Vous êtes sur la page 1sur 361


621(G) Fundamentals of Photonics

Franz X. Kaertner Spring Term 2009

1 Introduction 2 Classical Electromagnetism and Optics 2.1 Maxwells Equations of Isotropic Media . . . . . . . . . . . . 2.1.1 Helmholtz Equation . . . . . . . . . . . . . . . . . . 2.1.2 Plane-Wave Solutions (TEM-Waves) and Complex Notation . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.3 Poynting Vectors, Energy Density and Intensity . . . 2.1.4 Classical Permittivity . . . . . . . . . . . . . . . . . . 2.1.5 Optical Pulses . . . . . . . . . . . . . . . . . . . . . . 2.1.6 Pulse Propagation . . . . . . . . . . . . . . . . . . . 2.1.7 Dispersion . . . . . . . . . . . . . . . . . . . . . . . . 2.1.8 Loss and Gain . . . . . . . . . . . . . . . . . . . . . . 2.1.9 Kramers-Kroenig Relations and Sellmeier Equation . 2.2 Electromagnetic Waves and Interfaces . . . . . . . . . . . . . 2.2.1 Boundary Conditions and Snells law . . . . . . . . . 2.2.2 Measuring Refractive Index with a Prism . . . . . . . 2.2.3 Fresnel Reection . . . . . . . . . . . . . . . . . . . . 2.2.4 Brewsters Angle . . . . . . . . . . . . . . . . . . . . 2.2.5 Total Internal Reection . . . . . . . . . . . . . . . . 2.3 Polarization of Electromagnetic Waves . . . . . . . . . . . . 2.3.1 Polarization . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Jones Calculus . . . . . . . . . . . . . . . . . . . . . 2.4 Interference and Interferometers . . . . . . . . . . . . . . . . 2.4.1 Interference and Coherence . . . . . . . . . . . . . . . 2.4.2 TEM-Transmission Lines and TEM-Waves . . . . . . 2.4.3 Scattering and Transfer Matrix . . . . . . . . . . . . 2.4.4 Properties of the Scattering Matrix . . . . . . . . . . 3 1 13 . 13 . 15 . . . . . . . . . . . . . . . . . . . . . . 16 18 19 23 28 29 37 39 44 45 47 50 55 58 66 66 71 76 76 80 83 85

CONTENTS 2.4.5 Beamsplitter . . . . . . . . . . . . . . . . . . . . . . 2.4.6 Interferometers . . . . . . . . . . . . . . . . . . . . 2.4.7 Multipath Interference and Fabry-Perot Resonator . 2.4.8 Quality Factor of Fabry-Perot Resonances . . . . . 2.4.9 Thin-Film Filters . . . . . . . . . . . . . . . . . . . 2.5 Gaussian Beams and Resonators . . . . . . . . . . . . . . . 2.5.1 Paraxial Wave Equation . . . . . . . . . . . . . . . 2.5.2 Gaussian Beams . . . . . . . . . . . . . . . . . . . . 2.5.3 Optical Resonators . . . . . . . . . . . . . . . . . . 2.6 Waveguides and Integrated Optics . . . . . . . . . . . . . . 2.6.1 Planar Waveguides . . . . . . . . . . . . . . . . . . 2.6.2 Two-Dimensional Waveguides . . . . . . . . . . . . 2.6.3 Waveguide Coupling 143 2.6.4 Coupling of Modes . . . . . . . . . . . . . . . . . . 2.6.5 Switching by Control of Phase Mismatch . . . . . . 2.6.6 Optical Fibers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 87 91 96 100 102 102 104 111 123 124 143

. . 145 . . 150 . . 152 161

3 Quantum Nature of Light and Matter 3.1 Black Body Radiation . . . . . . . . . . . . 3.1.1 Rayleigh-Jeans-Law . . . . . . . . . . 3.1.2 Wiens Law . . . . . . . . . . . . . . 3.1.3 Plancks Law . . . . . . . . . . . . . 3.1.4 Thermal Photon Statistics . . . . . . 3.2 Photo-electric Eect . . . . . . . . . . . . . 3.3 Photodetection and Shot Noise . . . . . . . 3.3.1 Signal to Noise Ratio . . . . . . . . . 3.3.2 Noise Equivalent Power . . . . . . . . 3.4 Spontaneous and Induced Emission . . . . . 3.5 Matter Waves and Bohrs Model of an Atom 3.6 Wave Particle Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

161 165 166 166 168 172 173 176 177 177 181 186

3.7 Appendix A: Mode Counting . . . . . . . . . . . . . . . . . . . 186 3.8 Appendix B: Shot Noise Formula . . . . . . . . . . . . . . . . 189

CONTENTS 4 Schroedinger Equation 4.1 Free Motion . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Probability Conservation and Propability Currents . . . . . 4.3 Measureability of Physical Quantities (Observables) . . . . . 4.4 Stationary States . . . . . . . . . . . . . . . . . . . . . . . . 4.4.1 The One-dimensional Innite Box Potential . . . . . 4.4.2 The One-dimensional Harmonic Oscillator . . . . . . 4.5 The Hydrogen Atom . . . . . . . . . . . . . . . . . . . . . . 4.5.1 Ground State . . . . . . . . . . . . . . . . . . . . . . 4.5.2 Excited States . . . . . . . . . . . . . . . . . . . . . . 4.5.3 Energy Spectrum of Hydrogen . . . . . . . . . . . . . 4.5.4 Superposition States, Radiative Transitions and Selection Rules . . . . . . . . . . . . . . . . . . . . . . . . 4.6 Wave Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . 4.6.1 Position Statistics . . . . . . . . . . . . . . . . . . . . 4.6.2 Momentum Statistics . . . . . . . . . . . . . . . . . . 4.6.3 Energy Statistics . . . . . . . . . . . . . . . . . . . . 4.6.4 Arbitrary Observable . . . . . . . . . . . . . . . . . . 4.6.5 Eigenfunctions and Eigenvalues of Operators . . . . . 4.6.6 Appendix A: Spherical Harmonics . . . . . . . . . . . 4.6.7 Appendix B: Radial Wave Function . . . . . . . . . . 5 Interaction of Light and Matter 5.1 The Two-Level Model . . . . . . . . . . . . . . . . . 5.2 The Atom-Field Interaction In Dipole Approximation 5.3 Rabi-Oscillations . . . . . . . . . . . . . . . . . . . . 5.4 Energy- and Phase-Relaxation . . . . . . . . . . . . . 5.5 The Bloch Equations . . . . . . . . . . . . . . . . . . 5.6 Dielectric Susceptibility and Saturation . . . . . . . . 5.7 Rate Equations and Cross Sections . . . . . . . . . . 6 Lasers 6.1 The Laser (Oscillator) Concept . . . . . . 6.2 Laser Gain Media . . . . . . . . . . . . . . 6.2.1 Three and Four Level Laser Media 6.3 Types of Lasers . . . . . . . . . . . . . . . 6.3.1 Gas Lasers . . . . . . . . . . . . . . 6.3.2 Excimer Lasers: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5 193 . 196 . 201 . 204 . 206 . 207 . 209 . 213 . 215 . 218 . 226 . . . . . . . . . 229 233 234 234 235 236 237 239 241

247 . 247 . 249 . 252 . 256 . 261 . 262 . 265 273 . 273 . 276 . 276 . 279 . 279 . 280

6 6.3.3 Dye Lasers: . . . . . . . . . . . . . . . 6.3.4 Solid-State Lasers . . . . . . . . . . . . 6.3.5 Semiconductor Lasers . . . . . . . . . . 6.3.6 Quantum Cascade Lasers . . . . . . . . Lasers and its spectroscopic parameters . . . . Homogeneous and Inhomogeneous Broadening Laser Dynamics (Single Mode) . . . . . . . . . Continuous Wave Operation . . . . . . . . . . Stability and Relaxation Oscillations . . . . . Laser Eciency . . . . . . . . . . . . . . . . . "Thresholdless" Lasing 304 Short pulse generation by Q-Switching . . . . 6.11.1 Active Q-Switching . . . . . . . . . . . 6.11.2 Passive Q-Switching . . . . . . . . . . Short pulse generation by mode locking . . . . 6.12.1 Active Mode Locking . . . . . . . . . . 6.12.2 Passive Mode Locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

CONTENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 281 285 288 290 291 293 298 300 303

6.4 6.5 6.6 6.7 6.8 6.9 6.10 6.11


. . . . . .

. . . . . .

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

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

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

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

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

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

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

306 307 309 310 313 318 327 328 329 330 331 331 333 335 335 335 336 336 336 337 338 339 339 340 341

7 The Dirac Formalism and Hilbert Spaces 7.1 Hilbert Space . . . . . . . . . . . . . . . . . . . . 7.1.1 Scalar Product and Norm . . . . . . . . . 7.1.2 Vector Bases . . . . . . . . . . . . . . . . . 7.2 Linear Operators in Hilbert Spaces . . . . . . . . 7.2.1 Properties of Linear Operators . . . . . . . 7.2.2 The Dyadic Product . . . . . . . . . . . . 7.2.3 Special Linear Operators . . . . . . . . . . 7.2.4 Inverse Operators . . . . . . . . . . . . . . 7.2.5 Adjoint or Hermitian Conjugate Operators 7.2.6 Hermitian Operators . . . . . . . . . . . . 7.2.7 Unitary Operators . . . . . . . . . . . . . 7.2.8 Projection Operators . . . . . . . . . . . . 7.3 Eigenvalues of Operators . . . . . . . . . . . . . . 7.4 Eigenvectors of Commuting Operators . . . . . . 7.5 Complete System of Commuting Operators . . . . 7.6 Product Space . . . . . . . . . . . . . . . . . . . . 7.7 Quantum Dynamics . . . . . . . . . . . . . . . . . 7.7.1 Schroedinger Equation . . . . . . . . . . .

CONTENTS 7.7.2 Schroedinger Equation in x-representation . . . . . . 7.7.3 Canonical Quantization . . . . . . . . . . . . . . . . 7.7.4 Schroedinger Picture . . . . . . . . . . . . . . . . . . 7.7.5 Heisenberg Picture . . . . . . . . . . . . . . . . . . . 7.8 The Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . 7.8.1 Energy Eigenstates, Creation and Annihilation Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.8.2 Matrix Representation . . . . . . . . . . . . . . . . . 7.8.3 Minimum Uncertainty States or Coherent States . . . 7.8.4 Heisenberg Picture . . . . . . . . . . . . . . . . . . . 7.9 The Kopenhagen Interpretation of Quantum Mechanics . . . 7.9.1 Description of the State of a System . . . . . . . . . 7.9.2 Description of Physical Quantities . . . . . . . . . . . 7.9.3 The Measurement of Observables . . . . . . . . . . . 8 Quantization of the Electromagnetic Field 8.1 Decompostion of the Free Field into Normal Modes 8.2 Plane Wave Expansion . . . . . . . . . . . . . . . . 8.3 Quantization of the Electromagnetic Field . . . . . 8.4 The 1-dimensional Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

i 341 342 343 344 345 346 349 349 351 352 352 352 353

357 . 357 . 361 . 361 . 362

Chapter 1 Introduction
The word Photonics has been coined over the last two decades with the goal to describe the eld of Optics, Optoelectronic phenomena and its applications by one word similar to the very successful eld of Electronics. As the word indicates, Photonics uses primarily photons to carry out its mission in contrast to Electronics which uses primarily electrons or other charged particles. But no strict separation of both elds is possible, at least not on an elementary level. Electrons in acceleration generate electromagnetic waves. Sometimes these waves are guided and over short distances the wave aspect can be neglected and one can talk about electronics only. Similar, pure photonics, i.e. the free electromagnetic eld in vaccum, without matter is not of very much interest and use either. Quantum Mechanics teaches us that there are elementary excitations (in energy) of these waves, photons with particle like properties. So far, we are used to thinking of electrons as classical particles but Quantum Mechanics equally assigns to them wave properties, matter waves, like it assigns to the electromagnetic waves particle like properties. Electrons are elementary excitations of matter waves just as photons are elementary excitations of electromagnetic waves. We will talk later in more depth about this wave particle duality, once we have developed the mathematical tools to analyze its meaning quantitatively. So far we have been educated using the language of classical physics. In the classical limit, it turns out that electrons (particles with spin 1/2) are particles and photons (particles with integral spin) behave like classical waves and the particle nature is of vanishing importance. This is the reason why Photonics eventually is experienced as more abstract than Electronics, especially if you went through three years of education primarily focused on 1


Electronics and its applications. The particle properties of electromagnetic waves become of importance when the energy of the photons considered, hf, where h is Plancks constant and f is the frequency, is larger than the thermal energy stored in an electromagnetic mode, kT , where k is Boltzmanns constant and T is temperature. At room temperature this is the case for frequencies greater than 6 THz. For lower temperatures this transition frequency from classical to quantum behaviour may already occur at GHz frequencies. Certainly, at room temperature, the particle properties are important in the near infrared (IR) and visible spectrum where currently the bulk of the photonic activities are carried out, see Figure 1.1.

Figure 1.1: Wavelength and frequency ranges of electromagnetic radiation and its use. An important task of Photonics is the development of coherent sources of radiation, which are in the optical range called LASERs (Light Amplication by Stimulated Emission of Radiation). The rst amplier making explicit use of the quantum properties of matter was the MASER (Microwave Amplication by Stimulated Emission of Radiation) invented by J. P. Gordon, C. H. Townes and Zeiger in 1954. The extension of the MASER principle to

3 the optical wavelength range was proposed by Schawlow and Townes in 1958 and the rst laser was demonstrated in 1960 by Maiman [5]. As in the case of Electronics, Communications is a very important area in Photonics, and is usually called Optical Communications, which is carried out primarily in the infrared to visible wavelength range, see Figure 1.1. The enormous bandwidth available at optical frequencies, roughly four orders of magnitude higher than typical microwave frequencies (10GHz), enables the corresponding advance in information transmission capacity when compared to what electronics can do over a single long distance transmission link at microwave frequencies. Our current society could not be sustained without the worldwide deployed optical cable network, see Figure 1.2, which can be considered the largest connected machinery on earth. Understanding of the basic working principles of the components used in such lightwave networks, like the optical ber itself, waveguide couplers, modulators, light sources is at the heart of this course.

Figure 1.2: Worldwide underwater optical cable systems The big advantage of Photonics versus Electronics in the area of information transmission is that photons do not interact with each other directly. Maxwells equations in the classical vacuum are linear. Whereas electrons,


which are charged particles, show very strong interaction via their Coulomb elds. This is one reason why Photonics is very powerful in information transmission; essentially no interference with other signals; whereas electronics is very strong in information processing which is based on nonlinear operations.

Figure 1.3: High-Index Contrast SiN/SiO2 /Air Filter using Scanning Electron Beam Lithograph. Courtesy of H. Smith MIT-Nanostructures Laboratory As in the case of Electronics, miniaturization of many components and integration on a common technology platform is one of the major challenges in photonics (Integrated Optics). Over the last few years a new eld called Silicon-Photonics came to life. Modern nanofabrication techniques such as Scanning Electron Beam Lithography enable the fabrication of optical components on the scale of the optical wavelength with a relative precision in the picometer range in the same material system like electronics, i.e. silicon/silicon nitride/silicon dioxide, see Figure 1.3. Fig. 1.4 shows a measured

5 lter characteristic useful for optical communications applications. Systems that integrate both optical and electronic subsystems on a single chip called Electronic-Photonic Integrated Circuits are on the horizon.


stage through





Figure 1.4: Schematic and measured two stage third order ring lter. Courtesy MIT-Optics and Quantum Electronics Group

As with Electronics, Photonics has not only left its mark on communications but is used in a wide variety of sensing applications specic to its wavelength range as well as optical imaging, test and measurement instrumentation, manufacturing and materials processing. High power lasers with average powers of tens of kilowatts can cut many centimeter thick steel plates at high speed. Laser systems reaching petawatts (1015 Watt) of peak power have been built at Lawrence Livermore Laboratory for the rst time in 1996, (see Fig. 1.5) and much more compact versions are under construction at several laboratories around the world. The enormous peak power (the average power consumption of the earth is on the order of a few Terawatt (1012 Watt)), even though only available over a fraction of a picosecond, will enable us to investigate new physics and fundamental interactions at extreme intensities, such as the scattering of photons with each other invoking vacuum nonlinearities (the Quantum Electrodynamic Vacuum), see Figure 1.6.


Figure 1.5: First petawatt laser system installed at LLNL using chirped pulse amplication and the NOVA amplier chain, see http://www.llnl.gov/str/MPerry.html.

Because of the higher operating frequencies, optical sources can provide much higher time resolution than microwave sources. Therefore, it is not surprising that the shortest controlled events that currently can be made are optical pulses that comprise only a few cycles of light. These events are so short that one can only use the pulses themselves to reveal its pulse width, for example by performing an autocorrelation of the pulse with a copy of itself using a nonlinear element, see Fig. 1.7.

Figure 1.6: Progress in peak intensity generation from lasers over the last 45 years. Courtesy Gerard Mourou.


Figure 1.7: Experimental setup for measureing an interferometric autocorrelation. A copy of the pulse is generated and interferometrically delayed with respect to the original pulse. The superposition of the two pulses undergo second harmonic generation in a crystal and the second harmonic light is detected. Figure 1.8 shows a measured interferometric second harmonic autocorrelation trace of a 5fs (fs=femtosecond, 1015 s) short pulse at a center frequency of 800nm [19].
8 6 IAC 4 2 0 -20 -10 0 10 20 TIME DELAY, FS 30

Figure 1.8: Measured and tted interferometric autocorrelation function of a 5fs pulse [19]. Femtosecond pulses have been used to understand the carrier dynamics in semiconductors, that limit the speed of electronic devices, the dynamics

9 of chemical reactions such as photosynthesis or the early stages of vision. As in any discipline the development does not stop and most recently pulses as short as 250 attoseconds (1018 s) have been generated in the soft-x-ray regime [4]. Laser like sources potentially generating radiation up into the hard x-ray regime are being planned to be built over the next years [5]. The hope is that such sources enable the spatially and temporally resolved direct imaging of molecules and atoms undergoing reactions, which would advance medical diagnostics and drug discovery. As these examples show, Photonics is a booming eld. The basis for its success is fundamental and, therefore, it will continue to grow in importance with impact in our everyday lifes as has been the case with (Micro-) Electronics, just think about your high-speed internet connection or the CD-player. The course is organized as follows: In Chapter 2 we will review some concepts of Classical Electromagnetism from 6.013, such as plane electromagnetic waves in isotropic media, the related energy ow, and their properties upon reections at interfaces. This will enable us to study various important optical devices such as prisms, interferometers, Fabry-Perot resonators and thin lm interference coatings. Optics in crystalline materials and related phenomena such as birefringence and its multiple use in optical components for polarization manipulation will be discussed. The plane electromagnetic wave is an idealization similar to the perfect monochromatic sin or cosine in signal theory. We will use the plane waves to understand beams of nite size and discuss them as solutions to the paraxial wave equation, the Gaussian beams which in the limit of large beam size become optical rays. Then we will look into guided optical beams such as in optical bers and waveguides from the point of view of ray optics and from the view point of full solutions to Maxwells equations. The concept of optical modes and resonances will be discussed. This basic training in wave physics puts us in an excellent position to understand Quantum Mechanics, which introduces matter waves. We will revisit in chapter III the expermimental facts appearing at the turn of the 19th century that lead to the discovery of Quantum Mechanics. In chapter IV, we will nd the Schroedinger Equation, a wave equation for matter waves, and introduce the formalism of Quantum Mechanics at the example of the harmonic oscillator and the hydrogen atom. Both systems are model systems for understanding photons and energy eigenspectra of quantum systems that can be fully solved analytically. Chapter V is devoted to the basic understanding of the interaction of light



and matter and sets the basis for the study of the laser and its fundamental properties in chapter VI. In chapter VII the modulation of light using electro and acousto-optic eects is discussed. If time permits an introduction to the quantum theory of light elds and photo detection is also treated.

[1] T. H. Maimann, "Stimulated optical radiation in ruby", Nature 187, 493-494, (1960). [2] M. D. Perry et al., "Petawatt Laser Pulses," Optics Letters, 24, 160-163 (1999). [3] R. Ell, U. Morgner, F.X. Krtner, J.G. Fujimoto, E.P. Ippen, V. Scheuer, G. Angelow, T. Tschudi: Generation of 5-fs pulses and octave-spanning spectra directly from a Ti:Sappire laser, Opt. Lett. 26, 373-375 (2001) [4] H. Hentschel, R. Kienberger, Ch. Spielmann, G. A. Reider, N. Milosevic, T. Brabec, P. Corkum, U. Heinzmann, M. Drescher, F. Krausz: "Attosecond Metrology," Nature 414, 509-513 (2001). [5] John Arthur, "Status of the LCLS x-ray FEL program," Review of Scientic Instruments, 73, 1393-1395 (2002).




Chapter 2 Classical Electromagnetism and Optics

The classical electromagnetic phenomena are completely described by Maxwells Equations. The simplest case we may consider is that of electrodynamics of isotropic media


Maxwells Equations of Isotropic Media

Maxwells Equations are D + J, t B , E = t D = , B = 0. D =


H =

(2.1a) (2.1b) (2.1c) (2.1d)

The material equations accompanying Maxwells equations are: + P, (2.2a) (2.2b) B = 0 H + M.

Here, E and H are the electric and magnetic eld, D the dielectric ux, B the magnetic ux, J the current density of free chareges, is the free charge density, P is the polarization, and M the magnetization. 13



where is the Nabla operator and the Laplace operator, we obtain ! E P E 0 j+ 0 + = M + E (2.3) t t t t and hence ! 2 2 1 j P + M + E . 2 2 E = 0 + c0 t t t2 t with the vacuum velocity of light c0 = s 1 0

Note, it is Eqs.(2.2a) and (2.2b) which make electromagnetism an interesting and always a hot topic with never ending possibilities. All advances in engineering of artical materials or nding of new material properties, such as superconductivity and meta-materials, bring new life, meaning and possibilities into this eld. By taking the curl of Eq. (2.1b) and considering E = E E,



For dielectric non magnetic media, which we often encounter in optics, with no free charges and currents due to free charges, there is M = 0, J = 0, = 0. One can also show that the electric eld can be decomposed into a longitudinal and tranasversal component EL and ET , which are characterized by [6] EL = 0 and ET = 0 (2.6)

If there are no free charges, the longitudinal component is zero and only a transversal component is left over. Therefore, for the purpose of this class (and most of optics) the wave equation greatly simplies to 2 1 2 (2.7) 2 2 E = 0 2 P . c0 t t This is the wave equation driven by the polarization of the medium in which the eld propagates.




Helmholtz Equation

In general, the polarization in dielectric media may have a nonlinear and non local dependence on the eld. For linear media the polarizability of the medium is described by a dielectric susceptibility (r, t) Z Z (2.8) dr0 dt0 (r r0 , t t0 ) E (r0 , t0 ) . P (r, t) = 0 The polarization in media with a local dielectric suszeptibility can be described by Z (2.9) P (r, t) = 0 dt0 (r, t t0 ) E (r, t0 ) .

This relationship further simplies for homogeneous media, where the susceptibility does not depend on location Z (2.10) P (r, t) = 0 dt0 (t t0 ) E (r, t0 ) . which leads to a dielectric response function or permittivity (t) = and with it to D(r, t) = Z
0 ( (t)

+ (t))

(2.11) (2.12)

dt0 (t t0 ) E (r, t0 ) .

If the medium is linear and has only an induced polarization, completely described in the time domain (t) or in the frequency domain by its Fourier transform, the complex susceptibility ( ) = r ( ) 1 with the relative permittivity r ( ) = ()/ 0 , we obtain in the frequency domain with the Fourier transform relationship
+ Z e E (r, ) = E (r, t)ejt dt,


where, the tildes denote the Fourier transforms in the following. Substituted into (2.7) 2 e e + 2 E (r, ) = 2 0 0 ()E (r, ), (2.15) c0

e P (r, ) =

( )E (r, ), 0




we obtain


with the refractive index n() and 1 + ( ) = n ()2 results in the Helmholtz equation 2 e + 2 E (r, ) = 0, (2.17) c where c( ) = c0 /n () is the velocity of light in the medium. This equation is the starting point for nding monochromatic wave solutions to Maxwells equations in linear media, as we will study for dierent cases in the following. So far we have treated the susceptibility ( ) as a real quantity, which may not always be the case as we will see later in detail.


Plane-Wave Solutions (TEM-Waves) and Complex Notation

The real wave equation (2.7) for a linear medium has real monochromatic plane wave solutions Ek (r, t), which can be be written most eciently in terms of the complex plane-wave solutions E k (r, t) according to i n o 1h E k (r, t) + E k (r, t) = <e E k (r, t) , (2.18) Ek (r, t) = 2 with (2.19) E k (r, t) = E k ej(tkr) e(k). Note, we explicitly underlined the complex wave to indicate that this is a complex quantity. Here, e(k) is a unit vector indicating the direction of the electric eld which is also called the polarization of the wave, and E k is the complex eld amplitude of the wave with wave vector k. Substitution of eq.(2.18) into the wave equation results in the dispersion relation, i.e. a relationship between wave vector and frequency necessary to satisfy the wave equation 2 |k|2 = = k ( )2 . (2.20) c()2 The relation between wave vector and frequency of a wave is called dispersion relation. Here, it is given by k () = n( ). (2.21) c0

2.1. MAXWELLS EQUATIONS OF ISOTROPIC MEDIA with the wavenumber k = 2/,



where is the wavelength of the wave in the medium with refractive index n, the angular frequency, k the wave vector. Note, the natural frequency f = /2 . From E = 0, for all time, we see that k e. Substitution of the electric eld 2.18 into Maxwells Eqs. (2.1b) results in the magnetic eld i 1h Hk (r, t) = H k (r, t) + H k (r, t) (2.23) 2 with H k (r, t) = H k ej(tkr) h(k). (2.24)


This complex component of the magnetic eld can be determined from the corresponding complex electric eld component using Faradays law (2.25) jk E k ej(tkr) e(k) = j0 H k (r, t), H k (r, t) = E k j(tkr) e k e = H k ej(tkr) h 0 h(k) = k e(k) |k | (2.26)

with (2.27)


|k | 1 E . Ek = 0 ZF k The characteristic impedance of the TEM-wave is the ratio between and magnetic eld strength r 0 1 ZF = 0 c = = ZF0 n 0 r with the refractive index n = r and the free space impedance r 0 377 . ZF0 = Hk =

(2.28) electric



Note that the vectors e, h and k form an orthogonal trihedral, e h, k e, k h. (2.31)



That is why we call these waves transverse electromagnetic (TEM) waves. We consider the electric eld of a monochromatic electromagnetic wave with frequency and electric eld amplitude E0 , which propagates in vacuum k along the z-axis, and is polarized along the x-axis, (Fig. 2.1), i.e. |k = ez , | and e(k) = ex . Then we obtain from Eqs.(2.18) and (2.19) E (r, t) = E0 cos(t kz ) ex , and similiar for the magnetic eld H (r, t) = see Figure 2.1.
x E c


E0 cos(t kz ) ey , ZF0


Figure 2.1: Transverse electromagnetic wave (TEM) [6] Note, that for a backward propagating wave with E (r, t) = E ejt+jkr ex , and H (r, t) = H ej(t+kr) ey , there is a sign change for the magnetic eld H= |k| E, 0 (2.34)

so that the (k, E, H ) always form a right handed orthogonal system.


Poynting Vectors, Energy Density and Intensity

The table below summarizes the instantaneous and time averaged energy content and energy transport related to an electromagnetic eld

2.1. MAXWELLS EQUATIONS OF ISOTROPIC MEDIA Quantity Electric and magnetic energy density Poynting vector Poynting theorem Intensity Real elds ED = 1 E2 we = 1 2 2 0 r wm = 1 H2 H B = 1 2 2 0 r w = we + wm S = E H divS + E j + I = S = cw
w t



Complex elds 2 E hwe i = 1 4 0 r 2 H hwm i = 1 4 0 r hwi = hwe i + hwm i T= 1 E H 2 Ej + divT + 1 2 +2j (hwm i hwe i) = 0 I = Re{T } = c hwi

Table 2.1: Poynting vector and energy density in EM-elds For a plane wave with an electric eld E (r, t) = Eej(tkz) ex in a lossless medium, i.e. r = real,we obtain for the time averaged energy density in units of [J/m3 ] 1 hwi = r 0 |E |2 , (2.35) 2 the complex Poynting vector T = 1 |E |2 ez , 2ZF (2.36)

and the intensity in units of [W/m2 ] I= 1 1 |E |2 = ZF |H |2 . 2ZF 2 (2.37)


Classical Permittivity

In this section we want to get insight into propagation of an electromagnetic wavepacket in an isotropic and homogeneous medium, such as a glass optical ber due to the interaction of radiation with the medium. The electromagnetic properties of a dielectric medium is largely determined by the electric polarization P (t) induced by an electric eld in the medium. The polarization is dened as the total induced dipole moment per unit volume. If p is the dipole moment of the elementary unit (atom, molecule, ...) constituting the medium and N is density of elementary units, then the polirization is P (t) = dipole moment = N p(t), volume (2.38)



In the frequency domain, i.e. after Fourier transformation we obtain dipole moment e P () = = N e p() = volume e( )E (), 0 e (2.39)

assuming again a linear plarizability proportional to the electric eld. Or expressed in a dierent way the dielectric suszeptibility is the frequency response function of the polarization in a medium to an applied eld e( ) = N e p( ) . e 0 E ( ) (2.40)

Atoms and Molecules are composed of charge particles, such as a possitively charged nucleus and electrons bound to the nucleus by the Coulomb force. As a rst approximation towards the interaction of light and matter we consider a simple model for matter, where the elementary units, the atoms or molecules are modelled by a positively charge nucleus and an electron bound to it by a force increasing linear with distance between nucleus and electron, see Figure 2.2.

Figure 2.2: Classical harmonic oscillator model for radiation matter interaction As it turns out (justication later) this simple model correctly describes many aspects of the interaction of light with matter at very low electric eld strength, i.e. the elds do not change the electron distribution in the atom considerably or even ionize the atom. This model is called Lorentz model after the famous physicist A. H. Lorentz (Dutchman) studying electromagnetic phenomena at the turn of the 19th century. He also found the Lorentz Transformation and Invariance of Maxwells Equations with respect to these transformation, which showed the path to Special Relativity.



The equation of motion for such a unit is the damped harmonic oscillator driven by an electric eld in one dimension, x. At optical frequencies, the distance of elongation, x, is much smaller than an optical wavelength (atoms have dimensions on the order of a tenth of a nanometer, whereas optical elds have wavelength on the order of microns) and, therefore, we can neglect the spatial variation of the electric eld during the motion of the charges within an atom (dipole approximation, i.e. E (r, t) = E (rA , t) = E (t)ex . The equation of motion is then. m d2 x 0 dx + 2 m + m2 0 x = e0 E (t), 2 dt Q dt (2.41)

jt . Here, m is the mass of the electron assuming the that where E (t) = Ee the rest atom has innite mass, e0 the charge of the electron, 0 is the resonance frequency of the undamped oscillator and Q the quality factor of the resonance, which determines the damping of the oscillator. By using the trial solution x(t) = x ejt , we obtain for the complex amplitude of the dipole moment p with the time dependent response p(t) = e0 x(t) = p ejt p =
e2 0 m 2 ) +

(2 0

0 2j Q



Note, that we included ad hoc a damping term in the harmonic oscillator equation. At this point it is not clear what the physical origin of this damping term is and we will discuss this at length later in chapter 4. For the moment, we can view this term simply as a consequence of irreversible interactions of the atom with its environment. The simplest damping mechanism is that an accelerated electron radiates, i.e. radiation damping. We then obtain from (2.39) for the susceptibility e( ) = e( ) =
0 Nm

e2 1

0 2 (2 0 ) + 2j Q



2 p
0 2 (2 0 ) + 2j Q


2 where p is called the plasma frequency dened as 2 p = Ne0 /m 0 . The meaning of the plasma frequency will be further elucidated in recitations



and on problem sets. Figure 2.3 shows the real and imaginary part of the resulting classical dielectric susceptibility, e( ) = er ( ) + j ei ( ), (2.44), normalized to the absolute value of the imaginary part at = 0 , which is 2 2 p Q/ (20 ).

Figure 2.3: Real part (dashed line) and imaginary part (solid line) of the susceptibility of the classical oscillator model for the dielectric polarizability. Note, that there is a small resonance shift (almost invisible) due to the loss. O resonance, the imaginary part approaches zero very quickly. Not 2 so the real part, which approaches a constant value 2 p /0 below resonance for 0, and approaches zero far above resonance, but much slower than the imaginary part. As we will see later, this is the reason why there are low loss, i.e. transparent, media with refractive index very much dierent from 1, because as discussed above the real part of the dielectric susceptibility contributes to the refractive index in a material. The negative imaginary part describes obsorption of the electromagnetic wave in the medium made up of damped harmonic oscillators driven by the electric eld of the electromagnetic wave. Strong absorption occurs when the driving frequency is on resonance with the damped harmonic oscillator, i.e. the absorption resonance of the medium. After having a model for the dielectric susceptibility of a medium, we can



study how optical signals may propagate in such a medium, most importantly optical pulses, such as those used in optical communications.


Optical Pulses

Optical pulses are wave packets constructed by a continuous superposition of monochromatic plane waves. Consider a TEM-wavepacket, i.e. a superposition of waves with dierent frequencies, polarized along the x-axis and propagating along the z-axis Z d e E (r, t) = E ()ej(tK ()z) ex . (2.45) 2 0 Correspondingly, the magnetic eld is given by Z d e()ej(tK ()z) ey E H (r, t) = 2 Z ( ) F 0 (2.46)

Again, the physical electric and magnetic elds are real and related to the complex elds by 1 E (r, t) + E (r, t) (2.47) E (r, t) = 2 1 H (r, t) + H (r, t) . (2.48) H (r, t) = 2 ()|ej() is the complex wave amplitude of the electromagnetic wave Here, |E at frequency and K () = /c() = n()/c0 the wavenumber, where, n () is again the refractive index of the medium n 2 () = 1 + (), (2.49)

c and c0 are the velocity of light in the medium and in vacuum, respectively. The planes of constant phase propagate with the phase velocity c() of the wave depending on frequency in a dispersive medium. The wavepacket consists of a superposition of many frequencies with the spectrum shown in Fig. 2.4. At a given point in space, for simplicity z = 0, the complex eld of a pulse is given by (Fig. 2.4) Z 1 ()ejt d. E (z = 0, t) = E (2.50) 2 0



)| |E(

0 )| |A(

Figure 2.4: Spectrum of an optical wave packet described in absolute and relative frequencies

Optical pulses often have relatively small spectral width compared to the center frequency of the pulse 0 , as it is illustrated in the upper part of Figure 2.4. For example typical pulses used in optical communication systems for 10Gb/s transmission speed are 20 ps in lenght and have a center wavelength of = 1550 nm. Thus the spectral width is only on the order of 50 GHz, whereas the center frequency of the pulse is 200 THz, i.e. the bandwidth is 4000 smaller than the center frequency. In such cases it is useful to separate the complex electric eld in Eq. (2.50) into a carrier frequency 0 and an envelope A(t) and represent the absolute frequency as = 0 + . We can then rewrite Eq.(2.50) as Z 1 ( 0 + )ej(0 +)t d E (z = 0, t) = E 2 0 Z 1 j0 t = E ( 0 + )ejt d e 2 0 j 0 t A(t)e .



2.1. MAXWELLS EQUATIONS OF ISOTROPIC MEDIA The envelope, see Figure 2.8, is given by Z 1 ( )ejt d A(t) = A 2 0 Z 1 ( )ejt d, = A 2


(2.53) (2.54)

( ) = 0 for ( ) = E (0 + ) is the spectrum of the envelope with, A where A 0 . To be physically meaningful, the spectral amplitude A( ) must be zero for negative frequencies less than or equal to the carrier frequency, see Figure 2.8. Note, that waves with zero frequency can not propagate, since the corresponding wave vector is zero. The pulse and its envelope are shown in Figure 2.5.

Figure 2.5: Electric eld and envelope of an optical pulse. Table 2.2 shows pulse shape and spectra of some often used pulses as well as the pulse width and time bandwidth products. The pulse width and bandwidth are usually specied as the Full Width at Half Maximum (FWHM) 2 of the intensity in the time domain, |A(t)|2 , and the spectral density A ( ) in the frequency domain, respectively. Pulse shapes and corresponding spectra to the pulses listed in Table 2.2 are shown in Figs 2.6 and 2.7.


CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS Pulse Shape A(t) Fourier Transform R () = a(t)ejt dt A 1 2 2 2 e 2 sech 2 2
/2) sin( /2

Gaussian: e 2 2 Hyperbolic Secant: t sech( ) Rect-function: 1, |t| /2 0, |t| > /2 1 Lorentzian: 1+(t/ )2 t Double-Exp.: e| |


Pulse Width t 2 ln 2 1.7627 1.287 ln2

Time-Bandwidth Product t f 0.441 0.315 0.886 0.142 0.142

2 e| |
1+( )2

Table 2.2: Pulse shapes, corresponding spectra and time bandwidth products.

Figure 2.6: Fourier transforms to pulse shapes listed in table 2.2 [6].



Figure 2.7: Fourier transforms to pulse shapes listed in table 2.2 continued [6].




Pulse Propagation

Having a basic model for the interaction of light and matter at hand, via section 2.1.4, we can investigate what happens if an electromagnetic wave packet, i.e. an optical pulse propagates through such a medium. We start from Eqs.(2.45) to evaluate the wave packet propagation for an arbitrary propagation distance z Z 1 ()ej(tK ()z) d. E E (z, t) = (2.55) 2 0 Analogous to Eq. (2.51) for a pulse at a given position, we can separate an optical pulse into a carrier wave at frequency 0 and a complex envelope A(z, t), E (z, t) = A(z, t)ej(0 tK (0 )z) . (2.56) By introducing the oset frequency , the oset wavenumber k() and spec( ) trum of the envelope A = 0 , k() = K ( 0 + ) K ( 0 ), ( = 0 + ), () = E A we can rewrite Eq.(2.55) as Z 1 ()ej(tk()z) d ej(0 tK (0 )z) . E (z, t) = A 2 (2.57) (2.58) (2.59)


Thus the envelope at propagation distance z , see Fig.2.8, is expressed as Z 1 ( )ej(tk()z) d, (2.61) A A(z, t) = 2 with the same constraints on the spectrum of the envelope as before, i.e. the spectrum of the envelope must be zero for negative frequencies beyond the carrier frequency. In the frequency domain Eq.(2.61)) corresponds to (z = 0, )ejk()z . (z, ) = A A (2.62)

Depending on the dispersion relation k( ), (see Fig. 2.9),.the pulse will be reshaped during propagation as discussed in the following section.



Figure 2.8: Electric eld and pulse envelope in time domain.

Figure 2.9: Taylor expansion of dispersion relation at the center frequency of the wave packet.



The dispersion relation K () or dierential propagation constant, k ( ), indicates how much phase shift each frequency component experiences during



propagation. These phase shifts, if not linear with respect to frequency, will lead to distortions of the pulse. If the dispersion relation K () is only slowly varying over the pulse spectrum, it is useful to represent it or by its Taylor expansion, see Fig. 2.9, k ( ) = k 0 + k00 2 k(3) 3 + + O(4 ). 2 6 (2.63)

If the refractive index depends on frequency, the dispersion relation is no longer linear with respect to frequency, see Fig. 2.9 and the pulse propagation according to (2.61) can be understood most easily in the frequency domain (z, ) A (z, ). = jk( )A z Transformation of Eq.(2.64) into the time domain gives n X A(z, t) k(n) A(z, t). = j j z n ! t n=1 (2.64)


If we keep only the rst term, the linear term, in Eq.(2.62), then we obtain for the pulse envelope from (2.61) (z, ) = A (z = 0, )ejk0 z . A (2.66)

According to the shift theorem for Fourier Transfroms, a linear phase over the spectrum corresponds to a shift in the time domain, A(z, t) = A(0, t z/g0 ), (2.67)

Thus the derivative of the dispersion relation at the carrier frequency determines the propagation velocity of the envelope of the wave packet or group velocity, whereas the ratio between propagation constant and frequency determines the phase velocity of the carrier 1 K (0 ) p0 = 0 /K ( 0 ) = . (2.69) 0

where we introduced the group velocity at frequency 0 !1 1 dk ( ) dK ( ) g0 = 1/k0 = = d =0 d =0




To get rid of the trivial motion of the pulse envelope with the group velocity, we introduce the retarded time t0 = t z/vg0 . With respect to this retarded time the pulse shape is invariant during propagation, if we approximate the dispersion relation by the slope at the carrier frequency A(z, t) = A(0, t0 ). (2.70)

Note, if we approximate the dispersion relation by its slope at the carrier frequency, i.e. we retain only the rst term in Eq.(2.65), we obtain A(z, t) 1 A(z, t) + = 0, z g0 t (2.71)

and (??) is its solution. If, we transform this equation to the new coordinate system z 0 = z, t0 = t z/g0 , with 1 , = 0 z z g0 t0 = t t0 the transformed equation is A(z 0 , t0 ) = 0. z 0 (2.76) (2.74) (2.75) (2.72) (2.73)

Thus we see that the pulse shape doesnt change during propagation. The propagation can be simply accounted for by introducting a retarded time t0 = t z/g0 . Since z is equal to z 0 we keep z in the following. If the spectrum of the pulse is broad enough or the propagation distance long enough, so that the second order term in (2.63) becomes important, the pulse will no longer keep its shape. When keeping in the dispersion relation terms up to second order it follows from (2.65) and (2.72,2.73) A(z, t0 ) k00 2 A(z, t0 ) . =j z 2 t0 2 (2.77)



This is the rst non trivial term in the wave equation for the envelope. Because of the superposition principle, the pulse can be thought of to be decomposed into wavepackets (sub-pulses) with dierent center frequencies. Now, the group velocity depends on the spectral component of the pulse, see Figure 2.10, which will lead to broadening or dispersion of the pulse.
vg1 vg2 vg3 1 2 k 2

) A(

1 0 1

Figure 2.10: Decomposition of a pulse into wave packets with dierent center frequency. In a medium with dispersion the wavepackets move at dierent relative group velocity. Fortunately, for a Gaussian pulse, the pulse propagation equation 2.77 can be solved analytically. The initial pulse is then of the form E (z = 0, t) = A(z = 0, t)ej0 t 1 t02 0 A(z = 0, t = t ) = A0 exp 2 2 (2.78) (2.79)

Eq.(2.77) is most easily solved in the frequency domain since it transforms to (z, ) A k00 2 (2.80) = j A(z, ), z 2 with the solution 00 2 k (z = 0, ) exp j (z, ) = A z . A 2 (2.81)

Dispersion Relation




The pulse spectrum aquires a parabolic phase. Note, that here is the Fourier Transform variable conjugate to t0 rather than t. The Gaussian pulse has the advantage that its Fourier transform is also a Gaussian 1 2 2 (2.82) A(z = 0, ) = A0 2 exp . 2 Note, in the following we will often use the Gaussian Integral 1 2 Z e 2 ejx dx = e 2 for Re { } 0
x2 2


In the spectral domain the solution at an arbitray propagation distance z is 2 2 1 00 (z, ) = A0 2 exp + jk z . (2.84) A 2 With the Gaussian integral (2.83) the inverse Fourier transform is 1/2 t02 2 1 0 A(z, t ) = A0 exp ( 2 + jk00 z ) 2 ( 2 + jk00 z ) (2.85)

The exponent can be written as real and imaginary part and we nally obtain " # 1/2 2 2 t02 1 t02 1 00 0 + j k z exp A(z, t ) = A0 ( 2 + jk00 z ) 2 4 + (k00 z )2 2 4 + (k00 z )2 (2.86) As we see from Eq.(2.86) during propagation the FWHM of the Gaussian determined by # " 2 ( 0F W HM /2)2 = 0.5 (2.87) exp 4 + (k00 z )2 changes from F W HM = 2 ln 2 (2.88) at the start to 0F W HM s 00 2 k L = 2 ln 2 1 + 2 s 00 2 k L = F W HM 1 + 2




The strongly dispersed pulse has a width equal to the dierence in group delay over the spectral width of the pulse. Figure 2.11 shows the evolution of the magnitude of the Gaussian wave packet during propagation in a medium which has no higher order dispersion in normalized units, i.e. ([z ] =m, [t] =s, [kj] =s2 /m, kj = 2). The pulse spreads continuously.

at z = L. For large stretching this result simplies to 00 00 k L k L 0 for F W HM = 2 ln 2 2 1.


1 0.8

0.6 0.4 0.2

0 0.5 Distance z 1 1.5 -6 -4 -2 2 4 6

0 Time

Figure 2.11: Magnitude of the complex envelope of a Gaussian |A(z, t0 )| , in a dispersive medium.


As discussed before, the origin of this spreading is the group velocity dispersion (GVD), k 00 6= 0. The group velocity varies over the pulse spectrum signicantly leading to a group delay dispersion (GDD) after a propagation distance z = L of k 00 L 6= 0, for the dierent frequency components. This leads to the build-up of chirp in the pulse during propagation. We can understand this chirp by looking at the parabolic phase that develops over the pulse in

2.1. MAXWELLS EQUATIONS OF ISOTROPIC MEDIA time at a xed propagation distance. The phase is, see Eq.(2.86) 00 1 00 1 t02 k L 0 . (z = L, t ) = arctan + L k 2 2 2 4 + (k00 L)2
(a) Front k'' < 0 Time t k'' > 0 Phase Back




Instantaneous Frequency k'' < 0

Time t k'' > 0

Figure 2.12: (a) Phase and (b) instantaneous frequency shift of a Gaussian pulse during propagation through a medium with positive or negative dispersion. This parabolic phase, see Fig. 2.12 (a), can be understood as a localy varying frequency in the pulse, i.e. the derivative of the phase gives the instantaneous frequency shift in the pulse with respect to the center frequency (z = L, t0 ) = k 00 L 0 0 ( L, t ) = 2 t 0 4 00 t + (k L) (2.92)

see Fig.2.12 (b). The instantaneous frequency indicates that for a medium with positive GVD, ie. k00 > 0, the low frequencies are in the front of the



pulse, whereas the high frequencies are in the back of the pulse, since the sub-pulses with lower frequencies travel faster than sub-pulses with higher frequencies. The opposite is the case for negative dispersive materials. It is instructive for later purposes, that this behaviour can be completely understood from the center of mass motion of the sub-pulses, see Figure 2.10. Note, we can choose a set of sub-pulses, with such narrow bandwidth, that dispersion does not matter. In the time domain, these pulses are of course very long, because of the time bandwidth relationship. Nevertheless, since they all have dierent carrier frequencies, they interfere with each other in such a way that the superposition is a very narrow pulse. This interference, becomes destroyed during propagation, since the sub-pulses propagate at dierent speed, i.e. their center of mass propagates at dierent speed.
k">0 k"<0

Time, t

1/v vg2 g2

vg1 vg3

-5 0.0





Propagation distance, z

Figure 2.13: Pulse spreading by following the center of mass of sub-pulses according to Fig. 2.10. For z < 1, the pulses propagate in a medium with positive dispersion and for z > 1 in a medium with negative dispersion. The dierential group delay Tg ( ) = k00 L of a sub-pulse with its center frequency dierent from 0, is due to its dierential group velocity 2 00 vg ( ) = vg0 Tg ()/Tg0 = vg 0 k . Note, that Tg 0 = L/vg 0 . This is illustrated in Figure 2.13 by ploting the trajectory of the relative motion of the center of mass of each sub-pulse as a function of propagation distance, which asymptotically approaches the formula for the pulse width of the highly dis-



persed pulse Eq.(2.90). If we assume that the pulse propagates through a negative dispersive medium following the positive dispersive medium, the group velocity of each sub-pulse is reversed. The sub-pulses propagate towards each other until they all meet at one point (focus) to produce again a short and unchirped initial pulse, see Figure 2.13. This is a very powerful technique to understand dispersive wave motion and is also the connection between ray optics and physical optics.


