 242 plane of incidence, 74
magnitude, 1 Plane Wave Propagation in Crystals,
matrix multiplication, 11 119
Maxwell’s Adjustment to Ampere’s Plane Wave Solutions to the Wave
Law, 33 Equation, 45
Maxwell’s equations, 27 plane waves, 45, 47
Maxwell, James, 34 plasma frequency, 53
Michelson Interferometer, 203 Poisson’s spot, 260
Michelson, Albert, 205 polarizability, 63
mirage, 230 Polarization Effects of Reflection and
Mueller matrix, 163 Transmission, 156
Multilayer Coatings, 105 polarization of a medium, 36
multimode, 289 polarization of light, 145
Polarization of Materials, 36
narrowband, 172
polarizer, 145
negative crystal, 124
Polaroid, 149
Newton, Isaac, 172
positive crystal, 124
normal to a surface, 6
power spectrum, 177
object, 235 Poynting vector, 56, 58
obliquity factor, 262, 270 Poynting Vector in a Uniaxial Crystal,
optic axes of a crystal, 123 125
optical activity, 166 Poynting’s Theorem, 55
optical axis, 228, 235 Poynting’s theorem, 56, 192
optical path length, 232 Poynting, John Henry, 56
ordinary, 124 principal axes, 119, 122
oscillator strength, 53 principal planes, 228, 244
Principal Planes for Complex Optical
p-polarized light, 74, 91
INDEX 335
Systems, 244 Savart, Felix, 30

principal value, 197 scalar diffraction, 258
pulse chirping, 183 Scalar Diffraction Theory, 260
Pulse Chirping in a Grating Pair, 189 scalar Helmholtz equation, 261
pulse stretching, 183 senkrecht, 74
Setup of a Fabry-Perot Instrument,
Quadratic Dispersion, 183 100
quarter-wave plate, 154, 157 signal front, 192
sine complex representation, 7
radiometry, 60
skin depth, 50
Radiometry, Photometry, and Color,
slow axis, 153
60
Snell’s law, 75, 124
radius of curvature, 239
Snell, Willebrord, 75
ray, 227, 231
spatial coherence, 203, 211, 212, 215
ray diagram, 243
Spatial Coherence for a Continuous
ray tracing, 235
Source, 216
Rayleigh criterion, 281
spatial filter, 313
Rayleigh range, 290
Spectrometers, 285
Rayleigh, Lord, 176
spectrum, 177
real image, 243
spherical aberration, 249
real part, 8
spherical interface, 238
rectangular aperture, 263, 265
spherical surface, 237
reflectance, 78
spherical wave, 258, 305
Reflectance and Transmittance, 78
Stability of Laser Cavities, 245
reflection, 237
Stefan-Boltzmann Law, 316
Reflection and Refraction at Curved
stochastic phase, 214
Surfaces, 237
Stokes parameters, 161
reflection from a curved surface, 238
Stokes vector, 159, 162
Reflections from Metal, 83
Stokes’ theorem, 6
refraction, 75
Stokes, George Gabriel, 159
Refraction at a Uniaxial Crystal Sur-
Strutt, John William, 176
face, 124
subluminal, 188
Refraction at an Interface, 73
superluminal, 171, 188
Repeated Multilayer Stacks, 109
surface figure, 302
reshaping delay, 187
susceptibility, 49
resolution, 275, 280
susceptibility tensor, 118
Resolution of a Telescope, 279
Sylvester’s theorem, 12, 109
resolving power, 104, 287
Symmetry of Susceptibility Tensor,
retarder, 153
128
right-hand rule, 3
ring cavity, 246 Table of Integrals and Sums, 20
Roemer, Ole, 43 Taylor’s series, 7
Rotation of Coordinates, 129 temporal coherence, 203, 204
Temporal Coherence of Continuous
s-polarized light, 74, 91
Sources, 209
336 INDEX
Testing Optical Components, 302