Loss and Gain

If the medium considered has loss, described by the imaginary part of the dielectric susceptibility, see (2.44) and Fig. 2.3, we can incorporate this loss into a complex refractive index n () = nr () + jni () via n () = q 1+ e(). (2.93)


For an optically thin medium, i.e. e 1 the following approximation is very useful e() n () 1 + . (2.95) 2 As one can show (in Recitations) the complex susceptibility (2.44) can be approximated close to resonance, i.e. 0 , by the complex Lorentzian lineshape j0 e() = , (2.96) 0 1 + jQ 0
p where 0 = Q 2 2 will turn out to be related to the peak absorption of the 0 resonance, which is proportional to the density of atoms. 0 is the center 0 is the half width half maximum (HWHM) linewidth frequency and = Q of the transition. The real and imaginary part of the complex Lorentzian are

er () =

0 ) 0 ( 0 2 , 1 + 0 ei () = 0 2 , . 1 +

(2.97) (2.98)



Since we introduced a complex wavenumber, we have to redene the group velocity as the inverse derivative of the real part of the wavenumber with respect to frequency. At line center, we obtain Kr () 1 0 0 1 = 1 . (2.100) g = 0 c0 2
0 Thus, for a narrow absorption line, 0 > 0 and 1, the absolute value of the group velocity can become much larger than the velocity of light in vacuum. The opposite is true for an amplifying medium, 0 < 0. There is nothing wrong with this nding, since the group velocity only describes the motion of the peak of a Gaussian wave packet, which is not a causal wave packet. A causal wave packet is identical to zero for some earlier time t < t0 , in some region of space. A Gaussian wave packet lls the whole space at any time and can be reconstructed by a Taylor expansion at any time. Therefore, the motion of the peak of such a signal with a speed greater than the speed of light does not contradict special relativity. The imaginary part in the wave vector (2.99) leads with K = c to ab0 sorption K (2.101) () = e (). 2 i In the envelope equation (2.64) for a wavepacket with carrier frequency 0 = 0 0 and wave number K0 = the loss leads to a term of the form c0 (z, ) K0 /2 A (z, ) = 0 = (0 + )A (z, ). (2.102) 2 A z 1 + (loss)

In the derivation of the wave equation for the pulse envelope (2.65) in section 2.1.7, there was no restriction to a real refractive index. Therefore, the wave equation (2.65) also treats the case of a complex refractive index. If we assume a medium with the complex refractive index (2.95), then the wavenumber is given by 1 () = K 1 + ( ei ()) . (2.99) e () + j c0 2 r

In the time domain, we obtain up to second order in the inverse linewidth A(z, t0 ) 0 K0 1 2 A(z, t0 ), = (2.103) 1+ z (loss) 2 2 t2



K0 where g = 02 is the peak gain at line center per unit length and g is the HWHM linewidth of a transition providing gain.

which corresponds to a parabolic approximation of the line shape at line center, (Fig. 2.3). As we will see later, for an amplifying optical transition we obtain a similar equation. We only have to replace the loss by gain A(z, t0 ) 1 2 = g 1 + 2 2 A(z, t0 ), (2.104) z (gain) g t


Kramers-Kroenig Relations and Sellmeier Equation

The linear susceptibility is the frequency response or impulse response of a linear system to an applied electric eld, see Eq.(2.42). For a real physical system this response is causal, and therefore real and imaginary parts obey Kramers-Kroenig Relations 2 r () = Z

i ( ) d = n2 r () 1, 2 2 r () d. 2 2


2 i () =



For optical media these relations have the consequence that the refractive index and absorption of a medium are not independent, which can often be exploited to compute the index from absorption data or the other way around. The Kramers-Kroenig Relations also give us a good understanding of the index variations in transparent media, which means the media are used in a frequency range far away from resonances. Then the imaginary part of the susceptibility related to absorption can be approximated by X Ai ( i ) (2.107) i () =

and the Kramers-Kroenig relation results in the Sellmeier Equation for the refractive index X X i n2 () = 1 + Ai 2 = 1 + ai 2 . (2.108) 2 i 2 i i i



This formula is very useful in tting the refractive index of various media over a large frequency range with relatively few coecients. For example Table 2.3 shows the Sellmeier coecients for fused quartz and sapphire.

a1 a2 a3 2 1 2 2 2 3

Fused Quartz 0.6961663 0.4079426 0.8974794 4.679148103 1.3512063102 0.9793400102

Sapphire 1.023798 1.058364 5.280792 3.77588103 1.22544102 3.213616102

Table 2.3: Table with Sellmeier coecients for fused quartz and sapphire.

In general, each absorption line contributes a corresponding index change to the overall optical characteristics of a material, see Fig. 2.14. A typical situation for a material having resonances in the UV and IR, such as glass, is shown in Fig. 2.15. As Fig. 2.15 shows, due to the Lorentzian line shape, outside of an absorption line the refractive index is always decreasing as a function of wavelength. This behavior is called normal dispersion and the opposite behavior abnormal dispersion typically only occurs inside of absorption lines.

dn < 0 : normal dispersion (blue refracts more than red) d dn > 0 : abnormal dispersion d

This behavior is also responsible for the mostly positive group delay dispersion (GVD) or group delay dispersion (GDD) over the transparency range of a material, which are closely related to dn . d



Figure 2.14: Each absorption line must contribute to an index change via the Kramers-Kroenig relations.

Figure 2.15: Typcial distribution of absorption lines in a medium transparent in the visible.

Fig. 2.16 shows the transparency range of some often used media.



Figure 2.16: Transparency range of some materials according to [6], p. 175.

GVD and GDD are dened as the variation of the inverse group velocity or group delay as a function of frequency, see Eqs.() , i.e. d2 k( ) d 1 = (2.109) d 2 =0 d g ( ) =0 L d d d2 k( ) Tg ( )|=0 (2.110) L = = 2 d =0 d g ( ) =0 d

GV D = GDD =

where Tg ( ) = L/g ( ) is the group delay of a wave packet with relative center frequency . Often theses quantities need to be calculated from the refractive index given by the Sellmeier equation, i.e. n(). The corresponding quantities are listed in Table 2.4. The computations are done by substituting the frequency with the wavelength.

2.1. MAXWELLS EQUATIONS OF ISOTROPIC MEDIA Dispersion Characteristic medium wavelength: n medium wavenumber: kn phase velocity: p group velocity: g group velocity dispersion: GV D group delay: Tg =
L g


n 2 n k d 2 ; d = 2 d dk c0 d2 k d 2 kL) d = d(d d 2 (kL) dTg = d d 2 d

Comp. from n()

n() 2 n() c0 n() c0 dn 1 1 n n d 3 d2 n 2 2 2c0 d n dn 1 L c0 n d 3 d2 n 2L 2c2 0 d

d d

group delay dispersion: GDD

Table 2.4: Table with important dispersion characteristics and how to compute them from the wavelength dependent refractive index n() using |d | = 2c0 /2 d||.




Electromagnetic Waves and Interfaces

Many microwave and optical devices are based on the characteristics of electromagnetic waves undergoing reection or transmission at interfaces between media with dierent electric or magnetic properties characterized by and , see Fig. 2.17. Without restriction we can assume that the interface is the (x-y-plane) and the plane of incidence is the (x-z-plane). An arbitrary incident plane wave can always be decomposed into two components. One component is called the transverse electric (TE)-wave or also s-polarized wave. It has its electric eld perpenticular to the plane of incidence. The other component is called the TM-wave or also p-polarized wave. It has its electric eld in the plane of incidence. The most general case of an incident monochromatic TEM-wave is a linear superposition of a TE and a TM-wave.

Reflection of TE-Wave

Reflection of TM-Wave






ki i r kr



ki i r kr


kt Ht

kt Et Ht

Figure 2.17: a) Reection of a TE-wave at an interface, b) Reection of a TM-wave at an interface

The elds for both cases are summarized in table 2.5

2.2. ELECTROMAGNETIC WAVES AND INTERFACES TE-wave ey E i =E i e j(tki r) H i =H i e hi E r =E r ej(tkr r) ey H r =H r ej(tkr r) hr E t =E t ej(tkt r) ey H t =H t ej(tkt r) ht
j(tki r)


TM-wave E i =E i ej(tki r) ei H i =H i ej(tki r) ey

r E r =E r er er j(tkr r) ey H r =E r e E t =E t ej(tkt r) et H t =H t ej(tkt r) ey

j(tk r)

Table 2.5: Electric and magnetic elds for TE- and TM-waves. with wave vectors of the waves given by ki = kr = k0 1 1 , kt = k0 2 2 , ki,t = ki,t (sin i,t ex + cos i,t ez ) , kr = ki (sin r ex cos r ez ) , and unit vectors dening the directions of the transmitted and reected magnetic and electric elds, respectively hi,t = cos i,t ex + sin i,t ez ,

hr = cos r ex + sin r ez , ei,t = hi,t = cos i,t ex sin i,t ez , er = hr = cos r ex sin r ez .


Boundary Conditions and Snells law

From 6.013, we know that Stokes and Gauss Law for the electric and magnetic elds require constraints on some of the eld components at media boundaries. In the absence of surface currents and charges, the tangential electric and magnetic elds as well as the normal dielectric and magnetic uxes have to be continuous when going from medium 1 into medium 2 for all times at each point along the surface, i.e. the surface z = 0 Ei,x/y ej(tki,x x) + Er,x/y ej(tkr,x x) = Et,x/y ej(tkt,x x) . Hi,x/y ej(tki,x x) + Hr,x/y ej(tkr,x x) = Ht,x/y ej(tkt,x x) . (2.111) (2.112)



This equation can only be fullled at all times if and only if the x-component of the k-vectors for the reected and transmitted wave are equal to (match) the corresponding component of the incident wave (2.113) This phase matching condition is shown in Fig. 2.18 for the case 2 2 > 1 1 or kt > ki . The semi-circles dene the dispersion relation for the waves of the same frequency in the dierent media. The x-components of the wave vectors need to be matched. ki,x = kr,x = kt,x



i r t




Figure 2.18: Phase matching condition for reected and transmitted wave The phase matching condition Eq(2.113) results in r = i = 1 and Snells law for the angle t = 2 of the transmitted wave 1 1 sin i (2.114) sin t = 2 2 or for the case of non magnetic media with 1 = 2 = 0 n1 sin t = sin i n2





Measuring Refractive Index with a Prism

Snells law can be used for measuring the refractive index of materials. Consider a prism prepared from a material with unknown refractive index n(), see Fig. 2.19 (a).

Figure 2.19: (a) Beam propagating through a prism. (b) For the case of minimum deviation [3] p. 65. The prism is mounted on a rotation stage as shown in Fig. 2.20. The angle of incidence is then varied with a xed incident beam path and the transmitted light is observed on a screen. If one starts of with normal incidence on the rst prism surface one notices that after turning the prism one goes through a minium for the deection angle of the beam. This becomes obvious from Fig. 2.19 (b). There is an angle of incidence where the beam path through the prism is symmetric. If the input angle is varied around this point, it would be identical to exchange the input and output beams. From that we conclude that the deviation must go through an extremum at the symmetry point, see Figure 2.21. It can be shown (Recitations), that the refractive index is then determined by n=
min sin + 2 , sin 2


where min is the minimum deection angle. If the measurement is repeated for various wavelength of the incident radiation the complete wavelength dependent refractive index is characterized, see for example, Fig. 2.22.



Figure 2.20: Refraction of a Prism with n=1.731 for dierent angles of incidence alpha. The angle of incidence is stepwise increased by rotating the prism clockwise. The angle of transmission rst increases. After the angle for minimum deviation is reached the transmission angle starts to decrease [3] p67.



Figure 2.21: Deviation versus incident angle [1]

Figure 2.22: Refractive index as a function of wavelength for various media transmissive in the visible [1], p42.




Fresnel Reection

After understanding the direction of the reected and transmitted light, formulas for how much light is reected and transmitted are derived by evaluating the boundary conditions for the TE and TM-wave. According to Eqs.(2.111) and (2.113) we obtain for the continuity of the tangential E and H elds: TE-wave (s-pol.) E i +E r = E t H i cos i H r cos r =H t cos t TM-wave (p-pol.) E i cos i E r cos r =E t cos t Hi + Hr = Ht

(2.117) q

0 1/2 , Introducing the characteristic impedances in both half spaces Z 1/2 = 0 1/2 and the impedances that relate the tangential electric and magnetic eld T E/T M components Z 1/2 in both half spaces the boundary conditions can be rewritten in terms of the electric or magnetic eld components. Note, that for complex susceptibilities the wave impedances become complex.

TE-wave (s-pol.)
E ZT 1/2 = E i/t H i/t cos i/t

TM-wave (p-pol.) =
Z 1/2 cos 1/2 E i/t cos i/t = H i/t T M 2 H i H r = Z M Ht ZT 1 M ZT 1/2 =

Z 1/2 cos 1/2

E i +E r = E t E i E r =
E ZT 1 E T Z2 E t


Hi + Hr = Ht

Amplitude Reection and Transmission Coecients These equations can be easily solve for the amplitude reection and transmission coecients of the waves. By dividing both equations by the incident wave amplitudes we obtain for the amplitude reection and transmission coecients. Note, we have dened the reection and transmission coecients in terms of the electric elds for the TE-wave and in terms of the magnetic elds for the TM-wave. TE-wave (s-pol.) r

TM-wave (p-pol.) rT M = 1 1+r

Hr T M t ;t =H Hi Hi T M Z TM rT M = Z 2 TM t 1 TM TM


r t =E ; tT E = E Ei Ei TE TE



1 rT E =

E ZT 1 tT E T Z2 E




Note, that the pair of equations for the amplitude reection and transmission coecients are similar for the TE- and TM-wave. Thus the amplitude transmission and reection coecients are t

= 1+

Z 1/2 Z 2/1

2Z 2/1 Z1


+ Z2




Z 2/1 Z1


Z 1/2 + Z2




Despite the simplicity of these formulas, they describe an enormous wealth of phenomena. To get some insight, consider the case of dielectric lossless media characterized by its real refractive indices n1 and n2 . Then Eqs.(2.120) and (2.121) for the TE and TM case can be written as TE-wave (s-pol.)
E ZT 1/2

TM-wave (p-pol.)
M ZT 1/2 = Z 1/2 cos 1/2 =

Z1/2 cos 1/2

Z0 n1/2 cos 1/2

rT E = tT E =

n1 cos 1 n2 cos 2 n1 cos 1 +n2 cos 2 2n1 cos 1 n1 cos 1 +n2 cos 2

rT M = tT M =

n2 n cos1 cos 2 1 n2 n + cos1 cos 2 1 n 2 cos2 2 n2 n + cos1 cos 2 1

Z0 n1/2

cos 1/2 (2.122)



Figure 2.23 shows the evaluation of Eqs.(2.122) for the case of a reection at the interface of air and glass with n2 > n1 and (n1 = 1, n2 = 1.5). For TEpolarized light the reected light changes sign with respect to the incident light (reection at the optically more dense medium). Note, that this phase shift always happens for reection from optically less dense media independent from the angle of incidence. This is not so for TM-polarized light The phase shift only occurs for angles larger than B , which is called the Brewster angle. So for TM-polarized light the amplitude reection coecient is zero at Brewsters angle. This phenomena will be discussed in more detail later.

Amplitude refl. and transm. coefficients

1.2 1.0 0.8 0.6 0.4 0.2 0.0 -0.2 -0.4 -0.6 -0.8 -1.0 0 15



r r



30 45 60 Incident angle 1 (deg.) 75 90

Figure 2.23: The amplitude coecients of reection and transmission as a function of incident angle. These correspond to external reection n2 > n1 at an air-glas interface (n1 = 1, n2 = 1.5).



This behavior changes drastically if we consider the opposite arrangement of media, i.e. we consider the glass-air interface with n1 > n2 , see Figure 2.24. Then the TM-polarized light experiences a -phase shift upon reection close to normal incidence. For increasing angle of incidence this 0 reection coecient goes through zero at the Brewster angle B dierent from before. However, for large enough angle of incidence the reection coefcient reaches magnitude 1 and stays there. This phenomenon is called total internal reection and the angle where this occurs rst is the critical angle for total internal reection, tot . Total internal reection will be discussed in more detail later.

Amplitude refl. and transm. coefficients

2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 -0.2 0



r r




15 30 33.7 41.8 45 Incident angle 1 (deg.)

Figure 2.24: The amplitude coecients of reection and transmission as a function of incident angle. These correspond to internal reection n1 > n2 at a glas-air interface (n1 = 1.5, n2 = 1).



Power reection and transmission coecients Often we are not interested in the amplitude but rather in the optical power reected or transmitted in a beam of nite size, see Figure 2.25.

Figure 2.25: Reection and transmission of an incident beam of nite size [1]. Note, that to get the power in a beam of nite size, we need to integrated the corresponding Poynting vector over the beam area, which means multiplication by the beam crosssectional area for a homogenous beam. Since the angle of incidence and reection are equal, i = r = 1 this beam crosssectional area drops out in reection RT E/T M T E/T M T E/T M 2 T E/T M 2 Z Z 1 = 2 = T E/T M = r (2.123) T E/T M T E/T M Z1 Ii A cos r + Z2 Ir

A cos i

However, due to the dierent angles for the incident and the transmitted beam t = 2 6= 1 , we arrive at T


A cos t T E/T M Ii A cos i ( 1 Z 2/1


(2.124) ) Re ( 1 Z 1/2 )1 T E/T M 2 t .

cos 2 = Re cos 1

2.2. ELECTROMAGNETIC WAVES AND INTERFACES Using in the case of TE-polarization

Z 1/2 cos 1/2


E = ZT 1/2 and analogously for TM-

M polarization Z 1/2 cos 1/2 = Z T 1/2 , we obtain )1 ( T E/T M 4Z 2/1 1 T E/T M T = Re Re 2 T E/T M T E/T M T E/T M Z 1/2 + Z2 Z 1


Note, for the case where the characteristic impedances are complex this can not be further simplied. If the characteristic impedances are real, i.e. the media are lossless, the transmission coecient simplies to T

To summarize for lossless media the power reection and transmission coefcients are TE-wave (s-pol.)
TE Z1 /2 =

= 2 T E/T M T E/T M Z1 + Z2 .






TM-wave (p-pol.)
TM Z1 /2 = Z1/2 cos 1/2 = n n1 2 cos2 cos TM 2 1 R = n2 n cos + cos1 2 1 Z0 n1/2


T TE =

2 cos 2 n1 cos 1 2 = n n1 cos 1 +n2 cos 2

4n1 cos 1 n2 cos 2 |n1 cos 1 +n2 cos 2 |2

Z1/2 cos 1/2

Z0 n1/2 cos 1/2

cos 1/2 (2.127)

T TM =

T T E + RT E = 1

T TM + R

2 n2 n cos + cos1 2 1 TM

n1 n 4 cos2 2 cos 1


A few phenomena that occur upon reection at surfaces between dierent media are especially noteworthy and need a more indepth discussion because they enhance or enable the construction of many optical components and devices.


Brewsters Angle

As Figures 2.23 and 2.24 already show, for light polarized parallel to the plane of incidence, p-polarized light, the reection coecient vanishes at a given angle B , called the Brewster angle. Using Snells Law Eq.(2.115), n2 sin 1 = , n1 sin 2 (2.128)



we can rewrite the reection and transmission coecients in Eq.(2.127) only in terms of the angles. For example, we nd for the reection coecient RT M n2 cos 2 2 sin 1 n1 cos 1 2 = n2 cos 2 = sin 1 n + cos sin + sin 2 1 1 2 sin 21 sin 22 = sin 21 + sin 22 sin 21,B sin 22,B = 0 or 21,B = 22,B or 1,B + 2,B = cos 2 2 cos 1 cos 2 cos 1 (2.129) (2.130)

where we used in the last step in addition the relation sin 2 = 2 sin cos . Thus by forcing RT M = 0, the Brewster angle is reached for (2.131)

(2.132) 2 This relation is illustrated in Figure 2.26. The reected and transmitted beams are orthogonal to each other, so that the dipoles induced in the medium by the transmitted beam, shown as arrows in Fig. 2.26, can not radiate into the direction of the reected beam. This is the physical origin of the zero in the reection coecient, only possible for a p-polarized or TM-wave. The relation (2.132) can be used to express the Brewster angle as a function of the refractive indices, because if we substitute (2.132) into Snells law we obtain sin 1 n2 = sin 2 n1 sin 1,B sin 1,B = = tan 1,B , cos 1,B sin 2 1,B tan 1,B = n2 . n1 (2.133)


Using the Brewster angle condition one can insert an optical component with a refractive index n 6= 1 into a TM-polarized beam in air without having reections, see Figure 2.27. Note, this is not possible for a TE-polarized beam.


Reflection of TM-Wave at Brewsters Angle



Er 1, 1, kr





Figure 2.26: Conditions for reection of a TM-Wave at Brewsters angle. The reected and transmitted beams are orthogonal to each other, so that the dipoles excited in the medium by the transmitted beam can not radiate into the direction of the reected beam.

Figure 2.27: A plate under Brewsters angle does not reect TM-light. The plate can be used as a window to introduce gas lled tubes into a laser beam without insertion loss (ideally), [6] p. 209.




Total Internal Reection

Another striking phenomenon, see Figure 2.24, occurs for the case where the beam hits the surface from the side of the optically denser medium, i.e. n1 > n2 . There is a critical angle of incidence, beyond which all light is reected. How can that occur? This is easy to understand from the phase matching diagram at the surface, see Figure 2.18, which is redrawn for this case in Figure 2.28.

Figure 2.28: Phase matching diagram for total internal reection. There is no real wavenumber in medium 2 possible as soon as the angle of incidence becomes larger than the critical angle for total internal reection i > tot with sin tot = n2 . n1 (2.134)


Figure 2.29 shows the angle of refraction and incidence for the two cases of external (n1 < n2 ) and internal reection (n1 > n2 ), when the angle of incidence approaches the critical angle.



Figure 2.30: Relation between angle of refraction and incidence for external refraction and internal refraction ([1], p. 81).

Figure 2.29: Relation between angle of refraction and incidence for external refraction and internal refraction ([6], p. 11). Total internal reection enables broadband reectors. Figure 2.30 shows



again what happens when the critical angle of reection is surpassed. Figure 2.31 shows how total internal reection can be used to guide light via reection at a prism or by multiple reections in a waveguide.

Figure 2.31: (a) Total internal reection, (b) internal reection in a prism, (c) Rays are guided by total internal reection from the internal surface of an optical ber ([6] p. 11). Figure 2.32 shows the realization of a retro reector, which always returns a parallel beam independent of the orientation of the prism (in fact the prism can be a real 3D-corner so that the beam is reected parallel independent from the precise orientation of the corner cube). A surface patterned by little corner cubes constitute a "cats eye" used on trac signs.

Figure 2.32: Total internal reection in a retro reector. More on reecting prisms and its use can be found in [1], pages 131-136.



What is the eld in medium 2 when total internal reection occurs? Is it identical to zero? It turns out phase matching can still occur if the propagation constant in z-direction becomes imaginary, k2z = j2z , because then we can fulll the wave equation in medium 2. This is equivalent to the dispersion relation
2 2 2 k2 x + k2z = k2 ,

or with k2x = k1x = k1 sin 1 , we obtain for the imaginary wavenumber q 2 2 = k1 sin2 1 k2 , p = k1 sin2 1 sin2 tot .


(2.136) (2.137)

The electric eld in medium 2 is then, for the example for a TE-wave, given by E t = E t ey ej(tkt r) , E t ey ej(tk2,x x) e2z z .

(2.138) (2.139)

Thus the wave penetrates into medium 2 exponentially with a 1/e-depth , given by = 1 1 = p 2z k1 sin2 1 sin2 tot (2.140)

Figure 2.33 shows the penetration depth as a function of angle of incidence for a silica/air interface and a silicon/air interface. The gure demonstrates that light from inside a semiconductor material with a relatively high index around n=3.5 is mostly captured in the semiconductor material (Problem of light extraction from light emitting diodes (LEDs)), see problem set 3.



Penetration depth, m


n1=3.5 n2=1

n1=1.45 n2=1


0.0 0 20 40 60 o Angle of incidence, 80

Figure 2.33: Penetration depth for total internal reection at a silica/air and a silicon/air interface for = 1550nm.

As the magnitude of the reection coecient is 1 for total internal reection, the power owing into medium 2 must vanish, i.e. the transmission is zero. Note, that the transmission and reection coecients in Eq.(2.122) can be used beyond the critical angle for total internal reection. We only have to be aware that the electric eld in medium 2 has an imaginary dependence in the exponent for the z-direction, i.e. k2z = k2 cos 2 = j2z . Thus cos 2 in all formulas for the reection and transmission coecients has to be replaced by the imaginary number

cos 2 =

k2z k1 p 2 = j sin 1 sin2 tot k2 k2 n1 p 2 = j sin 1 sin2 tot n2 s 2 sin 1 = j 1. sin tot


2.2. ELECTROMAGNETIC WAVES AND INTERFACES Then the reection coecients in Eq.(2.122) change to all-pass functions TE-wave (s-pol.) rT E = r
TE n1 cos 1 n2 cos 2 n1 cos 1 +n2 cos 2 r 2 n sin 1 cos 1 +j n2 1 sin tot 1 r cos 1 j n2
1 n


TM-wave (p-pol.) rT M = r
n2 n cos1 cos 2 1 n2 n1 + cos cos 2 1 r n 2 n 2


T E 2

1 n2 cos 1 n1

sin tot r

sin 1

T M 2

cos 1 +j n1 cos 1 j n1

sin 1 sin tot

sin 1 sin tot sin 1

2 2

1 1

1 tan

1 n1 cos 1 n2

sin tot r

sin 1 sin tot


Thus the magnitude of the reection coecient is 1. However, there is a non-vanishing phase shift for the light eld upon total internal reection, denoted as T E and T M in the table above. Figure 2.34 shows these phase shifts for the glass/air interface and for both polarizations.
200 Phase in Reflection,

n1=1.45 n2=1 100 Brewster angle 0 0 20 40 60 o Angle of incidence, 80 TE-Wave (s-polarized) TM-Wave (p-polarized)

Figure 2.34: Phase shifts for TE- and TM- wave upon reection from a silica/air interface, with n1 = 1.45 and n2 = 1. Goos-Haenchen-Shift So far, we looked only at plane waves undergoing reection at surface due to total internal reection. If a beam of nite transverse size is reected from



such a surface it turns out that it gets displaced by a distance z , see Figure 2.35 (a), called Goos-Haenchen-Shift.

Figure 2.35: (a) Goos-Haenchen Shift and related beam displacement upon reection of a beam with nite size; (b) Accumulation of phase shifts in a waveguide. (c) Goos-Haenchen Shift by reection at a virtual layer located at the penetration depth. Detailed calculations show, that the displacement is given by z = 2 T E/T M tan 1 , (2.143)

as if the beam was reected at a virtual layer with depth T E/T M into medium

2.2. ELECTROMAGNETIC WAVES AND INTERFACES 2. It turns out, that for TE-waves T E = ,



where is the penetration depth according to Eq.(2.140) for evanescent waves. But for TM-waves (2.145) T M = 2 n1 2 1 + n2 sin 1 1

These shifts accumulate when the beam is propagating in a waveguide, see Figure 2.35 (b) and is important to understand the dispersion relations of waveguide modes. The Goos-Haenchen shift can be observed by reection at a prism partially coated with a silver lm, see Figure 2.36. The part reected from the silver lm is shifted with respect to the beam reected due to total internal reection, as shown in the gure.

Figure 2.36: Experimental proof of the Goos-Haenchen shift by total internal reection at a prism, that is partially coated with silver, where the penetration of light can be neglected. [3] p. 486. Frustrated total internal reection Another proof for the penetration of light into medium 2 in the case of total internal reection can be achieved by putting two prisms, where total



internal reection occurs back to back, see Figure 2.37. Then part of the light, depending on the distance between the two interfaces, is converted back into a propagating wave that can leave the second prism. This eect is called frustrated internal reection and it can be used as a beam splitter as shown in Figure 2.37.

Figure 2.37: Frustrated total internal reection. Part of the light is picked up by the second surface and converted into a propagating wave.


Polarization of Electromagnetic Waves

So far we have discussed linearly polarized electromagnetic waves, where the electric eld of a TEM-wave propagating along the z direction was either polarized along the x or y axis. The most general TEM-wave has simultaneously electric elds in both polarizations and the direction of the electric eld in space, i.e. its polarization, can change during propagation. A description of polarization and polarization evolution in optical systems can be accomplished using Jones vectors and matrices.


E 0x E (z, t) = E 0y ej(tkz) , 0

A general complex TEM-wave propagating along the z direction is given by (2.146)



where E 0x = E0x ejx and E 0y = E0y ejy are the complex eld amplitudes of the x and y polarized components of the wave. The real electric eld is given by E0x cos (t kz + x ) E (z, t) = E0y cos t kz + y , (2.147) 0 Both components are periodic functions in t kz = (t z/c) . Linear Polarization If the phases of the complex eld amplitudes along the x and y axis are equal, i.e. E 0x = E0x ej and E 0y = E0y ej then the real electric eld E0x E (z, t) = E0y cos (t kz + ) 0 (2.148)

always oscillates along a xed direction in the x-y-plane, see Figure 2.38

Figure 2.38: Linearly polarized light. (a) Time course at a xed position z. (b) A snapshot at a xed time t, [6], p. 197. The angle between the polarization direction and the x-axis, , is given by = arctan (E0y /E0x ) . If there is a phase dierence of the complex eld amplitudes along the x and y axis, the direction and magnitude of the electric eld amplitude changes periodically in time at a given position z.



Circular Polarization Special cases occur when the magnitude of the elds in both linear polarizations are equal E0x = E0y = E0 , but there is a phase dierence = in 2 both components. Then we obtain ej (2.149) E (z, t) = E0 Re ej () ej(tkz) 0 cos (t kz + ) = E0 sin (t kz + ) . (2.150) 0

For this case, the tip of the electric eld vector describes a circle in the x y plane, as
2 |Ex (z, t)|2 + |Ey (z, t)|2 = E0 for all z, t,


see Figure 2.39.

Figure 2.39: Trajectories of the tip of the electric eld vector of a right and left circularly polarized plane wave. (a) Time course at a xed position z. (b) A snapshot at a xed time t. Note, the sense of rotation in (a) is opposite to that in (b) [6], p. 197.



Right Circular Polarization If the tip of the electric eld at a given time, t, rotates counter clockwise with respect to the phase fronts of the wave, here in the positive z direction, then the wave is called right circularly polarized light, i.e. cos (t kz + ) 1 Erc (z, t) = E0 Re j ej(tkz+) = E0 sin (t kz + ) . 0 0 (2.152) A snapshot of the lines traced by the end points of the electric-eld vectors at dierent positions is a right-handed helix, like a right-handed screw pointing in the direction of the phase fronts of the wave, i.e. k vector see Figure 2.39 (b). Left Circular Polarization If the tip of the electric eld at a given xed time, t, rotates clockwise with respect to the phase fronts of the wave, here in the again in the positive z direction, then the wave is called left circularly polarized light, i.e. cos (t kz + ) 1 Elc (z, t) = E0 Re j ej(tkz+) = E0 sin (t kz + ) . 0 0 (2.153) Eliptical Polarization The general polarization case is called eliptical polarization, as for arbitrary E 0x = E0x ejx and E 0y = E0y ejy , we obtain for the locus of the tip of the electric eld vector from E0 cos (t kz + x ) E (z, t) = E0y cos t kz + y . (2.154) 0 the relations Ey = cos t kz + y = cos t kz + x + y x (2.155) E0y (2.156) = cos (t kz + x ) cos y x sin (t kz + x ) sin y x .

70 and


Ex = cos (t kz + x ) . (2.157) E0x These relations can be combined to Ey Ex cos y x = sin (t kz + x ) sin y x (2.158) E0y E0x s 2 Ex sin (t kz + x ) = 1 (2.159) E0x Substituting Eq.(2.159) in Eq.(2.158) and building the square results in 2 ! 2 Ex Ex Ey sin2 y x . cos y x = 1 E0y E0x E0x


This is the equation of an ellipse making an angle with respect to the x-axis given by 2E0x E0y cos y x . (2.162) tan 2 = 2 2 E0 x E0y see Figure 2.40.

After reordering of the terms we obtain 2 2 Ex Ey Ex Ey + 2 cos y x = sin2 y x . E0x E0y E0x E0y


Figure 2.40: (a) Rotation of the endpoint of the electric eld vector in the x-y-plane at a xed position z . (b) A snapshot at a xed time t [6], p. 197.



Elliptically polarized light can also be understood as a superposition of a right and left cicular polarized light, see Figure 2.41.

Figure 2.41: Elliptically polarized light as a superposition of right and left circularly polarized light [1], p. 223.


Jones Calculus

As seen in the last section, the information about polarization of a TEM-wave can be tracked by a vector that is proportional to the complex electric-eld vector. This vector is called the Jones vector E 0x Vx V = : Jones Vector (2.163) E 0y Vy Jones Matrix Figure 2.42 shows a light beam that is normally incident on a retardation plate along the z axis with a polarization state described by a Jones vector. A retardation plate is made up of a crystaline material, such as for example Quartz, which has unisotropic dielectric behavior. For example, it shows dierent refractive index depending on the alignment of the polarization with the crystal axes. For retardation plates, there are two principle axis, s for slow and f for fast axis, i.e. the refractive index along the slow axis is larger than along the fast axis. Let ns and nf be the refractive index of the slow and fast principle axis, respectively, then ns > nf .



Figure 2.42: A retardation plate rotated at an angle about the z-axis. f("fast") and s("slow") are the two principal dielectric axes of the crystal for light propagating along the z axis [2], p. 17. Usually, the principle axis (s and f axis) of the retardation plate are rotated by an angle with respect to the x and y axis. The polarization state of the emerging beam in the crystal coordinate system is thus given by 0 jk n L Vs e o s Vs 0 = , (2.164) Vf0 0 ejko nf L Vf The phase retardation is dened as the phase dierence between the two components (2.165) = (ns nf ) ko L. In birefringent crystals the dierence in refractive index is much smaller than the index itself, |ns nf | ns , nf , therefore parallel to the evolving dierential phase a large absolute phase shift occurs. Taking the mean phase shift 1 = (ns + nf ) ko L, (2.166) 2 out, we can rewrite (2.164) as j /2 0 Vs e Vs 0 j =e . (2.167) Vf0 Vf 0 ej /2



The matrix connecting the Jones vector at the input of an optical component with the Jones vector at the output is called a Jones matrix. If no coherent additon with another eld is planned at the output of the system, the average phase can be dropped. With the rotation matrix, R, connecting the (x, y ) coordinate system with the (s, f ) coordinate system cos sin , (2.168) R () = sin cos we nd the Jones matrix W describing the propagation of the eld components through the retardation plate as 0 Vx Vx =W . (2.169) Vy0 Vy with W = R () W0 R () . and W0 = ej /2 0 j /2 0 e . (2.170) (2.171)

Carrying out the matix multiplications leads to j /2 e cos2 () + ej /2 sin2 () j sin sin (2) 2 . W = j sin sin (2) ej /2 sin2 () + ej /2 cos2 () 2 (2.172) Note that the Jones matrix of a wave plate is a unitary matrix, that is W W = 1. Unitary matrices have the property that they transform orthogonal vectors into another pair of orthogonal vectors. Thus two orthogonal polarization states remain orthogonal when propagating through wave plates. Polarizer A polarizer is a device that absorbs one component of the polarization vector. The Jones matrix of polarizer along the x-axis or y-axis is 1 0 0 0 Px = , and Py = . (2.173) 0 0 0 1



Half-Wave Plate A half-wave plate has a phase retardation of = , i.e. its thickness is t = /2(ne no ). The corresponding Jones matrix follows from Eq.(2.172) cos(2) sin (2) . (2.174) W = j sin (2) cos(2) For the special case of = 45o , see Figure 2.43, the half-wave plate rotates a linearly polarized beam exactly by 900 , i.e. it exchanges the polarization axis. It can be shown, that for a general azimuth angle , the half-wave plate will rotate the polarization by an angle 2, see problem set. When the incident light is circularly polarized a half-wave plate will convert righthand circularly polarized light into left-hand circularly polarized light and vice versa, regardless of the azimuth angle .

Figure 2.43: The eect of a half-wave plate on the polarziation state of a beam, [2], p.21. Quarter-Wave Plate A quarter-wave plate has a phase retardation of = /2, i.e. its thickness is t = /4(ne no ). The corresponding Jones matrix follows again from Eq.(2.172)



W =

1 2

[1 j cos(2)] 1 j sin (2) 2

1 j sin (2) 2 1 [1 + j cos(2)] 2


see Figure 2.44.

and for the special case of = 45o , see Figure 2.43 we obtain 1 1 j , W= 2 j 1


Figure 2.44: The eect of a quarter wave plate on the polarization state of a linearly polarized input wave [2], p.22. If the incident beam is vertically polarized, i.e. Vx 0 = , 1 Vy


the eect of a 45o -oriented quarter-wave plate is to convert vertically polarized light into right-handed circularly polarized light. 0 j Vx 1 . (2.178) = Vy0 2 j



If the incident beam is horizontally polarized Vx 1 , = 0 Vy


the outgoing beam is a left-handed circularly polarized, see Figure 2.44. 0 1 Vx 1 . (2.180) = Vy0 2 j


Interference and Interferometers

A phenomenon characteristic for waves is interference. Many optical devices are based on the concept of interfering waves, such as interferometers, low loss dielectric mirrors and other multilayer optical coatings. After having a quick look into the phenomenon of interference, we will develope a powerful matrix formalism that enables us to evaluate eciently many optical (also microwave) systems based on interference.


Interference and Coherence

Interference Interference of waves is a consequence of the linearity of the wave equation (2.7). If we have two individual solutions of the wave equation E1 (r, t) = E1 cos( 1 t k1 r + 1 ) e1 , (2.181) (2.182)

E2 (r, t) = E2 cos( 2 t k1 r + 2 ) e2 ,

with arbitrary amplitudes, wave vectors and polarizations, the sum of the two elds (superposition) is again a solution of the wave equation E (r, t) = E1 (r, t) + E2 (r, t). (2.183)

If we look at the intensity, which is proportional to the amplitude square of the total eld 2 2 (2.184) E (r, t) = E1 (r, t) + E2 (r, t) ,

2.4. INTERFERENCE AND INTERFEROMETERS we nd E (r, t)2 = E1 (r, t)2 + E2 (r, t)2 + 2E1 (r, t) E2 (r, t) with E2 E1 (r, t) = 1 2 E2 E2 (r, t)2 = 2 2



1 + cos 2( 1 t k1 r + 1 ) , 1 + cos 2( 2 t k2 r + 2 ) , cos(2 t k2 r + 2 )

(2.186) (2.187)

E1 (r, t) E2 (r, t) = (e1 e2 ) E1 E2 cos( 1 t k1 r + 1 ) (2.188)

Since at optical frequencies neither our eyes nor photo detectors, can ever follow the optical frequency itself and certainly not twice as large frequencies, we drop the rapidly oscillating terms. Or in other words we look only on the cycle-averaged intensity, which we denote by a bar E (r, t)2 =
2 E1 E2 + 2 + (e1 e2 ) E1 E2 2 2 cos ( 1 2 ) t k1 k2 r + (1 2 ) (2.191)

1 E1 (r, t) E2 (r, t) = (e1 e2 ) E1 E2 2 cos (1 2 ) t k1 k2 r + (1 2 ) + cos ( 1 + 2 ) t k1 + k2 r + (1 + 2 )

(2.189) (2.190)

Depending on the frequencies 1 and 2 and the deterministic and stochastic properties of the phases 1 and 2 , we can detect this periodically varying intensity pattern called interference pattern. Interference of waves can be best visuallized with water waves, see Figure 2.45. Note, however, that water waves are a scalar eld, whereas the EM-waves are vector waves. Therefore, the interference phenomena of EM-waves are much richer in nature than for water waves. Notice, from Eq.(2.191), it follows immediately that the interference vanishes in the case of orthogonally polarized EM-waves, because of the scalar product involved. Also, if the frequencies of the waves are not identical, the interference pattern will not be stationary in time.



Figure 2.45: Interference of water waves from two point sources in a ripple tank [1] p. 276.

If the frequencies are identical, the interference pattern depends on the wave vectors, see Figure 2.46.


Lines of constant differential phase


Figure 2.46: Interference pattern generated by two monochromatic plane waves.

2.4. INTERFERENCE AND INTERFEROMETERS The interference pattern has itself a wavevector given by k1 k2 and therefore varies periodically with period 2 . = k1 k2





The ability of waves to generate an interference pattern is called coherence. Coherence can be quantied both temporally or spatially. For example, if we are at a certain position r in the interference pattern described by Eq.(2.191), we will only have stationary conditions over a time interval Tcoh << 2 . 1 2

Thus the spectral width of the waves determines the temporal coherence. However, it depends very often on the expermental arrangement whether a given situation can still lead to interference or not. Even so the interfering light may be perfectly temporally coherent, i.e. perfectly monochromatic, 1 = 2 ,yet the wave vectors may not be stable over time and the spatial inteference pattern may wash out, i.e. there is insucient spatial coherence. So for stable and maximum interference three conditions must be fullled: stable and identical polarization small change in the relative phase between the beams involved over the observation time, temporal coherence, often achieved by using narrrow linewidth light stable beam propagation or guiding of light in a waveguide to achieve spatial coherence. It is by no means trivial to arrive at a light source and an experimental setup that enables good coherence and strong interference of the beams involved. Interference of beams can be used to measure relative phase shifts between them which may be proportional to a physical quantity that needs to be



measured. Such phase shifts between two beams can also be used to modulate the light output at a given position in space via interference. In 6.013, we have already encountered interference eects between forward and backward traveling waves on transmission lines. This is very closely related to what we use in optics, therefore, we quickly relate the TEM-wave progagation to the transmission line formalism developed in Chapter 5 of 6.013.


TEM-Transmission Lines and TEM-Waves

The voltage V and current I along a TEM transmission line with an inductance L0 and a capacitance C 0 per unit length satisfy the following set of dierential equations also called transmission line equations I (t, z ) V (t, z ) = L0 z t I (t, z ) V (t, z ) = C 0 z t (2.194) (2.195)

Substitution of these equations into each other results in wave equations for either the voltage or the current 1 2 V (t, z ) 2 V (t, z ) = 0, z 2 c2 t2 1 2 I (t, z ) 2 I (t, z ) = 0, z 2 c2 t2 (2.196) (2.197)

where c = 1/ L0 C 0 is the speed of wave propagation on the transmission line. The ratio between voltage pand current for monochomatic waves is the characteristic impedance Z = L0 /C 0 . The equations of motion for the electric and magnetic eld of a x-polarized TEM wave according to Figure 2.1, with E eld along the x-axis and H elds along the y- axis follow directly from Faradays and Amperes law H (t, z ) E (t, z ) = , z t H (t, z ) E (t, z ) = , z t (2.198) (2.199)

which are identical to the transmission line equations (2.194) and (2.195). Substitution of these equations into each other results again in wave equations



For a homogeneous medium or transmission line the equations of the forward and backward traveling waves are uncoupled. Then the dynamics is very simple. Note, we introduced that cross section Aef f such that |a|2 is proportional to the total power carried by the wave. Clearly, the solutions are a(t, z ) = f (t z/c0 ), b(t, z ) = g(t + z/c0 ), (2.204) (2.205)

for electric and magnetic elds propagating pat the speed of light c = 1/ and with characteristic impedance ZF = /. Thus the tansmission line equations describe the same phenomena than a one dimensional TEM-wave. The solutions of the wave equation are forward and backward traveling waves, which can be decoupled by transforming the elds to the forward and backward traveling waves r 1 Aef f a(t, z ) = (E (t, z ) + ZF H (t, z )) , (2.200) 2 ZF r 1 Aef f (E (t, z ) ZF H (t, z )) , (2.201) b(t, z ) = 2 ZF which fulll the equations 1 + a(t, z ) = 0, (2.202) z c t 1 b(t, z ) = 0. (2.203) z c t

which resembles the DAlembert solutions of the wave equations for the electric and magnetic eld s ZF o E (t, z ) = (a(t, z ) + b(t, z )) , (2.206) Aef f s 1 (a(t, z ) b(t, z )) . (2.207) H (t, z ) = 2ZF o Aef f Here, the forward and backward propagating elds are already normalized such that the Poynting vector multiplied with the eective area gives already the total power transported by the elds in the eective cross section Aef f P = S (Aef f ez ) = Aef f E (t, z )H (t, z ) = |a(t, z )|2 |b(t, z )|2 . (2.208)



along the transmission line or corresponding electric and magnetic elds in one dimensional wave propagation is described by a generalized complex impedance Z (z ) that obeys certain transformation rules, see Figure 2.47 (a).