Thermodynamic Derivation of the
Stefan-Boltzmann Law, 324
thin lens, 239
Total Internal Reflection, 81
transmittance, 79
Two-Interface Transmittance at Sub
Critical Angles, 93
uniaxial, 123
unit vector, 1
unpolarized light, 145, 159
Van Cittert-Zernike Theorem, 217

vector, 1
Vector Calculus, 1
vector multiplication, 2
virtual image, 253
voltage, 31
Wave Equation, The, 37

wave number, 46
wave plate, 145, 153
wavelength, 46
Wien, Wilhelm, 320
Young’s Two-Slit Setup and Spatial

Coherence, 211
Young, Thomas, 214
Physical Constants
Constant Symbol Value
Permittivity ²0 8.8542 × 10−12 C2 /N · m2

µ0 4π × 10−7 T · m/A (or kg · m C2 )
±
Permeability
Speed of light in vacuum c 2.9979 × 108 m/s
Charge of an electron qe 1.602 × 10−19 C
Mass of an electron me 9.108 × 10−31 kg
Boltzmann’s constant kB 1.380 × 10−23 J/K
Plancks constant h 6.626 × 10−34 J · s
ħ 1.054 × 10−34 J · s
Stefan-Boltzmann constant σ 5.670 × 10−8 W/m2 · K4

Physics of Light and Optics PDF

Uploaded by

Copyright:

Available Formats

Physics of Light and Optics PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Physics of Light and Optics PDF

Uploaded by

Copyright:

Available Formats

What topics are covered in the book?

What topics are covered in the book?

What mathematical tools are reviewed in Chapter 0?

What mathematical tools are reviewed in Chapter 0?

Physics of Light and Optics

2 Plane Waves and Refractive Index 45

3 Reflection and Refraction 73

3.1 Refraction at an Interface . . . . . . . . . . . . . . . . . . . . . . . . 73

4 Multiple Parallel Interfaces 89

5 Propagation in Anisotropic Media 117

Review, Chapters 1–5 139

6 Polarization of Light 145

7 Superposition of Quasi-Parallel Plane Waves 171

8 Coherence Theory 203

Review, Chapters 6–8 223

9 Light as Rays 227

11 Diffraction Applications 275

12 Interferograms and Holography 301

Review, Chapters 9–12 309

13 Blackbody Radiation 315

Physical Constants 337

E (r) or B (r). An example of such a field is E (r) = q (r − r0 ) 4π²0 |r − r0 |3 , which

Solution: As mentioned above, the field is given by E (r) = q (r − r0 ) 4π²0 |r − r0 |3 .

Figure 0.1 The electric field vec-

where φ is the angle between the vectors k and r.

Proof of the final line of (0.2)

Another type of vector multiplication is the cross product , which is accom-

remembering its form.

Proof of cross-product properties

We will use several multidimensional derivatives in our study of optics, namely

the divergence, which applies to vector functions, is given by

and the curl, which also applies to vector functions, is given by

126-127 (New York: McGraw-Hill 1971).

x̂ = cos φρ̂ − sin φφ̂ and ŷ = sin φρ̂ + cos φφ̂

Meanwhile, the inverted form of the coordinate transformation is

from which we obtain the following derivatives:

Putting this all together, we arrive at

where we have used cos2 φ + sin2 φ = 1 (see Ex. 0.4).

We will sometimes need a multidimensional second derivative called the

This is possible because each unit vector is a constant in Cartesian coordinates.

Verification of (0.10) in Cartesian coordinates

From (0.6), we have

After some factorization, we obtain

where on the final line we invoked (0.4), (0.5), and (0.8).

We will also encounter several integral theorems3 involving vector functions.

Now we evaluate the right side of (0.11):

Another important theorem is Stokes’ theorem :

0.2 Complex Numbers

sine function is intrinsically present in this formula via the identity

cos α + β = cos α cos β − sin α sin β

This is a good formula to commit to memory, as well as the frequently used

p the well-known mathematician Johann

iφ −i φ the University of Basel at age thirteen.

2i versity of Basel. Under the auspices of

cos i γ = = cosh γ He successfully dened logarithms and

Finally, for cos2 φ + sin2 φ = 1 we have

As was mentioned previously, we will often be interested in waves of the

where the phase β is conveniently contained within the complex factor Ã ≡ Ae i β .

It is common (even conventional) to omit the explicit writing of Re { }. Thus,

cos i γ = = cosh γ He successfully dened logarithms and

Carl Friedrich Gauss (17771855, Ger-