In 6.013, it was shown that the relation between sinusoidal current and voltage waves V (t, z ) = Re V (z )ejt and I (t, z ) = Re I (z )ejt (2.209)

Figure 2.47: (a) Transformation of generalized impedance along transmission lines, (b) Transformation of generalized impedance accross free space sections with dierent characterisitc wave impedances in each section. Along the rst transmission line, which is terminated by a load impedance, the generalized impedance transforms according to Z 1 (z ) = Z 1 Z 0 jZ 1 tan (k1 z ) Z 1 jZ 0 tan (k1 z ) (2.210)

with k1 = k0 n1 and along the second transmission line the same rule applies as an example Z (L1 ) jZ 2 tan (k2 z ) Z 2 (z ) = Z 2 1 (2.211) Z 2 jZ 1 (L1 ) tan (k2 z )



with k2 = k0 n2 . Note, that the media can also be lossy, then the characteristic impedances of the transmission lines and the propagation constants are already themselves complex numbers. The same formalism can be used to solve corresponding one dimensional EM-wave propagation problems. Antireection Coating The task of an antireection (AR-)coating, analogous to load matching in transmission line theory, is to avoid reections between the interface of two media with dierent optical properties. One method of course could be to place the interface at Brewsters angle. However, this is not always possible. Lets assume we want to put a medium with index n into a beam under normal incidence, without having reections on the air/medium interface. The medium can be for example a lens. This is exactly the situation shown in Figure 2.47 (b). Z 2 describes the refractive index of the lense material, e.g. n2 = 3.5 for a silicon lense, we can deposit on the lens a thin layer of material with index n1 corresponding to Z 1 and this layer should match to the free space index n0 = 1 or impedance Z 0 = 377. Using (2.210) we obtain Z jZ 1 tan (k1 L1 ) (2.212) Z 2 = Z 1 (L1 ) = Z 1 0 Z 1 jZ 0 tan (k1 L1 ) If we choose a quarter wave thick matching layer k1 L1 = /2, this simplies to the important result Z2 = or n1 Z2 Z Z 1 , with Z 1 = 0 and Z 2 = 0 Z0 n1 n2 = n2 n0 and L1 = . 4n1 (2.213) (2.214)

Thus a quarter wave AR-coating needs a material which has an index corresponding to the geometric mean of the two media to be matched. In the current example this would be n2 = 3.5 1.87


Scattering and Transfer Matrix

Another formalism to analyze optical systems (or microwave circuits) can be formulated using the forward and backward propagating waves, which transform much simpler along a homogenous transmission line than the total



elds, i.e. the sum of forward and backward waves. However, at interfaces scattering of these waves occurs whereas the total elds are continuous. For monochromatic forward and backward propagating waves a(t, z ) = a(z )ejt and b(t, z ) = b(z )ejt (2.215)

propagating in z -direction over a distance z with a propagation constant k, we nd from Eqs.(2.202) and (2.203) a(z ) b(z ) = ejkz 0 0 ejkz a(0) b(0) . (2.216)

A piece of transmission line can be described as a two port. The matrix transforming the amplitudes of the waves at the input port (1) to those of the output port (2) is called the transfer matrix, see Figure 2.48

Figure 2.48: Denition of the wave amplitudes for the transfer matrix T.

For example, from Eq.(2.216) follows that the transfer matrix for free space propagation is a2 b2 =T a1 b1 , with T = ejkz 0 0 ejkz . (2.217)

This formalism can be expanded to arbitrary multiports, see Figure 2.49. The scattering matrix describes then the transformation from incoming and outgoing wave amplitudes of the multiport.



Figure 2.49: Scattering matrix and its port denition. The scattering matrix denes a linear transformation from the incoming to the outgoing waves b = Sa, with a = (a1, a2 , ...)T , b = (b1, b2 , ...)T . (2.218)

Note, that the meaning between forward and backward waves no longer coincides with a and b, a connection, which is dicult to maintain if several ports come in from many dierent directions. The transfer matrix T has advantages, if many two ports are connected in series with each other. Then the total transfer matrix is the product of the individual transfer matrices.


Properties of the Scattering Matrix

Physial properties of the system reect itself in the mathematical properties of the scattering matrix. Reciprocity A system with constant scalar dielectric and magnetic properties must have a symmetric scattering matrix (without proof) S = ST . (2.219)



Losslessness In a lossless system the total power owing into the system must be equal to the power owing out of the system in steady state 2 |a|2 = b . (2.220)


S+ S = 1 or S1 =S+ . The scattering matrix of a lossless system must be unitary. Time Reversal


To nd the scattering matrix of the time reversed system, we realize that incoming waves become outgoing waves under time reversal and the other way around, i.e. the meaning of a and b is exchanged and on top of it the waves become negative frequency waves. aej (tkz)
time reversal

aej (tkz) .


To obtain the complex amplitude of the corresponding positive frequency wave, we need to take the complex conjugate value. So to obtain the equations for the time reversed system we have to perform the following substitutions Original system Time reversed system . (2.223) a = Sb b = S1 a b = Sa



As an example, we look at the scattering matrix for a partially transmitting mirror, which could be simply formed by the interface between two media with dierent refractive index, which we analyzed in the previous section, see Figure 2.50. (Note, for brevity we neglect the reections at the normal surface input to the media, or we put an AR-coating on them.) In principle, this device has four ports and should be described by a 4x4 matrix. However, most often only one of the waves is used at each port, as shown in Figure 2.50.



Figure 2.50: Port denitions for the beam splitter The scattering matrix is determined by b = Sa, with a = (a1, a2 )T , b = (b1 , b2 )T and S= (2.224)

r jt jt r

, with r2 + t2 = 1.


The matrix S was obtained using the S -matrix properties described above. From Eqs.(2.122) we could immediately identify r as a function of the refractive indices, angle of incidence and the polarization used. Note, that the o-diagonal elements of S are identical, which is a consequence of reciprocity. That the main diagonal elements are identical is a consequence of unitarity for a lossless beamsplitter and furthermore t = 1 r2 . For a given frequency r and t can always be made real by choosing proper reference planes at the input and the output of the beam splitter. Beamsplitters can be made in many ways, see for example Figure 2.37.



Having a valid description of a beamsplitter at hand, we can build and analyze various types of interferometers, see Figure 2.51.



Figure 2.51: Dierent types of interferometers: (a) Mach-Zehnder Interferometer; (b) Michelson Interferometer; (c) Sagnac Interferometer [6] p. 66.

Each of these structures has advantages and disadvantages depending on the technology they are realized. The interferometer in Figure 2.51 (a) is called Mach-Zehnder interferometer, the one in Figure 2.51 (b) is called Michelson Interferometer. In the Sagnac interferometer , Figure 2.51 (c) both beams see identical beam path and therefore errors in the beam path can be balance out and only dierential changes due to external inuences lead to an output signal, for example rotation, see problem set 3. To understand the light transmission through an interferometer we analyze as an example the Mach-Zehnder interferometer shown in Figure 2.52. If we excite input port 1 with a wave with complex amplitude a0 and no input at port 2 and assume 50/50 beamsplitters, the rst beam splitter will



2 1

Figure 2.52: Mach-Zehnder Interferometer produce two waves with complex amplitudes
1 a b1 = 2 0 1 b2 = j 2 a0


During propagation through the interferometer arms, both waves pick up a phase delay 1 = kL1 and 2 = kL2 , respectively
1 a ej1 , a3 = 2 0 1 a ej2 . a4 = j 2 0


After the second beam splitter with the same scattering matrix as the rst one, we obtain j 1 ej2 , b3 = 1 a e 0 2 (2.228) a ej1 + ej2 . b4 = j 1 2 0 The transmitted power to the output ports is |b3 |2 = |b4 |2 =
|a0 |2 4 |a0 |2 4

The total output power is equal to the input power, as it must be for a lossless system. However, depending on the phase dierence = 1 2 between both arms, the power is split dierently between the two output ports, see Figure 2.53.With proper biasing, i.e. 1 2 = /2 + 0 , the dierence

1 ej (1 2 ) 2 = 1 + ej (1 2 ) 2 =

|a0 |2 2 |a0 |2 2

[1 cos (1 2 )] , [1 + cos (1 2 )] .




1.0 0.8 Output power 0.6 0.4 0.2 0.0 -3 -2 -1 0 1 2 3 |b3|


Figure 2.53: Output power from the two arms of an interfereometer as a function of phase dierence.

in output power between the two arms can be made directly proportional to the phase dierence 0 for 0 1. Opening up the beam size in the interferometer and placing optics into the beam enables to visualize beam distortions due to imperfect optical components, see Figures 2.54 and 2.55.

Figure 2.54: Twyman-Green Interferometer to test optics quality [1] p. 324.



Figure 2.55: Interference pattern with a hot iron placed in one arm of the interferometer ([1], p. 395).


Multipath Interference and Fabry-Perot Resonator

Interferometers can act as lters. The phase dierence between the interferometer arms depends on frequency, therefore, the transmission from input to output depends on frequency, see Figure 2.53. However, the lter function is not very sharp. The reason for this is that only a two beam interference is used. Much more narrowband lters can be constructed by multipath interferences such as in a Fabry-Perot Resonator, see Figure 2.56. To indicate again the these wave amplitudes are in the Fourier Domain, we added the tilde. The simplest Fabry Perot is described by a sequence of three layers where at least the middle layer has an index dierent from the other two layers, such that reections occur on these interfaces.



~ ~ n1

~ ~ 1 n2

~ ~ 2 ~ ~ ~ ~ S2 n3

~ ~

~ ~ S1

~ ~

Figure 2.56: Multiple intereferences in a Fabry Perot resonator. In the simplest implementation a Fabry Perot only consists of a sequence of three layers with dierent refractive index so that two reections occur with multiple interferences. Each of this discontinuites can be described by a scattering matrix. Any kind of device that has reections at two parallel interfaces may act as a Fabry Perot such as two semitransparent mirrors. A thin layer of material against air can act as a Fabry-Perot and is often called etalon. Given the reection and transmission coecients at the interfaces 1 and 2, we can write down the scattering matrices for both interfaces according to Eqs.(2.224) and (2.225). r1 jt1 a 1 r2 jt2 a 3 b1 b3 = and = . jt1 r1 a 2 jt2 r2 a 4 b2 b4 (2.230) If we excite the Fabry-Perot with a wave from the right with amplitude a 1 6= 0, then a fraction of that wave will be transmitted to the interface into the Fabry-Perot as wave b2 and part will be already reected into b1 ,
(0) 1 . b1 = r1 a


The transmitted wave will then propagate and pick up a phase factor ej/2 , with = 2k2 L and k2 = 2 n2 , 1 ej/2 . a 3 =jta (2.232)



After propagation it will be reected o from the second interface which has a reection coecient b3 2 = = r2 . (2.233) a 3
a4 =0

Then the reected wave b3 propagates back to interface 1, picking up another (1) 2 = phase factor ej/2 resulting in an incoming wave after one roundtrip of a 1 . Upon reection on interface 1, part of this wave is transmitted jt1 r2 ej a leading to an output (1) 1 . (2.234) b1 =jt1 jt1 r2 ej a The partial wave a2 is reected again and after another roundtrip it arrives (2) 1 . Part of this wave is transat interface 1 as a 2 = (r1 r2 ) ej jt1 r2 ej a mitted and part of it is reected back to go through another cycle. Thus in total if we sum up all partial waves that contribute to the output at port 1 of the Fabry-Perot lter, we obtain b1 = =
X (n) b1 X (1)

= r1 t2 1 r2 =


j r1 t2 1 r2 e

1 r1 r2 ej a a 1 (2.235)

r1 r2 ej a 1 r1 r2 ej 1

e 1 r1 r2 ej

n=0 j

Note, that the coecient in front of Eq.(2.235) is the coecient S11 of the scattering matrix of the Fabry-Perot. In a similar manner, we obtain b1 a 1 =S (2.236) a 4 b4 and 1 S= 1 r1 r2 ej r1 r2 ej t1 t2 ej/2 t1 t2 ej/2 r2 r1 ej (2.237)

In the following, we want to analyze the properties of the Fabry-Perot for the case of symmetric reectors, i.e. r1 = r2 and t1 = t2 . Then we obtain for



the power transmission coecient of the Fabry-Perot, |S21 |2 in terms of the power reectivity of the interfaces R = r2 1 R 2 (1 R)2 2 = |S21 | = (2.238) 1 Rej (1 R)2 + 4R sin2 (/2) Figure 2.57 shows the transmission |S21 |2 of the Fabry-Perot interferometer for equal reectivities |r1 |2 = |r2 |2 = R.
Fabry-Perot Transmission, |S21| 1.0 FSR 0.8 0.6 0.4 0.2 R=0.9 0.0 0.0 0.5 1.0 1.5 2.0 (f -fm)/ FSR 2.5 3.0 R=0.1


R=0.5 R=0.7

Figure 2.57: Transmission of a lossless Fabry-Perot interferometer with |r1 |2 = |r2 |2 = R At very low reectivity R of the mirror the transmission is almost everywhere 1, there is only a slight sinusoidal modulation due to the rst order interferences which are periodically in phase and out of phase, leading to 100% transmission or small reection. For large reectivity R, due to the then multiple interference operation of the Fabry-Perot Interferometer, very narrow transmission resonances emerge at frequencies, where the roundtrip phase in the resonator is equal to a multiple of 2 = 2f n2 2L = 2m, c0 (2.239)

2.4. INTERFERENCE AND INTERFEROMETERS which occurs at a comb of frequencies, see Figure 2.58 c0 . fm = m 2n2 L
1.0 0.8 0.6 0.4 0.2 0.0 fm-1 1 2fm fm f3 m+1 fm+2 4 frequency 5 am-1 am am+1 am+2



Figure 2.58: Developement of a set of discrete resonances in a dimensionsal resonator.

Fabry-Perot Transmission, |S21|


On a large frequency scale, a set of discrete frequencies, resonances or modes arise. The frequency range between resonances is called free spectral range (FSR) of the Fabry-Perot Interferometer F SR = c0 1 = , 2n2 L TR (2.241)

which is the inverse roundtrip time TR of the light in the one-dimensonal cavity or resonator formed by the mirrors. The lter characteristic of each resonance can be approximately described by a Lorentzian line derived from Eq.(2.238) by substituting f = fm + f with f F SR, |S21 |2 (1 R)2 f (1 R)2 + 4R sin2 m2 + 2 F /2 SR 1 2 , 2 R f 1 + 1R F SR = 1+ 1
f fF W HM /2


2 ,




where we introduced the FWHM of the transmission lter fF W HM = F SR , F (2.244)

with the nesse of the interferometer dened as R F = . 1R T


The last simplication is valid for a highly reecting mirror R 1 and T is the mirror transmission. From this relation it is immediately clear that the nesse has the additional physical meaning of the optical power enhancement inside the Fabry-Perot at resonance besides the factor of , since the power inside the cavity must be larger by 1/T , if the transmission through the Fabry-Perot is unity.


Quality Factor of Fabry-Perot Resonances

Another quantity often used to characterize a resonator or a resonance is its quality factor Q, which is dened as the ratio between the resonance frequency and the decay rate for the energy stored in the resonator, which is 1 also often called inverse photon lifetime, ph Q = ph fm . (2.246)

Lets assume, energy is stored in one of the resonator modes which occupies a range of frequencies [fm F SR/2, fm + F SR/2] as indicated in Figure 2.59. Then the fourier integral am (t) = Z
+F SR/2

b2 (f )ej 2(f fm )t df,


F SR/2

2 b2 (f ) is normalized such that it describes the power spectral density where of the forward traveling wave in the resonator gives the mode amplitude of the m-th mode and its magnitude square is the energy stored in the mode. Note, that we could have taken any of the internal waves a 2 , b2 , a 3 , and b3 . The time dependent eld we create corresponds to the eld of the forward or backward traveling wave at the corresponding reference plane in the resonator.


1.0 0.8 0.6 0.4 0.2 0.0 0.0 0.5 1.0 1.5 2.0 (f -fm)/ FSR 2.5 3.0 am-1(t) am(t) am+1(t)


Figure 2.59: Integration over all frequency components within the frequency range [fm F SR/2, fm + F SR/2] denes a mode amplitude a(t) with a slow time dependence We now make a "Gedanken-Experiment". We switch on the incoming waves a 1 ( ) and a 4 ( ) to load the cavity with energy and evaluate the in ternal wave b2 (). Instead of summing up all the multiple reections like we did in constructing the scattering matrix (2.236), we exploit our skills in analyzing feedback systems, which the Fabry-Perot lter is. The scattering equations set force by the two scattering matrices characterizing the resonator mirrors in the Fabry-Perot can be visuallized by the signal ow diagram in Figure 2.60

Transmission, |S21|

~ a _ 1 r ~ b _ 1 +

jt S1 jt

+ r

~ b2 _


~ a _ 3

jt r S2 jt

+ r

~ b4 _

~ a _ 2


~ b _ 3

~ a _ 4

Figure 2.60: Representation of Fabry-Perot resonator by a signal ow diagram



For the task to nd the relationship between the internal waves feed by the incoming wave only the dashed part of the signal ow is important. The internal feedback loop can be clearly recognized with a closed loop transfer function r2 ej , which leads to the resonance denominator 1 r2 ej in every element of the Fabry-Perot scattering matrix (2.236). Using Blacks formula from 6.003 and the superposition principle we immediately nd for the internal wave b2 = jt j/2 + re a a . 1 4 1 r2 ej (2.248)

Close to one of the resonance frequencies, = 2fm + , using t = 1 r2 , the round-trip phase can be approximated by = TR = 2m + TR , and the exponential of the round-trip phase by ej/2 = (1)m ejTR /2 , ej = ejTR = 1 jTR . Using those approximations in (2.248) and t = 1 R we obtain 1R j 4 ( ) , b2 ( ) a 1 () + r(1)m ejTR /2 a 1 R(1 jTR ) 1 j m jTR /2 ( ) + r ( 1) e a ( ) a 1 4 R 1 + j 1 TR T R 1 j a 4 ( ) 1 () + r(1)m ejTR /2 a 1 + jTR /T T j (1 + jTR /T ) b2 ( ) = a 4 ( ) 1 ( ) + r(1)m ejTR /2 a T (2.250) (2.251) (2.249)

(2.252) (2.253) (2.254)

for high reectivity R. Multiplication of this equation with the resonant denominator (2.255)



and inverse Fourier-Transform in the time domain, while recognizing that the internal elds vanish far o resonance, i.e. Z +F SR Z + jt am (t) = (2.256) b2 ( )e d = b2 ( )ejt d,

we obtain the following dierential equation for the mode amplitude slowly varying in time TR d am (t) = T am (t) + j T (a1 (t) + (1)m a4 (t TR /2)) dt Z
+ F SR


with the input elds a1/4 (t) = a 1/4 ( )ejt d. (2.258)


Despite the pain to derive this equation the physical interpretation is remarkably simple and far reaching as we will see when we apply this equation later to many dierent situations. Lets assume, we switch o the loading of the cavity at some point, i.e. a1/4 (t) = 0, then Eq.(2.257) results in am (t) = am (0)et/(TR /T ) And the power decays accordingly |am (t)|2 = |am (0)|2 et/(TR /2T ) (2.260) (2.259)

twice as fast as the amplitude. The energy decay time of the cavity is often called the cavity energy decay time, or photon lifetime, ph , which is here ph = TR . 2T

Note, the factor of two comes from the fact that each mirror of the FabryPerot has a transmission T per round-trip time. Thus the total transmission or loss form the cavity is 2T . For exampl a L = 1.5m long cavity with mirrors of 0.5% transmission, i.e. TR = 10ns and 2T = 0.01 has a photon lifetime of 1s. It needs hundred bounces on the mirror for a photon to be essentially lost from the cavity. Highest quality dielectric mirrors may have a reection loss of only 105...6 , this is not really transmission but rather scattering loss in the mirror. Such

100 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS high reectivity mirrors may lead to the construction of cavities with photon lifetimes on the order of milliseconds. Now, that we have an expression for the energy decay time in the cavity, we can evaluate the quality factor of the resonator Q = fm ph = m . 2T (2.261)

Again for a resonator with the same parameters as before and at optical frequencies of 300THz corresponding to 1m wavelength, we obtain Q = 2 108 .


Thin-Film Filters

Transfer matrix formlism is an ecient method to analyze the reection and transmission properties of layered dielectric media, such as the one shown in Figure 2.61. Using the transfer matrix method, it is an easy task to compute the transmission and reection coecients of a structure composed of layers with arbitrary indices and thicknesses. A prominent example of a thin-lm lter are Bragg mirrors. These are made of a periodic arrangement of two layers with low and high index n1 and n2 , respectively. For maximum reection bandwidth, the layer thicknesses are chosen to be quarter wave for the wavelength maximum reection occures, n1 d1 =0 /4 and n2 d2 =0 /4

~ a _ 1 n1 n2 n1 n2 n1 n2 .... d1 d2 d1 d2 d1 d2 ~ b _ 1

~ b2 _

~ a _ 2

Figure 2.61: Thin-Film dielectric mirror composed of alternating high and low index layers. As an example Figure 2.62 shows the reection from a Bragg mirror with n1 = 1.45, n2 = 2.4 for a center wavelength of 0 = 800nm. The layer thicknesses are then d1 =134nm and d2 =83nm.













0 500




900 lambda





Figure 2.62: Reectivity of an 8 pair quarter wave Bragg mirror with n1 = 1.45 and n2 = 2.4 designed for a center wavelength of 800nm. The mirror is embedded in the same low index material.



Gaussian Beams and Resonators

So far, we have only treated optical systems operating with plane waves, which is an idealization. In reality plane waves are impossible to generate because of there innite amount of energy required to do so. The simplest (approximate) solution of Maxwells equations describing a beam of nite size is the Gaussian beam. In fact many optical systems are based on Gaussian beams. Most lasers are designed to generate a Gaussian beam as output. Gaussian beams stay Gaussian beams when propagating in free space. However, due to its nite size, diraction changes the size of the beam and lenses are imployed to reimage and change the cross section of the beam. In this section, we want to study the properties of Gaussian beams and how they can be used to understand the spatial modes of optical resonators.


Paraxial Wave Equation

We start from the Helmholtz Equation (2.17) 2 e + k0 E (x, y, z, ) = 0, (2.262)

with the free space wavenumber k0 = /c0 . This equation can easily be solved in the Fourier domain, and one set of solutions are of course the plane 2 waves with wave vector |k|2 = k0 . We look for solutions which are polarized in x-direction e e(x, y, z ) ex . E (x, y, z, ) = E (2.263) We want to construct a beam with nite transverse extent into the x-y-plane and which is mainly propagating into the positive z -direction. As such we may try a superposition of plane waves with a dominant z -component of the k-vector, see Figure 2.63. The k-vectors can be written as kz q 2 2 k2 , = k0 kx y 2 2 kx + ky . k0 1 2 2k0


with kx , ky << k0 .



y x k z

Figure 2.63: Construction of a paraxial beam by superimposing many plane waves with a dominante k-component in z-direction. Then we obtain for the propagating eld Z + Z + e e0 (kx , ky ) E (x, y, z ) = E 2 2 + ky kx exp jk0 1 z jkx x jky y dkx dky , 2 2k0 Z + Z + e0 (kx , ky ) = E 2 2 kx + ky z jkx x jky y dkx dky ejk0 z , (2.265) exp j 2k0 e0 (kx , ky ) is the amplitude for the waves with the corresponding transwhere E verse k-component. This function should only be nonzero within a small range kx , ky k0 . The function 2 Z + Z + 2 kx + ky e e z jkx x jky y dkx dky E0 (x, y, z ) = E0 (kx , ky ) exp j 2k0 (2.266) is a slowly varying function in the transverse directions x and y , and it can be easily veried that it fullls the paraxial wave equation 2 e j 2 e0 (x, y, z ). E E0 (x, y, z ) = + (2.267) z 2k0 x2 x2

Note, that this equation is in its structure identical to the dispersive spreading of an optical pulse. The dierence is that this spreading occurs now in the two transverse dimensions and is called diraction.



Gaussian Beams

The Gaussian beam is often formulated in terms of the complex beam parameter or q -parameter. The propagation of the beam in free space and later even through optical imaging systems can be eciently described by a proper transformation of the q -parameter 2 r 1 e0 (r, z ) exp jk0 . (2.271) E q(z ) 2q (z ) Free space propagation is then described by q(z ) = z + jzR (2.272)

Eq.(2.266) can be rewritten as 2 Z + Z + 2 kx + ky e E0 (x, y, z ) (z + jzR ) jkx x jky y dkx dky , exp j 2k0 (2.269) 2 with the parameter zR = k0 /kT , which we will later identify as the Rayleigh range. Thus, Gaussian beam solutions with dierent nite transverse width in k-space and real space behave as if they propagate along the z-axis with dierent imaginary z-component zR . Carrying out the Fourier transformation results in the Gaussian Beam in real space 2 2 + y x j e0 (x, y, z ) . (2.270) exp jk0 E z + jzR 2(z + jzR )

Since the kernel in Eq.(2.266) is quadratic in the transverse k-components using a two-dimensional Gaussian for the amplitude distribution leads to a beam in real space which is also Gaussian in the radial direction because of the resulting Gaussian integral. By choosing for the transverse amplitude distribution 2 2 kx + ky e0 (kx , ky ) exp , (2.268) E 2 2kT

Using the inverse q -parameter, decomposed in real and imagniary parts, 1 1 = j 2 . q (z ) R(z ) w (z ) (2.273)

2.5. GAUSSIAN BEAMS AND RESONATORS leads to p 2 2 4 Z P r r F 0 e0 (r, z ) = exp 2 E jk0 + j (z ) . w (z ) 2R(z ) w(z )



Thus w(z ) is the waist of the beam and R(z ) is the radius of the phase fronts. We the beam such that the Gaussian beam intensity normalized 2 e I (z, r) = E0 (r, z ) /(2ZF0 ) expressed in terms of the power P carried by the beam is given by 2r2 2P exp 2 , I (r, z ) = w2 (z ) w (z ) Z Z 2 I (r, z ) rdr d. i.e. P =
0 0

(2.275) (2.276)

The use of the q-parameter simplies the description of Gaussian beam propagation. In free space propagation from z1 to z2 , the variation of the beam parameter q is simply governed by q2 = q1 + z2 z1 . (2.277)

where q2 and q1 are the beam parameters at z1 and z2 . If the beam waist, at which the beam has a minimum spot size w0 and a planar wavefront (R = ), is located at z = 0, the variations of the beam spot size and the radius of curvature of the phase fronts are explicitly expressed as " 2 #1/2 z w(z ) = wo 1 + , (2.278) zR and z 2 R , R(z ) = z 1 + z (2.279)

where zR is called the Rayleigh range. The Rayleigh range is the distance over which the cross section of the beam doubles. The Rayleigh range is determined by the beam waist and the wavelength of light according to zR =
2 wo .


106 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS Intensity Figure 2.64 shows the intensity of the Gaussian beam according to Eq.(2.275) for dierent propagation distances.

Figure 2.64: The normalized beam intensity I/I0 as a function of the radial distance r at dierent axial distances: (a) z=0, (b) z=zR , (c) z=2zR . The beam intensity can be rewritten as 2 w0 2P 2r2 I (r, z ) = I0 2 . exp 2 , with I0 = 2 w (z ) w (z ) w0


For z > zR the beam radius growth linearly and therefore the area expands quadratically, which brings down the peak intensity quadratically with propagation distance. On the beam axis (r = 0) the intensity is given by I (r, z ) = I0
2 w0 = w2 (z )


The normalized beam intensity as a function of propagation distance is shown in Figure 2.65

I0 2 .
z zR




Figure 2.65: The normalized Beam intensity I (r = 0)/I0 on the beam axis as a function of propagation distance z [6], p. 84.

Power The fraction of the total power contained in the beam up to a certain radius is Z 2 r0 P (r < r0 ) = I (r, z )rdr P P 0 Z r0 2r2 4 rdr (2.283) exp 2 = w2 (z ) 0 w (z ) 2 2r0 = 1 exp 2 . w (z ) Thus, there is a certain fraction of power within a certain radius of the beam P (r < w(z )) = 0.86, P P (r < 1.5w(z )) = 0.99. P Beam radius Due to diraction, the smaller the spot size at the beam waist, the faster the beam diverges according to 2.278 as illustrated in Figure ??. (2.284) (2.285)


Planes of const. Phase

Beam Waist

z/z R

Figure 2.66: Gaussian beam and its characteristics. Beam divergence The angular divergence of the beam is inversely proportional to the beam waist. In the far eld, the half angle divergence is given by = see Figure 2.66. Confocal parameter and depth of focus In linear microscopy, only a layer which has the thickness over which the beam is focused, called depth of focus, will contribute to a sharp image. In nonlinear microscopy (see problem set) only a volume on the order of beam cross section times depth of focus contributes to the signal. Therefore, the depth of focus or confocal parameter of the Gaussian beam, is the distance over which the beam stays focused and is dened as twice the Rayleigh range b = 2zR =
2 2wo .

, wo



The confocal parameter depends linear on the spot size (area) of the beam and is inverse to the wavelength of light. At a wavelength of 1m a beam with a radius of wo = 1cm,.the beam will stay focussed ove distances as long



600m. However, if the beam is stronlgy focussed down to wo = 10m the eld of depth is only 600m.

Phase The phase delay of the Gaussian beam is r2 (r, z ) = k0 z (z ) + k0 2R(z ) z (z ) = arctan . zR (2.288) (2.289)

On beam axis, there is the additional phase (z ) when the beam undergoes focussing as shown in Figure 2.67. This is in addition to the phase shift that a uniform plane wave already aquires.

Figure 2.67: Phase delay of a Gaussian beam relative to a uniform plane wave on the beam axis [6], p. 87. This phase shift is known as Guoy-Phase-Shift.

This eect is known as Guoy-Phase-Shift. The third term in the phase shift is parabolic in the radius and describes the wavefront (planes of constant phase) bending due to the focusing, i.e. distortion from the uniform plane wave.


Figure 2.68: The radius of curvature R(z ) of the wavefronts of a Gaussian beam [6], p. 89.

r The surfaces of constant phase are determined by k0 z (z ) + k0 2R = (z ) const. Since the radius of curvature R(z ) and the additional phase (z ) are slowly varying functions of z, i.e. they are constant over the radial variation of the wavefront, the wavefronts are paraboloidal surfaces with radius R(z ), see Figures 2.68 and 2.69.

Figure 2.69: Wavefronts of a Gaussian beam, [6] p. 88.

For comparison, Figure 2.70 shows the wavefront of (a) a uniform plane wave, (b) a spherical wave and (c) a Gaussian beam. At points near the beam center, the Gaussian beam resembles a plane wave. At large z, the beam behaves like a spherical wave except that the phase fronts are delayed by a quarter of the wavlength due to the Guoy-Phase-Shift.



Figure 2.70: Wavefronts of (a) a uniform plane wave;(b) a spherical wave; (c) a Gaussian beam [5], p. 89.


Optical Resonators

With the Gaussian beam solutions, we can nally construct optical resonators with nite transverse extent, i.e. real Fabry-Perots. To do that we insert into the Gaussian beam, see Figure 2.71, curved mirrors with the proper radius of curvature, such that the beam is imaged upon itself.


L R1 R2



Figure 2.71: Fabry-Perot resonator with nite beam cross section by inserting curved mirrors into the beam to back reect the beam onto itself. Curved-Flat Mirror Resonator We rst consider the simple resonator constructed by a curved and at mirror, i.e. only the left side of the gaussian beam in Figure 2.71, see Figure 2.72


z1 = L
Figure 2.72: Curved-Flat Mirror Resonator

From Eqs.(2.278) and (2.279) we immediately get an expression for the radius of curvature of the mirror necessary to generate a certain confocal 2 o parameter zR = w or spot size on the at mirror given a certain wavelength w1 = wo 1 + " L zR 2 #1/2 , (2.290)




Since the confocal parameter is a positive quantity it follows that we always need a mirror with a radius of curvature R1 > L, to form a stable gaussian mode in the resonator. The spot size on the at and curved mirrors in terms of the radius of curvature of the mirror and the distance of the mirror is s r R1 4 L L 1 , (2.292) wo = R1 R1 and w1 = r v u L R1 u R1 4 t . L 1 R 1


There is a resonator mode with nite size for 0 < L < R1 , i.e. the cavity is stable in this parameter range.. Figure 2.73 shows the normalized beam radii on both mirrors as a function of the normalized distance between the L mirrors R . 1

3 w1/sqrt(R1/) Beam Radii 2

1 w0/sqrt(R1/) 0 0.0 0.2 0.4 L/R1 0.6 0.8 1.0

Figure 2.73: Beam waists of the curved-at mirror resonator as a function of the cavity parameter g1 .

114 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS Note, that for a given curved mirror with radius of curvature (ROC ) = R1 , the waist of the beam goes to zero for both extremes of the distance between the mirrors where the cavity is stable. There is a large central range around L R1 /2, where the beam waist is maximum and insensitive to the exact distance between the mirrors. Two-Curved Mirror Resonator Now, we can analyze what happens for the case of a resonator with two curved mirrors as shown in Figure 2.71. The gaussian beam formulas Eqs.(2.278) and (2.279) give us now the following conditions for beam waists w1 = wo 1 + " " z1 zR z2 zR 2 #1/2 2 #1/2 2 # 2 # , , (2.294)

w2 = wo 1 + " "


R1 = z1 1 +

zR z1 zR z2


R2 = z2 1 + and


L = z1 + z2 .


The last three equations determine the confocal parameter zR and the distances between location of waist and mirrors z1 and z2 in dependence on the radii of curvature R1 , R2 and the distance between mirrors L. It is straight forward to eliminate the confocal parameter from Eqs.(2.296), (2.297) and to solve the remaining equation together with Eq.(2.298) for the distances z1 and z2 . The result is L(R2 L) z1 = , (2.299) R1 + R2 2L and by symmetry z2 = L(R1 L) = L z1 . R1 + R2 2L (2.300)



Resubstitution into Eq.(2.296) leads to an expression for the Rayleigh range of the beam and its waist

2 zR =L

(R1 L)(R2 L)(R1 + R2 L) . (R1 + R2 2L)2


and therefore 2 L (R1 L)(R2 L)(R1 + R2 L) = (2.302) L(R1 + R2 2L)2 2 L L L L L L L2 (1 R1 )(1 R2 )( R1 + R2 R1 R2 ) R1 R2 = . (2.303) L L L L 2 R1 R2 (R +R 2 R ) 1 2 1 R2

4 wo

For completeness the spot size on the mirrors is given by 2 L R1 R2 L = R1 L R1 + R2 L 2 L L L 1 R R1 R2 R1 R2 2 = . L L L L L 1 R + R R R R 1 1 2 1 2

4 w1

(2.304) (2.305)

By symmetry, we nd the spot size on mirror 2 by switching index 1 and 2: 2 R1 L R2 L = R2 L R1 + R2 L 2 L L L 1 R R1 R2 R1 R2 1 = . L L L L L 1 R +R R 2 R1 2 1 R2

4 w2

(2.306) (2.307)

The mode characteristics w0 , w1 , w2 , z1 and z2 for the two-mirror resonator are shown in Figure 2.74 for the case R1 = 10cm and R2 = 11cm as a function of the mirror distance L



/ ) w 0 / ( (R 1 *R 2 )

0.6 0.4 0.2 0.0 0 6 4 2 0 0 6 4 2 0 1.0 0.8 0.6 0.4 0.2 0.0 0 5 10 Cavity Length, L / cm 15 20 1.0 0.8 0.6 0.4 0.2 0.0 5 10 Cavity Length, L / cm 15 20 5 10 Cavity Length, L / cm 15 20

w 2 / ( R 2 / )


w 1 / ( R 1/ )


z2 / L

z1 / L

1.0 0 0.5 0.0 -0.5 -1.0 0

10 Cavity Length, L / cm



1.0 0.8 0.6 0.4 0.2 0.0


g 1, g


10 Cavity Length, L / cm



Figure 2.74: From top to bottom: two-mirror resonator mode characteristics w0 , w1 , w2 , z1 , z2 and cavity parameters, g1 , g2 , S for the case R1 = 10 cm and R2 = 11 cm.



Resonator stability As we can read of from Figure 2.74 for a two-mirror resonator with concave mirrors and R1 R2 , we obtain the general stability diagram as shown in Figure 2.75.

Figure 2.75: Stabile regions (black) for the two-mirror resonator.

There are two ranges for the mirror distance L, within which the cavity is stable, 0 L R1 and R2 L R1 + R2 .The stable and unstable parameter ranges can be expressed in a compact way in terms of the cavity parameters g1 and g2 dened by gi = (Ri L)/Ri , for i = 1, 2.These cavity parameters together and its product S = g1 g2 are shown in the last graph of Figure 2.74. The cavity mode has nite size, i.e. it is stable for

stable : 0 g1 g2 = S 1 and unstable for


unstable : g1 g2 0; or g1 g2 1.


The stability criterion can be easily interpreted geometrically. Of importance are the distances between the mirror mid-points Mi and the cavity end points, i.e. gi = (Ri L)/Ri = Si /Ri , as shown in Figure 2.76.


Figure 2.76: The stability criterion involves distances between the mirror mid-points Mi and the cavity end points. i.e. gi = (Ri L)/Ri = Si /Ri . The following rules for a stable resonator can be derived from Figure 2.76 using the stability criterion expressed in terms of the distances Si . Note, that without proof, the distances and radii can be positive and negative, where the latter indicate concave mirrors stable : 0 S1 S2 1. R1 R2 (2.310)

This results in the following rules for the geometry of stable two-mirror resonators:

A resonator is stable if the mirror radii, laid out along the optical axis, overlap. A resonator is unstable if the radii do not overlap or one lies within the other.

Figure 2.77 shows stable and unstable resonator congurations.



Figure 2.77: Illustration of stable and unstable resonator congurations.

Hermite-Gaussian-Beams (TEMmn -Beams) It turns out that the Gaussian Beams are not the only solution to the paraxial wave equation (2.267). The stable modes of the resonator reproduce themselves after one round-trip, " # 2x 2y w 0 e Gm Gn (2.311) Em,n (x, y, z ) = Am,n w(z ) w(z ) w(z ) 2 x + y2 + j (m + n + 1) (z ) exp jk0 2R(z ) where 2 u Gm [u] = Hm [u] exp , for m = 0, 1, 2, ... 2


120 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS are the Hermite-Gaussians with the Hermite-Polynomials

H0 [u] H1 [u] H2 [u] H3 [u]

= = = =

1, 2u, 4u2 1, 8u3 12u,


and (z ) is the Guoy-Phase-Shift according to Eq.(2.289). The lower order Hermite Gaussians are depicted in Figure 2.78

Figure 2.78: Hermite-Gauissians Gm (u) for m = 0, 1, 2 and 3.

and the intensity prole of the rst higher order resonator modes are shown in Figure 2.79.



Figure 2.79: Intensity prole of TEMmn -beams, [6], p. 103. Besides the dierent mode proles, the higher order modes experience greater phase advances during propogation, because they are made up of k-vectors with larger transverse components. Axial Mode Structure As we have seen for the Fabry-Perot resonator, the longitudinal modes are characterized by a roundtrip phase that is a multiple of 2. Back then, we did not consider transverse modes. Thus in a resonator with nite transverse beam size, we obtain an extended family of resonances, with distinguishable eld patterns. The resonance frequencies pmn are determined by the roundtrip phase condition plm = 2p, for p = 0, 1, 2, ... (2.314)

122 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS For the linear resonator according to Figure 2.71, the roundtrip phase of a Hermite-Gaussian Tplm -beam is plm = 2kL 2(l + m + 1) ( (z2 ) (z1 )) , (2.315)

where (z2 ) (z1 ) is the additional Guoy-Phase-Shift, when the beam goes through the focus once on its way from mirror 1 to mirror 2. Then the resonance circular frequencies are c plm = [p + (l + m + 1) ( (z2 ) (z1 ))] (2.316) L If the Guoy-Phase-Shift is not a rational number times , then all resonance frequencies are non degenerate. However, for the special case where the two mirrors have identical radius of curvature R and are spaced a distance L = R apart, which is called a confocal resonator, the Guoy-Phase-shift is (z2 ) (z1 ) = /2, with resonance frequencies 1 c p + (l + m + 1) . (2.317) fplm = 2L 2 In that case all even, i.e. l + m, transverse modes are degenerate to the longitudinal or fundamental modes, see Figure 2.80.

Figure 2.80: Resonance frequencies of the confocal Fabry-Perot resonator, [6], p. 128. The odd modes are half way inbetween the longitudinal modes. Note, in contrast to the plan parallel Fabry Perot all mode frequencies are shifted by half of the free spectral range, F SR = 2c , due to the Guoy-Phase-Shift. L




Waveguides and Integrated Optics

As with electronics, miniaturization and integration of optics is desired to reduce cost while increasing functionality and reliability. One essential element is the guiding of the optical radiation in waveguides for integrated optical devices and optical bers for long distance transmission. Waveguides can be as short as a few millimeters. Guiding of light with exceptionally low loss in ber (0.1dB/km) can be achieved by using total internal reection. Figure 2.81 shows dierent optical waveguides with a high index core material and low index cladding. The light will be guided in the high index core. Similar to the Gaussian beam the guided mode is made up of mostly paraxial plane waves that hit the high/low-index interface at grazing incidence and therefore undergo total internal reections. The concomittant lensing eect overcomes the diraction of the beam that would happen in free space and leads to stationary mode proles for the radiation. Depending on the index prole and geometry one distinguishes between dierent waveguide types. Figure 2.81 (a) is a planar slab waveguide, which guides light only in one direction. This case is analyzed in more detail, as it has simple analytical solutions that show all phenomena associated with waveguiding such as cuto, dispersion, single and multimode operation, coupling of modes and more, which are used later in devices and to achieve certain device properties. The other two cases show complete waveguiding in the transverse direction; (b) planar strip waveguide and (c) optical ber.

Saleh 239

Figure 2.81: Dark shaded area constitute the high index regions. (a) planar slab waveguide; (b) strip waveguide; (c) optical ber [6], p. 239. In integrated optics many components are fabricated on a single sub-

124 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS strate, see Figure 2.82 with fabrication processes similar to those in microelectronics.



Figure 2.82: (a) Schematic of an integrated polarization insensitive Optical Add-Drop Multiplexer (OADM). It integrates polarization splitters, rotators, and ring lter stages for each polarization to drop and add a corresponding wavelength channel for each polarization. The two polarziations are later recombined to form a single output into a ber. (b) The device utilizes SiN waveguides on silica using a dual mask lithography process. [11]. More advanced devices also contain modulators, lasers and photo detectors to perform complete transmitt and receive functions. But as the example in Fig. 2.82 shows, the most important passive component to understand in an integrated optical circuit are waveguides and couplers. Here, we treat the basics of these devices.


Planar Waveguides

To understand the basic physics and phenomena in waveguides, we look at a few examples of guiding in one transverse dimension. These simple cases can be treated analytically.



The planar mirror waveguide is composed of two ideal metal mirrors a distance d apart, see Figure 2.83

Figure 2.83: Planar mirror waveguide, [6], p. 240. We consider a TE-wave, whose electric eld is polarized in the y direction and that propagates in the z direction. The reections of the light at the ideal lossless mirrors will guide or conne the light in the xdirection. The eld will be homogenous in the y direction, i.e. will not depend on y . Therefore, we make the following trial solution for the electric eld of a monochromatic complex TE-wave E (x, z, t) = E y (x, z ) ejt ey . (2.318)

Note, this trial solution also satises the condition E = 0, see (2.6) Modes of the planar waveguide Furthermore, we are looking for solutions that do not change their eld distribution transverse to the direction of propagation and experience only a phase shift during propagation. We call such solutions modes of the waveguide, because they dont change its transverse eld prole. The modes of the above planar waveguide can be expressed as E y (x, z ) = u(x) ejz , (2.319) where is the propagation constant of the mode. This solution has to obey the Helmholtz Eq.(2.17) in the free space section between the mirrors 2 d2 2 2 2 u ( x ) = k = . u ( x ) with k dx2 c2 (2.320)

126 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS The presence of the metal mirrors requires that the electric eld vanishes at the metal mirrors, otherwise innitely strong currents would start to ow to shorten the electric eld. u(x = d/2) = 0 (2.321) .Note, that Eq.(2.320) is an eigenvalue problem to the dierential operator d2 u(x) = u(x) with u(x = d/2) = 0. (2.322) dx2 in a space of functions u, that satises the boundary conditions (2.321). The eigenvalues are for the moment arbitrary but constant numbers. Depending on the sign of the eigenvalues the solutions can be sine or cosine functions ( < 0) or exponentials with real exponents for ( > 0). In the latter case, it is impossible to satiesfy the boundary conditions. Therefore, the eigensolutions are q 2 cos (kx,m x) with , kx,m = m, m = 1, 3, 5, ..., even modes d qd um (x) = 2 sin (kx,m x) with , kx,m = d m, m = 2, 4, 6, ..., odd modes d (2.323) Propagation Constants The propagation constants for these modes follow from comparing (2.320) with (2.322) to be
2 2 2 m = k kx,m d2 dx2



m =

where = 0 /n(0 ) is the wavelength in the medium between the mirrors. This relationship is shown in Figure 2.84. The lowest order mode with index m = 1 has the smallest k -vector component in x-direction and therefore the largest k-vector component into z-direction. The sum of the squares of both components has to be identical to the magnitude sqaure of the k-vector in the medium k. Higher order modes have increasingly more nodes in the x-direction, i.e. largest kx -components and the wave vector component in z direction decreases, until there is no real solution anymore to Eq.(2.325) and the corresponding propagation constants m become imaginary. That is, the

s 2 2 2 2 m = m c2 d d




corresponding waves become evanescent waves, i..e they can not propagate in a waveguide with the given dimensions.
k 1 2 3 k

1 d

2 d

3 d

k 4 x d

Figure 2.84: Determination of propagation constants for modes Field Distribution The transverse electric eld distributions for the various TE-modes is shown in Figure 2.85

Figure 2.85: Field distributions of the TE-modes of the planar mirror waveguide [6], p. 244. Cuto Wavelength/Frequency For a given planar waveguide with separation d, there is a lowest frequency, i.e. longest wavelength, beyond which no propagating mode exists. This wavelenth/frequency is refered to as cuto

128 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS wavelength/frequency which is (2.326) cutof f = 2d c . (2.327) fcutof f = 2d The physical origin for the existence of a cuto wavelength or frequency is that the guided modes in the mirror waveguide are a superposition of two plane waves, that propagate under a certain angle towards the z-axis, see Figure 2.86

Figure 2.86: (a) Condition for self-consistency: as a wave reects twice it needs to be in phase with the previous wave. (b) The angles for which selfconsistency is achieved determine the x-component of the k-vectors involved. The corresponding two plane waves setup an interference pattern with an extended node at the position of the metal mirrors satisfying the boundary conditions, [6], p. 241. In order that the sum of the electric eld of the two plane waves fullls the boundary conditions, the phase of one of the plane waves after reection



on both mirrors needs to be inphase with the other plane wave, i.e. the x-component of the k-vectors involved, kx , must be a multiple of 2 2kx d = 2m. If we superimpose two plane waves with kx,m = m/d, we obtain an interference pattern which has nodes along the location of the metal mirrors, which obviously fullls the boundary conditions. It is clear that the minimum distance between these lines of nodes for waves of a given wavelength is /2, hence the separation d must be greater than /2 otherwise no solution is possible. Single-Mode Operation For a given separation d, there is a wavelength range over which only a single mode can propagate, we call this wavelength range single-mode operation. From Figure 2.84 it follows for the planar mirror waveguide <k< 2 d d or d < < 2d (2.329) (2.328)

Waveguide Dispersion Due to the waveguiding, the relationship between frequency and propagation constant is no longer linear. This does not imply that the waveguide core, i.e. here the medium between the plan parallel mirrors, has dispersion. For example, even for n = 1, we nd for phase and group velocity of the m-th mode r c 2 1 1 m ( ) = m = 1 (2.330) vp,m c d s 2 1 = 1 (2.331) m c 2d and 1 vg,m or = 1 d m ( ) = q 2 2 2 c 2 d 2 m c2 d vg,m vp,m = c2 . (2.332)


130 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS Thus dierent modes have dierent group and phase velocities. Figure 2.87 shows group and phase velocity for the dierent modes as a function of the normalized wave number kd/.
2.0 m=1 vg/c and v p/c 1.5 1.0 vg 0.5 0.0 0 1 2 kd/ 3 4 5 m=1 m=2 m=3 m=4 vp m=2 m=3 m=4

Figure 2.87: Group and phase velocity of propagating modes with index m as a function of normailzed wave number. TM-Modes The planar mirror waveguide does not only allow for TEwaves to propagate. There are also TM-waves, which have only a magnetic eld component transverse to the propagation direction and parallel to the mirrors, i.e. in y -direction H (x, z, t) = H y (x, z ) ejt ey , (2.334)

and now H (x, z ) has to obey the Helmholtz equation for the magnetic eld. The corresponding electric eld can be derived from Amperes law 1 E (x, z ) = H y (x, z ) ey (2.335) j 1 H y (x, z ) 1 H y (x, z ) (2.336) ex + ez . = j z j x The electric eld tangential to the metal mirrors has to vanish again, which leads to the boundary condition H y (x, z ) (x = d/2) = 0. x (2.337)



After an analysis very similar to the discussion of the TE-waves we nd for the TM-modes with H y (x, z ) = u(x) ejz ey , (2.338) the transverse mode shapes q 2 cos (kx,m x) with , kx,m = m, m = 0, 2, 4, 6, ..., even modes d d q um (x) = 2 sin (kx,m x) with , kx,m = m, m = 1, 3, 5, ..., odd modes d d (2.339) Note, that in contrast to the electric eld of the TE-waves, which is zero at the metal surface, the transverse magnetic eld of theTM-waves is at a maximum at the metal surface. Also there is no cuto wavelenght of the TM-modes. When the frequency goes to zero, the TM-wave becomes a TEM wave between the two conducting plates. We will not consider this case further, because otherwise the discussion of modes and waveguide dispersion can be worked out very analogous to the case for TE-modes. Multimode Propagation Depending on the boundary conditions at the input of the waveguide at z = 0 many modes may be excited. Eventually there are even excitations with such high transverse wavevectors kx present, that are below cuto. Depending on the excitation amplitudes of each mode, the total eld in the waveguide will be the superposition of all modes. Lets assume that there are only TE-modes excited, then the total eld is
X E (x, z, t) = am ej m z + bm ej m z um (x) ejt ey , m=1


where the amplitudes am and bm are the excitations of the m-th mode in forward and backward direction, respectively. It is easy to show that these excitation amplitudes are determined by the transverse electric and magnetic elds at z = 0 and t = 0. In many cases, the excitation of the waveguide will be such that only the forward propagating modes are excited. E (x, z, t) =

am um (x) ej m z ejt ey ,



When many modes are excited, the transverse eld distribution will change during propagation, see Figure 2.88


Figure 2.88: Variation of the intensity distribution in the transverse direction x at dierent distances z. Intensity prole of (a) the fundamental mode m = 1, (b) the second mode with m = 2 and (c) a linear combination of the fundamental and second mode, [6], p. 247. Modes which are excited below cuto will decay rapidly as evanescent waves. The other modes will propagate, but due to the dierent propagation constants these modes superimpose dierently at dierent propagation distances along the waveguide. This dynamic can be used to build many kinds of important integrated optical devices, such as multimode interference couplers (see problem set 5). Depending on the application, undesired multimode excitation may be very disturbing due to the large group delay dierence between the dierent modes. This eect is called modal dispersion. Mode Orthogonality It turns out that the transverse modes determined by the functions um (x) build an orthogonal set of basis functions into which any function in a certain function space can be decomposed. This is obvious for the case of the



planar-mirror waveguide, where the um (x) are a subset of the basis functions for a Fourier series expansion of an arbitrary function f (x) in the interval [d/2, d/2] which is fulllls the boundary condition f (x = d/2) = 0. It is



um (x) un (x) dx = mn , X f (x) = am um (x) with am = Z

m d/2

(2.342) (2.343) (2.344)

um (x) f (x) dx


From our familiarity with Fourier series expansions of periodic functions, we can accept these relations here without proof. We will return to these equations later in Quantum Mechanics and discuss in which mathematical sense Eqs.(2.342) to (2.343) really hold. Besides illustrating many important concepts, the planar mirror waveguide is not of much practical use at optical frequencies. More in use are dielectric waveguides.

Planar Dielectric Slab Waveguide

In the planar dielectric slab waveguide, waveguiding is not achieved by real reection on a mirror but rather by total internal reection at interfaces between two dielectric materials with refractive indices n1 > n2 , see Figure 2.89


y x

Figure 2.89: Symmetric planar dielectric slab waveguide with n1 > n2 . The light is guided by total internal reection. The eld is evanescent in the cladding material and oscillatory in the core, [6], p. 249. Waveguide Modes As in the case of the planar mirror waveguide, there are TE and TM-modes and we could nd them as a superposition of correspondingly polarized TEM waves propagating with a certain transverse k-vector such that total internal reection occurs. We do not want to follow this procedure here, but rather use immediately the Helmholtz Equation. We again write the electric eld E y (x, z ) = u(x) ejz ey . (2.345)

The eld has to obey the Helmholtz Eq.(2.17) both in the core and in the cladding 2 d2 2 2 2 2 u ( x ) = k = n1 , u ( x ) with k 1 1 dx2 c2 0 2 2 2 d2 2 2 u ( x ) = k = n2 u ( x ) with k cladding : 2 2 dx2 c2 0 core : (2.346) (2.347)

The boundary conditions are given by the continuity of electric and magnetic eld components tangential to the core/cladding interfaces as in section 2.2.



Since the guided elds must be evanescent in the cladding and oscillatory in the core, we rewrite the Helmholtz Equation as 2 d2 2 2 2 , u ( x ) = k u ( x ) with k = k x x 1 dx2 2 d2 2 2 u(x) = 2 cladding : x u(x) with x = k2 2 dx core :

(2.348) (2.349)

where x is the decay constant of the evanescent waves in the cladding. It is obvious that for obtaining guided modes, the propagation constant of the mode must be between the two propagation constants for core and cladding
2 2 < 2 < k1 . k2


Or by dening an eective index for the mode c0

= k0 nef f , with k0 =


we nd n1 > nef f > n2 , and Eqs.(2.348), (2.349) can be rewritten as 2 d2 2 u(x) k0 n1 n2 ef f u(x) = 0 2 dx 2 d2 2 n2 n2 cladding : 2 u(x) k0 ef f u(x) = 0 dx core : (2.352)

(2.353) (2.354)

For reasons, which will become more obvious later, we draw in Figure 2.90 the negative refractive index prole of the waveguide.



0 -n2 -neff -n1


Figure 2.90: Negative refractive index prole and shape of electric eld for the fundamental and rst higher order transverse TE-mode From Eq.(2.353) we nd that the solution has the general form A exp (x x) + B exp (x x) , for x < d/2 C cos (kx x) + D sin (kx x) , for |x| < d/2 (2.355) u(x) = E exp (x x) + F exp (x x) , for x > d/2

The coecients B and E in each case have to be determined from the boundary conditions. From the continuity of the tangential electric eld E y , and

For a guided wave, i.e. um (x ) = 0 the coecients A and F must be zero. It can be also shown from the symmetry of the problem, that the solutons are either even or odd (proof later) B exp (x x) , for x < d/2 (e) C cos (kx x) , for |x| < d/2 , u (x) = (2.356) E exp (x x) , for x > d/2 B exp (x x) , for x < d/2 (o) D sin (kx x) , for |x| < d/2 . u (x) = (2.357) E exp (x x) , for x > d/2



the tangential magnetic eld H z , which follows from Faradays Law to be H z (x) = 1 E y du j0 x dx (2.358)

we obtain the boundary conditions for u(x) u(x = d/2 + ) = u(x = d/2 ), du du (x = d/2 + ) = (x = d/2 ). dx dx (2.359) (2.360)

Note, these are four conditions determining the coecients B, D, E and the propagation constant or refractive index nef f . These conditions solve for the parameters of even and odd modes separately. For the case of the even modes, where B = E , we obtain d d = C cos kx (2.361) B exp x 2 2 d d B x exp x = Ckx sin kx (2.362) 2 2 or by division of both equations d . x = kx tan kx 2 Eqs.(2.348) and (2.349) can be rewritten as one equation 2 2 2 2 2 2 + 2 kx x = k1 k2 = k0 n1 n2 (2.364) (2.363)

Eq.(2.363) together with Eq.(2.364) determine the propagation constant via the two relations. d d d x = kx tan kx , and (2.365) 2 2 2 2 2 2 d d d kx + x = k0 NA (2.366) 2 2 2 where q 2 NA = (n2 1 n2 ) (2.367)

138 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS is called the numerical apperture of the waveguide. We will discuss the physical signicance of the numerical apperture shortly. A graphical solution of these two equations can be found by showing both relations in one plot, see Figure 2.91.

10 k0d/2 NA 8 m=0 xd/2 6 4 2 0 0 2 4 kxd/2 6 8 10 m=1 m=4 m=2 m=3 m=5

Each crossing in Figure 2.91 of a solid line (2.365) with a circle (2.366) NA represents an even guided mode. Similarly one nds for with radius k0 d 2 the odd modes from the boundary conditions the relation d d d x = kx cot kx , (2.368) 2 2 2 which is shown in Figure 2.91 as dotted line. The corresponding crossings with the circle indicate the existence of an odd mode. There are also TM-modes, which we dont want to discuss for the sake of brevity.

Figure 2.91: Graphical solution of Eqs.(2.365) and ( 2.366), solid line for even modes and Eq.(2.368) for the odd modes. d The dash dotted line shows (2.366) for dierent values of the product k0 2 NA

Numerical Aperture Figure 2.91 shows that the number of modes guided is determined by he product k0 d NA, where NA is the numerical apperture 2

2.6. WAVEGUIDES AND INTEGRATED OPTICS dened in Eq.(2.367) d M = Int k0 NA/(/2) + 1, 2 d = Int 2 NA + 1, 0


(2.369) (2.370)

where the function Int[x] means the largest integer not greater than x. Note, that there is always at least one guided mode no matter how small the sized and the refractive index contrast between core and cladding of the waveguide is. However, for small size and index contrast the mode may extend very far into the cladding and the connement in the core is low. The numerical apperture also has an additional physical meaning that becomes obvious from Figure 2.92.

Figure 2.92: Maximum angle of incoming wave guided by a waveguide with numerical aperture NA, [6], p. 262. The maximum angle of an incoming ray that can still be guided in the waveguide is given by the numerical apperture, because according to Snells Law n0 sin (a ) = n1 sin () , (2.371) where n0 is the refractive index of the medium outside the waveguide. The maximum internal angle where light is still guided in the waveguide by total internal reection is determined by the critical angle for total internal reection (2.135) , i.e. max = /2 tot with sin (tot ) = n2 . n1 (2.372)

140 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS Thus for the maximum angle of an incoming ray that can still be guided we nd s 2 n2 n0 sin (a,max ) = n1 sin (max ) = n1 1 = NA. (2.373) n1 Most often the external medium is air with n0 1 and the refractive index contrast is week, so that a,max 1 and we can replace the sinusoid with its argument, which leads to a,max = NA. (2.374) Field Distributions Figure 2.93 shows the eld distribution for the TE guided modes in a dielectric waveguide. Note, these are solutions of the second order dierential equations (2.353) and (2.354) for an eective index nef f , that is between the core and cladding index. These guided modes have a oscillatory behavior in those regions in space where the negative eective index is larger than the negative local refractive index, see Figure 2.90 and exponentially decaying solutions where the negative eective index is smaller than than the negative local refractive index.

Figure 2.93: Field distributions for TE guided modes in a dielectric waveguide. These results should be compared with those shown in Figure 2.85 for the planar-mirror waveguide [6], p. 254. Figure 2.94 shows a comparison of the guided modes in a waveguide with a Gaussian beam. In contrast to a the Gaussian beam which diracts,



in a waveguide diraction is balanced by the guiding action of the index discontinuity, i.e. total internal reection. Most importantly the cross section of a waveguide mode stays constant and therefore a waveguide mode can eciently interact with the medium constituting the core or a medium that is incorporated in the core.

Figure 2.94: Comparison of Gaussian beam in free space and a waveguide mode, [6], p. 255. Besides integration, this prolonged interaction distance is one of the major reasons for using waveguides. The interaction lenght can be arbitrarily long, only limited by the waveguide loss, in contrast to a Gaussian beam, which stays focused only over the confocal distance or Rayleigh range. As in the case of a planar-mirror waveguide, one can show that the transverse mode functions are orthogonal to each other. At rst, a striking difference here is that we have only a nite number of guided modes and one might worry about the completeness of the transverse mode functions. The answer is that in addition to the guided modes, there are unguided modes or leaky modes, which together with the guided modes from a complete set. Each initial eld can be decomposed into these modes. The leaky modes rapidly loose energy because of radiation and after a relatively short propagation distance only the eld of guided modes remains in the waveguide. We will not pursue this further in this introductory class. The interested reader should consult with [12].

142 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS Connement Factor A very important quantity for a waveguide mode is its connement in the core, which is called the connement factor R d/2 2 u (x) dx m = R0 2m . (2.375) um (x) dx 0

The connement factor quanties the fraction of the mode energy propagating in the core of the waveguide. This is very important for the interaction of the mode with the medium of the core, which may be used to amplify the mode or which may contain nonlinear media for frequency conversion. Waveguide Dispersion For the guided modes the eective refractive indices of the modes and therefore the dispersion relations must be between the indices or dispersion relations of core and cladding, see Figure 2.95

Figure 2.95: Dispersion relations for the dierent guided TE-Modes in the dielectric slab waveguide. The dierent slopes d/d for each mode indicate the dierence in group velocity between the modes. Note, that there is at least always one guided mode.




Two-Dimensional Waveguides

Both the planar-mirror waveguide and the planar dielectric slab waveguide conne light only in one direction. It is straight forward to analyze the modes of the two-dimensional planar-mirror waveguide, which you have already done in 6.013. Figure 2.96 shows various waveguides that are used in praxis for various devices. Here, we do not want to analyze them any further, because this is only possible by numerical techniques.

Figure 2.96: Various types of waveguide geometries: (a) strip: (b) embedded strip: (c) rib ro ridge: (d) strip loaded. The darker the shading, the higher the refractive index [6], p. 261.


Waveguide Coupling

The core size of a waveguide can range from a fraction of the free space wavelength to many wavelength for a multimode ber. For example a typical high-index contrast waveguide with a silicon core and a silica cladding for 1550 nm has a cross section of 0.2m 0.4m, single-mode ber, which we will discuss in the next section with an index contrast of 0.5-1% between core and cladding has a typical mode-eld radius of 6m. If the mode cross section is not prohibitively small the simplest approach to couple light into a waveguide is by using a proper lens, see Figure 2.97 (a) or direct butt coupling of the source to the waveguide if the source is a waveguide based device itself.


Figure 2.97: Coupling to a waveguide by (a) a lens; (b) direct butt coupling of an LED or laser diode, [6], p. 262

The lens and the beam size in free space must be chosen such that the spot size matches the size of the waveguide mode while the focusing angle in free space is less than the numerical aperture of the waveguide, (see problem set). Other alternatives are coupling to the evanscent eld by using a prism coupler, see Figure 2.98

Figure 2.98: Prism coupler, [6], p. 263.



Figure 2.99: Grating Coupler

The coupling with the prism coupler is maximum if the propagation constant of the waveguide mode matches the longitudinal component of the k-vector = knp cos p ,

Another way to match the longitudinal component of the k -vector of the incoming light to the propagation constant of the waveguide mode is by a grating coupler, see Figure 2.99


Coupling of Modes

If two dielectric waveguides are placed closely together their elds overlap. This situation is shown in Figure 2.100 at the example of the planar dielectric slab waveguide. Of course this situation can be achieved with any type of two dimensional dielectric waveguide shown in Figure 2.96


Figure 2.100: Coupling between the two modes of the dielectric slab waveguide, [6], p. 264.

Once the elds signicantly overlap the two modes interact. The shape of each mode does not change very much by the interaction. Therefore, we can analyze this situation using perturbation theory. We assume that in zero-th order the mode in each waveguide is independent from the presence of the other waveguide. We consider only the fundamental TE-modes in each of the waveguide which have excitation amplitudes a1 (z ) and a2 (z ), respectively. The dynamics of each mode can be understood in terms of this wave amplitude. In the absence of the second waveguide, each waveguide amplitude undergoes only a phase shift during propagation according to its dispersion relations da1 (z ) = j 1 a1 (z ), dz da2 (z ) = j 2 a2 (z ). dz (2.376) (2.377)

The polarization generated by the eld of mode 2 in waveguide 1 acts as a source for the eld in waveguide 1 and the other way arround. Therefore, coupling of modes can be described by adding a source term in each equation proportional to the free propagation of the corresponding wave in the other

2.6. WAVEGUIDES AND INTEGRATED OPTICS guide da1 (z ) = j 1 a1 (z ) j12 a2 (z ), dz da2 (z ) = j21 a1 (z ) j 2 a2 (z ). dz


(2.378) (2.379)

12 and 21 are the coupling constants of the modes. An expression in terms of waveguide properties can be derived using perturbation theory [6],. These coupled mode equations describe a wealth of phenomena and are of fundamental importance in many areas of physics and engineering. As we will see, there is only a signicant interaction of the two modes if the two propagation constants are not much dierent from each other (phase matching). Therefore, we write the propagation constants in terms of the average 0 and the phase mismatch 1/2 = 0 with 2 1 + 2 and = 1 . 0 = 2 2 (2.380) (2.381)

and we take the overall trivial phase shift of both modes out by introducing the slowly varying relative eld amplitudes a 1 (z ) = a1 (z )ej 0 z and a 2 (z ) = a2 (z )ej 0 z which obey the equation d 1 (z ) j12 a 2 (z ), a (z ) = j a dz 1 d a (z ) = j21 a 1 (z ) + j a 2 (z ). dz 2 Power conservation during propagation demands d |a 1 (z )|2 + |a 2 (z )|2 = 0 dz (2.383) (2.384) (2.382)


which requests that 21 = 12 , i.e. the two coupling coecients are not independent from each other (see problem set). Note, Eqs.(2.383) and (2.384) are a system of two linear ordinary dierential equations with constant coecients, which is straight forward to solve

148 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS (see problem set). Given the excitation amplitudes a 1 (0) and a 2 (0) = 0 at the input of the waveguides, i.e. no input in waveguide 2 the solution is a 1 (z ) = a sin z , (2.386) 1 (0) cos z j 21 1 (0) sin z, (2.387) a 2 (z ) = j a with q = 2 + |12 |2 .


The optical powers after a propagation distance z in both waveguides are then ! 2 2 P1 (z ) = |a 1 (z )| = P1 (0) cos2 z + sin2 z , (2.389) ! |21 |2 sin2 z. (2.390) P2 (z ) = P1 (0) 2 This solution shows, that depending on the dierence in phase velocity between the two-waveguides more or less power is coupled back and fourth between the two waveguides, see Figure 2.101. The period at which the power exchange occurs is L= . (2.391)

If both waveguides are identical, i.e. = 0 and = |12 |, the waves are phase matched, Eqs.(2.389) and (2.390) simplify to P1 (z ) = P1 (0) cos2 z P2 (z ) = P1 (0) sin2 z. (2.392) (2.393)

Complete transfer of power occurs between the two waveguides after a distance , (2.394) L0 = 2 |12 | see Figure 2.102



Figure 2.101: Periodic exchange of power between guides 1 and 2 [6], p. 266.

Figure 2.102: Exchange of power between guides 1 and 2 in the phasematched case, [6], p. 266. Depending on the length of the coupling region the coupling ratio can be chosen. A device with a distance L0 /2 and L0 achieves 50% and 100% power transfer into waveguide two, respectively, see Figure 2.103


Figure 2.103: Optical couplers: (a) 100% coupler, (b) 3dB coupler, [6], 267.


Switching by Control of Phase Mismatch

If we keep the interaction length of the waveguides xed at a length L0 = , then the power tranfer from waveguide 1 to waveguide 2 depends crit2|12 | ically on the phase mismatch

T ( ) =

P2 (z = L0 ) = P1 (z = 0)

2 2

s 2 1 2L0 sin c2 1+ , 2


where sinc(x) = sin(x)/(x). Figure 2.104 shows the transfer characteristic as a function of normalized phase mismatch. The phase mismatch between waveguides can be controlled for example by the linear electro-optic or Pockels eect, which we will investigate later.



Figure 2.104: Dependence of power transfer from waveguide 1 to waveguide 2 as a function of phase mismatch, [6], p. 267. The implementation of such a waveguide coupler switch is shown in Figure 2.105.

Figure 2.105: Integrated waveguide coupler switch, [6], p. 708



Optical Fibers

Optical bers are cylindrical waveguides, see Figure 2.106, made of low-loss materials such as silica glass.

Figure 2.106: Optical bers are cylindrical dielectric waveguides, [6], p. 273. Similar to the waveguides studied in the last section the most basic bers consist of a high index core and a lower index cladding. Today ber technology is a highly developed art which has pushed many of the physical parameters of a waveguide to values which have been thought to be impossible a few decades ago: Fiber with less than 0.16dB/km loss Photonic crystal ber (Nanostructured ber) Hollow core ber Highly nonlinear ber Er-doped ber for ampliers Yb-doped ber for ecient lasers and ampliers Raman gain ber Large area single mode bers for high power (kW) lasers. Figure 2.107 shows the ranges of attenuation coecients of silica glass single-mode and multimode ber.



Figure 2.107: Ranges of attenuation coecients of silica glass single-mode and multimode ber, [10], p. 298. For the purpose of this introductory class we only give an overview about the mode structure of the most basic ber, the step index ber, see Figure 2.108 (b)

Figure 2.108: Geometry, refractive index prole, and typical rays in: (a) a multimode step-index ber, (b) a single-mode step-index ber, (c) a multimode graded-index ber [6], p. 274

154 CHAPTER 2. CLASSICAL ELECTROMAGNETISM AND OPTICS Step-index ber is a cylindrical dielectric waveguide specied by its core and cladding refractive indices, n1 and n2 and the core radius a, see Figure 2.106. Typically the cladding is assumed to be so thick that the nite cladding radius does not need to be taken into account. The guided modes need to be suciently decayed before reaching the cladding boundary, which is usually strongly scattering or absorbing. In standard ber, the cladding indices dier only slightly, so that the relative refractive-index dierence = n1 n2 n1 (2.396)

is small, typically 103 < < 2 102 . Most bers currently used in medium to long optical communication systems are made of fused silica glass (SiO2 ) of high chemical purity. The increase in refractive index of the core is achieved by doping with titanium, germanium or boron, among others. The refractive index n1 ranges from 1.44 to 1.46 depending on the wavelength utilized in the ber. The acceptance angle of the rays coupling from free space into guided modes of the waveguide is determined by the numerical apperture as already discussed for the dielectric slab waveguide, see Figure 2.109 q 2 a sin(a ) = NA = n2 n n 2. (2.397) 1 1 2

Figure 2.109: The acceptance angle of a ber and numerical aperture NA [6], p. 276.



Again the guided waves can be found by looking at solutions of the Helmholtz equations in the core and cladding where the index is homogenous and by additionally requesting the continuity of the tangential electric and magnetic elds at the core-cladding boundary. In general the ber modes are not anylonger pure TE or TM modes but rather are hybrid modes, i.e. the modes have both transverse and longitudinal electric and magnetic eld components. Only the radial symmetric modes are still TE or TM modes. To determine the exact mode solutions of the ber is beyond the scope of this class and the interested reader may consult reference [2]. However, for weakly guiding bers, i.e. 1, the modes are actually very much TEM like, i.e. the longitudinal eld components are much smaller than the radial eld components. The linear in x and y directions polarized modes form orthogonal polarization states. The linearly polarized (l, m) mode is usually denoted as the LPlm -mode.The two polarizations of the mode with indices (l, m) travel with the same propagation constant and have identical intensity distributions. The generic solutions to the Helmhotz equation in cylindrical coordinates are the ordinary, Jm (kr), and modied, Km (kr), Bessel functions (analogous to the cos(x)/ sin(x) and exponential functions ex , that are solutions to the Helmholtz equation in cartesian coordinates). Thus, a generic mode function for a cylinder symmetric ber has the form cos(l) , for r < a, core Jl (kl,m r) sin(l) ul,m (r, ) = (2.398) cos(l) , for r > a, cladding Kl (kl,m r) sin(l)

For large r, the modied Bessel function approaches an exponential, Kl (kl,m r) cos(l) . The propagation constants for this two dimensional wavegeln m r sin(l) uide have to fulll the additional constraints
2 kl,m =

2 l,m
2 kl,m

2 l,m

2 2 n1 k0 2 , 2 = 2 n2 2 k0 , =
2 k0 NA2 .

(2.399) (2.400) (2.401)

Figure 2.110 shows the radial dependence of the mode functions


Figure 2.110: Radial dependence of mode functions u(r),[6], p.279. The transverse intensity distribution of the linearly polarized LP0,1 and LP3,4 modes are shown in Figure 2.111.

Figure 2.111: Intensity distribtuion of the (a) LP01 and (b) LP3,4 modes in the transverse plane. The LP01 has a intensity distribution similar to the Gaussian beam, [6], p. 283. Number of Modes It turns out, that as in the case of the dielectric slab waveguide the number of guided modes critically depends on the numerical aperture or more precisely

2.6. WAVEGUIDES AND INTEGRATED OPTICS on the V-parameter, see Eq.(2.369) a V = k0 N A. 2 Without proof the number of modes is M 4 2 V , for V 1. 2




which is similar to Eq.(2.369) for the one-dimensional dielectric slab waveguide, but the number of modes here is now related to the square of the V-parameter, because of the two-dimensional transverse connement of the modes in the ber. As in the case of the dielectric waveguide, there is always at least one guided mode (two polarizations). However, the smaller the V-parameter the more the mode extends into the cladding and the guiding properties become weak, i.e. small bending of the ber may already lead to high loss.


[1] Hecht and Zajac, "Optics," Addison and Wesley, Publishing Co., 1979. [2] B.E.A. Saleh and M.C. Teich, "Fundamentals of Photonics," John Wiley and Sons, Inc., 1991. [3] Bergmann and Schaefer, "Lehrbuch der Experimentalphysik: Opitk," 1993. [4] H. Kogelnik and T. Li, Laser Beams and Resonators, Appl. Opt. 5, pp. 1550 1566 (1966). [5] H. Kogelnik, E. P. Ippen, A. Dienes and C. V. Shank, Astigmatically Compensated Cavities for CW Dye Lasers, IEEE J. Quantum Electron. QE-8, pp. 373 379 (1972). [6] H. A. Haus, Fields and Waves in Optoelectronics, Prentice Hall 1984. [7] F. K. Kneubhl and M. W. Sigrist, Laser, 3rd Edition, Teubner Verlag, Stuttgart (1991). [8] A. E. Siegman, Lasers, University Science Books, Mill Valley, California (1986). [9] Optical Electronics, A. Yariv, Holt, Rinehart & Winston, New York, 1991. [10] Photonic Devices, Jia, Ming-Liu , Cambridge University Press, 2005. [11] T. Barwicz, et al., Polarization Transparent Microphotonics in the Strong Connement Limit, in Nature Photonics 2007. [12] T. Tamir, "Guided-Wave Optoelectronics," Springer, 1990.


Chapter 3 Quantum Nature of Light and Matter

We understand classical mechanical motion of particles governed by Newtons law. In the last chapter we examined in some detail the wave nature of electromagnetic elds. We understand the occurance of guided traveling modes and of resonator modes. There are characteristic dispersion relations or resonance frequencies associated with them. In this chapter, we want to summarize some experimental ndings that ultimately lead to the discovery of quantum mechanics and with it to a greatly improved understanding of the nature of matter and electromagnetic waves. It turns out that matter has in addition to its particle like properties which we are familiar with from classsical mechanics also wave properties and electromagnetic waves have in addition to its wave properties particle like properties. The nal theory, which will be developed in subsequent chapters is much more than just that because the quantum mechanical wave function has a dierent physical interpretation than a electromagnetic wave only the mathematical concepts used are in many cases very similar. However, familiarity with electromagnetic waves is a tremendous help and guideance in doing and nally understanding quantum mechanics.


Black Body Radiation

In 1900 the physicist Max Planck found the law that governs the emission of electromagnetic radiation from a black body in thermal equilibrium. More 161



specically Plancks law gives the energy stored in the electromagntic eld in a unit volume and unit frequency range, [f, f + df ] with df = 1Hz , when the electromagnetic eld is in thermal equilibrium with its surrounding that is at temperature T. A black body is simply dened as an object that absorbs all light. The best implementation of a black body is the Ulbricht sphere, see Figure 3.1.

Figure 3.1: The Ulbricht sphere, is a sphere with a small opening, where only a small amount of radiation can escape, so that the interior of the sphere is in thermal equilibrum with the walls, which are kept at a constant tremperature. The inside walls are typically made of diuse material, so that after multiple scattering of the walls any incoming ray is absorbed, i.e. the wall opening is black. Figure 3.2 shows the energy density w(f ) of electromagnetic radiation in a black body at temperature T . Around the turn of the 19th century w(f ) was measured with high precision and one was able to distinguish between various approximations that were presented by other researchers earlier, like the Rayleigh-Jeans law and Wiens law, which turned out to be asymptotic approximations to Plancks Law for low and high frequencies. In order to nd the formula describing the graphs shown in Figure 3.2 Planck had to introduce the hypothesis that harmonic oscillators with frequency f can not exchange arbitrary amounts of energy but rather only in discrete portions, so called quanta. Planck modelled atoms as classical



oscillators with frequency f . Therefore, the energy of an oscillator must be quantized in energy levels corresponding to these energy quanta, which he found to be equal to hf, where h is Plancks constant h = 6.62620 5 1034 Js. (3.1)


Black Body Energy Density w(f), fJs/m

Rayleigh-Jeans (T=5000K) 0.8 Wien (T=5000K)



T=5000K T=4000K T=3000K T=2000K


0.0 0 2 4 6 Frequency f, 100THz 8 10

Figure 3.2: Spectral energy density of the black body radiation according to Plancks Law. As a model for a black body, we neglect the presence of the opening, which is only necessary to observe the radiation and use a cavity with perfectly reecting walls, somewhat dierent from the Ulbricht sphere. One could make this opening so small that it does essentially not change the internal radiation eld. Then the radiation in the cavity is the sum over all possible resonator modes in the cavity. If the cavity is at temperature T all the modes are thermally excited by emission and absorption of energy quanta from the atoms of the wall.



For the derivation of Plancks law we consider a cavity with perfectly conducting walls, see Figure 3.3.

Figure 3.3: (a) Cavity resonator with metallic walls. (b) Resonator modes characterized by a certain k-vector. If we extend the analysis of the plan parallel mirror waveguide to nd the TE and TM modes of a three-dimensional metalic resonator, the resonator modes are TEmnp and THmnp modes characterized by its wave vector components in x, y , and z direction. The resonances are standing waves in three dimensions with wavevector components kx = m n p , ky = , kz = , for m, n, p = 0, 1, 2, ... Lx Ly Lz (3.2)

An expression for the number of modes in a frequency interval [f, f + df ] can be found by recognizing that this is identical to the number of points in Figure 3.3(b) that are in the rst octant of a spherical shell with thickness dk at k = 2f /c.The volume occupied by one mode in the space of wave 3 numbers k is V = L L L = with the volume V = Lx Ly Lz . Then the V x y z number of modes N (f ) in the frequency interval [f, f + df ] in volume V are k2 dk 4k2 dk = V , 3 2 8 V

N (f ) = 2


where the factor of 2 in front accounts for the two polarizations or T E and T H -modes of the resonator and the 8 in the denominator accounts for the



fact that only 1/8th of the sphere, an octand, is occupied by the positive wave vectors. With k = 2f/c and dk = 2df/c, we obtain for the number of modes nally 8 2 f df (3.4) c3 Note, that the same density of states is obtained using periodic boundary conditions in all three dimensions, i.e. then we can represent all elds in terms of a three dimensional Fourier series. The possible wave vectors would range from negative to positive values N (f ) = V kx = 2m 2n 2p , ky = , kz = f or m, n, p = 0, 1, 2. . . . Lx Ly Lz (3.5)

However, these wavevectors ll the whole sphere and not just 1/8th , which compensates for the 8-times larger volume occupied by one mode. If we imply periodic boundary conditions, we have forward and backward running waves that are independent from each other. If we use the boundary conditions of a resonator with perfectly conducting walls as before, the forward and backward running waves are identical and not independent and form standing waves. One should not be disturbed by this fact as all volume properties, such as the energy density, only depends on the density of states, and not on surface eects, as long as the volume is reasonably large.



The complex excitation amplitude a(t) of each mode obeys the equation of motion of a harmonic oscillator with a resonance frequency equal to the resonance frequency 0 of the resonator mode. a(t) = a(0)ej0 t = Q(t) + jP (t) with a (t) = j 0 a(t) (t) + 2 Q 0 Q(t) = 0 (t) = 0 P (t). Q (3.7) (3.6)


(3.8) (3.9)



Therefore, classically one expects that each mode is in thermal equilibrium excited with a thermal energy kT according to the equipartition theorem, where k is Boltzmanns constant with k = 1.38062 6 1023 J/K. (3.10)

If that is the case the spectral energy density is given by the Rayleigh-JeansLaw, see Figure 3.2. w(f ) = 1 N (f ) 8 kT = 3 f 2 kT. V df c (3.11)

As can be seen from Figure 3.2, this law describes very well the black body radiation for frequencies hf kT but there is an arbitrary large deviation for high frequencies. This formula can not be correct, because it predicts innite energy density for the high frequency modes resulting in an "ultraviolet catastrophy", i.e. the electromagnetic eld contains an innite amount of energy at thermal equilibrium.


Wiens Law

The high frequency or short wavelength region of the black body radiation was rst empirically described by Wiens Law 8hf 3 hf /kT e . (3.12) c3 Wiens law is surprisingly close to Plancks law, however it slightly fails to correctly predict the asympthotic behaviour at low frequencies or long wavelengths. w(f ) =


Plancks Law

In the winter of 1900, Max Planck found the correct law for the black body radiation by assuming that each oscillator can only exchange energy in discrete portions or quanta. We rederive it by assuming that each mode can only have the discrete energie values Es = s hf, for s = 0, 1, 2, ... (3.13)



where Z is a normalization factor such that the total propability of the oscillator to have any of the allowed energy values is
X s=0

Thus s is the number of energy quanta stored in the oscillator. If the oscillator is a mode of the electromagnetic eld we call s the number of photons. For the probability ps , that the oscillator has the energy Es we assume a Boltzmanndistribution Es 1 hf 1 ps = exp = exp s , (3.14) Z kT Z kT

ps = 1.


Note, due to the fact that the oscillator energy is proportional to the number of photons, the statistics are exponential statistics. From Eqs.(3.14) and (3.15) we obtain for the normalization factor X hf 1 hf , Z= (3.16) exp s = kT 1 exp kT s=0

which is also called the partition function. The photon statistics are then given by hf Es ps = exp 1 exp (3.17) kT kT
hf or with = kT

X 1 1 es , with Z ( ) = ps = es = . Z ( ) 1 e s=0


Given the statistics of the photon number, we can compute moments of the probability distribution, such as the average number of photons in the mode hsi =
X s=0

s ps .


This rst moment of the photon statistics can be computed from the partition function, using the "trick" 1 s = 1 1 Z ( ) = Z ( ) e , Z ( ) ( )1 (3.20)



which is hsi = exp

hf kT


With the average photon number hsi , we obtain for the average energy stored in the mode hEs i = hsi hf, (3.22) and the energy density in the frequency intervall [f, f + df ] is then given by w (f ) = N (f ) hEs i . V df (3.23)

With the density of modes from Eq.(3.4) we nd Plancks law for the black body radiation hf 8 f 2 , (3.24) w (f ) = hf 3 c exp kT 1 which was used to make the plots shown in Figure 3.2. In the limits of low and high frequencies, i.e. hf kT and hf kT , respectively Plancks law asympthotically approaches the Rayleigh-Jeans law and Wiens law.


Thermal Photon Statistics

It is interesting to further investigate the intensity uctuations of the thermal radiation emitted from a black body. If the wall opening in the Ulbricht sphere, see Figure 3.1, is small enough very little radiation escapes through it. If the Ulbricht sphere is kept at constant temperature the radiation inside the Ulbricht sphere stays in thermal equilibrium and the intensity of the radiation emitted from the wall opening in a frequency interval [f, f + df ] is I (f ) = c w (f ) . (3.25)

Thus the intensity uctuations of the emitted black body radiation is directly related to the photon statistics or quantum statistics of the radiation modes at freuqency f , i.e. related to the stochastic variable s : the number of photons in a mode with frequency f . This gives us direclty experimental access to the photon statistics of an ensemble of modes or even a single mode when proper spatial and spectral ltering is applied.



Using the expectation value of the photon number 3.21, we can rewrite the photon statistics for a thermally excited mode in terms of its average photon number in the mode as hsis 1 ps = s+1 = (hsi + 1) (hsi + 1) hsi (hsi + 1) s


The thermal photon statistics display an exponential distribution, see Figure 3.4. Before we move on, lets see how the average photon number in a given mode depends on temperature and the frequency range considered. Figure 3.5 shows the relationship between average number of photons in a mode with frequency f or wavelength and temperature T.

Figure 3.4: Photon statistics of a mode in thermal equilibrium with a mean photon number < s >= 10 (a) and < s >= 1000 (b).



Figure 3.5: Average photon number in a mode at frequency f or wavelength and temperature. Figure 3.5 shows that at room temperature and microwave frequencies large numbers of photons are present due to the thermal excitation of the mode. However, at high frequencies, which start for room temperature in the far to mid infrared range, on average much less than one photon is thermally excited. The variance of the photon number distribution is 2 2 (3.27) hsi2 . s = s By generalizing Eq.(3.16) to the m-th moment by replacing the exponent 1 by m hsm i = =
X s=0

sm ps

(3.28) (3.29)

1 m Z ( ) , Z ( )m

and therefore for the variance of the photon number using Eq.(3.27) is
2 2 s = hsi + hsi .

we obtain for the second moment 2 s = 2Z ( ) 2 e2 + Z ( ) e = 2 hsi 2 + hsi .





As expected from the wide distribution of photon numbers the variance is larger than the square of the expectation value. This means that if we look at the light intensity coming form a single mode in thermal equilibrium, the intensity is subject to extremly strong uctuations, i.e. its standard deviation, as large as its mean value. So why dont we see this rapid thermal uctuations when we look at the black body radiation coming, for example, from the surface of the sun? Well we dont look at a single mode but rather at a whole multitude of modes. Even when we restrict us to a certain narrow frequency range and spatial direction, there is a multitude of transverse modes presence. We obtain for the average total number of photons in a group of modes and its variance
N X i=1 N X i=1

hstot i = 2 tot =

hsi i , 2 i .

(3.32) (3.33)

Since these modes are independent identical systems, we have hstot i = N hsi . 2 1 hstot i2 + hstot i . 2 tot = N hsi + hsi = N (3.34)

Due to the averaging over many modes, the photon number uctuations in a large number of modes is reduced compared to its mean value SNR = hstot i2 = 2 tot 1 1 . + hstot i (3.35)

1 N

Thus, if one averages over many modes and has many photons in these modes the intensity uctuations become small. This is the reason while natural light, emitted from the sun or an incandescent lamp, both are thermal emitters, do not appear as highly uctuating sources. Our eyes are averaging over multiple modes which leads to a well dened average photon number phstot i detected in a certain time interval with a standard deviation tot hstot i much smaller than the average vale for N .




Photo-electric Eect

Another strong indication for the quantum nature of light was the discovery of the photoelectric eect by Lenard in 1903. He discovered that when ultra violet light is radiated on a photo cathode electrons are emitted, see Figure 3.6.
(a) (b) -V


Figure 3.6: Photo-electric eect: (a) Schematic setup and (b) dependence of the necessary grid voltage to supress the electron current as a funtion of light frequency. Lenard surrounded the photo cathode by a grid, which is charged by the emitted electrons up to a voltage V, which blocks the emission of further electrons. Figure 3.6 shows the blocking voltage as a function of the frequency of the incoming light. Depending on the cathode material there is a cuto frequency. For lower frequencies no electrons are emitted at all. This frequency as well as the blocking voltage does not depend on the intensity of the light. In 1905, this eect was explained by Einstein introducing the quantum hypothesis for radiation. According to him, each electron emission is caused by a light quantum, now called photon. This photon has an energy Eph = hf and this quantum energy must be larger than the work function We of the material. The remainder of the energy me v 2 /2 is transfered to the electron in form of kinetic energy. The resulting energy balance is 1 hf = We + me v 2 2 (3.36)

The kinetic energy of the electron can be used to reach the grid surrounding



the photo cathode until the charging energy due to the grid potential is equal to the kinetic energy of the electrons 1 eV = me v 2 2 or (3.37)

1 (3.38) V = (hf We ), for hf > We . e This equation explains the empirically found law by Lenard explaining the cuto frequency and the charge buildup as a function of light frequency. Einstein was rst to introduce the idea that the electromagnetic eld contains light quanta or photons.


Photodetection and Shot Noise

If a resistor is placed between the grid and the cathode, the voltage will drive an external current through the resistor and the arrangement is a photodetector that converts the energy of the incoming photons into electrical energy dissipated in a load resistor, see Fig. 3.7(a).

(a) -V


Figure 3.7: (a) Photodetector with external circuit. (b) Equivalent circuit. The current delivered from the photodetector is I = QE e P = OEE P. hf (3.39)



where QE is called the quantum eciency of the detector, e is the electron charge, P the incoming optical power hitting the detector and OEE the responsivity or optoelectronic eciency (OEE) of the photo detector. For an ideal photo detector QE = 1. The photon energy Eph depends on the wavelength according to Eph = hf = 1.24eV . [m] (3.40)

Thus the detector has at a wavelength of = 1.24m a responsivity of e = 1A/W, (3.41) hf i..e the detector delivers 1 Ampere electrical current for 1 Watt of optical power. Thus to rst order a photodetector can be modelled as a current source, that delivers a current proportional to the optical power absorbed in the detector, see Fig. 3.7(b). Without proof, it turns out that for an ideal photodector the statistics of the electrons making up the photo current is identical to the statistic of the indicident ux of photons. In real photodetectors this one to one relationship between photons and electrons maybe spoiled by additional statistical processes, like loss in the photodetector either on the optical or electrical side resulting in QE < 1 and due to subsequent amplication of the electrical signal. The statistical uctuations in the photon number that arrive in a given time interval carry over into statistical ucuations of the photo electrons making up the photo current. If the photo current I is made up of s photons detected in a given time interval T, then this photo current can be and a uctuating or often called noise decomposed into its average current I component In . The total current will be the average photo current and a noise photo current OEE = QE + In (t). I (t) = I (3.42)

For an ideal photodetector the average current times the observation interval T is identical to the average number of photons detected in that time interval I T = hsi = number of detected photons. e (3.43)

The noise current, which is a random variable is characterized by its power spectral density SIn In (f ). Each detector has a nite response time, i.e. the



time it takes for a photo electron to leave the photo detector, or in other words the photo current is made up of a sum of individual electrons emitted from the detector at random times ti I (t) = X

e h(t ti ) with

h(t ti ) dt = 1,


where h(t ti ) is the detector response due to a single incident photon leading to a photo electron and therefore a photo current. If there is no correlation in the arrival time of the photons, the number of photo electrons extracted from the photodetector in two dierent time intervals that are at least one response time of the detector apart must be independent from each other. This means that the current uctuation spectrum SIn In (f ) for frequencies shorter than the inverse detector response time, pd , must be constant or "white", SIn In (f ) = const., for |f | << 1/ pd (3.45)

because only then will the variance of the detected photons or generated photo electrons grow linearly with the total averaging time, i.e or inverse bandwidth of the measurment. The detailed calculation is in the appendix, here we argue qualitatively. We obtain for the uctuation in detected photon number sn (t) over a measurement interval [t, t + T ], by integrating over the noise current 1 sn (t) = e
t+T Z t

T 1 In (t0 )dt0 = eT

t+T Z t

In (t0 )dt0 .


As we know from 6.011, for a stationary random process In (t), the variance if it is the same as the variance of sn and the variance can be computed from the autocorrelation spectrum according to
f Z

2 sn sn

T2 = s2 n = 2 e

SIn In (f ) df


T SI I (f ) 2f. e2 n n




As in the case of averaging over many (innitly many) modes in the previous section, one can show that for a randomly arriving stream of particles, here photons, which have no correlations with each other and a given average ux (Poisson process), the variance of the number of particles/photons measured in a given time interval T should be equal to the average number of detected particles independent from the observation interval T 2 sn sn = I T2 S ( f )2 f = h s i = T, I I n n e2 e (3.49)

i.e. the power spectral density of the noise current is 1 = eI, SIn In (f ) = eI 2T f (3.50)

where we have used 2T f = 1 to derive at the correct noise formula for the shot noise due to a poison distributed current, see appendix. Note, that SIn In (f ) is the two-sided autocorrelation spectrum, i.e. dened over positive and negative frequencies. If only positive frequencies are used, we need to ). mulitply by a factor of two (SIn In (f ) = 2eI One may think that shot noise is rather due to the discreteness of the electrical charge than due to the presence of photons. However, this is not so. There exist optical sources that generate a photon stream which has less noise than the shot noise formula 3.50 predicts, and which can not be explained classically.


Signal to Noise Ratio

Thus if a constant optical power is applied to a photo detector the signal to noise ratio in terms of signal power to noise power when measured with a bandwdith f is SNR = 2 QE P I I = = . 2SIn In (f )f 2ef 2hf f (3.51)

The signal to noise ratio only growth linearly with the optical power, since also the noise grows with the applied optical power but less strong.




Noise Equivalent Power

Most often a real photo detector includes an additional amplication process which may add addtional noise. Then the signal to noise ratio changes to account for these additional noise sources. Manufacturers of photodetectors therfore specify a noise equivalent power NEP which has the units [W/ Hz]. It simply species the optical power level necessary on the photodetector to achieve a signal to noise ratio of 1.


Spontaneous and Induced Emission

The number of photons in a radiation mode may change via emission of a photon into the mode or absorption of a photon from the mode by atoms, molecules or a solid state material. Einstein introduced a phenomenological theory of these processes in order to explain how matter may get into thermal equilibrium by interaction with the modes of the radiation eld. He considered the interaction of a mode with atoms modeled by two energy levels E1 and E2 , see Figure 3.8.

Figure 3.8: Energy levels of a two level atom and populations. n1 and n2 are the population densities of these two levels considering a whole ensemble of these atoms. Transitions are possible in the atom between the two energy levels by emission of a photon at a frequency f= E2 E1 h (3.52)



Absorption of a photon is only possible if there is energy present in the radiation eld. Einstein wrote for the corresponding transition rates, which should be proportional to the population densities and the photon density at the transition frequency dn1 dn 2 = = B12 n1 w(f21 ). (3.53) dt Abs dt Abs The coecient B12 characterizes the absorption properties of the transition. Einstein had to allow for two dierent kind of processes for reasons that become clear a little later. Transitions induced by the already present photons or radiation energy as well as spontaneous transitions dn1 dn 2 = =B21 n2 w (f21 ) + A21 n2 . (3.54) dt Em dt Em

The coecient B21 describes the induced and A21 the spontaneous emissions. The latter transitions occur even in the absence of any radiation and the corresponding coecient determines the lifetime of the excited state
1 sp = A 21 ,


Using Eqs.(3.53) and (3.54) we nd

in the absence of the radiation eld. The total change in the population densities is due to both absorption and emission processes dni dni dni , for i = 1, 2 = + (3.56) dt dt Em dt Abs dn1 dn2 = = (B12 n1 B21 n2 ) w (f21 ) A21 n2 . dt dt (3.57)

In thermal equilibrium the energy density of the radiation eld must fulll the condition A21 /B12 , (3.58) w (f21 ) = n1 /n2 B21 /B12 while the atomic ensemble itself should also be in thermal equilibrium which again should be described by the Boltzmann statistics, i.e. the ratio between the population densities are determined by the Boltzmann factor E2 E1 n2 /n1 = exp . (3.59) kT

3.4. SPONTANEOUS AND INDUCED EMISSION And with it the energy density of the radiation eld must be w (f21 ) = A21 /B12 hf21 . exp kT B21 /B12 B21 = B12 , and A21 =



A comparison with Plancks law, Eq.(3.24), gives


3 8 hf21 B12 . (3.62) c3 Clearly, without the spontaneous emission process it is impossible to arrive at Plancks Law in equilibrium. The spectral energy density of the radiation eld can be rewritten with the average photon number in the modes at the transition frequency f21 as 2 8 f21 1 w (f21 ) = . hf21 hsi , with hsi = hsi = hf 3 21 c exp kT 1


A21 hsi . B12 With that relationship Eq.(3.54) can be rewritten as dn1 = dn2 =A21 n2 (hsi + 1) , dt Em dt Em w (f21 ) =

Or we can write


which indicates that the number of spontaneous emissions is equvalent to induced emissions caused by the presence of a single photon per mode. Having identied the coecients describing the transition rates in the atom interacting with the eld from equilibrium considerations, we can rewrite the rate equations also for the non equilibrium situation, because the coecients are constants depending only on the transition considered dn1 dn2 1 [(n2 n1 ) hsi + n2 ] . = = dt dt sp (3.65)

With each transition from the excited state of the atom to the ground state an emission of a photon goes along with it. From this, we obtain a change in the average photon number of the modes d hsi dn1 =V , dt dt (3.66)



which is

d hsi V = [(n2 n1 ) hsi + n2 ] . dt sp


Again the rst term describes the stimulated or induced processes and the second term the spontaneous processes. As we will see later, the stimulated emission processes are coherent with the already present radiation eld that is inducing the transitions. This is not so for the spontaneous emissions, which add noise to the already present eld. For n1 > n2 the stimulated processes lead to a decrease in the photon number and the medium is absorbing. In the case of inversion, n2 > n1 , the photon number increases exponentially. According to Eq.(3.59) inversion corresponds to a negative temperature, which is an indication for a non equilibrium situation that can only be maintained by additional means. It is impossible to achieve inversion by simple irradiation of the atoms with intense radiation. As we see from Eq.(3.67) in steady state the ratio between excited state and ground state population is n2 hsi = , n1 hsi + 1 (3.68)

which at most approaches equal population for very large photon number. However, such a process can be exploited in a three or four level system, see Figure 3.9, to achieve inversion.

Figure 3.9: Three level system: (a) in thermal equilibrium and (b) under optical pumping at the transition frequency f31 .



By optical pumping population from the ground state can be transfered to the excited level with energy E3 . If there is a fast relaxation from this level to level E2 , where level two in contrast has a long lifetime, it is conceivable that an inversion between level E2 and E1 can build up. If inversion is achieved radiation at the frequency f21 is amplied.


Matter Waves and Bohrs Model of an Atom

By systematic scattering experiments Ernest Rutherford showed in 1911, that the negative charges in an atom are homogenously distributed in contrast to the positive charge which is concentrated in a small nucleus about 10,000 times smaller than the atom itself. The nucleus also carries almost all of the atomic mass. Rutherford proposed a model of an atom where the electrons circle the nucleas similar to the planets circling the sun where the gravitational force is replaced by the Coulomb force between the electrons and the nucleus. This model had many short comings. How was it possible that the electrons, which undergo acceleration on their trajectory around the nucleus, do not radiate according to classical electromagnetism, loose energy and nally fall into the nucleus? Due to advances in optical instrumentation the light emitted from thermally excited atomic vapors was known to be in the form of discrete lines. Balmer found in 1885 that these lines could be expressed by the rule 1 = RH 1 1 2 2 2 n , with n = 3, 4, 5, ... (3.69)

where is the wavelength of light and RH = 10.968 m1 is the Rydberg constant for hydrogen. For n = 3 this corresponds to the red H -line at = 656.3nm, for n one obtains the wavelength of the limiting line in this series at = 364.6nm, see Figure 3.10.



Figure 3.10: Balmer series on a wave number scale. . In the subsequent spectroscopy work further sequences where found: 1. Lyman Series: 1 1 1 , with n = 2, 4, 5, ... (3.70) = RH 12 n2 2. Balmer Series: 1 = RH 3. Paschen Series: 1 = RH 4. Brackett Series: 1 = RH 5. Pfund Series: 1 = RH 1 1 42 n2 , with n = 5, 6, 7, ... (3.73) 1 1 2 2 2 n , with n = 3, 4, 5, ... (3.71)

1 1 2 2 3 n

, with n = 4, 5, 6, ...


1 1 2 2 5 n

, with n = 6, 7, 8, ...


The Lyman series in the UV-region of the spectrum, whereas the Pfund series is in the far infrared. These sequences can be represented as transitions between energy levels as shown in Figure 3.11.



Figure 3.11: Energy level diagram for the hydrogen atom. . In 1913, Niels Bohr found the quantization condition for the electron trajectories in the Hydrogen atom and he was able to derive from that the spectral series discussed above. He postulated that only those electron trajectories are allowed that within one rountrip around the nucleus have an action equal to a multiple of Plancks quantum of action h. I p ds = n h, with n = 1, 2, 3.... (3.75)

Second, he postulated that the electron can switch from one energy level or trajectory to another one by the emission or absorption of a photon with an



energy equivalent to the energy dierence between the two energy levels, see Figure 3.12. hf = E. (3.76)

Figure 3.12: Transition between dierent energy levels in the hydrogen atom. Assuming a circular trajectory of the electron with radius rn around the nucleus, the quantization condition for the electron trajectory (3.75) leads to mvn 2rn = nh, with n = 1, 2, 3... (3.77)

The other condition for radius and velocity of the electron around the nucleus is given by the equality of Coulomb and centrifugal force at radius rn , which leads to 2 e2 mvn = , (3.78) 2 40 rn rn or e2 2 vn . (3.79) = 40 rn m Substituting this value for the electron velocity in the squared quantization condition (3.77), we nd the radius of the electron trajectories 0 h2 2 n. (3.80) e2 m The radius of the rst trajectory, called Bohr radius is r1 = 0.529 1010 m. The velocities on the individual trajectories are rn = vn = e2 1 . 20 h n (3.81)



The highest velocity is found for the tightest trajectory around the nucleus, i.e. for n = 1, which can be expressed in terms of the velocity of light as 1 e2 v1 = c= c, 20 hc 137


1 where 2e0 hc = 137 is the ne structure constant. The energy of the electrons on these trajectories with the quantum number n is due to both potential and kinetic energy

Ekin = Epot or

1 2 me4 mvn = 2 2 2 , 2 80 h n 2 e me4 = = 2 2 2. 40 rn 40 h n

(3.83) (3.84)

En = Ekin + Epot me4 En = 2 2 2 . 80 h n

(3.85) (3.86)

Note, the energy of a bound electron is negative. For n , En = 0. The electron becomes detached from the atom, i.e. the atom becomes ionized. The lowest and most stable energy state of the electron is for n = 1 E1 = me4 = 13.53eV, 2 82 0h (3.87)

with correspondes to the ground state in hydrogen. When a transition between two of this energy eigenstates occurs a photon with the corresponding energy is released hf = Ek En , me4 1 1 = 2 2 . 80 h k2 n2 (3.88) (3.89)


Wave Particle Duality



Bohrs postulates were not able to explain all the intricacies of the observed spectra and they couldnt explain satisfactory the structure of the more complex atoms. This was only achieved with the introduction of wave mechanics. In 1923, de Broglie was the rst to argue that matter might also have wave properties. Starting from the equivalence principle of mass and energy by Einstein E0 = m0 c2 (3.90) 0, he associated a frequency with this energy accordingly f0 = m0 c2 0 /h. (3.91)

Since energy and frequency are not relativistically invariant quantities but rather components of a four-vector which has the particle momentum as its other components (E0 /c0 , px ,py , pz ) or ( 0 /c0 , kx ,ky , kz ), it was a necessity that with the energy frequency relationship E = hf = ~, (3.92)

there must also be a wave number associated with the momentum of a particle according to p = ~k. (3.93) In 1927, C. J. Davisson and L. H. Germer experimentally conrmed this prediction by nding strong diraction peaks when an electron beam penetrated a thin metal lm. The pictures were close to the observations of Laue in 1912 and Bragg in 1913, who studied the structure of crystaline and poly crystaline materials with x-ray diraction. With that nding the duality between waves and particles for both light and matter was established. Duality means that both light and matter have simultaneous wave and particle properties and it depends on the experimental arrangement whether one or the other property manifests itself strongly in the experimental outcome.


Appendix A: Mode Counting

It is interested to estimate the number of modes one is averaging over given a certain emitting surface and a certain measurement time, see Figure 3.13.



x Lx As y Lz Ly c Ac k AD z

Figure 3.13: Counting of longitudinal and transverse modes excited from a radiating surface of size As . If the area As is emitting light, it will couple to the modes of the free eld. To count the modes we put a large box (universe) over the experimental arrangement under consideration. The emitting surface is one side of the box. The light from this surface, i.e. specifying the transverse electric and magnetic elds, couples to the modes of the universe with wave vectors according to Eq.(3.5). Longitudinal Modes The number of longitudinal modes, that propagate along the positive zdirection in the frequency interval f can be derived from k = (2/Lz ) N and k = (2/c0 ) f Lz N = f, (3.94) c0 or using the propagation or measurement time over which the experiment extends = Lz /c0 , (3.95) we obtain for the number of longitudinal modes that are involved in the measurement that is carried out over a time intervall and a frequency range f N = f . (3.96)



Transverse Modes The free space modes that arrive at the detector area AD will not only have wave vectors with a z component, but also transverse components. Lets assume that the detector area is far from the emitting surface, and we consider only the paraxial plane waves. The wave vectors of these waves at a given frequency or free space wave number k0 can be approximated by 2m 2n with m, n = 0, 1, 2, ... (3.97) , , k0 kmn = Lx Ly where m and n are transverse mode indices. Then one mode occupies the volume angle c = 4 2 , 2 Lx Ly k0 = 2 0 /As .


If the modes are thermally excited, the radiation in individual modes is uncorrelated. Therefore, if there is a detector at a distance r then only the eld within an area (3.99) Ac = r2 c , is correlated. If the photodetector has an area Ad , then the number of transverse modes detected is Nt = Ad /Ac. The total number of modes detected is Ntot = Ad f = Ac Ad As f . r2 2 0 (3.101) (3.100)

Note, that there is perfect symmetry between the area of the emitting and receiving surface. The emitter and the receiver could both be black bodies. If one of them is at a higher temperature than the other, there is a net ow of energy from the warmer body to the colder body until equilibrium is reached. This would not be possible without interaction over the same number of modes. Thus the formula which is completely unrelated to thermodynamics is necessary to fulll one of the main theorems of thermodynamcis, that is that energy ows from warmer to colder bodies.




Appendix B: Shot Noise Formula

We obtain for the deviation in detected photon number sn over a measurement time T Z t+T In (t0 ) 0 sn (t) = dt (3.102) e t

which can be expressed as a linear ltering problem with the rectangular function G(t) T sn (t) = In (t) G(t) (3.103) e where (f ) = 1 (ejf T ejf T ) = sin(fT ) , (3.104) G j2f T (fT ) As we know from 6.011 the autocorrelation spectrum of a stationary random process at the output of a linear lter is given by T2 2 Ssn sn (f ) = 2 SIn In (f ) G(f ) (3.105) e and its variance is given by the integral over the autocorrelation spectrum 2 sn sn 2 T2 = sn = 2 e Z 2 SIn In (f ) G(f ) df (3.106)

T2 = 2 SIn In (f ) e

This variance should be equal to the total number of detected photons independent from the observation interval T 2 sn sn = or SIn In (f ) = eI. (3.109) T2 1 SIn In (f ) = hsi = 2 e T I T, e (3.108)

Z T2 1 2 G(f ) df = 2 SIn In (f ) . e T




[1] Fundamentals of Photonics, B.E.A. Saleh and M.C. Teich, [2] Optical Electronics, A. Yariv, Holt, Rinehart & Winston, New York, 1991. [3] Introduction to Quantum Mechanics, Griths, David J., Prentice Hall, 1995. [4] Quantum Mechanics I, C. Cohen-Tannoudji, B. Diu, F. Laloe, John Wiley and Sons, Inc., 1978.




Chapter 4 Schroedinger Equation

Einsteins relation between particle energy and frequency Eq.(3.92) and de Broglies relation between particle momentum and wave number of a corresponding matter wave Eq.(3.93) suggest a wave equation for matter waves. This search for an equation describing matter waves was carried out by Erwin Schroedinger. He was successful in the year 1926. The energy of a classical, nonrelativistic particle with momentum p that is subject to a conservative force derived from a potential V (r) is E= p2 + V (r) . 2m (4.1)

For simplicity lets begin rst with a constant potential V (r) = V0 = const. This is the force free case. According to Einstein and de Broglie, the dispersion relation between and k for waves describing the particle motion should be ~2 k2 ~ = (4.2) + V0 . 2m Note, so far we had a dispersion relation for waves in one dimension, where the wavenumber k(), was a function of frequency. For waves in three dimensions the frequency of the wave is rather a function of the three components of the wave vector. Each wave with a given wave vector k has the following time dependence ~k2 V0 ej(krt) , with = + (4.3) 2m ~ Note, this is a wave with phase fronts traveling to the right. In contrast to our notation used in chapter 2 for rightward traveling electromagnetic waves, we 193



switched the sign in the exponent. This notation conforms with the physics oriented literature. A superposition of such waves in kspace enables us to construct wave packets in real space Z (r, t) = k, ej(krt) d3 k d (4.4) The inverse transform of the above expression is Z 1 (r, t) ej(krt) d3 r dt, k, = (2 )4 with ! ~k2 V0 k, = (k) . 2m ~



Or we can rewrite the wave function in Eq.(4.4) by carrying out the trivial frequency integration over " ! #! Z ~k2 V0 (r, t) = (k) exp j kr + t d3 k. (4.7) 2m ~ Due to the Fourier relationship between the wave function in space and time coordinates and the wave function in wave vector and frequency coordinates k, (r, t) (4.8) we have (k, ) j (r, t) , t (4.9) (4.10) (4.11) (4.12) (4.13)

k (k, ) j (r, t) , k2 (k, ) (r, t) . where = ex + ey + ez , x y z 2 2 2 + + . x2 y 2 z 2

= 2 =

195 From the dispersion relation follows by multiplication with the wave function in the wave vector and frequency domain ~ (k, ) = ~2 k2 (k, ) + V0 (k, ) . 2m (4.14)

With the inverse transformation the corresponding equation in the space and tieme domain is j~ ~2 (r, t) = (r, t) + V0 (r, t) . t 2m (4.15)

Generalization of the above equation for a constant potential to the instance of an arbitrary potential in space leads nally to the Schroedinger equation j~ (r, t) ~2 = (r, t) + V (r) (r, t) . t 2m (4.16)

Note, the last few pages ar not a derivation of the Schroedinger Equation but rather a motivation for it based on the ndings of Einstein and deBroglie. The Schroedinger Equation can not be derived from classical mechanics. But classical mechanics can be rederived from the Schroedinger Equation in some limit. It is the success of this equation in describing the experimentally observed quantum mechanical phenomena correctly, that justies this equation. The wave function (r, t) is complex. Note, we will no longer underline complex quantities. Which quantities are complex will be determined from the context. Initially the magnitude square of the wave function | (r, t)|2 was interpreted as the particle density. However, Eq.(4.15) in one spatial dimension is mathematical equivalent to the dispersive wave motion Eq.(2.77), where the space and time variables have been exchanged. The dispersion leads to spreading of the wave function. This would mean that any initially compact particle, which has a well localized particle density, would decay, which does not agree with observations. In the framwork of the "Kopenhagen Interpretation" of Quantum Mechanics, whose meaning we will dene later in detail, | (r, t)|2 dV is the probability to nd a particle in the volume dV at position r , if an optimum measurement of the particle position is carried out at time t. The particle is assumed to be point like. (r, t) itself is then considered to be the probability amplitude to nd the particle at position r at time t. Note, that the measurement of physical observables like the position of a particle plays a central role in quantum theory. In contrast to classical



mechanics where the state of a particle is precisely described by its position and momentum in quantum theory the full information about a particle is represented by its wave function (r, t). (r, t) enables to compute the outcome of a measurement of any possible observable related to the particle, like its position or momentum. Before, we discuss this issue in more detail lets look at a few examples to get familiar with the mathematics of quantum mechanics.


Free Motion

Eq.(4.15) describes the motion of a free particle. For simplicity, we consider only a one-dimensional motion along the x-axis. Initially, we might only know the position of the particle with nite precision and therefore we use a Gaussian wave packet with nite width as the initial wave function x2 (x, t = 0) = A exp 2 + jk0 x . (4.17) 4 0 The probability density to nd the particle at position x is a Gaussian distribution x2 2 2 | (x, t = 0)| = |A | exp 2 , (4.18) 2 0 2 0 is the variance of the initial particle position. Since the probability to nd the particle somewhere must be one, we can determine the amplitude of the wave function by requiring Z 1 | (x, t = 0)|2 dx = 1 A = 4 2 0 (4.19)

The meaning of the wave number k0 in the wave function (4.17) becomes obvious by expressing the solution to the Schroedinger Equation by its Fourier transform
+ Z (x, t) = (k) exp [ j (kx (k) t)] dk


4.1. FREE MOTION or specically for t = 0

+ Z (x,0) = (k) e jkx dk ,



or 1 (k) = 2

For the initial Gaussian wavepacket of

+ Z (x,0) e jkx dx .


x2 (x, 0) = A exp 2 + jk0 x 4 0


we obtain

A0 2 . (4.24) (k ) = exp 2 0 (k k0 ) This is a Gaussian distribution for the wave number, and therefore momentum, of the particle with its center at k0 . With the dispersion relation =

~ k2 , (4.25) 2m with the constant potential V0 set to zero, the wave function at any later time is + Z ~k2 A0 2 2 exp 0 (k k0 ) j t + jkx dk. (4.26) (x, t) = 2m

This is exactly the same Gaussian integral we were studying for dispersive pulse propagation or the diraction of a Gaussian beam in chapter 2 which results in 2 ~ 2 2 2 2 0 k0 x 4 j 0 k0 x + j m t A . (4.27) (x, t) = q exp ~t ~ 2 1 + j m22 4 0 1 + j m22 t
0 0

As expected the wave packet stays Gaussian. The probability density is 2 k0 x ~m t |A|2 2 . (4.28) exp |(x, t)| = q 2 ~t ~t 2 1 + ( 2m 2 )) 2 0 1 + 2m2 0



With the value for the amplitude A according to Eq.(4.19), the wave packet remains normalized + Z |(x, t)|2 dx = 1. (4.29)

Using the probability distribution for the particle position, we obtain for its expected value + Z hxi = x |(x, t)|2 dx (4.30) ~ k0 t . (4.31) m Thus the center of the wave packet moves with the velocity of the classical particle ~ k0 0 = , (4.32) m which is the group velocity derived from the dispersion relation (4.2) E ( k ) (k ) 1 = . (4.33) 0 = k k=k0 ~ k k=k0 hxi = or

As we will see later, the expected value for the center of mass of the particle follows Newtons law, which is called Ehrenfests Theorem. For the uncertainty in the particle position q (4.34) x = hx2 i hxi2 follows for the freely moving particle s 2 ~t x = 0 1 + 2m2 0


The probability density for the particle position disperses over time. Asymptotically one nds ~t . ~t x = for 1 . (4.36) 2 2m0 2m2 0 Figure 4.1 (a) is a sketch of the complex wave packet and (b) indicates the temporal evolution of the average and variance of the particle center of mass



motion described by the complex wave packet. The wave packet disperses faster, if it is initially stronger localised.

Figure 4.1: Gaussian wave packet: (a) Real and Imaginary part of the complex wave packet. (b) width and center of mass of the wave packet. Example: Using this one dimensional model, we can estimate how rapidly an electron moves in a hydrogen atom. If we localize an electron in a box with a size similar to that of a hydrogen atom, i.e. 0 = a0 = 0.5 1010 m, without the presence of the proton that holds the electron back from escaping, it will only 2 31 take t = 2m 2 kg (0.5 1010 ) m2 /6.626 1034 Js = 46.5as 0 /~ = 2 9.81 10 (attoseconds=1018 sec) until its wave function disperses signicantly. This result indicates that electronic motion in atoms occurs on a attosecond time scale. Note, these time scales quickly become very long if macroscopic objects are described quantum mechanically. For example, for a particle with



a mass of 1g localized in a box with dimensions 1m, the equivalent time for signicant dispersion of the wave function is t = 2 1019 s = 2Million years. This result gives us a rst indication why we are far, far away from encountering quantum mechanical eects in our everyday life and why the mechanics of the micro cosmos, on an atomic or molecular level, is so different from our macroscopic experience. The reason is the smallness of the quantum of action h. The reason for this behaviour is that a well localized particle has a wider momentum or wave number distribution. This is in one to one analogy that an otpical pulse disperses faster in a medium with a given dispersion if it is shorter because of larger spectral width. The wave number spread is Z (k k0 )2 | (k)|2 dk Z . | (k)|2 dk 1 , 2 0 ~ . 2 0 (4.37)

(k)2 =

Here, we have k = or for the momentum spread p =



The position-momentum uncertainty product is then s 2 ~t ~ 1+ . p x = 2 2m2 0


The position-momentum uncertainty product is a minimum at t = 0 and steadily increases from this initial value. As we will show later it is in generally true that the position-momentum uncertainty product satises the condition ~ xi pi > . (4.41) 2 Note, that the index i indicates the coordinate x, y or z. This is Heisenbergs uncertainy relation between particle position and moment, which holds for

4.2. PROBABILITY CONSERVATION AND PROPABILITY CURRENTS201 each component individually. Later, we will nd other pairs of physical observavles, which are called conjugate observables and which satisfy similar uncertainty relations. The product of such quantities is always an action. This is for example also true for the product of energy and frequency and the resulting energy-time uncertainty relation is E t > ~ . 2


Note, whereas the position-momentum uncertainy is related to the choice of the particle state described by the wave function, the energy-time uncertainty relation is related to the dynamics of a quantum process. What it means is that a quantum system can only change its state signicantly within a time span t, if the state, the quantum system is in, has an energy uncertainty larger than E > 2~ . t Position and momentum variables that do not belong to the same degree of freedom, such as y, and px are not subject to an uncertainty relation.


Probability Conservation and Propability Currents

Max Born was the rst to introduce the propabilistic interpretation of the wave function found by Schroedinger, that is the propability to nd the center of mass of a particle at position r in a volume element dV is given by the magnitude square of the wave function multiplied by dV p (r, t) = | (r, t)|2 d V . (4.43)

If this interpretation makes sense, then the total propability that the particle can be found somewhere should by 1 and this normalization should not change during the dynamics. We found that this is true for the Gaussian wave packet undergoing free motion. Here, we want to show that this is true under the most general circumstances. We look at the rate of change of the



probability to nd the particle in an arbitrary but xed volume V = V ol Z d p (r, t) d3 r = (4.44) dt V ol Z | (r, t)|2 d3 r = dt ZV ol (r, t) (r, t) + (r, t) (r, t) d3 r = t t V ol Using the Schroedinger Equation (4.16) for the temporal change of the wave function we obtain Z d p (r, t) d3 r = dt V ol Z ~ j (r, t) V (r) (r, t) (r, t) d3 r (4.45) = j 2 m ~ V ol Z j ~ + (r, t) (r, t) + V (r) (r, t) d3 r j 2 m ~ V ol Since the potential V (r) is real the terms related to it cancel. The other two terms can be written as the divergence of a current density Z Z 3 p (r, t) d r = J (r, t) d3 r, (4.46) t V ol V ol with J (r, t) = ~ ( (r, t) ( (r, t)) (r, t) ( (r, t))) . j 2m (4.47)

Eq.(4.46) is true for any volume, i.e. Z p (r, t) + J (r, t) d3 r = 0, t V ol which is only possible if the integrand vanishes p (r, t) = J (r, t) . t



Clearly, J (r, t) has the physical meaning of a probability current. The probability in a volume element changes because of probablity owing out of

4.2. PROBABILITY CONSERVATION AND PROPABILITY CURRENTS203 this volume element. Note, this is the same local law that we have for the conservation of charge. In fact, if the particle is a charged particle, like an electron is, multiplication of J with e0 would result in the electrical current associated with the wave function (r, t) . Gausss theorem states Z Z 3 J (r, t) d r = J (r, t) dS, (4.50)
V ol S

where V ol is the volume over which the integration is carried out and S is the surface that encloses the volume with dS an outward pointing surface normal vector. With Gausss theorem the local conservation of probability can be transfered to a global result, since Z Z Z d 3 3 (4.51) p (r, t) d r = J (r, t) d r = J (r, t) dS. dt V ol V ol S
If we choose as the volume the whole space and if (r, t) and t (r, t) vanish rapidly enough for r such that the probability current vanishes at innity, the total probability is conserved. These ndings proove that the probabilty interpretation of the wave function is a valid interpretation not contradicting basic laws of probability. If the wave function properly normalized at the beginning it will stay normalized.

Example The Gaussian wave packet satises the condition that the probability current decays rapidly enough at the surface of a large enough chosen volume so that the normalization is preserved. A monochromatic plane wave does not satisfy this condition. However, the probability current density gives a physical meaning to it. The wave function corresponding to a plan wave (r, t) = ej(krt) , with = ~k 2 V0 + 2m ~ (4.52)

which is not normalizable, results in a homogenous probability current J (r, t) ~ ( (r, t) ( (r, t)) (r, t) ( (r, t))) (4.53) j 2m p ~k = | (r, t)|2 = | (r, t)|2 = v | (r, t)|2 , m m =



that is identical to the classical velocity of the particle. Thus a plane wave describes a particle with a precise velocity or momentum but completely unknown position, therefore the related probability current density is completely homogenous but directed into the direction of v. Such waves describe the initial state in a scattering experiment, where we shoot particles with a 2 ~p2 ~k2 precisely dened velocity v or momentum p or energy E = mv =2 =2 2 m m onto another object described by a scattering potential, see problem set. The position of these particles is completely unspecied, i.e. | (r, t)|2 =const.


Measureability of Physical Quantities (Observables)

The reason for the more intricate description necessary for microscopic processes in comparison with macroscopic processes is simply the fact that these systems are so small that the interaction of the system with an eventual measurement apparatus can no longer be neglected. It turns out this fact is not to overcome by choosing more and more sophisticated measurement apparati but rather is a principle limitation. If this is so, then it eventually doesnt make sense or it becomes even inconsistent to attribute to a system more precisely dened physical quantities than actually can be retrieved by measurements. This is the physical reason behind the introduction of the wave function instead of the precisely dened position and momentum of the particle that we used to deterministically predict the trajectory of a particle in an external eld. It is impossible to assign to a microscopic particle a precise position and momentum at the same time. To demonstrate this, we consider the following (Heisenberg) microscope to measure the exact position of a particle. We use light with wavelength and focus it strongly with a lense of some focal distance d, see Figure 4.2. From our construction of the Gaussian beam in section 2.5.2, we found that if we generate a focused beam with a waist wo having a Rayleigh range 2 o , the beam is composed of plane waves which have a Gaussian distrizR = w 2 bution in its transverse k-vector, which has a variance kT /2, see Eq.(2.268).
2 kT (4.54) 2 The Rayleigh range of the beam is related to the transverse wave number

k =


2 spread of the beam by zR = k0 /kT , with k0 = 2/, see (2.269) and there2 after. Note, the intensity prole of the beam has a variance wo /4. If a particle crosses the focus of the beam and scatters a single photon, which we detect with the surrounding photo detector arrangement, then it is reasonable to assume that we know the position of the particle in the x-direction, with an uncertainty equal to the uncertainty in the transverse photon or intensity distribution of the beam, i.e. x = wo /2.


Weak particle beam with precise momentum p

Figure 4.2: Determination of particle position with an optical microscope. A weak particle beam with precisely dened moment p of the particles is directed towards the focus of the Gaussian beam. In the focus the particle scatters at least one photon. Detection of the scattered photon with the surrounding photodetector signals, that the position of the particle in xdirection has been determined within the beam waist of the Gaussian beam. However, due to the scattering of the photon a momentum uncertainty has been introduced to the particle state. During the measurement, the photon recoil induces a momentum kick with an uncertainty px = ~kT / 2. So even if the momentum of the particle was perfectly know before the measurement, after the additional determination of its position with a precision x it has at least aquired an uncertainty in its momentum of magnitude px . The product of the uncertainties in postion and momentum after the measurement is



~ wo ~ ~kT wo px x = = = . (4.55) wo 2 2 2 2 Note, this result is exact and is independent of focusing. Tighter focusing will enable us to more precisely determine the position of the particle, but we will introduce more momentum uncertainty due to the photon recoil; the opposite is true for less focusing. Since we can not determine, and therefore, prepare a particle in a state with its position and momentum more precisely determined than this uncertainty product allows, there is no such state and (4.55) is the minimum uncertainty product achievable. The experimental setup can easily be extended to measure the momentum and position of a particle in all three dimensions. For example one can use three focused laser beams at dierent wavelength, which are orthogonal to each other. Once a particle will y through the focus and scatters three photons, each of dierent color. If we knew its momentum initially precisely, we would know afterwards its 3-dimensional position with a position and momentum spread as described by Eq.(4.55).


Stationary States

One of the great mysteries before the advent of quantum mechanics was the orgin of the discrete energy spectra observed in spectroscopic investigations and empirically described by the Bohr-Sommerfeld model of the atom. This mystery is easily explained by the Schroedinger Equation (4.16) (r, t) ~2 = (r, t) + V (r) (r, t) . t 2m It allows for solutions (r, t) = (r) ej t , j~ which have a time independent probability density, i.e. | (r, t)|2 = | (r)|2 = const., (4.58) (4.56) (4.57)

which is the reason for calling these states stationary states. Since the right side of the Schroedinger Equation is equal to the total energy of the system, these states correspond to energy eigenstates of the system with energy eigenvalues E = ~. (4.59)



These energy eigenstates (r) are eigen solutions to the stationary or time independent Schroedinger Equation ~2 (r) + V (r) (r) = E (r) . 2m (4.60)

We get familiar with this equation by considering a few one-dimensional examples, before we apply it to the Hydrogen atom.


The One-dimensional Innite Box Potential

A simple example for a quantum mechanical system is an electron that can freely move in one dimension x but only over a nite distance a. Such a situation closely describes an electron that is strongly bound to a molecule with a cigar like shape with length a. The potential describing this situation is the one-dimensional box potential 0, for |x| < a/2 V (x) = , (4.61) , for |x| a/2 see Figure 4.3.

Figure 4.3: One dimensional box potential with innite barriers. In the interval [a/2, a/2] the stationary Schroedinger equation is ~2 d2 (x) = E (x) . 2m dx2 (4.62)



For |x| a/2 the wave function must vanish, otherwise the energy eigenvalue can not be nite, i.e. (x = a/2) = 0. This is analogous to the electric eld solutions for the TE-modes for a planar mirror waveguide and we nd r 2 nx n (x) = cos for n = 1, 3, 5 . . . , (4.63) a a r 2 nx n (x) = sin for n = 2, 4, 6 . . . . (4.64) a a The corresponding energy eigenvalues are En = n2 2 ~2 . 2ma2 (4.65)

We also nd that the stationary states constitute an orthogonal system of functions + Z m (x) n (x) dx = mn . (4.66)

In fact this system is complete. Any function in the interval [a/2, a/2] can be expanded in a superposition of the basis functions n (x), which is a Fourier series X cn n (x) (4.67) f (x) =


cm =


m (x) f (x) dx,



which is a consequence of the orthogonality relation (4.66). Example: If we approximate the binding potential of a hydrogen atom by a one-dimensional box potential with a width equal to twice the Bohr radius a = 2a0 = 1010 m, the energy eigenvalues are En = n2 35eV. Clearly, the spacing of the energy eigenvalues does not conform with what has been observed experimentaly, compare with section 3.5, however the energy scale is within an order of magnitude. The ionization potential of the hydrogen atom is 13.5eV .




The One-dimensional Harmonic Oscillator

The most important example of a quantum system is the one-dimensional harmonic oscillator. It is the most basic mechanical and electrical system and it describes the dynamics of a mode of the radiation eld, see Figure 4.4.

Figure 4.4: Elastically bound particle Mechanically, a harmonic oscillation comes about by the elastic force obeying Hooks law F (x) = Kx, (4.69) that pulls back a particle with mass m in its equilibrium position. This force is conservative and can be derived from a potential by F (x) = with V (x) = d V (x) , dx (4.70)

1 Kx2 . 2 Newtons law results in the classical equation of motion mx = F (x) ,


(4.72) (4.73)

or x + 2 0 x = 0, with the oscillation frequency 0 = r K m (4.74)



The corresponding stationary Schroedinger Equation is d2 (x) 2m 1 2 (x) = 0. + 2 E Kx dx2 ~ 2


This equation is well known in mathematical physics and we want to bring it into standardized form by the scale transformation, i.e. introducing a normalized distance = ax, (4.76) with the scale factor a= mK ~2 1 4 = r 0m = ~ r K . ~ 0 (4.77)

In addition we introduce the energy scale factor = 2E . ~0 (4.78)

Then the stationary Schroedinger Equation for the harmonic oscillator is d2 ( ) + 2 ( ) = 0. 2 d (4.79)

It turns out [1][6], that this equation has only solutions that are bounded, i.e. ( ) = 0, if the normalized energies are n = 2n + 1. And the corresponding eigensolutions are the Hermite Gaussians, n ( ) = const. Hn ( ) e 2



which we discovered already as solutions of the paraxial wave equation, see Eqs.(2.312) and (2.313), i.e. Hn ( ) = (1)n e H0 ( ) = 1 , H1 ( ) = 2 , H2 ( ) = 4 2 2

dn 2 e d n


H3 ( ) = 8 3 12 , H4 ( ) = 16 4 48 2 + 12 , H5 ( ) = 32 5 160 3 + 120 .




After denormalization of x and normalization of the stationary wave functions are r a 1 a2 x2 2 n (x) = H ( ax ) e . (4.84) n n 2 n! Again, we nd that the Hermite Gaussians constitute an orthogonal system of functions such that + Z m (x) n (x) dx = mn . (4.85)

Figure 4.5 shows the rst six stationary states or energy eigenstates of the harmonic oscillator.

Figure 4.5: First six stationary states of the harmonic oscillators. The energy eigenvalues of the stationary states are 1 ~ 0 . En = n + 2


Note, that the energy eigenvalues are equidistant and the dierence between two energy eigenstates follows the ndings of Planck. An oscillator has discrete energy levels which dier by energy quanta of size ~ 0 , see Figure 4.6.



Figure 4.6: Lowest order wavefunctions of the harmonic oscillator and the corresponding energy eigenvalues [3]. The only dierence is, that the whole energy scale is shifted by the energy of half a quantum, which is the lowest energy eigenvalue. Thus the minimum energy, or ground state energy, of a harmonic oscillator is not zero but E0 = 1 ~0 .It is obvious, that an oscillator can not have zero energy because its 2 energy is made up of kinetic and potential energy E= 1 p2 + Kx2 . 2m 2 (4.87)

, Since every state has to fulll Heisenbergs uncertainty relation p x ~ 2 one can show that the state with minimum energy possible has an energy E0 = 1 ~ 0 , which is true for the ground state 0 (x) according to Eq.(4.84). 2 The stationary states of the harmonic oscillator correspond to states with precisely denied energy but completely undened phase. If we assume a classical harmonic oscillator with a well dened energy E = 1 Kx2 0 . Note, 2 that during a harmonic oscillation the energy is periodically converted from potential energy to kinetic energy. Then the oscillator oscillates with a xed ampltiude x0 x(t) = x0 cos ( 0 t + ) . (4.88) If the phase is assumed to be random in the interval [-, ], one nds for the

4.5. THE HYDROGEN ATOM probability density of the position x to be d 2 1 p (x) = p() dx 2 = 2 x0 sin (0 t + ) 1 p . = 2 x2 0x



Figure 4.7 shows this probability density corresponding to an energy eigenstate n (x) with quantum large quantum number n = 10.

Figure 4.7: Probability density |10 |2 of the harmonic oscillator containing exactly 10 energy quanta. On average, the quantum mechanical probability density agrees with the classical probability density, which is some form of the correspondence principle, which says that for large quantum numbers n the wave functions resume classical properties.


The Hydrogen Atom

The simplest of all atoms is the Hydrogen atom, which is made up of a positively charged proton with rest mass mp = 1.6726231 1027 kg, and a negatively charged electron with rest mass me = 9.1093897 1031 kg. Therefore, the hydrogen atom is the only atom which consists of only two



Figure 4.8: Bohr Sommerfeld model of the Hydrogen atom. particles. This makes an analytical solution of both the classical as well as the quantum mechanical dynamics of the hydrogen atom possible. All other atomes are composed of a nucleus and more than one electron. According to the Bohr-Somerfeld model of hydrogen, the electron circles the proton on a planetary like orbit, see Figure 4.8.The stationary Schroedinger Equation for the Hydrogen atom is 2m0 (E V (r)) (r) = 0 (4.90) ~2 The potential is a Coulomb potential between the proton and the electron such that e2 0 (4.91) V (r) = 4 0 |r| (r) + and the mass is actually the reduced mass m0 = mp me mp + me (4.92)

that arises when we transform the two body problem between electron and proton into a problem for the center of mass and relative coordinate motion. Due to the large, but nite, mass of the proton, i.e. the proton mass is 1836 times the electron mass, both bodies circle around a common center of mass. The center of mass is very close to the position of the proton and the reduced mass is almost identical to the proton mass. Due to the spherical symmetry of the potential the use of spherical coordinates is advantageous. The Laplace operator in spherical coordinates

4.5. THE HYDROGEN ATOM is 2 2 2 1 1 1 = , + 2 sin + + r2 r r r sin sin2 2



and the stationary Schroedinger Equation takes on the form 2 2 1 2 1 2m0 e2 1 0 + 2 E+ + sin + = 0. + r2 r r r2 sin ~ 4 0 r sin2 2 (4.94)


Ground State

Before, we look at the general solutions for the stationary Schroedinger Equation for the hydrogen atom, lets nd the most simple solution. We look rst for a radially symmetric solution (r, , ) = 1 (r) , which leads to 2 e2 2 2m0 0 (r) + 2 E + 1 (r) = 0. + r2 r r 1 ~ 4 0 r 2 2m0 1 (r) + 2 E 1 (r) = 0. 2 r ~


Asymptotically for large r, this equation approaches the form (4.96)

Since, the energy eigenvalues E, which we expect, will be negative, the solution of this equation is an exponential and we choose a decaying exponential that will lead to a normalizable wave function for a bound electron. r 2m0 E 1 (r) = er , with = 2 (4.97) ~ Substituting this trial solution into 4.96, we nd in addtion 1 m0 e2 0 1 (r) = 0. 2 + r 2 0 ~2 This equation is only fullled for all r if = m0 e2 0 1 = r1 , 0 h2 (4.99)




0 h where r1 = e 0.529 1010 m, is the Bohr radius from Bohrs models 2m 0 of the atom, see Eq.(3.80). And the energy of the ground state follows from (4.97) to be

E1 =

~2 me4 = = 13.53eV 2 2 2m0 r1 82 0h


in agreement with (3.87). We nally normalize the wave function, i.e. we introduced a normalization constant C 1 (r) = C er , determined such that Z which leads to C= s 1 . 3 r1 (4.103)


|1 (r)|2 4r2 dr = 1,


Thus in total the properly normalized groundstate wave function of hyrdogen is 1 (r) = s 1 r/r1 e , 3 r1 (4.104)

Figure 4.9 shows the magnitude square of the hydrogen ground state wave function |1 (r)|2 and the corresponding propability density to nd the electron in a radial interval [r, r + dr].




Figure 4.9: Sketch of the magnitude square of the hydrogen ground state. wave function and of the propability density to nd an electron in a spherical shell with radius r and thickness dr.




Excited States

To nd the excites states of the hydrogen atom, we need to allow for a more general solution than a spherical symmetric solution. We nd a general solution of the stationary Schroedinger Equation by using separation of variables (r, , ) = R (r) Y (, ) . (4.105)

Substituting this trial wave function into the Schroedinger Equation using d2 R 2 dR 1 2 + = 2 r R , dr2 r dr r r r multiplying by r2 sin2 and deviding by R (r) Y (, ) , we obtain from 4.93 1 e2 2m0 r2 0 2 E + R + r = R r r ~2 4 0 r 2Y 1 Y 1 1 sin + (4.106) Y sin sin2 2 The left side of this equation is only a function of the radius r, and the right side of the angles and . Therefore, this equation can only be fullled if each side is equal to a constant number C, that is e2 2m0 r2 1 0 2 E + r R + = C, (4.107) R r r ~2 4 0 r 2Y 1 1 Y 1 sin + = C. (4.108) Y sin sin2 2 Spherical Harmonics The solutions to the angular part of the Schroedinger equation (4.108) 2Y Y 1 1 sin + + CY = 0. (4.109) sin sin2 2 are called spherical harmonics, see Appendix A to this chapter s (2l + 1) (l |m|)! m jm P1 (cos ) e . (4.110) Y1m (, ) = (1)m 4 (l + |m|)!



In this solution, l and m are quantum numbers, whose physical meaning we will discuss later, but from the solution, we see immediately, that m determines the azimutal part of the spherical harmonic and l the polar part. These numbers may take on the following values

l = 0, 1, 2, ... positive whole number,


m = 0, 1, 2, ..., l.


m The functions P1 (cos ) are called associated Legendre Polynomials. The m Y1 (, ) are called normalized spherical harmonics and play an important role whenever a partial dierential equation that contains the Laplace operator is solved in spherical coordinates. The spherical harmonics form a system of orthogonal functions on the full volume angle 4, i.e. [0, ] and [, ] 2 Z Z 0 0

Ylm (, ) Ylm (, ) sin d d = ll0 , mm0 . 0


Therefore, a function of the angular variable (, ) can be expanded in spherical harmonics. The spherical harmonics with negative azimuthal number -m can be expressed in terms of those with positive azimuthal number m. Y1m (, ) = (1)m (Ylm (, ))


The lowest order spherical harmonics are listed in Table 4.1. Figure 4.10 shows a cut through the spherical harmonics Y1m (, ) along the meridional plane.


CHAPTER 4. SCHROEDINGER EQUATION q q Y11 (, ) = 83 sin ej ,

1 Y0 0 (, ) = 4 ,

Y10 (, ) =

3 4

cos ,

q q q 1 5 15 j , Y2 (, ) = 15 sin2 e2j , 2 ( , ) = (3 cos 1) , Y ( , ) =sin cos e Y0 2 2 2 16 8 32 Y0 3 Y2 3 q 7 (, ) = 16 (5 cos3 3 cos ) , q 105 (, ) = 32 sin2 cos ej2 , Y1 3 Y3 3 q 35 (, ) =- 64 sin3 ej3 q 21 (, ) =- 64 sin (5 cos2 1) ej , .

Table 4.1: Lowest order spherical harmonics

Figure 4.10: Lowest order spherical harmonics Y1m (, ) , along the meridional plane, i.e. = 0.

4.5. THE HYDROGEN ATOM Radial Hydrogen Wave Functions


From the sperical harmonics one nds, that the constant C depends only on the polar quantum number l C = l(l + 1), and the radial equation becomes d2 R 2 dR 2m0 E l (l + 1) m0 e2 0 + R = 0. + + dr2 r dr ~2 20 ~2 r r2 (4.115)


The radial equation has in addition to the 1/r -Coulomb potential aquired an additional term that can be traced back to the additional centrifugal energy of the electron, related to the angular momentum of the electron orbiting around the nucleus. Erot = ~2 l (l + 1) L2 = . 2m0 r2 2m0 r2 (4.117)

One nds that the radial equation (4.116) has only bound state solutions, i.e. E < 0, for discrete values of the energy equal to m0 e4 1 E = En = 2 2 2 , 80 h n with n = 1, 2, ...and n l + 1. The corresponding radial wave functions take on the form s 2 (n l 1)! 3/2 /2 r1 1 L21+1 , Rn1 (r) = 2 nl1 () e n [(n + l)!]3

Obviously, the spherical harmonics are related to the angular momentum L of the particle, because Erot is the energy of a particle due to its angu p lar momentum L = l (l + 1)~ and the moment of inertia m0 r2 . Thus quantum mechanically, the particle can no longer access arbitrary values for the angular momentum. The angular momentum can only have values p L = l (l + 1)~ with l = 0, 1, 2, .... (4.118)




2r where r1 is the Bohr radius, = nr is the normalized radius and L21+1 nl1 () 1 is the Laguerre polynomial [5], which can be expressed as

Lr s (x) =

s X q =0


xq (s + r)!2 . (sq ) ! (r + q )! q !


The forefactors in the radial wave functions are chosen such that they are normalized when weighted by the radial weight r2 Z

Rnl (r) Rn0 l (r) r2 dr = n,n0 ,


The lowest order Laguerre Polynomials are summarized in Table 4.2 The

L1 0 (x) = 1 ,

L1 1 (x) = 4 2x , ,

2 L1 2 (x) = 18 18x + 3x

2 3 L1 3 (x) = 96 144x + 48x 4x

L2 0 (x) = 2 , ,

L2 1 (x) = 18 6x , L3 3 (x) = 6 , L4 0 (x) = 24 .

2 L2 2 (x) = 144 96 + 12x

L3 1 (x) = 96 24x ,

Table 4.2: Lowest order Laguerre Polynomials

radial wave functions of the hydrogen atom are listed in Table 4.3 and plots of the lowest order radial wave functions are presented in Figure 4.11



2 R10 (r) = er/r1 , 3 r1

R20 (r). =

1 3 2 2 r1

r r1


R21 (r) = R30 (r) = R31 (r) =

1 r r/2r1 3 e r1 2 6 r1

1 3 81 3 r1

2 r 27 18 r1 + 2 rr1 er/3r1 6
r r1

4 3 81 6 r1

r r/3r1 e , r1

R32 (r) =

4 3 81 30 r1

r r1


Table 4.3: Lowest order radial wavefunctions Rn,l (r).

Figure 4.11: Radial wavefunctions Rnl (r) of the hydrogen atom.

In total we found the stationary states, or the energy eigenfunctions, of



the hydrogen atom. Those are

nlm (r, , ) = Rnl (r) Ylm (, ) .


The lower order wave functions are listed in Table 4.4 and plots of the resulting probability densities of the lowest order energy eigenstates of the hydrogen atom are shown in Figure 4.12

100 (r, , ) = 200 (r, , ) = 210 (r, , ) =

1 r/r1 3e r1 1 3 4 2 r1

r r1


1 r r/2r1 3 e r1 4 2 r1 1 r r/2r1 3r e 1 8 r1

cos sin ej ,

211 (r, , ) = 300 (r, , ) =

1 3 81 3 r1

2 r 27 18 r1 + 2 rr1 er/3r1

Table 4.4: Lowest order hydrogen wavefunctions n,l,m (r, , ).

For each value of l, there are 2l + 1 possible values for the azimutal quantum number m. These are l, (l 1), ...., 1, 0, 1, ..., l 1, l. As a result, there are n2 dierent stationary states for each value of n. For historic reasons the states with l = 0 are called s-states (sharp), the states with l = 1 are called p-states (principle), the states with l = 2 are called d-states (diuse) and the states with l = 3 are called f-states (fundamental). Figure 4.13 shows the surfaces of equal propability density for the rst three excited states (n=2) of the hydrogen atom in 3 dimensions.



Figure 4.12: Probability densities of the lowest order hydrogen wavefunctions. (The density is presented along the meridial plane).



310 (r, , ) =

1 3 81 r1

r r1

er/3r1 cos
r r/3r1 e r1

311 (r, , ) = 320 (r, , ) =

1 3 81 r1

r r1

sin ej

1 r2 r/3r1 3 2e r1 81 6 r1 1 r2 r/3r1 3 r2 e 81 r1 1

(3 cos2 1) sin cos ej , sin2 e2j

321 (r, , ) = 322 (r, , ) =

1 r2 r/3r1 3 2e r1 162 3 r1

Table 4.5: Lowest order hydrogen wavefunctions n,l,m (r, , ).continued.




Figure 4.13: Surfaces of constant propability density for the rst three excited states (n=2) of the hydrogen atom.


Energy Spectrum of Hydrogen

We havent yet discussed the energy eigenspectrum of hydrogen following from the solution of the stationary Schroedinger Equation. From Eqs.(??) and (3.87) we nd this to be E= m0 e4 1 , 2 2 8 2 0h n (4.123)



which also agrees with the energy spectrum of the Bohr-Sommerfeld model, see section 3.5. The lowest energy eigenstate is

E1 =

m0 e4 = 13.7eV. 2 8 2 0h


The energy eigenvalues constitute a sequence that converges for large n towards 0, which corresponds to removing the electron from the atom. The energy to do so is E E1 = 13.7eV. Figure 4.15 shows the energy levels and the term diagram of the hydrogen atom and how the Lyman, Balmer, Paschen, Brackett and Pfund series arise from it. Each wavefunction is uniquely described by the set of quantum numbers (n,l,m). The rst quantum number n species the energy eigen value En . As we will show in problem sets, the second quantum number l determines the eigenvalue of the squared angular momentum operator L2 with eigenvalues

L2 nlm (r, , ) = l(l + 1)~2 nlm (r, , ) ,


which determines the rotational energy of the corresponding stationary state. And the third quantum number m detemines the eigenvalue of the operator describing the z-component of the angular momentum operator

Lz nlm (r, , ) = m~ nlm (r, , ) .


The 2l + 1 dierent values for m, describe the possible values for the z component of the angluar momentum for a given value of l. Figure 4.14. shows the possible values for the z-component of the angular moment vector for l = 1 and l = 2.



Figure 4.14: Possible values for the z-component of the angular momentum vector for the cases l = 1 and l = 2.

Figure 4.15: Energy levels and term diagram for the hydrogen atom [3] In fact, the description of the electron wave functions is not yet complete, because the electron has an internal degree of freedom, that is its spin. The spin is an internal angular momentum of the electron that carries a magnetic moment with it. The Stern-Gerlach experiment shows that this degree of



freedom has two eigenstates, i.e. the spin can be oriented parallel or antiparallel to the direction of an applied magnetic eld. The values of the internal angluar mometum with respect to the quantization axis dened by an external eld, that shall be chosen along the z-axis, are s = ~/2. Thus the energy eigenstates of an electron in hydrogen are uniquely characterized by four quantum numbers, n, l, m, and s. As Figure 4.15 shows, the energy spectrum is degenerate, i.e. for n > 1, there exist to each energy eigenvalue several eigenfunctions, that are only uniquely characterized by the additional quantum numbers for angular momentum and spin. This is called degeneracy because there exist to a given energy eigenvalue several physically dierent eigen states.


Superposition States, Radiative Transitions and Selection Rules

If the atom is in one of its energy eigenstates it is obvious by symmetry that the average position of the electron calculated via Z hri = r | (r, t)|2 d3 r = 0 (4.127)

vanishes. Then also the average dipole moment p = e hri vanishes and, therfore, the atom does not radiate in a stationary state as postulated in the Bohr model. Here, it is a natural outcome. No postulate is needed. However, when the atom is in a superposition state, for example between ground state and the rst excites states, the average position of the electron or dipole moment of the atom does not any longer vanish and the atom has a dipolement that radiates like a classical dipole does. For example Figure 4.16 shows the charge distribution as a function of time for an atom in the superposition state between the 1s-groundstate and the 2p(m = 0) excite state 1 1 1s (r, t) + 2p,m=0 (r, t) = (100 (r, t) + 210 (r, t)) = . (4.128) 2 2 1 r r/2r1 1 r/r1 jE1 t/~ jE2 t/~ . e + e cos e = p 3 e 4 2 r1 2 r1 In the propability density, i.e. the magnitude square of the wave function the contributions between the ground state and excited state interfer positively



or negatively depending on the relative phase between the two wavefunctions, which depends on the phase angle Et/~, with E = E2 E1 ..

Figure 4.16: Charge distribution in the superposition state 1 1s (r, t) + 2p,m=0 (r, t) for dierent times. The atom have a dipole 2 moment in the z-direction [6]. Figure 4.16 shows the charge distribution as a function of time for an atom in the superposition state between the 1s-groundstate and the 2p(m = 1) excite state. Instead of an oscillating charge distribution, the atom shows now a rotating dipole, which emittes a circular polarized electromagnetic wave. 1 1 1s (r, t) + 2p,m=1 (r, t) = (100 (r, t) + 211 (r, t)) . (4.129) 2 2 1 r r/2r1 1 r/r1 jE1 t/~ j jE2 t/~ . e + e sin e e = p 3 e 8 r1 2 r1

Figure 4.17: Charge distribution in the 1 ( r, t ) + for dierent times. 1s 2p,m=1 (r, t) 2 dipole moment rotating in the x-y plane [6].

superposition state The atoms show a



Figure 4.18 summarizes the polarization of the electromagnetic eld emitted from a superposition state with an oscillating dipole in the z-direction (m = 0) (a) and a rotating dipol in the x-y-plane (m = 1)(b). The transition with (m = 1) emittes polarized light with the opposite circularlity then the transition with (m = 1).

Figure 4.18: Polarization of the associated electromagnetic radiation emitted during the transition of the atom for (a) an oscillating dipole and (b) from a rotating dipole [6]. If an atomic gas in termal equilibrium no orientation of the atom is prefered and it emittes light from all possible transitions equally and, therefore, the light is not polarized. However, if the atom is put into a magnetic eld, which is oriented in the z-direction, the energy levels for the dierent p-



states split depending on the magnetic quantum number m. This eect is called Zeeman eect. Because, with the orbital angular momentum L of the electron state there is an electrical current associated with it, that leads to a magnetic moment of that particular state and with it to en energy shift of the corresponding energy level, see Figure 4.19. Then the light emitted at the dierent frequencies is polarized correspondingly. Or light with a given polarization interacts only with the corresponding transition, where it can induce the corresponding dipolemoment in the atom.

Figure 4.19: Zeeman eect and related energy levels for the m=0 and m=+/1 transitions.



As we will see in more detail later, the interaction strength between light and an atom via dipole transistions between energy eigenstates a and b is determined by the transition matrix element Z 3 (4.130) Mab = e0 a (r ) r b (r ) d r

As we have seen for the possible dipole transitions between states with principle quantum number n=1 and n=2, the transistions are related to a change in orbital quantum number l = 1 and m = 0, 1. It turns out that this is in generel the selection rule for dipole radiation. Finally, we schould mention, that also higher order transitions are possible, such as quadrupole transitions. However, at optical frequences these transitions are much weaker than dipole transitions.


Wave Mechanics

In this section, we generalize the concepts we have learned in the previous sections. The goal here is to give a broader description of quantum mechanics in terms of wave functions that are solutions to the Schroedinger Equation. In classical mechanics the particle state is determined by its position and momentum and the state evolution is determined by Newtons law. In quantum mechanics the particle state is completely described by its wave function and the state evolution is determined by the Schroedinger equation. The wave function as a complete description of the particle enables us to compute expected values of physical quantities of the particle when a corresponding measurement is performed. The measurement results are real numbers, like the energy, or position or momentum the particle has in this state. The physically measureable quantities are called observables. In classical mechanics these observables or real variables like x for position, p for momentum or functions thereof, like the energy, which is called the Hamilp2 tonian H (p, x) = 2 + V (x) in classical mechanics. For simplicity, we state m the results only for one-dimensional systems but it is straight forward to extend these results to multi-dimensional sytems. In quantum mechanics these



observables become operators: position operator ~ p = : momentum operator j x ~2 2 H (p, x) = + V (x) : Hamiltonian operator 2m x2 x : (4.131) (4.132) (4.133)

If we carry out measurements of these observables, the result is a real number in each measurement and after many measurements on identical systems we can make a statistics of these measurements and the statistics is completely described by the moments of the observable.


Position Statistics

The statistical interpretation of quantum mechanics enables us to compute the expected value of the position operator or any of its moments according to Z hxi = (x, t) x (x, t) dx (4.134) Z m (x, t) xm (x, t) dx (4.135) hx i =

The expectation value of functions of operators can always be evaluated by dening the operator by its Taylor expansion Z hf (x)i = (x, t) f (x) (x, t) dx (4.136) + * X 1 f (n) (0) xn = n ! n=0 Z X 1 (n) n f (0) (x, t) x (x, t) dx = n! n=0


Momentum Statistics

The momentum statistics is then



~ (x, t) dx (4.137) j x which can be written in terms of the wave function in the wave number space, which we dene now for symmetry reasons as the Fourier transform of the wave function where the 2 is symmetrically distributed between Fourier and inverse Fourier transform Z 1 (k, t) = (x, t) ejkx dx, (4.138) 2 Z 1 (k, t) ejkx dk. (4.139) (x, t) = 2 hpi = (x, t) Using the dierentiation theorem of the Fourier transform and the generalized Parseval relation Z Z 1 (k) 2 (k) dk = (4.140) 1 (x) 2 (x) dx

we nd hpi = =

(k, t) ~k (k, t) dk ~k | (k, t)|2 dk.

(4.141) (4.142)

The introduction of the symmetrically dened expectation value of an operator according Eq.(4.134), where x can stand for any operator can be carried out using the wave function in the position space or the wave number space using the corresponding represenation of the wave function and of the operator.


Energy Statistics

The analysis for the measurement of position or moment carries over to every observable in an analogous way. Thus the expectation value of the energy is hH (x, p)i = (x, t) H (x, p) (x, t) dx (4.143) Z 1 2 = (x, t) + V (x) (x, t) dx. (4.144) 2m~2 x2 Z



If the system is in an energy eigenstate, i.e. (x, t) = n (x) ej n t with H (x, p) n (x) = En n (x) , we obtain hH (x, p)i = Z

(4.145) (4.146) (4.147)

(x, t) En (x, t) dx = En .

If the system is in a superposition of energy eigenstates (x, t) = we obtain hH (x, p)i =

X n=0

cn n (x)ej n t .


X n=0

En |cn |2 .



Arbitrary Observable

There may also occur observables that are not simple to translate from the classical to the quantum domain, such as the product pcl xcl = xcl pcl (4.150)

Classically it does not matter which variable comes rst. However, if we tranfer this expression into quantum mechanics, the corresponding operator depends on the odering, for example pqm xqm (x, t) = ~ (x (x, t)) = j x ~ ~ (x, t) + x (x, t) , = j j x ~ + xqm pqm (x, t) . = j (4.151) (4.152) (4.153)

The decision of which expression represents the correct quantum mechanical operator or eventually even a linear combination of the possible expressions, has to be based on a close examination of the actual measurement apparatus



that would measure the corresponding observable. Finally, the expression also has to deliver results that are in agreement with experimental ndings. If we have an operator that is a function of x and p and we have decided on a unique expression in terms of a power expansion in x and p g (x, p) gop (x, ~ ) j x (4.154)

then we can compute its expected value either in the space domain or the wave number domain Z ~ ) (x, t) dx (4.155) (x, t) gop (x, hgop i = j x Z (4.156) (k, t) gop (j , ~k) (k, t) dk = k That is this operator can be represented either in real space or in k-space as gop (x, ~ ) or gop (j k , ~k). j x


Eigenfunctions and Eigenvalues of Operators

A dierential operator has in general eigenfunctions and corresponding eigenvalues gop (x, ~ ) (x) = gn n (x) , j x n (4.157)

where gn is the eigenvalue to the eigenfunction n (x) . An example for a dierential operator is the Hamiltonian operator describing a partical moving in a potential 1 2 + V (x) (4.158) 2m~2 x2 the corresponding eigenvalue equation is the stationary Schroedinger Equation (4.159) Hop n (x) = En n (x) . Hop = Thus the energy levels of a quantum system are the eigenvalues of the corresponding Hamiltonian operator. The operator for which

238 Z


(x) (Hop m (x)) dx =

(Hop n (x)) m (x) dx,


for arbitrary wave functions n and m is called a hermitian operation. From this equation we nd immediately that the expected values of a hermitian operator are real, which also has the consequence that the eigenvalues of hermitian operators are real. This is important since operators that represent observables must have real expected values and real eigenvalues since these are results of physical measurements, which are real. Thus observables are represented by hermitian operators. This is easy to proove. Lets assume we have found two eigenfunctions and the corresponding eigen values gop m = gm m , gop n = gn n . Then Z n gop m dx = gm Z nm. (4.161) (4.162) (4.163)

Thus, if there is no degneracy, the eigenfunctions of a hermitian operator are orthogonal to each other. If there is degeneracy, one can always choose an orthogonal set of eigenfunctions. If the eigenfunctions are properly normalized R n n dx = 1, then the eigenfunctions build an orthonormal system Z (4.167) n m dx = nm ,

, i.e. If n = m the integral can not vanish and Eq.(4.165) enforces gn = gn the corresponding eigenvalues are real. If n 6= m and the corresponding eigenvalues are not degenerate, i.e. dierent eigenfunctions have dierent eigenvalues, then Eq.(4.165) enforces that the eigenfunctions are orthogonal to each other Z (4.166) n m dx = 0, for n 6= m.

The right sides of Eqs.(4.163) and (4.164) must be equal Z (gm gn ) n m dx = 0

By taking advantage of the fact that the operator is hermitian we can also write Z Z Z (4.164) n gop m dx = (gop n ) m dx = gn n m dx (4.165)



and are complete, i.e. any arbitrary function f (x) can be expressed as a superposition of the orthonormal basis functions n (x) f (x) =
X n=0

cn n (x) .


Thus we can freely change the basis in which we describe a certain physical problem. To account fully for this fact, we no longer wish to use wave mechanics, ie. express the wave function as a function in position space or in k-space. Instead we will utilize a vector in an abstract function space, i.e. a Hilbert space. In this way, we can formulate a physical problem, without using a xed representation for the state of the system (wave function) and the corresponding operator representations. This description enables us to make full use of the mathematical structure of Hilbert spaces and the algebraic properties of operators.


Appendix A: Spherical Harmonics

It is obvious, that after multiplication with sin2 the equation can be separted again in two equations for the polar and azimutal functions Y (, ) = () () with d2 + m2 = 0 , d2 d 1 d m2 = 0. sin + C sin d d sin2

In this appendix, we present a brief derivation of the spherical harmonics, which are a solution of the equation 2Y 1 Y 1 + CY = 0. (4.169) sin + sin sin2 2


(4.171) (4.172)

Since the wavefunction has to be periodic in the azimutal angle , with period 2, m must be a whole number and the solutions are either sines or cosines or the complex solutions () = const. ejm , with m = . . . 2, 1, 0, 1, 2 . . . . (4.173)



Now, we are left to solve the polar equation (4.172), which can be transformed into Legendres dierential equation with the substitution = cos (4.174)

It turns out, that this equation has only bounded solutions on the interval [1, 1], if the constant C is a whole number of the form C = l (l + 1) with l = 0, 1, 2 . . . . For a given number l, there are only 2l + 1 dierent solutions for m m = l, l + 1, . . . 1, 0, 1 . . . l 1, l

into the Legendres dierential equation, well known from electrostatics with spherical boundary conditions. 2 d m2 2 d 1 =0 . (4.175) 2 + C d d 2 1 2



For m = 0, Eq.(4.175) is the regular Legendres dierential equation and the solutions are the Legendre-Polynomials [5] P0 ( ) = 1 , P1 ( ) = , 2 1 , P2 ( ) = 3 2 2 P3 ( ) = 5 3 3 , 2 2 4 15 2 P4 ( ) = 35 +3 , 8 4 8 63 5 35 3 15 P5 ( ) = 8 4 + 8 .


For m 6= 0, Eq.(4.175) is the associated Legendres dierential equation and the solutions are the associated Legendre-Polynomials, which can be generated from the Legendre-Polynomials by m/2 dm P1 ( ) m P1 ( ) = 1 2 . d m (4.179)

Overall the angular functions can be combined to form the spherical harmonics s (2l + 1) (l |m|)! m jm P1 (cos ) e Y1m (, ) = (1)m . (4.180) 4 (l + |m|)!



The fore factor is chosen such that they are normalized over the full volume angle 4, i.e. [0, ] and [, ] taking the polar weighting factor sin into account
2 Z Z 0 0

Ylm (, ) Ylm (, ) sin d d = ll0 , mm0 . 0


The spherical harmonics form a system of orthonormal functions on the full volume angle. Any function of the angular variables (, ) can be expanded in spherical harmonics. The spherical harmonics with negative azimuthal number -m can be expressed in terms of those with positive azimuthal number m. Y1m (, ) = (1)m (Ylm (, )) . (4.182)


Appendix B: Radial Wave Function

After choosing the spherical harmonic with indices l, m the radial Equation (4.107) becomes d2 R 2 dR m0 e2 2m0 E l (l + 1) 0 R = 0. (4.183) + + + dr2 r dr ~2 20 ~2 r r2 It is advantageous to introduce normalized quantities for the energy and the radial variable E = E1 n2 2r , = nr1 (4.184) (4.185)

which transforms the radial equation into the dimensionless equation d2 R 2 dR 1 n l (l + 1) R = 0. (4.186) + + + d2 d 4 2 We rst consider who the solution to this equation has to behave asymptotically for . For large , we end up with the equation d2 R 1 R = 0, d2 4 (4.187)



which has only one useful solution to construct a bound state solution, that is e/2 . Therefore, it makes sense to look for solutions to the full radial equation (4.186) of the form R() = s w() e/2 . This trial solution leads to d2 w dw + [2 ( s + 1) ] + [(n s 1) + s(s + 1) l(l + 1)] w = 0. d2 d (4.189) This equation only works for = 0, if s = l or s = (l + 1). The latter choice is not possible, because otherwise the radial solution diverges at the origin. We are then left over to solve the equation 2 2 dw d2 w + [(n l 1)] w = 0. + [2 (l + 1) ] 2 d d (4.190) (4.188)

We try a polynomial solution w() = b0 + b1 + b2 2 + .... (4.191)

and after substitution into Eq.(4.191) and comparing coecients we nd the recursive relation bk+1 = For n=p+l+1 (4.193) the series stops after the term with index p. If this is not the case, the series does not stop and the coecients asymptotically approach bk+1 = bk /k for k . But this would indicate that w()e for large , which would lead to an unbounded wave function. Therefore, n must be a positive whole number and because of condition (4.193) n l + 1. It turns out that the function the dierential equation (4.190) is related to Laguerres dierential equation [5] and its solutions can be expressed in terms k+l+1n bk . (k + 1)(k + 2l + 2) (4.192)



of the Laguerre polynomial L21+1 nl+1 () , such that the radial wave functions can be written as /2 Fnl () = 1 L21+1 . (4.194) nl1 () e The Laguerre polynamials can be expressed as Lr s xq (s + r)!2 (x) = (1) . (sq) ! (r + q)! q ! q =0
q s X


The radial functions again form an orthogonal system of functions Z


2n [(n + l)!]3 nn0 . Fnl () Fn0 l () d = (n l 1)!



After substituting back the normalization of the radial coordinate from Eq.(4.184), we nd the radial wave function to be Rn1 (r) = Nnl Fnl () . The normalization factor is determined by Z


Rnl (r) Rn0 l (r) r2 dr = n,n0 , s


which gives 2 Nnl = 2 n

(n l 1)! 3/2 r1 . [(n + l)!]3




[1] Introduction to Quantum Mechanics, Griths, David J., Prentice Hall, 1995 [2] Quantum Mechanics I, C. Cohen-Tannoudji, B. Diu, F. Laloe, John Wiley and Sons, Inc., 1978. [3] The Physics of Atoms and Quanta, Haken and Wolf, Springer Verlag 1994. [4] Practical Quantum Mechanics, S. Flgge, Springer Verlag, Berlin, 1999. [5] Handbook of Mathematical Functions, Abramowitz and Stegun, Dover Publications, NY 1970. [6] Introduction to Modern Optics, G. R. Fowles, Dover Publications, NY 1989.




Chapter 5 Interaction of Light and Matter

Usually, there are innitely many energy eigenstates in an atomic, molecular or solid-state medium and the spectral lines of the emission are associated with transitions between two of these energy eigenstates.As we have seen form the treatment of the hydrogen atom, atomic and also molecular gases in low concentration show discrete energy eigen spectra. For many physical considerations it is already sucient to take only two of these possible energy eigenstates into account, for example those which are related to the laser transition. The pumping of the laser can be later introduced by phenomenological relaxation processes that describe the pumping of the laser via other levels. The resulting simple model is often called a two-level atom, which is mathematically also equivalent to a spin 1/2 particle in an external magnetic eld, because the spin can only be parallel or anti-parallel to the eld, i.e. it has two energy levels and energy eigenstates [4]. The interaction of the two-level atom with the electric eld of an electromagnetic wave is described by the Bloch equations, which where originally used to describe nuclear magnetic resonance phenomena.


The Two-Level Model

An atom with only two energy eigenvalues is described by a two-dimensional state space spanned by the two energy eigenstates with wave functions g (r) and e (r). The two states constitute a complete orthonormal system. The corresponding energy eigenvalues are Eg and Ee for ground and excited state, see Fig. 5.1. Fig. 5.1 shows a one-dimensional potential of nite depth that 247



Figure 5.1: One dimensional model for a two-level atom. has only two bound energy eigenstates, which can serve as a model for the following discussion. The Hamiltonian operator of the two-level atom shell be denoted HA , with HA e (r) = Ee e (r) HA g (r) = Eg g (r). An arbitrary state is a superposition state of ground and excited state (r, t) = cg (t) g (r) + ce (t) e (r). (5.3) The coecients cg (t) and ce (t) are the propability amplitudes to nd the atom in the ground or excited state, respectively: |cg |2 : |ce |2 : propability to nd the atom in the ground state propability to nd the atom in the excited state (5.4) (5.5) (5.1) (5.2)

In this two-dimensional state space the temporal dynamics is described by the time dependence of the coecients. The time dependence follows from the Schroedinger Equation: (5.6) j ~ (r, t) = HA (r, t) t j~ c g (t) g (r) + c e (t) e (r) = Eg cg (t) g (r) + Ee ce (t) e (r) (5.7) .

5.2. THE ATOM-FIELD INTERACTION IN DIPOLE APPROXIMATION249 By multiplication of this equation from the left with the complexe conjugate ground state or the excited state and integration over x using the orthogonality relations for the energy eigenstates, we obtain to separate equations for the time dependence of the coecients: c e = je ce , with e = Ee /~, c g = jg cg , with g = Eg /~. (5.8) (5.9)

As we will discuss in more detail later, this procedure is equivalent to projecting the Schroedinger Equation onto the energy eigenstates. The time dependent solution of the Schroedinger Equation of the free atom with initial conditions cg (0) and ce (0) is then (r, t) = cg (0)ejg t g (r) + ce (0)eje t e (r). (5.10)

The question that arises now is how does the atomic dynamics change in the presence of an external electro-magnetic eld and environmental perturbations?


The Atom-Field Interaction In Dipole Approximation

The strongest interaction between atoms or molecules and electromagnetic elds is due to the induced dipole moment in the atom or molecule in the presence of an electric eld. The dipole moment of an atom d is determined by the position d of the electron from the nucleus via d = e0 r. (5.11)

A dipole in an elecric eld E has the additional energy Hd = dE. Therefore, we may argue that the Schroedinger Equation for an atom in an electro magnetic eld taking the dipole interaction into account may be described by the total hamiltonian operator HAF = HA d E (rA , t), (5.12)

where rA is the position of the atom and E (rA , t) = E (t) the eld at the position of the atom. The new Schroedinger Equation using the hamiltonian



(5.12) leads to new equations of motion for the propability amplitudes when projected onto the corresponding energy eigenstates, as done before to derive eqs. (5.8) and (5.9). During these projections the following matrix elements of the dipolmoment of the atom arise Z Z Mee = e (r) d e (r) dr = e0 (5.13) e (r ) r e (r ), Z Z Meg = (5.14) e (r ) d g (r ) dr = e0 e (r ) r g (r ), Z (5.15) Mge = g (r ) d e (r ) dr = Mg , Z Z (5.16) Mgg = g (r ) d g (r ) dr = e0 g (r ) r g (r ). As we have seen when discussing the hydrogen atom in the last section, due R energy eigenstates, the matrix dipole elements R to the symmetry of atomic e (r) d e (r) dr and g (r) d g (r) dr, vanish, i.e. there is no atomic dipolmoment if the atom is in an energy eigenstate. This might not be the case for energy eigenstates in a solid. The atoms consituting the solid are oriented in a lattice, which may break the symmetry. If so, R there are permanent dipole moments and consequently the matrix elements e (r ) d e (r ) dr and R g (r) d g (r) dr would not vanish. Here, we assume atomic wave functions, i.e. symmetric or anti-symmetric wave functions and therefore the new equations of motion for the propability amplitudes become Z 1 c e = j e ce + jcg (5.17) e (r) d g (r) dr E (t), ~ Z 1 c g = j g cg + jce g (r) d e (r) dr E (t). (5.18) ~ Separating the electric eld into its polarization vector e and excitation amplitude E (t) = E (t) e, (5.19) the Schroedinger Equation becomes c e = j e ce + jcg c g Meg e E (t), ~ e Meg = j g cg + jce E (t). ~ (5.20) (5.21)

5.2. THE ATOM-FIELD INTERACTION IN DIPOLE APPROXIMATION251 The expected value for the dipole moment of an atom in state (5.3) can also be expressed in terms of the dipole matrix elements D E d = |ce |2 Mee + |cg |2 Mgg + c e cg Meg + cg ce Mge = c e cg Meg + c.c.,


where, the last line again follows from the symmetry of the wavefunctions. The abreviation c .c . stands for the complex conjugate expression to the expression immediatly preceding. This equation shows again that the atom posesses only a dipole moment if it is in a superposition of energy eigenstates. The Schroedinger equation, expressed as Eqs. (5.20) and (5.21), indicates that an external eletric eld is capable of coupling two energy eigenstates, i.e. induces a dipole moment in an atom. To obtain more insight into the dynamics induced by the electric eld in the atom, we look at the case of a monochromatic electric eld at the position of the atom E (t) = 1 jt E 0 ejt + E , 0e 2 (5.23)

where E 0 is the complex electric eld amplitude. We expect strong interaction between the eld and the atom if the atomic transition frequency between the states, eg = e g , is close to the frequency of the driving eld, i.e. eg . It is advantageous to transform to new probability amplitudes, that take some trivial oscillations already into account
e + g + Ce = ce ej( 2 t) e + g C = c ej( 2 t)

(5.24) (5.25)

which leads to the new equations of motion e + g + e + g + + + Meg e e g e = j C je ce ej( 2 t) + jcg E (t) ej( 2 t) , 2 ~ + + M e g = j e + g j g cg ej( e 2g t) + jce eg E (t) ej( e 2g t) . C 2 ~ Introducing the detuning between the atomic transistion and the electric eld frequencies eg = (5.26) 2



and the Rabi-frequency

Meg e j2t r = E0 + E , 0e ~


we obtain the following coupled mode equations for the probability amplitudes. d Ce = jCe + j r Cg , dt 2 d r Cg = +jCg + j Ce . dt 2 (5.28) (5.29)

If the Rabi-frequency is small |r | << eg , the Rotating-Wave Approximation (RWA) [3], can be used, where we only keep the slowly varying components in the interaction, i.e. r
e Meg E 0 = const.. ~




Note, equations (5.28) and (5.29) are identical to the coupled mode equations between two waveguide modes as studied in section 2.6.4. But now the coupling is between modes in time, i.e. resonances. The modes are electronic ones instead of photonic modes. But otherwise what has been said in section 2.6.4 applies in the same way. For example conservation of power ow becomes now conservation of probability. For the case of vanishing detuning the exchange in probability, excitation, between the modes is the strongest. For that case it is especially easy to eliminate one of the variables and we arrive at d2 |r |2 Ce = Ce dt2 4 d2 |r |2 Cg . Cg = dt2 4 (5.31) (5.32)

The solution to this set of equations are oscillations. If the atom is at time t = 0 in the ground-state, i.e. Cg (0) = 1 and Ce (0) = 0, respectively, we

5.3. RABI-OSCILLATIONS arrive at |r | t Cg (t) = cos 2 |r | Ce (t) = j sin t . 2


(5.33) (5.34)

Then, the probabilities for nding the atom in the ground or excited state are |r | 2 2 |cg (t)| = cos t (5.35) 2 |r | 2 2 |ce (t)| = sin t , (5.36) 2 as shown in Fig. 5.2. For the expectation value of the dipole operator under the assumption of a real dipole matrix element Meg = Meg we obtain D E d = Meg ce c g + c.c. (5.37) (5.38)

= Meg sin (|r | t) sin (eg t) .

The coherent external eld drives the population of the atomic system between the two available states with a period Tr = 2/r . Applying the eld only over half of this period leads to a complete inversion of the population. These Rabi-oscillations have been observed in various systems ranging from gases to semiconductors. Interestingly, the light emitted from the coherently driven two-level atom is not identical in frequency to the driving eld. If we look at the Fourier spectrum of the polarization according to Eq.(5.38), we obtain lines at frequencies = eg r . This is clearly a nonlinear output and the sidebands are called Mollow-sidebands [2]. Most important for the existence of these oscillations is the coherence of the atomic system over at least one Rabi-oscillation. If this coherence is destroyed fast enough, the Rabi-oscillations cannot happen and it is then impossible to generate inversion in a two-level system by interaction with light. This is the case for a large class of situations in light-matter interaction and especially for typical laser materials. So we are interested in nding out what happens in the case of loss of coherence in the atomic system due to additional interaction of the atoms with its environment.



Figure 5.2: Evolution of occupation probabilities of ground and excited state and the average dipole moment of a two-level atom in resonant interaction with a coherent classical eld. The coherent external eld drives the population of the atomic system between the two available states with a period Tr = 2/r . Applying the eld only over half of this period leads to a complete inversion of the population. The population inversion is dened as w = Pe Pg = |ce |2 |cg |2 (5.39)

These Rabi-oscillations have been observed in various systems ranging from gases to semiconductors. Interestingly, the light emitted from the coherently driven two-level atom is not identical in frequency to the driving eld. If we look at the Fourier spectrum of the polarization according to Eq.(5.38), we obtain lines at frequencies = eg |r | . This is clearly a nonlinear output and the sidebands are called Mollow-sidebands [2]. Most important for the existence of these oscillations is the coherence of the atomic system over at least one Rabi-oscillation. If this coherence is destroyed before signicant population between the levels is exchanged, the Rabi-oscillations cannot happen and it is then impossible to generate inversion in a two-level system by interaction with light. This is the case for a large class of situations



in light-matter interaction and especially for typical laser materials. So we are interested in nding out what happens in the case of loss of coherence in the atomic system due to additional interaction of the atoms with its environment. These non energy preserving processes can not be easily included in the Schroedinger Equation formalism. However, we can treat these processes phenomenologically in the equations of motion for the expectation values of the dipol moment and the population inversion, which are of interest because those qunatities feed back into Maxwells Equations as a driving term. From the equations of motion for the coecients of the wave function, Eqs. (5.28) and (5.29), we derive equations of motion for the complex slowly varying dipole moment dened as
jt d = c = Ce Cg . e cg e


and by applying the product rule we nd d d = dt d d C Cg + Ce Cg dt e dt r r = jCe Cg j Cg Cg + jCe Cg + j Ce Ce 2 2 r = jd + j w 2

(5.41) (5.42) (5.43)


d w = dt

d d + c.c. (5.44) Ce Ce . Cg Cg dt dt r + c.c. (5.45) + j r Cg Ce jCg Cg j Ce Cg = jCe Ce 2 2 = +j (5.46) r d + c.c

For the monochromatic wave of Eq.(5.23), we nd for the dynamics of a two=level atom interacting with a coherent driving eld, normalized to a frequency, the Rabi-Frequency r d d = j2d + j w dt 2 d w = +j r d + c.c dt (5.47) (5.48)




Energy- and Phase-Relaxation

In reality it is extremely dicult to completely isolate an atom from its environment. For example, an atom in free space does not only interact with an external electric eld that is intentionally incident upon an atom but also with the electric eld from all the free space modes of the surrounding electromagnetic eld. If the atom is in a solid, it interacts with the lattice vibrations of the solid. This random interaction leads to a thermalization and decoherence of the atom. As an example for the interaction of an atom with its environment in thermal equilibrium, we consider the interaction with the free space electromagnetic eld, that is in thermal equilibrium at a given temperature T, because we studied already the statistics of the energy uctuations of that eld in section 3.1, i.e. black body radiation. Now, we consider, that the electric eld amplitude in the Bloch equations (5.47) and (5.48) is a random quantitiy and represents the eld of the black body radiation
ebb Meg E bb (t), (5.49) ~ Where E bb (t) is the random eld and ebb the random polarization of the black body radiation. Note, that in the process of deriving the Rabi frequency the eld is related to the eld at the transition frequency eg . This thermal Rabi frequency is a random process with autocorrelation function + * M e 2 eg bb hbbr (t + ) bbr (t)ith = (5.50) hE bb (t + )E bb (t)i . ~

bbr (t) =

The details of the computation of this autocorrelation function and the approximations used can be found in Appendix A, the result is hbbr (t + ) bbr (t)ith = 2 ( ), T1 2 Meg 2 3 1 1 eg ~ with nth ( eg ) + , = T1 3~2 c3 2 and nth (eg ) = 1/(exp(~ eg /kT ) 1)


(5.52) (5.53)

|Meg | This result can be easily interpreted. The rst factor 3~ comes from the average of the projection of the dipol matix element onto a unit vector



when averaged over every possible polarization direction. The second factor 43 eg ~ 1 n comes from expressing the power spectral density of the ( ) + th eg 3 c 2 electric eld amplitudes at the transisition frequency eg by the spectral energy density of the black body radiation according to section 3.1. However, we did not only include in the energy the part due to the thermal photon population of the mode, but also its ground state energy ~eg /2, which we did not include in section 3.1 but found when treating the harmonic oscillator 1 quantum mechanically in section 4.4.2. Thus even at temperature T 0, T 1 stays nite, The white noise property helps us to nd an equation of motion for the inversion in the presence of this uctuating eld. The solution to Eq. (5.47) can be written as Z j t (j )(tt0 ) d(t) = e bbr (t0 ) w(t0 )dt0 . (5.54) 2 Here, is a decay term that ensures convergence of integrals. As we will see, the nal results will not depend on . The equation for the inversion is then Z d 1 t (j )(tt0 ) w= e bbr (t)bbr (t0 ) w(t0 )dt0 + c.c (5.55) dt 2 Not, after performing this statistical average, the inversion w does no longer describe the inversion of a specic atom but rather the average inversion of an ensemble of such atoms. We perform now an average over the random thermal eld and obtain Z d 1 t (j )(tt0 ) w = e hbbr (t)bbr (t0 )ith w(t0 )dt0 + c.c (5.56) dt 2 Z 1 t (j )(tt0 ) 2 = e (t t0 ) w(t0 )dt0 + c.c (5.57) 2 T1 1 1 w(t) + c.c = w(t). (5.58) = 2T1 T1 Note, the evaluation of the integral over the delta-function is ambiguous, since the singularity of the delta-function was at the integration boundary. However, the delta-function is the limit of a regular function. Here, we used the limit of a symmetric function, which contributes 1/2 after the execution of the integral. The random eld uctuations of the electromagnetic vacuum



lead to an exponential decay of the population inversion. Note, a similar computation could be carried out, integrating rst the equation for the inversion and substituting it into the equation for the dipole moment, which will lead to an exponential decay term according to d 1 d. d = jd dt 2T1 (5.59)

As for the inversion, this equation describes now the average dipole moment in an ensemble of identical atoms, rather then for a specic one. Note, so far we only averaged the equations of motion over the random excitation eld assuming that no interactions existed between the random eld and the atomic system at the start of the integration. This is not true, of course in steady state the atomic system should also be in thermal equilibrium with the thermal population inversion w0 . To account for this fact, the equation for the inversion has to be changed to w w0 d w= . dt T1 (5.60)

This rate equation for the population dierence can be compared with our phenomenolocial discussion of how thermal equilibrium between thermal radiation and a two-level system is reached using Einsteins A and B coecients in section 3.4. The approach here and back then should give the same result. In section 3.4 we used as variables the densities for the ground and excited state n1 and n2 . But normalization of these densities to the total density of atoms gives the propability to be in the excited or ground state, respectively. Eq.(3.65), which is given here again dn1 dn2 1 = = [(n2 n1 ) hsi + n2 ] . dt dt sp can be transfered to dPg dPe 1 = = (Pe Pg ) nth +Pe . dt dt sp dPg dPe = =e Pe a Pg . dt dt (5.62) (5.61)

or rewritten as


5.4. ENERGY- AND PHASE-RELAXATION with abbreviations e = a see Figure 5.3. 1 (nth + 1), sp 1 = nth . sp


(5.64) (5.65)

E Pe Ee



Figure 5.3: Two-level atom with transistion rates due to induced and spontaneous emission and absorption. For the inversion we obtain then d d 2 d w = Pe Pg = (Pe Pg ) nth +Pe dt dt dt sp 1 2nth + 1 w+1 2 wnth + w+ = . = sp 2 sp 2nth + 1 Note, here we used that Pe +Pg = 1 and, therefore, Pe = coecients between Eqs.(5.60) and (5.67) we nd
w+1 . 2

(5.66) (5.67)


1 2nth + 1 = = e + a T1 sp ~ eg a e 1 = tanh . = w0 = a + e 2nth + 1 2kT

(5.68) (5.69)



For zero temperature the decay time T1 approaches the spontaneous lifetime of the atom due to the zero-point uctuations of the electromagnetic eld 1 1 = = T1 sp 2 Meg 3 eg 3 ~c3

This is an expression for the spontaneous lifetime of an atom in terms of the dipole matrix element and the density of modes in the electromagnetic eld at the transition frequency eg . In summary, the equation for the dipole moment d and the inversion w due to its interaction with the environment can be written as = (j ( eg ) 1 )d, d T2 w w0 , w = T1 (5.70) (5.71)

The time constant T1 denotes the energy relaxation in the two-level system and T2 the phase relaxation. T2 is the correlation time between amplitudes ce and cg . The coherence between the excited and the ground state described by the dipole moment is destroyed by the interaction of the two-level system with the environment. In this basic model, the energy relaxation rate is exactly half the phase relaxation rate or T2 = 2T1 . (5.72) The atoms in a real medium do not only interact with the electromagnetic eld, but also with phonons, i.e. acoustic vibrations of the host lattice or with themselves through collisions in a gas laser. All these processes must be considered when determining the energy and phase relaxation rates. Thus there might not only be radiative transistions that lead to a nite energy relaxation time T1 . Some of the processes are elastic, i.e. there is no energy relaxation but only the phase is inuenced during the collision. Therefore, these processes reduce T2 but have no inuence on T1 . In real systems the phase relaxation time is most often much shorter than twice the energy relaxation time. (5.73) T2 2T1 .



If the inversion deviates from its equilibrium value, w0 , it relaxes back into equilibrium with a time constant T1 . Eq. (5.69) shows that for all temperatures T > 0 the inversion is negative, i.e. the lower level is stronger populated than the upper level. Thus with incoherent thermal light, inversion in a two-level system cannot be achieved. Inversion can only be achieved by pumping with incoherent light, if there are more levels and subsequent relaxation processes into the upper laser level. Due to these relaxation processes the rate a deviates from the equilibrium expression (5.65), and it has to be replaced by the pump rate . If the pump rate exceeds e , the inversion corresponding to Eq. (5.69) becomes positive, w0 = e . + e (5.74)

If we allow for articial negative temperatures, we obtain with T < 0 for the ratio of relaxation rates
~ eg e 1 + nth = = e kT < 1. a nth


Thus the pumping of the two-level system drives the system away from thermal equilibrium. Now, we have a correct description of an ensemble of atoms in thermal equilibrium with its environment, which is a much more realistic description of media especially of typical laser media.


The Bloch Equations

Thus, the total dynamics of the two-level system including the pumping and dephasing processes from Eqs.(5.70) and (5.71) is given by = ( 1 j (eg ))d + j r w, d T2 2 w w0 + j w = r d jr d . T1 (5.76) (5.77)

These equations are called the Bloch Equations. They describe the dynamics of a statistical ensemble of two-level atoms interacting with a classical electric eld. Together with the Maxwell-Equations, where the polarization of the medium is related to the expectation value of the dipole moment of the atomic ensemble these result in the Maxwell-Bloch Equations.




Dielectric Susceptibility and Saturation

The Bloch Equations are nonlinear. However, for moderate eld strength E 0 , i.e. the magnitude of the Rabi-frequency is much smaller than the optical frequency, |r | << , the inversion does not change much within an optical cycle of the eld. We assume that the inversion w of the atom will only be slowly changing and it adjusts itself to a steady state value ws . For a constant eld strength E 0 Eqs.(5.76) and (5.77) reach the steady state values e ws M eg j ds = (5.78) E 2~ 1/T2 + j( eg ) 0 w0 . (5.79) ws = 2 1 /T 2 |Meg e| 1 2 1+ T | E | 2 2 2 0 ~ (1/T2 ) +( eg ) We introduce the normalized lineshape function, which is in this case a Lorentzian, (1/T2 )2 L( ) = , (5.80) (1/T2 )2 + ( eg )2

and connect the square of the eld |E 0 |2 to the intensity I of a propagating 1 plane wave, according to Eq. (2.37), I = 2Z |E 0 |2 , F ws = w0 . 1 + IIs L() (5.81)

Thus the stationary inversion depends on the intensity of the incident light. Therefore, w0 is called the unsaturated inversion, ws the saturated inversion and Is ,with 1 2T1 T2 ZF 2 Is = |Meg e| , (5.82) ~2 is the saturation intensity. The expectation value of the dipole operator (5.22) is then given by D E (5.83) d = Meg d ejt + c.c.

Multiplication with the number of atoms per unit volume, N, relates the dipole moment of the atom to the macroscopic polarization P . As the electric eld also the polarization can be written in terms of complex quantities



P (t) =

1 P 0 ejt + P 0 ejt 2 = N Meg ds ejt + c.c.

(5.84) (5.85)

or P 0 = 2N Meg ds , With the denition of the complex susceptibility P0 =

0 ( )eE 0 ,



and comparison with Eqs. (5.86) and Eq. (5.78), we obtain for the linear susceptibility of the medium
+ ( ) = Meg Meg

jN ws , ~ 0 1/T2 + j( eg )


which is a tensor. In the following we assume that the direction of the atom is random, i.e. the alignment of the atomic dipole moment, Meg , and the electric eld is random. Therefore, we have to average over the angle enclosed between the electric eld of the wave and the atomic dipole moment, which results in 2 Megx 0 0 Megx Megy Megx Megz Megx Megx 1 2 Megy Megx = 0 Megy Megy Megy Megz Megy 0 = |Meg |2 1. 3 2 Megz Megx Megz Megy Megz Megz Megz 0 0 (5.89) How to arrive at this average over the orientation is also discussed in Appendix A. Thus, for homogeneous and isotropic media the susceptibility tensor shrinks to a scalar jN ws 1 ( ) = |Meg |2 . 3 ~ 0 1/T2 + j( eg ) Real and imaginary part of the susceptibility ( ) = 0 ( ) + j00 ( ) (5.91) (5.90)



are then given by

2 ( eg ) |Meg |2 Nws T2 L( ), 3~ 0 |Meg |2 Nws T2 L(). 00 () = 3~ 0

0 () =

(5.92) (5.93)

If the incident radiation is weak, i.e. I L( ) 1 Is


we obtain ws w0 . For optical transitions there is no thermal excitation of the excited state and w0 = 1. For an inverted system, w0 > 0, the real and imaginary parts of the susceptibility are shown in Fig. 5.4. The shape of the susceptibility computed quantum mechanically compares well with the classical susceptibility (2.44) derived from the harmonic oscillator model close to the transistion frequency eg for a transition with reasonably high Q = T2 eg . Note, the quantum mechanical susceptibility is identical to the complex Lorentzian introduced in Eq.(2.96). There is an appreciable deviation, however, far away from resonance. Far o resonance the rotating wave approximation should not be used. The physical meaning of the real and imaginary part of the susceptibility is of course identical to section 2.1.8. The propagation constant k of a TEMwave in such a medium is related to the susceptibility by p 1 k = 0 0 (1 + ( )) k0 1 + () , 2 for || 1. Under this assumption we obtain k = k0 (1 + 00 0 ) + jk0 . 2 2 (5.96) with k0 = 0



1.0 T 2 eg =10 0.5 0.6 0.4
'( ) / '' max


'' ( ) / '' max

0.2 0.0 -0.2 -0.4 0.0 0.0 0.5 1.0 / 1.5


-0.6 2.0

Figure 5.4: Real and imaginary part of the complex susceptibility for an inverted medium ws > 0. The positive imaginary susceptibility indicates exponential growth of an electromagnetic wave propagating in the medium. The real part of the susceptibility contributes to the refractive index n = 1 + 0 /2. In the case of 00 < 0, the imaginary part leads to an exponential damping of the wave. For 00 > 0 amplication takes place. Amplication of the wave is possible for w0 > 0, i.e. an inverted medium. The phase relaxation rate 1/T2 of the dipole moment determines the width of the absorption line or the bandwidth of the amplier. The amplication can not occur forever, because the amplier saturates when the intensity reaches the saturation intensity. This is a strong deviation from the linear susceptibility we derived from the classical oscillator model. The reason for this saturation is two fold. First, the light can not extract more energy from the atoms then there is energy stored in them, i.e. energy conservation holds. Second the induced dipole moment in a two-level atom is limited by the maximum value of the matrix element. In contrast the induced dipole moment in a classical oscillator grows proportionally to the applied eld without limits.


Rate Equations and Cross Sections

In many cases the fastest process in the atom-eld interaction dynamics is the dephasing of the dipole moment, i. e. T2 0. For example, in semi-



conductors T2 < 50fs. In those cases the magnitude of the dipole moment relaxes instantaneously into the steady state and follows the slowly varying eld envelope E0 (t) electromagntic eld, which evolves on a much slower time scale. We obtain with the quasi steady state solution for the dipole moment (5.78), which may now have a slow time dependence due to the slowly varying eld envelope E0 (t), for the time dependent inversion in the atomic system w = w(t) w0 w(t) L( )I (t), T1 T1 Is (5.97)

where I (t) = |E0 (t)|2 /(2ZF ) is the intensity of the electromagntic wave interacting with the two-level atom. In this limit the Bloch Equations are replaced by a simple rate equation for the population. We only take care of the counting of population dierences due to spontaneous and stimulated emissions. The interaction of an atom with light at a given transition with the stream of photons on resonance, i.e. = eg is often discribed by the mass action law. That is, the number of induced transistions from the excited to the ground state, is proportional to the product of the number of atoms in the excited state and the photon ux density Iph = I/~eg w |induced = wIph = w I. T1 Is (5.98)

This denes an interaction cross section that can be expressed in terms of the saturation intensity as = ~eg T1 Is 2eg T2 ZF |Meg e|2 . = ~ (5.99) (5.100)

In this chapter, we found the most important spectroscopic quantities that characterize an atomic transition, which are the lifetime of the excited state or often called upper-state lifetime or longitudinal lifetime T1 , the phase relaxation time or transverse relaxation time T2 which is the inverse halfwidth at half maximum of the line and the interaction cross-section that only depends on the dipole matrix element and the linewidth of the transition.

5.7. RATE EQUATIONS AND CROSS SECTIONS Appendix: A. Derivation of Field Autocorrelation Function


For understanding the relaxation processes in a two-level system subject to a thermal radiation eld, we need to calculate the following expectation value + * M e 2 bb eg hbbr (t + ) bbr (t)ith = hE bb (t + )E bb (t)i . ~ 2 Meg ebb ,


We rst compute the average

which is the projection of a vector onto a random unit vector, see Figure 5.5.

Meg ebb

Figure 5.5: Projection onto a unit vector and averaging over all possible anges. This average is over the full volume angle, whereby it does not depend on the polar angle . Z 2 1 2 = (5.102) Meg ebb Meg cos2 () sin () d 2 0 2 M eg 1 2 3 M = cos ( ) = eg 0 6 3

Next is the eld autocorrelation function hE bb (t + )E bb (t)i , which is the inverse Fourier transform of the power spectral density of the electric eld.



For a TEM wave the electric eld amplitude is directly related to the energy density in the wave, see Table 2.1 1 (5.103) |E |2 . 2 Thus the power spectral density of the electric eld of a thermal noise source is directly related to the spectral energy density of a thermal radiator, which is given by Plancks law w= SEE ( ) = SEE ( ) = 2 w (f ) = , 1 1 . hsi , with hsi = hf exp kT 1 exp
hf kT

2 8 f c3


(5.104) (5.105)

4 ~ 3 c3

In the derivation of Plancks law we only included for the energy of a mode, which is identical to a harmonic oscillator, the exchangable energy ~ hsi , where hsi is the average photon number. However, from the quantum mechanical treatment of the harmonic oscillator in section 4.4.2, we know that the ground state energy, i.e. corresponding to zero photon number is not zero, but rather ~/2. If we want to describe the uctuations in the electric eld properly also for zero temperature, we need to include the zero point energy, which describes the random eld uctuations in the ground state 4 ~ 3 1 SEE ( ) = nth ( ) + . (5.106) c3 2 As we have discussed before, and is shown in 5.6, the atom only interacts strongly with elds in the spectral range close to the transistion frequency, eg . Therefore, we replace the power spectral density for the elds in this frequency range by 3 4 ~ eg 1 SEE () = nth (eg ) + , (5.107) c3 2 since only those modes will interact with the atom. The constant spectral density simplies thel eld autorcorrelation function in the time domain to 3 4 ~ eg 1 hE bb (t + )E bb (t)i = nth (eg ) + ( ) , (5.108) c3 2 1 with nth (eg ) = eg exp kT 1

5.7. RATE EQUATIONS AND CROSS SECTIONS hbbr (t + ) bbr (t)ith = with 1 T1


2 ( ), (5.109) T1 2 Meg 23 1 eg ~ nth ( eg ) + , (5.110) = 3~2 c3 2



[1] I. I. Rabi: "Space Quantization in a Gyrating Magnetic Field," Phys. Rev. 51, 652-654 (1937). [2] B. R. Mollow, "Power Spectrum of Light Scattered by Two-Level Systems," Phys. Rev 188, 1969-1975 (1969). [3] P. Meystre, M. Sargent III: Elements of Quantum Optics, Springer Verlag (1990). [4] L. Allen and J. H. Eberly: Optical Resonance and Two-Level Atoms, Dover Verlag (1987).




Chapter 6 Lasers
After having derived the quantum mechanically correct suszeptibility for an inverted atomic system that can provide gain, we can use the rate equations to study the laser and its dynamics. After discussing the laser concept briey we will investigate various types of gain media, gas, liquid and solid-state, that can be used to construct lasers and optical ampliers. Then the dynamics of lasers, threshold behavior, steady state behavior and relaxation oscillations are discussed. A short introduction in the generation of high energy and ultrashort laser pulses using Q-switching and mode locking will be given at the end.


The Laser (Oscillator) Concept

Since the invention of the vacuum amplier tube by Robert von Lieben and Lee de Forest in 1905/06 it was known how to amplify electromagnetic waves over a broad wavelength range and how to build oscillators with which electromagnetic waves could be generated. This was extended into the millimeter wave region with advances in amplier tubes and later solid-state devices such as transistors. Until the 1950s thermal radiation sources were mostly used to generate electromagnetic waves in the optical frequency range. The generation of coherent optical waves was only made possible by the Laser. The rst amplier based on discrete energy levels (quantum amplier) was the MASER (Microwave Amplication by Stimulated Emission of Radiation), which was invented by Gordon, Townes and Zeiger 1954. In 1958 Schawlow and Townes proposed to extend the MASER principle to the optical regime. 273



The amplication should arise from stimulated emission between discrete energy levels that must be inverted, as discussed in the last section. Ampliers and oscillators based on this principle are called LASER (Light Amplication by Stimulated Emission of Radiation). Maiman was rst to demonstrate a laser. It was based on the solid-state laser material Ruby.

Figure 6.1: Theodore Maiman with the rst Ruby Laser in 1960 and a cross sectional view of the rst device [4]. The rst gas laser, a HeNe-Laser followed in 1961. It is a gas laser built by Ali Javan at MIT, with a wavelength of 632.8 nm and a linewidth of only 10kHz. The basic principle of an oscillator is a feedback circuit that is unstable, i.e. there is positive feedback at certain frequencies or certain frequency ranges, see Figure 6.2. It is the feedback circuit that determines the frequency of oscillation. Once the oscillation starts, the oscillation will build up to an intensity approaching, or even surpassing, the saturation intensity of the amplier medium by many times, until the amplier gain is reduced to a value equal to the losses that the signal experiences after one roundtrip in the feedback loop, see Figure 6.3



Figure 6.2: Principle of an oscillator circuit: an amplier with positive feedback [6] p. 495.

Figure 6.3: Saturation of amplication with increasing signal power leads to a stable oscillation [6], p. 496. In the radio frequency range the feedback circuit can be an electronic feedback circuit. At optical frequencies we use an optical resonator, which is in most cases well modeled as a one-dimensional Fabry-Perot resonator, which we analysed in depth in section 6.6. We already found back then that the transfer characterisitcs of a Fabry-Perot resonator can be understood as a feedback structure. All we need to do to construct an oscillator is provide amplication in the feedback loop, i.e. to compensate in the resonator for eventual internal losses or the losses due to the output coupling via the mirrors of the Fabry-Perot, see Figure 6.4.We have already discussed in section



2.5.3 various optical resonators, which have Gaussian beams as the fundamental resonator modes. One can also use waveguides or bers that have semitransparent mirrors at its ends or form rings as laser resonators. In the latter ones output coupling of radiation is achieved with waveguide or ber couplers in the rings. Today, lasers generating light continuosly or in the form of long, nanosecond, or very short, femtosecond pulses can be built. Typically these lasers are Q-switched or mode-locked, respectively. The average power level can vary from microwatt to kilowatts and peak powers over femtosecond pulse durations can reach petawatts. The estimated average power consumption of the world in 2001 was about 2 Terawatt.

Figure 6.4: A laser consists of an optical resonator where the internal losses and/or the losses due to partially reecting mirrors are compensated by a gain medium inside the resonator [6], p. 496.


Laser Gain Media

Important characteristics of laser gain media are how inversion can be achieved and what the spectroscopic parameters of the transistion involved in the laser 2 process are, i.e. upperstate lifetime, L = T1 , linewdith fF W HM = T and 2 the cross-section for stimulated emission.


Three and Four Level Laser Media

As we discussed before inversion can not be achieved in a two level system by optical pumping with incoherent light. The coherent regime is typically inaccesible by typcial optical pump sources. Inversion by optical pumping



can only be achieved when using a three or four-level system, see Figures 6.5 and 6.6
a) 2 N2 2 b) N2

Rp 1 0

21 N1 1 0


21 N1





Figure 6.5: Three-level laser medium.

3 2 Rp 1 0 10 32

N2 21 N1 N0

Figure 6.6: Four-level laser medium. If the medium is in thermal equilibrium, typically only the ground state is occupied. By optical pumping with an intense lamp (ash lamp), or another laser, one can pump a signicant fraction of the atoms from the ground state with population N0 into the excited state N2 , or N1 for the three level laser operating according to the schemes shown in Figure 6.5 (a) and (b) or N3 in the case of the four level laser, see Figure 6.6. For the cases depicted in



277 (a) and Figure 6.6, if the relaxation rate 10 is very fast compared to 21 , inversion can be achieved, i.e. N2 > N1 . For the four level laser the relaxation rate 32 should also be fast in comparison to 21 . These systems are easy to analyze in the rate equation approximation, where the dipole moments are already adiabatically eliminated. For example, for the three level system in Figure 6.5 a). we obtain the rate equations of the three level system in analogy to the two-level system d N2 = 21 N2 21 (N2 N1 ) Iph + Rp (6.1) dt d N1 = 10 N1 + 21 N2 + 21 (N2 N1 ) Iph (6.2) dt d (6.3) N0 = 10 N1 Rp dt Here, 21 is the cross section for stimulated emission between the levels 2 and 1 and Iph is the photon ux at the transition frequency f21 .In most cases, there are plenty of atoms available in the ground state such that optical pumping can never deplete the number of atoms in the ground state N0 = const. That is why we can assume a constant pump rate Rp . If the relaxation rate 10 is much faster than 21 and the number of possible stimulated emission events that can occur 21 (N2 N1 ) Iph , then we can set N1 = 0 and obtain only a rate equation for the upper laser level Rp d 21 N2 Iph . N2 = 21 N2 (6.4) dt 21

This equation is identical to the equation for the inversion of the two-level Rp is the equilibrium upper state population in system, see Eq.(5.97). Here, 21 1 the absence of photons, 21 = L is the inverse upper state lifetime due to radiative and non radiative processes. Note, a similar analysis can be done for the three level laser operating according to the scheme shown in Figure 6.5 (b). Then the relaxation rate from level 3 to level 2, which is now the upper laser level has to be fast. But in addition the optical pumping must be so strong that essentially all the ground state levels are depleted. Undepleted groundstate populations would always lead to absorption of laser radiation. In the following we want to discuss the electronic structure of a few often encountered laser media. A detail description of laser media can be found in [7].




Types of Lasers
Gas Lasers

Helium-Neon Laser The HeNe-Laser is the most widely used noble gas laser. Lasing can be achieved at many wavelength 632.8nm (543.5nm, 593.9nm, 611.8nm, 1.1523m, 1.52m, 3.3913m). Pumping is achieved by electrical discharge, see Figure 6.7.

Figure 6.7: Energy level diagram of the transistions involved in the HeNe laser [9]. The helium atoms are excited to a metastable state by electron impact through an electric discharge occuring at about 1000V through electrodes in



a vaccum tube. The energy is then transfered to Neon by collisions. The rst HeNe laser operated at the 1.1523m line [8]. HeNe lasers are used in many applications such as interferometry, holography, spectroscopy, barcode scanning, alignment and optical demonstrations. Argon and Krypton Ion Lasers Similar to the HeNe-laser the Argon ion gas laser is pumped by electric discharge and emitts light at wavelength: 488.0nm, 514.5nm, 351nm, 465.8nm, 472.7nm, 528.7nm. It is used in applications ranging from retinal phototherapy for diabetes, lithography, and pumping of other lasers. The Krypton ion gas laser is analogous to the Argon gas laser with wavelength: 416nm, 530.9nm, 568.2nm, 647.1nm, 676.4nm, 752.5nm, 799.3nm. Also pumped by electrical discharge. When mixed with argon, it can be used as a "white-light" laser for light shows. Carbon Lasers In the carbon dioxide (CO2 ) gas laser the laser transistions are related to vibrational-rotational excitations of the CO2 -molecule. CO2 lasers are highly ecient approaching 30%. The main emission wavelengths are 10.6m and 9.4m. They are pumped by transverse (high power) or longitudinal (low power) electrical discharge. It is heavily used in the material processing industry for cutting, and welding of steel and in the medical area for surgery. Carbon monoxide (CO) gas laser: Wavelength 2.6 - 4m, 4.8 - 8.3m pumped by electrical discharge are also used in material processing such as engraving and welding and in photoacoustic spectroscopy. Output powers as high as 100kW have been demonstrated.


Excimer Lasers:

Chemical lasers emitting in the UV: 193nm (ArF), 248nm (KrF), 308nm (XeCl), 353nm (XeF) excimer (excited dimer). These are molecules that exist only if one of the atoms is electronically excited. Without excitation the two atoms repell each other. Thus the electronic groundstate is not stable and is therefore not populated, which is ideal for laser operation. These lasers are used for ultraviolet lithography in the semiconductor industry and laser surgery.




Dye Lasers:

The laser gain medium are organic dyes in solution of ethyl, methyl alcohol, glycerol or water. These dyes can be excited optically using for example Argon lasers that emit at 390-435nm (stilbene), 460-515nm (coumarin 102), 570-640 nm (rhodamine 6G) and many others. These lasers have been widely used in research and spectroscopy because of there wide tuning ranges. Unfortunately, dyes are carcinogenic and as soon as tunable solid state laser media became available dye laser became extinct.


Solid-State Lasers

Ruby Laser The rst laser was indeed a solid-state laser: Ruby emitting at 694.3nm [5].

Figure 6.8: Energy level diagram for Ruby, [2], p. 13. Ruby consists of the naturally formed crystal of aluminum oxide (Al2 O3 ) called corundum. In that crystal some of the Al3+ ions are replaced by Cr3+ ions. Its the chromium ions that give Ruby the pinkish color, i.e. its



ourescence, which is related to the laser transisitons, see the level structure in Figure 6.8. Ruby is a three level laser. Today, for the manufacturing of ruby as a laser material, articially grown crystals from molten material which crystalizes in the form of sapphire is used. The liftetime of the upper laser level is 3ms. Pumping is usually achieved with ashlamps, see Figure 6.1. Neodymium YAG (Nd:YAG) Neodymium YAG consists of Yttrium-Aluminium-Garnet (YAG) Y3 Al5 O12 in which some of the Y3+ ions are replaced by Nd3+ ions. Neodymium is a rare earth element, where the active electronic states are shielded inner 4f states. Nd:YAG is a four level laser, see Figure 6.9.

Figure 6.9: Energy level diagram for Nd:YAG, [3], p. 370. The main emission of Nd:YAG is at 1.064m. Another line with considerable less gain is at 1.32m. Initially Nd:YAG was ashlamp pumped. Today, much more ecient pumping is possible with laser diodes and diode arrays. Diode pumped versions which can be very compact and ecient are becoming a competition for CO2 -lasers in material processing, range nding, surgery, pumping of other lasers in combination with frequency doubling to produce a green 532nm beam). Neodymium can also be doped in a host of other crystals such as YLF (Nd:YLF) emitting at 1047m, YVO4 (Nd:YVO4 ) emitting at 1.064m, glass



(Nd:Glass) at 1.062m (Silicate glasses), 1.054m (Phosphate glasses). Glass lasers have been used to build extremely high power (Petawatt), high energy (Megajoules) multiple beam systems for inertial connement fusion. The big advantage of glass is that it can be fabricated on meter scale which is hard or even impossible to do with crystalline materials. Other rare earth elements are Er3+ , Tm3+ , Ho3+ , Er3+ , which have emmission lines at 1.53m and in the 2-3m range. Ytterbium YAG (Yb:YAG) Ytterbium YAG is a quasi three level laser, see Figure 283 emitting at 1.030m. The lower laser level is only 500-600cm1 (60meV) above the ground state and is therefore at room temperature heavily thermally populated. The laser is pumped at 941 or 968nm with laser diodes to provide the high brighness pumping needed to achieve gain.

Figure 6.10: Energy level diagram of Yb:YAG, [3], p. 374. Yb:YAG has many advantages over other laser materials: Very low quantum defect, i.e. dierence between the photon energy


CHAPTER 6. LASERS necessary for pumping and photon energy of the emitted radiation, (hfP hfL ) /hfP 9%.

long radiative lifetime of the upper laser level, i.e. much energy can be stored in the crystal.

high doping levels can be used without upper state lifetime quenching broad emission bandwidth of fF W HM = 2.5THz enabling the generation of sub-picosecond pulses

with cryogenic cooling Yb:YAG becomes a four level laser.

Due to the low quantum defect and the good thermal properties of YAG, Yb:YAG lasers approaching an optical to optical eciency of 80% and a wall plug eciency of 40% have been demonstrated.

Titanium Sapphire (Ti:sapphire) In contrast to Neodymium, which is a rare earth element, Titanium is a transition metal. The Ti3+ ions replace a certain fraction of the Al3+ ions in sapphire (Al2 O3 ). In transistion metal lasers, the laser active electronic states are outer 3s electrons which couple strongly to lattice vibrations. These lattice vibrations lead to strong line broadening. Therefore, Ti:sapphire has an extremely broad amplication linewidth fF W HM 100THz. Ti:sapphire can provide gain from 650-1080nm. Therefore, this material is used in todays highly-tunable or very short pulse laser systems and ampliers. Once Ti:sapphire was developed it rapidly replaced the dye laser systems. Figure 6.11 shows the absorption and emission bands of Ti:sapphire for polarization along its optical axis ( polarization).



Figure 6.11: Absorption and ourescence spectra of Ti:sapphire, [10]


Semiconductor Lasers

An important class of solid-state lasers are semiconductor lasers. Depending on the semiconductor material used the emission wavelength can be further rened by using bandstructure engineering, 0.4 m (GaN) or 0.63-1.55 m (AlGaAs, InGaAs, InGaAsP) or 3-20 m (lead salt). The AlGaAs based lasers in the wavelength range 670nm-780 nm are used in compact disc players and therefore are the most common and cheapest lasers in the world. In semiconductor lasers the electronic bandstructure is exploited, which arises from the periodic crystal potential, see problem set 10. The energy eigenstates can be characterized by the periodic crystal quasi momentum vector k, see Figure 6.12. Since the momentum carried along by an optical photon is very small compared to the momentum of the electrons in the crystal lattice, transistions of an electron from the valence band to the conduction band occur essentially vertically, see Figure 6.13 (a).



Conduction band

Band Gap

Valence band

Figure 6.12: (a) Energy level diagram of the electronic states in a crystaline solid-state material. There is usually a highest occupied band, the valence band and a lowest unoccupied band the conduction band. Electronics states in a crystal can usually be characterized by their quasi momentum k. b) The valence and conduction band are separated by a band gap.

Optical Transition

Unoccupied states Holes

Figure 6.13: (a) At thermal equilibrium the valence band is occupied and the conduction band is unoccupied. Optical transistions occur vertically under momentum conservation, since the photon momentum is negligible compared to the momentum of the electrons. (b) To obtain amplication, the medium must be inverted, i.e. electrones must be accumulated in the conduction band and empty states in the valence band. The missing electron behave as a positively charged particles called holes.



Inversion, i.e. electrons in the conduction band and empty states in the valence band, holes, see Figure 6.13 (b) can be achieved by creating a pnjunction diode and forward biasing, see Figure 6.14.

P - doped

N - doped

Acceptors Energy

Conduction band

Valence band



Figure 6.14: Forward biased pn-junction laser diode. Electrons and holes are injected into the space charge region of a pn-junction and emit light by recombination. When forward-biased electrons and holes are injected into the space charge region. The carriers recombine and emit the released energy in the form of photons with an energy roughly equal to the band gap energy. A sketch of a typical pn-junction diode laser is shown in Figure 6.15.

Figure 6.15: Typical broad area pn-homojunction laser, [3], p. 397. The devices can be further rened by using heterojunctions so that the



carriers are precisely conned to the region of the waveguide mode, see Figure 6.16.

Figure 6.16: a) Refractive index prole. b) transverse beam prole, and c) band structure (shematic) of a double-heterostructure diode laser, [3], p. 399. .


Quantum Cascade Lasers

A new form of semiconductor lasers was predicted in the 70s by two russian physicists, Kazarinov and Suris, that is based only on one kind of electrical



carriers. These are most often chosen to be electrons because of there higher mobility. This laser is therefore a unipolar device in contrast to the conventional semiconductor laser that uses both electrons and holes. the transitions are intraband transistions. A layout of a quantum cascade laser is shown in Figure 6.17.

Figure 6.17: Quantum Cascade laser layout. Like semiconductor lasers these lasers are electrically pumped. The rst laser of this type was realized in 1994 by Federico Capassos group at Bell Laboratories [9], 23 years after its theoretical prediction. The reason for this delay is the necessary high quality semiconductor growth, which is only possible using advanced semiconductor growth capabilities such as molecular beam epitaxy (MBE) and more recently metal oxide chemical vapor depostion (MOCVD). Lasers have been demostrated from the few THz range [13] up to the 3.5m region.




Lasers and its spectroscopic parameters

Some of the most important spectroscopic parameters of the laser transitions of often used laser media, some of which we just discussed, are summarized in table 6.1. Wavelength Laser Medium 0 (nm) 3+ Nd :YAG 1,064 Nd3+ :LSB 1,062 3+ Nd :YLF 1,047 3+ Nd :YVO4 1,064 Nd3+ :glass 1,054 3+ Er :glass 1,55 Ruby 694.3 3+ Ti :Al2 O3 660-1180 Cr3+ :LiSAF 760-960 3+ Cr :LiCAF 710-840 3+ Cr :LiSGAF 740-930 He-Ne 632.8 + Ar 515 CO2 10,600 Rhodamin-6G 560-640 semiconductors 450-30,000 Cross Section (cm2 ) 4.1 1019 1.3 1019 1.8 1019 2.5 1019 4 1020 6 1021 2 1020 3 1019 4.8 1020 1.3 1020 3.3 1020 1 1013 3 1012 3 1018 3 1016 1014 Upper-St. Lifetime L (s) 1,200 87 450 50 350 10,000 1,000 3 67 170 88 0.7 0.07 2,900,000 0.0033 0.002 Linewidth fF W HM = 2 (THz) T2 0.210 1.2 0.390 0.300 3 4 0.06 100 80 65 80 0.0015 0.0035 0.000060 5 25 Refr. index n 1.82 1.47 (ne) 1.82 (ne) 2.19 (ne) 1.5 1.46 1.76 1.76 1.4 1.4 1.4 1 1 1 1.33 3-4

Typ H H H H H/I H/I H H H H H I I H H H/I

Table 6.1: Wavelength range, cross-section for stimulated emission, upperstate lifetime, linewidth, typ of lineshape (H=homogeneously broadened, I=inhomogeneously broadened) and index for some often used solid-state laser materials, and in comparison with semiconductor and dye lasers.

It is surprizing, that such a wide variety of dierent materials can be characterized by essentially three spectroscopy parameters that describe to a large extend the laser dynamics properly when these materials are used as laser gain media.




Homogeneous and Inhomogeneous Broadening

Laser media are also distinguished by the line broadening mechanisms involved. Very often it is the case that the linewidth observed in the absorption or emission spectrum is not only due to dephasing processes that are acting on all atoms in the same, i.e. homogenous way. For example, lattice vibrations that lead to a line broadening of electronic transisitions of laser ions in the crystal act in the same way on all atoms in the crystal. Such mechanisms are called homogeneous broadening. However, It can be that in an atomic ensemble there are groups of atoms with a dierent center frequency of the atomic transistion. The overall ensemble therefore may eventually show a very broad linewidth but it is not related to actual dephasing mechanism that acts upon each atom in the ensemble. This is partially the case in Nd:silicate glass lasers, see table 6.1 and the linewidth is said to be inhomogeneously broadened. Wether a transistion is homogenously or inhomogeneously broadened can be tested by using a laser to saturate the medium. In a homogenously broadened medium the loss or gain saturates homogenously, i.e. the whole line is reduced. In an inhomogenously broadened medium spectral hole burning occurs, i.e. only that sub-group of atoms that are suciently in resonance with the driving eld saturate and the others not, which leads to a spectral hole in the inversion of the atoms depending on the center frequency of the atom. Figure 6.18 shows the impact of an inhomogeneously broadened gain medium on the continous wave output spectrum of a laser. Inhomogenous broadening leads to lasing of many longitudinal laser modes because of inhomogenous saturation of the gain. In the homogenously broadened medium the gain saturates homogenously and only one or a few modes located in frequency at the peak of the gain can lase. An often encountered inhomogenous broadening mechanism in gases is doppler broadening. Due to the motion of the atoms in a gas relative to an incident electromagnetic beam, the center frequency of each atomic transistion is doppler shifted according to its velocity by v f = 1 f0 , (6.5) c where the plus sign is correct for an atom moving towards the beam and the minus sign for a atom moving with the beam. The velocity distribution of an ideal gas with atoms or molecules of mass m in thermal equilibrium is given



Figure 6.18: Laser with inhomogenously broaden laser medium (Nd:silicate glass) and homogenously broadened laser medium (Nd:phosphate glass), [14] by the Maxwell-Boltzman distribution r 2 m mvx p(vx ) = exp . 2kT 2kT


This means that p(vx )dvx is equal to the probability that the atom or molecule has a velocity in the interval [vx , vx + dv ]. Here, vx is the component of the velocity that is in the direction of the beam. If the homogenous linewidth of the atoms is small compared to the doppler broading, we obtain the lineshape of the inhomogenously broadened gas simply by substituting the velocity by the induced frequency shift due to the motion vz = c Then the lineshape is a Gaussian g(f ) = r " 2 # mc mc2 f f0 . exp 2kT f0 2kT f0 (6.8) f f0 f0 (6.7)

The full width at half maximum of the line is r 8 ln 2 kT fF W HM = f0 . mc2





Laser Dynamics (Single Mode)

In this section, we study the single mode laser dynamics. A laser, when pumped over threshold, typically starts to lase in a few closely spaced longitudinal modes, which are incoherent with each other and the dynamics is to a large extent similar to the dynamics of a single mode that carries the power of all lasing modes. To do so, we complement the rate equations for the populations in the atomic medium, that can be reduced to the population of the upper laser level Eq.(6.4), as discussed before, with a rate equation for the photon population in the laser mode. There are two dierent kinds of laser cavities, linear and ring cavities, see Figure 6.19

Figure 6.19: Possible cavity congurations. (a) Schematic of a linear cavity laser. (b) Schematic of a ring laser. The laser resonators can be modelled as Fabry Perot resonators as dis-



cussed in section . Typically the laser is constructed to avoid lasing of transverse modes and only the longitudinal modes are of interest. The resonance frequencies of the longitudinal modes are determined by the round trip phase to be a multiple of 2 (m ) = 2m. (6.10) neighboring modes are spaced in frequency by the inverse roundtrip time ( 0 + ) = (0 ) + TR = 2m. TR is the round trip time in the resonator, which is TR = 2 L , g (6.12) (6.11)

where g is the group velocity in the cavity in the frequency range considered. L is the cavity length of the linear or ring cavity, where 2 = 1 for the ring cavity and 2 = 2 for the linear cavity. In the case of no dispersion, the longitudinal modes of the resonator are multiples of the inverse roundtrip timeand 1 fm = m . (6.13) TR The mode spacing of the longitudinal modes is f = fm fm1 = 1 TR (6.14)

If we assume frequency independent cavity loss and Lorentzian shaped gain (see Fig. 6.20). Initially when the laser gain is larger then the cavity loss, many modes will start to lase. To assure single frequency operation a lter (etalon) can be inserted into the laser resonator, see Figure 6.21. If the laser is homogenously broadened the laser gain will satured to the loss level and only the mode at the maximum of the gain will lase. If the gain is not homogenously broadend and in the absence of a lter many modes will lase. For the following we assume a homogenously broadend laser medium and only one cavity mode is able to lase. We want to derive the equations of motion for th population inversion, or population in the upper laser level and the photon number in that mode, see Figure 6.22.



Figure 6.20: Laser gain and cavity loss spectra, longitudinal mode location, and laser output for multimode laser operation.

Figure 6.21: Gain and loss spectra, longitudinal mode locations, and laser output for single mode laser operation.



Figure 6.22: Rate equations for a laser with two-level atoms and a resonator. The intensity I in a mode propagating at group velocity vg with a mode volume V is related to the number of photons NL or the number density nL = NL /V stored in the mode with volume V = Aef f L by NL 1 (6.15) vg = hfL nL vg , 2V 2 where hfL is the photon energy and as before, 2 = 2 for a linear laser resonator (then only half of the photons are [propagating in one direction), and 2 = 1 for a ring laser (all photons propagate into the same direction). In this rst treatment we consider the case of space-independent rate equations, i.e. we assume that the laser is oscillating on a single mode and pumping and mode energy densities are uniform within the laser material. With the interaction cross section for stimulated emission dened as hfL , (6.16) = Is L I = hfL and Eq. (6.4) with the number of atoms in the upper laser level N2 in the mode, we obtain d N2 vg N2 = N2 NL + Rp . (6.17) dt L V Here, vg nL is the photon ux, is the stimulated emission cross section, L = 21 the upper state lifetime and Rp is the pumping rate into the upper laser level. A similar rate equation can be derived for the total number of photons in the mode d NL vg NL = N2 (NL + 1) . + dt p V (6.18)



Here, p is the photon lifetime in the cavity or cavity decay time. The 1 in the term (NL + 1) in Eq.(6.18) accounts for the spontaneously photons into the laser mode, which is equivalent to stimulated emission by one photon occupying the mode, as we have seen in section 3.4 Eq.(3.64). Note, this is not the spontaneous emmission into all the free space modes, which is include in the upper state lifetime L . For a laser cavity with a semi-transparent mirror with power transmission T , see section 2.4.8, producing a power loss 2l = T per round-trip in the cavity, the cavity decay time is 1/ p = 2l/TR , if TR = 2 L/vg is the roundtrip-time in linear cavity with optical length 2L or a ring cavity with optical length L. Eventual internal losses can be treated in a similar way and contribute to the cavity decay time. Note, the decay rate for the inversion in the absence of a eld, 1/ L , is not only due to spontaneous emission into free space modes, but is also a result of non radiative decay processes. So the two rate equations are d N2 vg N2 = N2 NL + Rp dt L V d NL vg NL = N2 (NL + 1) . + dt p V (6.19) (6.20)

Experimentally, the photon number and the inversion in a laser resonator are not the quantities directly measured. We therefore introduce the circulating intracavity power and the roundtrip gain P = I Aef f = hfL g = vg N2 TR . 2V NL , TR (6.21) (6.22)

The intracavity power is directly proportional to the output power from the laser Pout = T P. (6.23) From Eqs.(6.19) and (6.20) for inversion and photon number, we obtain d g g0 gP g = dt L Esat 1 d 2g P = P+ (P + Pvac ) , dt p TR

(6.24) (6.25)



where we introduced the following laser parameters, dened as the saturation energy of the gain medium, Esat , the saturation power of the gain medium, Psat the "vacuum"-power, Pvac , and the small signal gain, g0 , of the laser Esat = Psat Pvac g0 hfL V 1 = Is Aef f L vg TR 2 = Esat / L = hfL /TR Rp = 2 L , 2Aef f (6.26) (6.27) (6.28) (6.29)

Note, the factor of two in front of gain and loss is due to the fact, that we dened g and l as gain and loss with respect to amplitude. Eq.(6.29) elucidates that the gure of merit that characterizes the small signal gain achievable in a laser connected with the spectroscopic parameters of the laser gain medium is the L -product.


Continuous Wave Operation

In many technical applications, where the dimensions of the laser is large compared to the wavelength and Pvac P Psat = Esat / L , we can neglect the spontaneous emission, i.e. Pvac . If the laser power is initially small, on the order of Pvac , and the gain is unsaturated, g = g0 , we obtain from Eq.(6.25), dt dP = 2 (g0 l) P TR or P (t) = Pvac e
2(g0 l) Tt



The laser power builts up from vaccum uctuations, once the small signal gain surpasses the laser losses, g0 > gth = l, see Figure 6.23, until it reaches the saturation power. Saturation sets in within the built-up time TB = Psat Aef f TR TR TR ln ln = . 2 (g0 l) Pvac 2 (g0 l) L (6.32)



Figure 6.23: Built-up of laser power from spontaneous emission noise. Some time after the built-up phase the laser reaches steady state, with a certain saturated gain and steady state power resulting from Eqs.(6.246.25), neglecting in the following the spontaneous emission, Pvac = 0, and for d =0: dt gs = Ps g0 =l s 1 + PP sat g 0 1 , = Psat l (6.33) (6.34)

Figure 6.24 shows output power and gain as a function of small signal gain g0 , which is proportional to the pump rate Rp . Below threshold, the output power is zero and the gain increases linearly with in crease pumping. After reaching threshold the gain stays clamped at the threshold value determined by gain equal loss and the output power increases linearly. The threshold condition is again gth = l, 2lAef f Rp,th = . 2 L (6.35) (6.36)

Thus the pump rate to reach threshold is proportional to the optical loss of the mode per roundtrip, the mode cross section (in the gain medium) and inverse proportional to the L product.



Output Power, Gain


P=0 g0=gth=l Small signal gain g0

Figure 6.24: Output power and gain of a laser as a function of pump power.


Stability and Relaxation Oscillations

How does the laser reach steady state, once a perturbation has occured? g = gs + g P = Ps + P Substitution into Eqs.(6.24-6.25) and linearization leads to dP Ps = +2 g (6.39) dt TR dg gs 1 = P g (6.40) dt Esat stim 1 Ps = 1 1 + is the inverse stimulated lifetime. The stimulated where stim P sat L lifetime is the lifetime of the upper laser state in the presence of the optical eld. The perturbations decay or grow like P0 P = est . (6.41) g g0 which leads to the system of equations (using gs = l) ! Ps s 2T P0 P0 R A = = 0. TR 1 Esat stim s g0 g0 2 p (6.37) (6.38)




which determines the relaxation rates or eigen frequencies of the linearized system s 2 1 1 Ps s1/2 = . (6.44) 2 stim 2 stim Esat p
s , which tells us how often we Introducing the pump parameter r = 1 + PP sat pump the laser over threshold, the eigen frequencies can be rewritten as s ! 1 4 (r 1) stim 1j s1/2 = 1 , (6.45) 2 stim r p s 2 (r 1) r r = j (6.46) 2 L L p 2 L

There is only a solution, if the determinante of the coecient matrix vanishes, i.e. 1 Ps s +s + = 0, (6.43) stim Esat p

There are several conclusions to draw: (i): The stationary state (0, g0 ) for g0 < l and (Ps , gs ) for g0 > l are always stable, i.e. Re{si } < 0. (ii): For lasers pumped above threshold, r > 1, and long upper state r < 1p , lifetimes, i.e. 4 L the relaxation rate becomes complex, i.e. there are relaxation oscillations 1 s1/2 = j R . (6.47) 2 stim with a frequency R approximately equal to the geometric mean of inverse stimulated lifetime and photon life time s 1 R . (6.48) stim p If the laser can be pumped strong enough, i.e. r can be made large enough so that the stimulated lifetime becomes as short as the cavity decay time, relaxation oscillations vanish.



The physical reason for relaxation oscillations and instabilities related to it is, that the gain reacts to slow on the light eld, i.e. the stimulated lifetime is long in comparison with the cavity decay time. Example: diode-pumped Nd:YAG-Laser 0 = 1064 nm, = 4 1019 cm2 , Aef f = (100m 150m) , r = 50 L = 1.2 ms, l = 1%, TR = 10ns From Eq.(6.16) we obtain: Isat = stim hfL kW = 0.4 2 , Psat = Isat Aef f = 0.18 W, Ps = 9.2W L cm s L 1 = 24s, p = 1s, R = = = 2 105 s1 . r stim p

Figure 6.25 shows the typically observed uctuations of the output of a solid-

Figure 6.25: Relaxation oscillations in the time and frequency domain. state laser in the time and frequency domain. Note, that this laser has a long upperstate lifetime of several 100 s



One can also dene a quality factor for the relaxation oscillations by the ratio of the imaginary to the real part of the complex eigen frequencies 6.46 s 4 L (r 1) Q= . (6.49) p r2 The quality factor can be as large a several thousand for solid-state lasers with long upper-state lifetimes in the millisecond range.


Laser Eciency

An important measure for a laser is the eciency with which pump power is converted into laser output power. To determine the eciency we must review the important parameters of a laser and the limitations these parameters impose. From Eq.(6.34) we found that the steady state intracavity power Ps of a laser is 2g0 1 , (6.50) Ps = Psat 2l

where 2g0 is the small signal round-trip power gain, Psat the gain saturation power and 2l is the power loss per round-trip. Both parameters are expressed in Eqs.(6.26)-(6.29) in terms of the fundamental pump parameter Rp , L product and mode cross section Aef f of the gain medium. For this derivation it was asummed that all pumped atoms are in the laser mode with constant intensity over the beam cross section 2g0 = 2 Psat Rp L , Aef f hfL = Aef f 2 L (6.51) (6.52)

The power losses of lasers are due to the internal losses 2lint and the transmission T through the output coupling mirror. The internal losses can be a signicant fraction of the total losses. The output power of the laser is then 2g0 1 (6.53) Pout = T Psat 2lint + T

304 The pump power of the laser is given by Pp = Rp hfP ,



where hfP is the energy of the pump photons. In discussing the eciency of a laser, we consider the overall eciency = Pout Pp (6.55)

which approaches the dierential eciency D if the laser is pumped many times over threshold, i.e. r = 2g0 /2l D = Pout = (r ) Pp 2 T = L Psat 2lint + T Aef f hfP hfL T . = 2lint + T hfP (6.56) (6.57) (6.58)

Thus the eciency of a laser is fundamentally limited by the ratio of output coupling to total losses and the quantum defect in pumping. Therefore, one would expect that the optimum output coupling is achieved with the largest output coupler, however, this is not true as we considered the case of operating many times above threshold, see problem set 9.


"Thresholdless" Lasing

So far we neglected the spontaneous emission into the laser mode. This is justied for large lasers where the density of radiation modes in the laser medium is essentially the free space mode density and eects very close to threshold are not of interest. For lasers with small mode volume, or for a laser operating very close to threshold, the spontaneous emission into the laser mode can no longer be neglected and we should use the full rate equations

6.10. "THRESHOLDLESS" LASING (6.24) and (6.25)


g g0 gP d g = dt L Esat d 2g 1 (P + Pvac ) , P = P+ dt p TR

(6.59) (6.60)

where Pvac is the power of a single photon in the mode. The steady state conditions are

gs =

g0 , (1 + Ps /Psat ) 0 = (2gs 2l) P + 2gs Pvac .

(6.61) (6.62)

Substitution of the saturated gain condition (6.61) into (6.62) and using the pump parameter r = 2g0 /2l, leads to a quadratic equation for the normalized intracavity steady state power p=Ps /Psat in terms of normalized vacuum power pv = Pvac /Psat = L vg /V. This equation has the solutions

r 1 + rpv p= 2

r 1 + rpv 2

+ (rpv )2 .


where only the solution with the plus sign is of physical signicance. Note, the typical value for the L -product of the laser materials in table 6.1 is L = 1023 cm2 s. If the volume is measured in units of wavelength cubed, V = 3 , we obtain pv = 0.3/ for = 1m, and vg = c. Figure 6.26 shows the behavior of the intracavity power as a function of the pump parameter for various values of the normalized vacuum power on a linear and logarithmic scale.



Intracavity power, p

1 0.5




p v =10 0.0 0.0 0.5 1.0 Pump parameter r 1.5



10 10
Intracavity power, p

0 -2 -4 -6 -8

1 10

10 10 10



p v =10





1.0 Pump parameter r



Figure 6.26: Intracavity power as a function of pump parameter r on a linear scale (a) and a logarithmic scale (b) for various values of the normalized vacuum power pv . Figure 6.26 shows that for lasers with small mode volumes, i.e. mode volumes of the size of the wavelength cubed, the threshold is no longer well dened.


Short pulse generation by Q-Switching

The energy stored in the laser medium can be released suddenly (exponentially fast) when switching the laser from a high loss situation above threshold by reducing the loss suddently, i.e. increasing the Q-value of the cavity. This



can be done actively, for example by quickly moving one of the resonator mirrors in place or passively by placing a saturable absorber in the resonator [1, 8]. Hellwarth rst suggest this method only one year after the invention of the laser. As a rough orientation for a solid-state laser the following relation for the relevant time scales that govern the laser dynamics is generally valid L TR p . (6.64)


Active Q-Switching

High losses, laser is below threshold t (b)



Build-up of inversion by pumping t


Losses Gain

t (d)
Gain Losses

In active Q-switching, the losses are redu after the laser medium is pumped for as lo upper state lifetime. Then the loss is redu and laser oscillation starts.

"Q-switched" Laserpuls

Laser emission stops after the energy sto the gain medium is extracted. t
Length of pump pulse

Figure 6.27: Gain and loss dynamics of an actively Q-switched laser. Fig. 6.27 shows the principle dynamics of an actively Q-switched laser. The laser is pumped by a pump pulse with a length on the order of the upper-state



lifetime, while the intracavity losses are kept high enough so that the laser can not reach threshold. At this point, the laser medium acts as energy storage with the energy slowly relaxing by spontaneous and nonradiative transitions. Intracavity loss is suddenly reduced, for example by a rotating cavity mirror. In the high-Q situation, the laser is pumped way above threshold and the light eld builts up exponentially until the pulse reaches the saturation energy of the gain medium. The gain saturates and its energy is extracted, causing the laser to be shut o by the pulse itself. A typical actively Q-switched pulse is asymmetric: The rise time is proportional to the net gain after the Q-value of the cavity is actively switched to a high value. The light intensity grows proportional to 2g0 /TR . When the gain is depleted, the fall time mostly depends on the cavity decay time p , see Figure 6.28. For short Q-switched pulses a short cavity length, high gain and a large change in the cavity Q is necessary. If the Q-switch is not fast, the pulse width may be limited by the speed of the switch. Typical time scales for electro-optical and acousto-optical switches are 10 ns and 50 ns, respectively


~ g 0 /TR

~1/ p

8 12 16 Time (ns)


Figure 6.28: Asymmetric actively Q-switched pulse. For example, with a diode-pumped Nd:YAG microchip laser [15] using an electro-optical switch based on LiT aO3 Q-switched pulses as short as 270 ps at repetition rates of 5 kHz, peak powers of 25 kW at an average power of 34 mW, and pulse energy of 6.8 J have been generated (Figure 6.29).



Figure 6.29: Q-switched microchip laser using an electro-optic switch. The pulse is measured with a sampling scope [15]


Passive Q-Switching

In the case of passive Q-switching, the intracavity loss modulation is performed by a saturable absorber, which introduces large losses for low intensities of light and small losses for high intensities. Relaxation oscillations are due to a periodic exchange of energy stored in the laser medium by the inversion and the light eld. Without the saturable absorber these oscillations are damped. If for some reason there is too much gain in the system, the light eld can build up quickly. Especially for a low



gain cross section the back action of the growing laser eld on the inversion is weak and it can grow further. This growth is favored in the presence of loss that saturates with the intensity of the light. The laser becomes unstable and the eld intensity growth as long as the gain does not saturate below the net loss, see Fig.6.30.

Loss Pulse Gain

Figure 6.30: Gain and loss dynamics of a passively Q-switched laser The saturable absorber leads to a destabilization of the relaxation oscillations resulting in "giant" laser pulses. Once, the gain is depleted below threshold the laser shuts itself o. If continuously pumped, the gain gradually increases again until reaching threshold and the giant-pulse emission starts all over again. This system describing the photon and upper state atomic population in the presence of a saturable absorber is similar to the Lotka-Volterra equation know in population dynamics describing the cyclic behavior of predator and prey populations.


Short pulse generation by mode locking

Q-switching is a single mode pheonomenon, i.e. the pulse build-up and decay occurs over many round-trips via build-up of energy and decay in a single (or a few) longitudinal mode. If one could get several longitudinal modes lasing in a phase coherent fashion with resepct to each other, these modes would



be the Fourier components of a periodic pulse train emitted from the laser. Then a single pulse is traveling inside the laser cavity. This is called mode locking and the pulses generated are much shorter than the cavity round-trip time due to interference of the elds from many modes. The electric eld can be written as a superpostion of the longitudinal modes. We assume a given polarization. The electric eld in this polarization is then " X

E (z, t) = <

m ej(m tkm z+m ) , E mc ,

(6.65a) (6.65b) (6.65c)

m = 0 + m = 0 + m . km = c Equation (6.65a) can be rewritten as (

E (z, t) = < ej0 (tz/c)

with the complex envelope z X Em ej(m(tz/c)+m ) = complex envelope (slowly varying). = A t c m (6.67) ej0 (tz/c) is the carrier wave (fast oscillation). Here, both the carrier and the envelope travel with the same speed (no dispersion assumed). The envelope function is periodic with period T = 2 2 L = = , c c (6.68)

= < A(t z/c)ej0 (tz/c)


m ej(m(tz/c)+m ) E

(6.66a) (6.66b)

where L is the optical round-trip length in the cavity. If we assume that N modes with equal amplitudes Em = E0 and equal phases m = 0 are lasing, the envelope is given by A(z, t) = E0
(N 1)/2 m=(N 1)/2

ej(m(tz/c)) .


312 With
q 1 X


am =


1 aq , 1a


we obtain

The laser intensity I is proportional to E (z, t)2 averaged over one optical cycle: I |A(z, t)|2 . At z = 0, we obtain I (t)
t sin2 N 2 |E0 |2 t sin2 2

sin N t z 2 c . A(z, t) = E0 z sin t 2 c



The periodic pulses given by Eq. (6.72) have

the period: T = 1/f = L/c pulse duration: t =

2 N

1 N f

peak intensity N 2 |E0 |2 average intensity N |E0 |2 peak intensity is enhanced by a factor N .

If the phases of the modes are not locked, i.e. m is a random sequence then the intensity uctuates randomly about its average value ( N |E0 |2 ), which is the same as in the mode-locked case. Figure 6.31 shows the intensity of a modelocked laser versus time,if the relative phases of the modes to each other is (a) constant and (b) random



Figure 6.31: Laser intensity versus time from a mode-locked laser with (a) perfectly locked phases and (b) random phases.

We are distinguishing between active modelocking, where an external loss modulator is inserted in the cavity to generate modelocked pulses, and passive modelocking, where the pulse is modulating the intracavity itself via a saturable absorber.


Active Mode Locking

Active mode locking was rst investigated in 1970 by Kuizenga and Siegman [16] and later by Haus [17] . In the approach by Haus, modelocking is treated as a pulse propagation problem.



Figure 6.32: Actively modelocked laser with an amplitude modulator (Acousto-Optic-Modulator). The pulse is shaped in the resonator by the nite bandwidth of the gain and the loss modulator, which periodically varies the intracavity loss according to q (t) = M (1 cos( M t)). The modulation frequency has to be precisely tuned to the resonator round-trip time, M = 2/TR , see Fig.6.32. The mode locking process is then described by the master equation for the slowly varying pulse envelope A 2 = g (T ) + Dg 2 l M (1 cos( M t)) A. (6.73) TR T t This equation can be interpreted as the total pulse shaping due to gain, loss and modulator within one roundtrip, see Fig.6.33. If we x the gain in Eq. (6.73) at its stationary value, what ever it might be, Eq.(6.73) is a linear p.d.e, which can be solved by separation of variables. The pulses, we expect, will have a width much shorter than the round-trip time TR . They will be located in the minimum of the loss modulation where the cosine-function can be approximated by a parabola and we obtain A 2 2 (6.74) = g l + Dg 2 Ms t A. TR T t Ms is the modulation strength, and corresponds to the curvature of the loss modulation in the time domain at the minimum loss point g , (6.75) Dg = 2 g M2 M Ms = . (6.76) 2



Figure 6.33: Schematic representation of the master equation for an actively mode-locked laser. The dierential operator on the right side of (6.74) corresponds to the SchrdingerOperator of the harmonic oscillator problem. Therefore, the eigen functions of this operator are the Hermite-Gaussians An (T, t) = An (t)en T /TR , s 2 Wn t2 2 Hn (t/ a )e a , An (t) = 2n n! a (6.77) (6.78)

where a determines the pulse width of the Gaussian pulse. The width is given by the fourth root of the ratio between gain dispersion and modulator strength q a =

Dg /Ms .


Note, from Eq. (6.77) we see that the gain per round-trip of each eigenmode is given by n (or in general the real part of n ), which is given by 1 n = gn l 2Ms 2 a (n + ). 2 The corresponding saturated gain for each eigen solution is given by gn = 1 1+
Wn Psat TR





where Wn is the energy of the corresponding solution and Psat = Esat / L the saturation power of the gain. Eq. (6.80) shows that for a given g the eigen solution with n = 0, the ground mode, has the largest gain per roundtrip. Thus, if there is initially a eld distribution which is a superpostion of all eigen solutions, the ground mode will grow fastest and will saturate the gain to a value gs = l + Ms 2 a. (6.82)

such that 0 = 0 and consequently all other modes will decay since n < 0 for n 1. This also proves the stability of the ground mode solution [17]. Thus active modelocking without detuning between resonator round-trip time and modulator period leads to Gaussian steady state pulses with a FWHM pulse width tF W HM = 2 ln 2 a = 1.66 a . The spectrum of the Gaussian pulse is given by 0 ( ) = A = and its FWHM is fF W HM = 1.66 . 2 a (6.86) Z


A0 (t)eit dt

(6.84) (6.85)

q ( a )2 Wn a e 2 ,

Therfore, the time-bandwidth product of the Gaussian is tF W HM fF W HM = 0.44. (6.87)

The stationary pulse shape of the modelocked laser is due to the parabolic loss modulation (pulse shortening) in the time domain and the parabolic ltering (pulse stretching) due to the gain in the frequency domain, see Figs. 6.34 and 6.35. The stationary pulse is achieved when both eects balance. Since external modulation is limited to electronic speed and the pulse width does only scale with the inverse square root of the gain bandwidth actively modelocking typically only results in pulse width in the range of 10-100ps.



Figure 6.34: Loss modulation leads to pulse shortening in each roundtrip

Figure 6.35: The nite gain bandwidth broadens the pulse in each roundtrip. For a certain pulse width there is balance between the two processes. For example: Nd:YAG; 2l = 2g = 10%, g = fF W HM = 0.65 THz, M = 0.2, fm = 100 MHz, Dg = 0.24 ps2 , Ms = 4 1016 s1 , p 99 ps. With the pulse width (6.79), Eq.(6.82) can be rewritten in several ways gs = l + Ms 2 a = l+ Dg 1 1 Dg = l + Ms 2 , a+ 2 a 2 2 2 a (6.88)



which means that in steady state the saturated gain is lifted above the loss level l, so that many modes in the laser are maintained above threshold. There is additional gain necessary to overcome the loss of the modulator due to the nite temporal width of the pulse and the gain lter due to the nite bandwidth of the pulse. Usually gs l Ms 2 a = 1, l l (6.89)

since the pulses are much shorter than the round-trip time. The stationary pulse energy can therefore be computed from gs = 1 1+

= l.


The name modelocking originates from studying this pulse formation process in the frequency domain. Note, the term M [1 cos( M t)] A generates sidebands on each cavity mode present according to M [1 cos( M t)] exp(j n0 t) 1 1 = M exp(jn0 t) exp(j (n0 t M t)) exp(j ( n0 t + M t)) 2 2 1 1 = M exp(jn0 t) + exp(jn0 1 t) + exp(jn0 +1 t) 2 2 if the modulation frequency is the same as the cavity round-trip frequency. The sidebands generated from each running mode is injected into the neighboring modes which leads to synchronisation and locking of neighboring modes, i.e. mode-locking, see Fig.6.36


Passive Mode Locking

Electronic loss modulation is limited to electronic speeds. Therefore, the curvature of the loss modulation that determines the pulse length is limited. It is desirable that the pulse itself modulates the loss. The shorter the pulse the sharper the loss modulation and eventually much shorter pulses can be reached.




f n0-1

f n0

f n0+1

Figure 6.36: Modelocking in the frequency domain: The modulator transvers energy from each mode to its neighboring mode, thereby redistributing energy from the center to the wings of the spectrum. This process seeds and injection locks neighboring modes. The dynamics of a laser modelocked with a fast saturable absorber can also be easily understood using the master equation (6.73) and replacing the loss due to the modulator with the loss from a saturable absorber. For a fast saturable absorber, the losses q react instantaneously on the intensity or power P (t) = |A(t)|2 of the eld q0 q(A) = , (6.91) A|2 1 + |P A where PA is the saturation power of the absorber. Such absorbers can be made out of semiconductor materials or nonlinear optical eects can be used to create articial saturable absorption, such as in Kerr lens mode locking. There is no analytic solution of the master equation (6.73) when the loss modulation is replaced by the absorber response (6.91). We can however make expansions on the absorber response to get analytic insight. If the absorber is not saturated, we can expand the response (6.91) for small intensities q(A) = q0 |A|2 , (6.92)

with the saturable absorber modulation coecient = q0 /PA . The constant nonsaturated loss q0 can be absorbed in the losses l0 = l + q0 . The resulting master equation is A(T, t) 2 2 TR (6.93) = g l0 + Dg 2 + |A| A(T, t). T t



Up to the imaginary unit, this equation is similar to a type of nonlinear Schroedinger Equation, i.e. the potential depends on the wave function itself. One nds as a possible stationary solution t As (T, t) = As (t) = A0 sech . (6.94) Note, there is d sechx = tanh x sechx, dx d2 sechx = tanh2 x sechx sech3 x, dx2 = sechx 2 sech3 x . (6.95)


Substitution of the solution (6.94) into the master equation (6.93), and assuming steady state, results in t Dg 2 0 = (g l0 ) + 2 1 2sech t t A0 sech . (6.97) + |A0 |2 sech2 Comparison of the coecients with the sech- and sech3 -expressions leads to a condition for the peak pulse intensity and pulse width, , and for the saturated gain Dg 1 |A0 |2 , = 2 2 Df g = l0 2 . From Eq.(6.98) and with the pulse energy of a sech pulse W = follows = Z

(6.98) (6.99)

2|As (t)|2 dt = 2|A0 |2 , 4Dg . W




321 (6.102)

Equation (6.99) together with (6.101) determines the pulse energy g0 Dg = l0 2 W 1 + PL TR (6.103)

gs (W ) =

(W )2 = l0 16Dg

Figure 6.37 shows the time dependent variation of gain and loss in a laser modelocked with a fast saturable absorber on a normalized time scale.

Figure 6.37: Gain and loss in a passively modelocked laser using a fast saturable absorber on a normalized time scale x = t/ . The absorber is assumed |A|2 to saturate linearly with intensity according to q (A) = q0 1 A2 .

Here, we assumed that the absorber saturates linearly with intensity up to a maximum value q0 = A2 0 . If this maximum saturable absorption is completely exploited, see Figure 6.38.



Figure 6.38: Saturation characteristic of an ideal saturable absorber The minimum pulse width achievable with a given saturable absorption q0 results from Eq.(6.98) Dg q0 = , (6.104) 2 2 to be = r 2gs 1 . q0 g


Note that in contrast to active modelocking the achievable pulse width is now scaling with the inverse gain bandwidth. This gives much shorter pulses. Figure 6.37 can be interpreted as follows: In steady state, the saturated gain is below loss, by about one half of the exploited saturable loss before and after the pulse. This means that there is net loss outside the pulse, which keeps the pulse stable against growth of instabilities at the leading and trailing edge of the pulse. If there is stable mode-locked operation, there must always be net loss far away from the pulse, otherwise, a continuous wave signal running at the peak of the gain would experience more gain than the pulse and would break through. From Eq.(6.98) it follows, that one third of the exploited saturable loss is used up during saturation of the aborber and actually only one sixth is used to overcome the lter losses due to the nite gain bandwidth. Note, there is a limit to the mimium pulse width. This limit is due to the saturated gain (6.99), gs = l + 1 q . Therefore, from 2 0 Eq.(6.105), if we assume that the nite bandwidth of the laser is set by the gain, we obtain for q0 l 1 min = (6.106) g



for the linearly saturating absorber model. This corresponds to mode locking over the full bandwidth of the gain medium, as for a sech-shaped pulse, the time-bandwidth product is 0.315, and therefore, fF W HM = 0.315 g = . 1.76 min 1.76 (6.107)

As an example, for the Ti:sapphire laser this corresponds to g = 240 THz, min = 3.7 fs, F W HM = 6.5 fs, which is in good agreement with experimentally observed results [19]. This concludes the introdcution into ultrashort pulse generation by mode locking.



[1] R. W. Hellwarth, Eds., Advances in Quantum Electronics, Columbia Press, New York (1961). [2] A. E. Siegman, Lasers, University Science Books, Mill Valley, California (1986). [3] O. Svelto, "Principles of Lasers," Plenum Press, NY 1998. [4] http://www.llnl.gov/nif/library/aboutlasers/how.html [5] T. H. Maimann, "Stimulated optical radiation in ruby", Nature 187, 493-494, (1960). [6] B.E.A. Saleh and M.C. Teich, "Fundamentals of Photonics," John Wiley and Sons, Inc., 1991. [7] M. J. Weber, "Handbook of Lasers", CRC Press, 2000. [8] A. Javan, W. R. Bennett and D. H. Herriott, "Population Inversion and Continuous Optical Maser Oscillation in a Gas Discharge Containing a He-Ne Mixture," Phys. Rev. Lett. 6, (1961). [9] W. R. Bennett, Applied Optics, Supplement 1, Optical Masers, 24, (1962). [10] W. Koechner, "Solid-State Lasers," Springer Verlag (1990). [11] R F Kazarinov and R A Suris, "Amplication of electromagnetic waves in a semiconductor superlattice," Sov. Phys. Semicond. 5, p. 707 (1971). [12] J. Faist, F. Capasso, D. L. Sivco, C. Sirtori, A. L. Hutchinson, A. Y. Cho, "Quantum cascade laser," Science 264, p. 553 (1994). 325



[13] B. S. Williams, H. Callebaut, S. Kumar, Q. Hu and J. L. Reno, "3.4-THz quantum cascade laser based on longitudinal-optical-phonon scattering for depopulation," Appl. Phys. Lett. 82, p. 1015-1017 (2003). [14] D. Kopf, F. X. Krtner, K. J. Weingarten, M. Kamp, and U. Keller, Diode-pumped mode-locked Nd:glass lasers with an antiresonant Fabry-Perot saturable absorber, Optics Letters 20, p. 1169 (1995). [15] J. J. Zayhowski, C. Dill, Diode-pumped passively Q-switched picosecond microchip lasers, Opt. Lett. 19, pp. 1427 1429 (1994). [16] D. J. Kuizenga and A. E. Siegman, FM and AM modelocking of the homogeneous laser - part I: theory, IEEE J. Qunat. Electron. 6, pp. 694 701 (1970). [17] H. A. Haus, A Theory of Forced Mode Locking, IEEE Journal of Quantum Electronics QE-11, pp. 323 - 330 (1975). [18] H. A. Haus, Theory of modelocking with a fast saturable absorber, J. Appl. Phys. 46, pp. 3049 3058 (1975). [19] R. Ell, U. Morgner, F.X. Krtner, J.G. Fujimoto, E.P. Ippen, V. Scheuer, G. Angelow, T. Tschudi: Generation of 5-fs pulses and octave-spanning spectra directly from a Ti:Sappire laser, Opt. Lett. 26, 373-375 (2001)

Chapter 7 The Dirac Formalism and Hilbert Spaces

So far we treated quantum mechanics by using wave functions dened in position space. We identied the Fourier transform of the wave function in position space as a wave function in the wave vector or momentum space. Expectation values of operators that represent observables of the system can be computed using either representation of the wavefunction. Obviously, the physics must be independent whether represented in position or wave number space. P.A.M. Dirac was rst to introduce a representation-free notation for the quantum mechanical state of the system and operators representing physical observables. He realized that quantum mechanical expectation values could be rewritten. For example the expected value of the Hamiltonian can be expressed as Z with |i = Hop |i . (7.3)

(x) Hop (x) dx = h| Hop |i , = h| i ,

(7.1) (7.2)

Here, |i and |i are vectors in a Hilbert-Space, which is yet to be dened. For example, complex functions of one variable, (x), that are square inte327


(x) (x) dx < ,


form the Hilbert-Space of square integrable functions denoted as L2 . In Dirac notation this is Z (7.5) (x) (x) dx = h| i . Orthogonality relations can be rewritten as Z m (x) n (x) dx = h m | n i = mn . (7.6)

As the above expressions look like a bracket, he called the vector |n i a ketvector and hm | a bra-vector. The meaning of these vectors clearly need a precise mathematical denition.


Hilbert Space

A Hilbert Space is a linear vector space, i.e. if there are two elements |i and |i in this space the sum of the elements must also be an element of the vector space |i + |i = | + i . The sum of two elements is commutative and associative Commutative : Associative : |i + |i = |i + |i , . (7.8) (7.9) (7.7)

|i + | + i = | + i + |i

The product of the vector with a complex quantity c is again a vector of the Hilbert-Space c |i |ci . (7.10) The product between vectors and numbers is distributive Distributive : c | + i = c |i + c |i . (7.11)

In short, every linear combination of vectors in a Hilbert space is again a vector in that Hilbert space.




Scalar Product and Norm

There is a bilinear form dened by two elments of the Hilbert Space |i and |i , which is called a scalar product resulting in a complex number h| i = a . (7.12)

Ths scalar product obtained by exchanging the role of |i and |i results in the complex conjugate number h |i = h |i = a . The scalar product is distributive Distributive : h |1 + 2 i = h |1 i + h |2 i h |ci = c h |i And from Eq.(7.13) follows hc| i = h| ci = c h| i . (7.16) . . (7.14) (7.15) (7.13)

Thus if the complex number is pulled out from a bra-vector it becomes its complex conjugate. The bra- and ket-vectors are hermitian, or adjoint, to each other. The adjoint vector is denoted by the symbol+ (|i)+ = h| , (h|)+ = |i . (7.17) (7.18)

The vector spaces of bra- and ket-vectors are dual to each other. This means both sets of vectors form each of them a vector space and to each element in one space corresponds a unique element in the other space. To transform an arbitrary expression into its adjoint, one has to replace all operators and vectors by the adjoint operator or vector and in addition the order of the elements must be reversed. For example (c|i)+ = c h| , h| i+ = h| i = h| i . (7.19)


330 CHAPTER 7. THE DIRAC FORMALISM AND HILBERT SPACES This equation demands that a scalar product of a vector with itself is always real. Here, we even demand that it is positive h |i > 0, real . (7.21)

The equal sign in Eq.(7.21) is only fullled for the null vector, which is dened by |i + 0 = |i . (7.22) Note, we denote the null vector not with the symbol |0i but rather with the scalar 0. Because the symbol |0i is reserved for the ground state of a system. If the scalar product of a vector with itself is always positive, Eq.(7.21), then one can derive from the scalar product the norm of a vector according to p kk = h |i . (7.23) For vectors that are orthogonal to each other we have h |i = 0 without having one of them be the null vector. (7.24)


Vector Bases

The dimensions of a Hilbert space are countable, i.e. each dimension can be assigned a whole number and thereby all dimensions are referenced in a unique way with 1, 2, 3, ..... For example, the Hermite-Gaussian functions, which we found to be the energy eigenstates of the harmonic oscillator form a complete basis for square integrable functions on the interval [, +]. A vector space that is a Hilbert space has the following additional properties. Completeness: If there is a sequence of vectors in a Hilbert space |1 i , |2 i , |3 i , |4 i , ... that fullls Cauchys convergence criterion then the limit vector |i is also an element of the Hilbert space. Cauchys convergence criterion states that if kn m k < , for some n, m > N () the sequence converges uniformly [2].



The Hilbert space is separable. This indicates that for every element |i in the Hilbert space there is a sequence with |i as the limit vector. Every vector in the Hilbert space can be decomposed into linear independent basis vectors |n i . The number of basis vectors can be innite X cn |n i . (7.25) |i =

The components cn of the vector |i with respect to the basis |n i are complex numbers denoted with index n. It is advantageous to orthonormalize the basis vectors hn | m i = nm . (7.26) The components of the vector |i can then be determined easily by hm | i = or This leads to |i = X

cn hm | n i =


cn mn ,


cm = hm | i , X

(7.28) (7.29)

|n i hn | i .


Linear Operators in Hilbert Spaces

Properties of Linear Operators

An operator L is dened as a mapping of a vector |i onto another vector |i of the Hilbert space L |i = |i . (7.30) A linear operator L has the property that it maps a linear combination of input vectors to a linear combination of the correponding individual maps

L (a |1 i + b |2 i) = (a L |1 i + b L |2 i) = a |1 i + b |2 i , for a, b C.


332 CHAPTER 7. THE DIRAC FORMALISM AND HILBERT SPACES The sum of two linear operators is dened as (L + M) |i = L |i +M |i . And the product of two operators is dened as L M |i = L (M |i) . (7.33) (7.32)

The null element and 1-element of the operators is denoted as 0, and 1. Often we will not bold face these operators, especially in products, where a scalar has the same meaning. The two operators are dened by their action on arbitrary vectors of the Hilbert space 0 |i = 0, |i , 1 |i = |i , |i In general, the multiplication of two operators is not commutative L M |i 6= ML |i , |i , or in short L M 6= M L. The expression [L, M] = L M M L (7.38) (7.37) (7.36) (7.34) (7.35)

is therefore called the commutator between L and M. If [L, M] = 0, the operators commute. The following rules for commutators apply: [L, M] = [M, L] , [L, L] = 0 , [L, 1] = 0 , L, L1 = 0 , (7.39) (7.40) (7.41) (7.42) (7.43) (7.44) (7.45)

[L,aM] = a [L, M] , [L1 + L2 , M] = [L1 , M] + [L2 , M] , [L, M] = [M, L]



[L1 L2 , M] = [L1 , M] L2 + L1 [L2 , M] , [M, L1 L2 ] = [M, L1 ] L2 + L1 [M, L2 ] . Often the anticommutator is also used. It is dened as [L, M]+ = L M + M L . If [L, M]+ = 0, the operators are called anti-commuting.

(7.46) (7.47) (7.48)


The Dyadic Product

Two vectors in the Hilbert space can not only be used to build a scalar product but rather what is called a dyadic product, which is an operator |i h | . (7.49)

The dyadic product is the formal product between a ket- and a bra-vector. If applied to a vector, it projects the vector onto the state | i and generates a new vector in parallel to |i with a magnitude equal to the projection |i h | i = h | i |i . (7.50) Note, the word projection has here the same meaning as in conventional vector-calculus. There we project a vector u onto another vector v by building the scaler product between the two vectors, i.e. v+ u. As we have seen from Eq.(7.29), if the vectors |n i build a complete orthonormal basis, then X |i = |n i hn | i , |i , . (7.51)

Eq.(7.35) implies


If an operator is multiplied from the left and right side by a unit matrix, the operator is expressed in the base |n i ! X X 1L1 = |m i hm | L |n i hn | (7.53) = XX
m n m n


|n i hn | .


Lmn |m i hn |

334 CHAPTER 7. THE DIRAC FORMALISM AND HILBERT SPACES with the matrix elements Lmn = hm | L |n i . Ket-vector |a i = X


an |n i

Bra-vector X h| = a n h n |

Column vector a1 a2 . . .

Inner product ha |b i = . X = a n bn

Row vector a1 a2 a m bn h m | n i

m n

Scalar product b1 a1 a2 b2 . . .
=a 1 b1 + a2 b2 +

Operator L= X

Lmn |m i hn |

Dyadic Product |a i hb | = X

Matrix L11 L12 L21 L22 . . ... . . . . a1 a2 b1 . . . a1 b 1 a1 b2 = a2 b1 a2 b2 . . . . . . b 2 ... =

am b n | m i h n |

The matrix elments Lmn represent the operator in the chosen base |n i . Once we choose a base and represent vectors and operators in terms of this base, the components of the vector and the operator can be collected in a column vector and a matrix, respectively. The table below shows a comparison between a representation based on Hilbert space vectors and operators in term of vectors and matrices in an euclidian vector space. Initially matrix mechanics was



developed by Heisenberg independently from Schroedingers wave mechanics. The Dirac representation in terms of bra- and ket-vectors unies them and shows that both forms are isomorph to each other.

7.2.3 7.2.4

Special Linear Operators Inverse Operators

The operator inverse to a given operator L is denoted as L1 |i , |i = L |i = |i = L1 |i , LL1 = 1 The inverse of a product is the product of the inverse in inverse order (ML)1 = L1 M1 (7.57) (7.55)

which leads to



Adjoint or Hermitian Conjugate Operators


here |i and |i are arbitrary vectors in the Hilbert space. Note, that if the adjoint of an expression is formed, each component gets conjugated and the order is reversed. For example (L |i)+ = h| L+ , + + L |i = h| L (7.59) (7.60)

The adjoint or hermitian conjugate operator L+ is dened by h| L |i = L+ i = h| L+ |i ,

Given a certain base |n i , the matrix elements of the adjoint operator in that basis follow directly from the denition (7.58)
+ L+ mn = h m | L | n i = h n | L | m i = Lnm


The following rules apply to adjoint operators + + L = L,


336 CHAPTER 7. THE DIRAC FORMALISM AND HILBERT SPACES (aL)+ = a L+ , (L + M)+ = L+ +M+ , (L M)+ = M+ L+ . (7.63) (7.64) (7.65)


Hermitian Operators

If the adjoint operator L+ is identical to the operator itself, then we call the operator hermitian L = L+ . (7.66) Hermitian operators have real expected values. As discussed already during our exploration of wave mechanics, physical observables are represented by hermitian operators.


Unitary Operators
U1 = U+ ,

If the inverse of an operator U is the adjoint operator (7.67)

then this operator is called a unitary operator and U+ U = UU+ = 1. (7.68)

If the operator U is unitary and H is a hermitian operator, then the product UHU1 is also a hermitian operator. + + UHU1 = U1 H+ U+ = UHU1 . (7.69)


Projection Operators

The dyadic product Pn = |n i hn | , (7.70) is a projection operator Pn that projects a given state |i onto the unit vector |n i Pn |i = |n i hn | i . (7.71)

7.3. EIGENVALUES OF OPERATORS If |i is represented in the orthonormal base |n i X |i = cn |n i ,




we obtain Pn |i = cn |n i . (7.73) By construction, projection operators are hermitian operators. Besides operators that project on vectors, there are also operators that project on subspaces of the Hilbert space X PU = |n i hn | . (7.74)

Here, the orthonormal vectors |n i span the sub-space U . Projection operators are idempotent (7.75) Pk n = Pn , for k > 1.


Eigenvalues of Operators

In chapter 4, we studied the eigenvalue problem of dierential operators. Here, we want to formulate the eigenvalue problem of operators in a Hilbert space. An operator L in a Hilbert space with eigenvectors |n i fullls the equations L |n i = Ln |n i , (7.76)

with eigenvalues Ln . If there exist several dierent eigenvectors to the same eigenvalue Ln , this eigenvalue is called degenerate. For example, the energy eigenfunctions of the hydrogen atom are degenerate with respect to the indices l and m. The set of all eigenvalues is called the eigenvalue spectrum of the operator L. As shown earlier the eigenvalues of hermitian operators are real and the eigenvectors to dierent eigenvalues are orthogonal to each other, because hm | L |n i = Ln h m |n i = Lm hm |n i , or (Ln Lm ) hm |n i = 0. (7.78) (7.77)

338 CHAPTER 7. THE DIRAC FORMALISM AND HILBERT SPACES If the eigenvectors of the operator L form a complete base of the Hilbert space, the operator L is represented in this base by a diagonal matrix Lmn = hm | L |n i = Ln hm |n i = Ln mn The operator can then be written in its spectral representation X X L= Ln |n i hn | = Ln Pn .
n n




Eigenvectors of Commuting Operators

Two operators, A and B, that commute with each other have a common set of eigenvectors. To proove this theorem, we assume that the eigenvalue spectrum of the operator A is non degenerate. The eigenvectors and eigenvalues of operator A are |n i and An , respectively A |n i = An |n i . Using [A, B] = 0, we nd X

(7.81) (7.82)

hm | A

|n i hn | B BA |n i

hm | AB BA |n i !

= 0 = 0


(Am An ) hm | B |n i =

(Am An ) Bmn = 0 .

Since the eigenvalues are assumed to be not degenerate, i.e. Am 6= An , the matrix Bmn must be diagonal, which means that the vector |n i has also to be an eigenvector of operator B. If the eigenvalues are degenerate, one can always choose, in the sub-space that belongs to the degenerate eigenvalue, a base that are also eigenvectors of B. The operator B thus eventually has no degeneracies in this sub-space and therefore, the eigenvalues of B may help to uniquely characterize the set of joint eigenvectors. Also the reverse is true. If two operators have a joint system of eigenvectors, they commute. This is easy to see from the spectral representation of both operators.



Example: We dene the parity operator Px which, when applied to a wave function of a particle in one dimension (x), changes the sign of the position x Px (x) = (x). V (x) = V (x), (7.84)

The Hamiltonian of a particle in an inversion symmetric potential V (x), i.e. (7.85)

commutes with the parity operator. Then the eigenfunctions of the Hamiltonian are also eigenfunctions of the parity operator. The eigenfunctions of the parity operator are the symmetric or antisymmetric functions with eigenvalues 1 and 1, respectively. Therefore, the eigenfunctions of a Hamiltonian with symmetric potential has symmetric and antisymmetric eigenfunctions, see box potential and harmonic oscillator potential.


Complete System of Commuting Operators

In the case of the hydrogen atom, we had to use three qunatum numbers n, l, and m to characterize the energy eigenfunctions completely. Without proof, the indices l and m characterize the eigenvalues of the square of the angular momentum operator L2 , and of the z -component of the angular moment Lz with eigenvalues l (l + 1) ~2 and m~, respectively. One can show, that the Hamilton operator of the hydrogen atom, the square of the angular momentum operator and the z -component of the angular moment operator commute with each other and build a complete system of commuting operators (CSCO), whose eigenvalues enable a unique characterization of the energy eigenstates of the hydrogen atom.


Product Space

Very often in quantum mechanics one deals with interacting systems, for example system A and system B . The state space of each isolated system is Hilbert space HA and Hilbert space HB spanned by a complete base |n i A and | n i B , respectively. LA and MB are operators on each of the Hilber

340 CHAPTER 7. THE DIRAC FORMALISM AND HILBERT SPACES spaces of the individual systems. The Hilbert space of the total system is the product space H = HA HB . (7.86) The vectors in this Hilbert space are given by the direct product of the individual vectors and one could choose as a base in the product space |nm i = |n iA |m iB = |n iA |m iB . (7.87)

Operators that only act on system A can be extended to operate on the product space by L = LA 1B . or similar for operators acting on system B M = 1A MB . The product of both operators is then LM = LA MB . (7.90) (7.89) (7.88)

An operator in this product space acts on a vector in the following way LM |nm i = L |n iA M |m iB . (7.91)

Since, the vecotrs |n iA and |m iB build a complete base for system A and B , respectively, the product vectors in Eq.(7.87) build a complete base for the interacting system and each state can be written in terms of this base |i = X

amn |nm i =


amn |n iA |m iB .



Quantum Dynamics

In chapter 4, we derived the Schroedinger Equation in the x-represenation. The stationary Schroedinger Equation was written as an eigenvalue problem to the Hamiltonian operator, which was then a dierential operator. With the Dirac formulation we can rewrite these equations without refering to a special representation.




Schroedinger Equation
| (t)i = H | (t)i . t

In the Dirac notation the Schroedinger Equation is j~ (7.93)

H is the Hamiltonian operator; it determines the dynamics of the quantum system. p 2 + V(x ). (7.94) 2m The Hamiltonian operator is the generator of motion in a quantum system. Here p and x and functions of them are operators in the Hilbert space. | (t)i is the Hilbert space vector describing fully the systems quantum state at time t. When looking for states that have a harmonic temporal behaviour H= | (t)i = ejEn t/~ |n i , we obtain the stationary Schroedinger Equation H |n i = En |n i , (7.96) (7.95)

that determines the energy eigenstates of the system. If the |n i build a complete basis of the Hilbert space, H , the system is dynamically evolving, the most general time dependent solution of the Schroedinger Equation is then a superposition of all energy eigenstates X an ejEn t/~ |n i . (7.97) | (t)i =


Schroedinger Equation in x-representation

We can return to wave mechanics by rewriting the abstract Schroedinger Equation in the eigenbase |xi of the position operator. For simplicity in notation, we only consider the one dimensional case and dene that there exists the following eigenvectors x |x0 i = x0 |x0 i , with the orthogonality relation hx |x0 i = (x x0 ). (7.99) (7.98)

342 CHAPTER 7. THE DIRAC FORMALISM AND HILBERT SPACES Note, since the position operator has a continuous spectrum of eigenvalues the orthogonality relation is a dirac delta function rather than a delta-symbol. The completness relation using this base is expressed in the unity operator Z 1 = |x0 i hx0 | dx0 , (7.100) rather then a sum as in Eq.(7.52). Inserting this unity operator in the Schroedinger Equation (7.93) and projecting from the left with hx|, we obtain Z (7.101) j ~ hx | (t)i = hx| H |x0 i hx0 | (t) i dx0 . t The expression hx | (t)i is the probability amplitude that a position measurement on the system in state | (t)i yields the value x, which is precisely the meaning of the wave function (x, t) = hx | (t)i (7.102)

in chapter 4. Using the eigenvalue property of the states and the orthogonality relations we obtain from Eq.(7.101) j~ ~ (x, t) = H (x, p = ) (x, t) . t j x (7.103)


Canonical Quantization

Thus the dynamics of a quantum system is fully determined by its Hamiltonian operator. The Hamiltonian operator is usually derived from the classical Hamilton function according to the Hamilton-Jacobi formulation of Classical Mechanics [3]. The classical Hamilton function H ({qi } , {pi }) is a function of the position coordinates of a particle xi or generalized coordinates qi and the corresponding momentum coordinates pi . The classical equations of motion are found by H ({qi } , {pi }) , pi p i (t) = H ({qi } , {pi }) . qi q i (t) = (7.104) (7.105)

In quantum mechanics the Hamiltonian function and the position and momentum coordinates become operators and quantization is achieved by



imposing on position and momentum operators that are related to the same degree of freedom, for example the x-coordinate of a particle and the associate momentum px , canonical commutation relations H ({qi } , {pi }) H({qi } , {pi }), [qi , pj ] = j ~ ij . (7.106) (7.107)

Imposing this commutation relation implies that position and momentum related to one degree of freedom can not be measured simultaneously with arbitrary precision and Heisenbergs uncertainty relation applies to the possible states the system can take on.


Schroedinger Picture

In the Schroedinger picture the quantum mechanical state of the system is evolving with time. If there is no explicit time dependence in the operators then the operators stay time independent. The Schroedinger Equation (7.93) j~ plus the initial state | (t = 0)i = | (0)i , (7.109) | (t)i = H | (t)i , t (7.108)

unquely determine the dynamics of the system. The evolution of the quantum state vector can be described as a mapping of the initial state by a time evolution operator. | (t)i = U(t) | (0)i . (7.110) If this solution is substituted into the Schroedinger Equation (7.108) it follows that the time evolution operator has to obey the equation j~ U(t) = H U(t). t (7.111)

For a time independent Hamiltonian Operator the formal integration of this equation is U(t) = exp [jHt/~] . (7.112)

344 CHAPTER 7. THE DIRAC FORMALISM AND HILBERT SPACES The time evolution operator is unitary U1 (t) = U+ (t) , (7.113)

because the Harmiltonian operator is hermitian, and therefore the norm of an initial state is preserved. The initial value for the time evolution operator is U(t = 0) = 1. (7.114) The expected value of an arbitrary operator A is given by h (t)| A | (t)i = h (0)| U+ (t)AU(t) | (0)i . (7.115)


Heisenberg Picture

Since the physically important quantities are the expected values, i.e. the outcome of experiments, Eq.(7.115) can be used to come up with an alternative formulation of quantum mechanics. In this formulation, called the Heisenberg picture, the operators are evolving in time according to AH (t) = U+ (t)AU(t), (7.116)

and the state of the system is time independent and equal to its initial state |H (t)i = |S (0)i . (7.117)

Clearly an expected value for a time dependent operator using the Heisenberg state (7.117) is identical with Eq.(7.113). This is identical to describing a unitary process in an eucledian vector space. Scalar products between vectors are preserved, if all vectors are undergoing a unitary transformation, i.e. a rotation for example. An alternative description is that the vectors are time independent but the coordinate system rotates in the opposite direction. When the coordinate system changes, the operators described in the time dependent coordinate system become timedependent themselves. From the denition of the time evolution operator we nd immediately an equation of motion for the time dependent operators of the Heisenberg picture AH (t) U(t) U+ (t) + j~ = j~ AS U(t) + U (t)AS j ~ (7.118) t t t AS U(t) +U+ (t) j~ t




AH (t) = U+ (t)H+ AS U(t) + U+ (t)AS HU(t) t AS + U(t) +U (t) j~ t


Using the relation U+ (t)U(t) = U(t)U+ (t) = 1 and inserting it between the Hamiltonian operator and the operator A, we nally end up with the Heisenberg equations of motion for the Heisenberg operators A j~ AH (t) = HH AH + AH HH + j~ (7.120) t t H A = [AH , HH ] + j ~ (7.121) t H with HH = U+ (t)HS U(t) = HS for conservative systems, i.e. HS 6= HS (t) (7.122) (7.123)

Note, that the last term in Eq.(7.121) is only present if the Schroedinger operators do have an explicit time dependence, a case which is beyond the scope of this class.


The Harmonic Oscillator

To illustrate the beauty and eciency in describing the dynamics of a quantum system using the dirac notation and operator algebra, we reconsider the one-dimensional harmonic oscillator discussed in section 4.4.2 and described by the Hamiltonian operator H= with [x, p] = j~. (7.125) p2 1 + K x2 , 2m 2 (7.124)



Energy Eigenstates, Creation and Annihilation Operators

It is advantageous to introduce the following normalized position and momentum operators X = K x ~ 0 1 P = p m~0 r (7.126) (7.127)

q with 0 = K . The Hamiltonian operator and the commutation relationship m of the normalized position and momentum operator resume the simpler forms ~ 0 2 P + X2 , 2 [X, P] = j . H =

(7.128) (7.129)

Algebraically, it is very useful to introduce the nonhermitian operators 1 a = (X + jP) , 2 1 a+ = (XjP) , 2 which satisfy the commutation relation a, a+ = 1. aa+ = 1 2 1 + a a = 2 2 j X + P2 [X, P] = 2 2 j 2 X + P + [X, P] = 2 1 2 1 2 2 X + P2 + 1 , (7.132) (7.130) (7.131)

We nd

(7.133) (7.134)

and the Hamiltonian operator can be rewritten in terms of the new operators a and a+ as

2 X + P2 1 ,



H =

~ 0 + a a + aa+ 2 1 + . = ~0 a a+ 2 N = a+ a,

(7.135) (7.136)

We introduce the operator (7.137) which is a hermitian operator. Up to an additive constant 1/2 and a scaling factor ~ 0 equal to the energy of one quantum of the harmonic oscillator it is equal to the Hamiltonian operator of the harmonic oscillator. Obviously, N is the number operator counting the number of energy quanta excited in a harmonic oscillator. We assume that the number operator N has eigenvectors denoted by |ni and corresponding eigenvalues n N |ni = a+ a |ni = n |ni . (7.138) We also assume that these eigenvectors are normalized and since N is hermitian they are also orthogonal to each other hm |ni = mn . (7.139)

Multiplication of this equation with the operator a and use of the commutation relation (7.132) leads to a a+ a |ni = na |ni + a a + 1 a |ni = na |ni N a |ni = (n 1) a |ni (7.140) (7.141) (7.142)

Eq.(7.140) indicates that if |ni is an eigenstate to the number operator N then the state a |ni is a new eigenstate to N with eigenvalue n 1. Because of this property, the operator a is called a lowering operator or annihilation operator, since application of the annihilation operator to an eigenstate with n quanta leads to a new eigenstate that contains one less quantum a |ni = C |n 1i , (7.143)

where C is a yet undetermined constant. This constant follows from the normalization of this state and being an eigenvector to the number operator. hn| a+ a |ni = |C |2 , n. C = (7.144) (7.145)



Clearly, if there is a state with n = 0 application of the annihilation operator leads to the null-vector in this Hilbert space, i.e. a |0i = 0, (7.147)

and there is no other state with a lower number of quanta. Thus, the ns must be whole numbers and |0i is the ground state of the harmonic oscillator, the state with the lowest energy, otherwise the iteration (7.146) would not stop. If a is an annihilation operator for energy quanta, a+ must be a creation operator for energy quanta, otherwise the state |ni would not fulll the eigenvalue equation Eq.(7.138) a+ a |ni = n |ni a+ n |n 1i = n |ni n |ni a+ |n 1i = or a+ |ni = n + 1 |n + 1i . (7.148) (7.149) (7.150)


Starting from the energy ground state of the harmonic oscillator |0i with energy ~ 0 /2 we can generate the n-th energy eigenstate by n-fold application of the creation operator a+ and proper normalization + n 1 |ni = p |0i , a (n + 1)! H |ni = En |ni , and En 1 = ~ 0 n+ . 2 (7.154) (7.152)






Matrix Representation

We can express the normalized position and momentum operators as functions of the creation and annihilation operators 1 X = 2 j P = 2 + a +a , + a a . (7.155) (7.156)

These operators do have the following matrix representations n m,n1 , hm| a+ |ni = n + 1 m,n+1 , (7.157) hm| a |ni = hm| aa+ |ni = (n + 1) m,n , (7.158) hm| a+ a |ni = n m,n , 1 (7.159) n + 1 m,n+1 + n m,n1 , hm| X |ni = 2 j n + 1 m,n+1 n m,n1 , hm| P |ni = (7.160) 2 p n(n 1) m,n2 , p (n + 1) (n + 2) m,n+2 , hm| a+2 |ni = p 1 (2n + p 1) m,n + n(n 1) m,n2 2 , hm| X |ni = 2 + (n + 1) (n + 2)m,n+2 p 1 (2n + p 1) m,n n(n 1) m,n2 2 . hm| P |ni = 2 (n + 1) (n + 2) m,n+2 hm| a2 |ni = (7.161) (7.162) (7.163) (7.164)


Minimum Uncertainty States or Coherent States

From the matrix elements calculated in the last section, we nd that the energy or quantum number eigenstates |ni have vanishing expected values for position and momentum. This also follows from the x-representation n (x) = hx |ni studied in section 4.4.2 hn| X |ni = 0 , 1 , 2 hn| P |ni = 0 , 1 . 2 (7.165) and the uctuations in position and momentum are then simply hn| X2 |ni = n + hn| P2 |ni = n + (7.166)

350 CHAPTER 7. THE DIRAC FORMALISM AND HILBERT SPACES The minimum uncertainty product for the uctuations q 1 X = hn| X2 |ni hn| X |ni2 = n + , 2 q 1 hn| P2 |ni hn| P |ni2 = n + . P = 2 is then X P = n +

(7.167) (7.168)

1 . (7.169) 2 Only the ground state n = 0 is a minimum uncertainty wave packet, since it satises the eigenvalue equation a |0i = 0, where (7.170)

1 (7.171) a = (X + jP) , . 2 In fact we can show that every eigenstate to the annihilation operator a |i = |i , for C (7.172)

is a minimum uncertainty state. We obtain for expected values of position or momentum in these states h| a |i = , h| a+ |i = , 1 j h| X |i = ( + ) , h| P |i = ( ) , 2 2 and for its squares h| a+ a |i = ||2 , h| aa+ |i = ||2 + 1 , h| a+2 |i = 2 , h| a2 |i = 2 , 1 2 1 h| X2 |i = + 2 + 2 + 1 = h| X |i2 + , 2 2 1 2 1 2 2 2 + 2 + 1 = h| P |i + . h| P |i = 2 2 (7.175) (7.176) (7.177) (7.178)

(7.173) (7.174)

Thus the uncertainty product is at its minimum



1 C. (7.179) 2 In fact one can show that the statistics of a position or momentum measurement for a harmonic oscillator in this state follows a Gaussian satistics with the average and variance given by Eqs.(7.173), (7.177) and (7.178). This can be represented pictorially in a phase space diagram as shown in Figure 7.1 X P =

P <P>

2X 2P


Figure 7.1: Representation of a minimum uncertainty state of the harmonic oscillator as a phase space distribution.


Heisenberg Picture

The Heisenberg equations of motion for a linear system like the harmonic oscillator are linear dierential equations for the operators, which can be easily solved. From Eqs.(7.121) we nd j~ aH (t) = [aH , H] t = ~ 0 aH , aH (t) = ej0 t aS . (7.180) (7.181) (7.182)

with the solution Therefore, the expectation values for the creation, annihilation, position and momentum operators are identical to those of Eqs.(7.173) to (7.178); we only

352 CHAPTER 7. THE DIRAC FORMALISM AND HILBERT SPACES need to subsitute ej0 t . We may again pictorially represent the time evolution of these states as a probability distribution in phase space, see Figure 7.2.
P 2X <P> 0 e
-jO t



Figure 7.2: Time evolution of a coherent state in phase space.


The Kopenhagen Interpretation of Quantum Mechanics

Description of the State of a System

At a given time t the state of a system is described by a normalized vector |(t)i in the Hilbert space, H . The Hilbert space is a linear vector space. Therefore, any linear combination of vectors is again a possible state of the system. Thus superpositions of states are possible and with it come interferences.


Description of Physical Quantities

Measurable physical quantities, observables, are described by hermitian operators A = A+ .



The Measurement of Observables

An observable has a spectral representation in terms of eigenvectors and eigenvalues, which can be discrete or continuous, here we discuss the discrete case X A= An |An i hAn | , (7.183)

The eigenvectors are orthogonal to each other and the eigenvalues are real hAn | An0 i = n,n0 .


Upon a measurement of the observable A of the system in state |(t)i the outcome can only be one of the eigenvalues An of the observable and the probability for that event to occur is pn = | hAn | (t)i|2 . (7.185)

If the eigenvalue spectrum of the operator A is degenerate, the probabilities of the probabilities of the dierent states to the same eigenvector need to be added. After the measurement the system is in the eigenstate |An i corresponding to the eigenvalue An found in the measurement, which is called the reduction of state[1]. This unphysical reduction of state is only necessary as a shortcut for the description of the measurement process and the fact that the system becomes entangled with the state of the macroscopic measurement equipment. This entanglement leads to a necessary decoherence of the superposition state of the measured system, which is equivalent to assuming a reduced state.


[1] Introduction to Quantum Mechanics, Griths, David J., Prentice Hall, 1995. [2] Functional Analysis, G. Bachman and L. Naricci, Academic Press, 1966. [3] Classical Mechanics, H. Goldstein, Addison and Wesley series in physics, 1959. [4] Quantum Mechanics I, C. Cohen-Tannoudji, B. Diu, F. Laloe, John Wiley and Sons, Inc., 1978.