Volker Perlick Ray Optics, Fermat’s Principle, and Applications to General Relativity 6 SpringerVolker Perlick Ray Optics, Fermat’s Principle, and Applications to General Relativity 6 SpringerAuthor Volker Perlick Technische Universitit Berlin Sekr. PN 7-1 Hardenbergstrasse 36 10623 Berlin, Germany Library of Congress Cataloging in-Publication Data Perlick, Volker, 1956- ay oes, Fema’ prncil, napisy e Vole etic ‘Dotes in physics. Monographs, ISSN 0940-7677 ; v. m61) Includes bibhographial references and index ISBN 3540668585 (alk paper) jght-~Transmission—Mathematical models. 2. Maxwell equations. 3. General relativity (Physies) 1. Title. Il Lecture notes in physics. New series m, Monographs : ml. 389 P37 2000 01'S3~de2 99-089303 ISSN 0940-7677 (Lecture Not ISBN 3-540-66898-5 Springer. Physics. Monographs) ferlag Berlin Heidelberg New York This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of itystrations recitation, broadcasting, reproduction on microflm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use mus always be obtained from Springer-Verlag, Violations are liable for prosecution under the German Copyright Law. ‘© Springer-Verlag Berlin Heidelberg 2000 Printed in Germany ‘The use of general descriptivenames, registered names, trademarks, etc. inthis publication the absence ofa specific statement, that such names are exempt fom the relevant protective laws and regulations and therefore free for general use. ‘Typesetting: Camera-ready by the author Cover design: design & production, Heidelberg Printed on acid-free paper SPIN:10644482 ss/siga/du-5 43210Preface All kind of information from distant celestial bodies comes to us in the form of electromagnetic radiation. In most cases the propagation of this radiation ccan be described, as a reasonable approximation, in terms of rays. This is ‘rue not only in the optical range but also in the radio range of the electro- ‘magnetic spectrum. For this reason the laws of ray optics are of fundamental importance for astronomy, astrophysics, and cosmology. According to general relativity, ight rays are the light-like geodesics of a Lorentzian metric by which the spacetime geometry is described. This, how- ever, is true only as long as the light rays are freely propagating under the only influence of the gravitational field which is coded in the spacetime ge- cometry. Ifa light ray is influenced, in addition, by an optical medium (e.¢., by a plasma), then it will not follow a light-like geodesic of the spacetime metric. It is true that for electromagnetic radiation traveling through the universe usually the influence of a medium on the path of the ray and on the frequency is small. However, there are several cases in which this infiu- cence is very well measurable, in particular in the radio range. For example, the deflection of radio rays in the gravitational field of the Sun is consider- ably influenced by the Solar corona. Moreover, current and planned Doppler experiments with microwaves in the Solar system reach an accuracy in the frequency of Aw/w * 10-5 which makes it necessary to take the influence of the interplanetary medium into account. Finally, even in cases where the ‘quantitative influence of the medium is negligibly small it is interesting to ask in which way the qualitative aspects of the theory are influenced by the ‘medium. The latter remark applies, in particular, to the intriguing theory of gravitational lensing. Unfortunately, general-relativistc light propagation in media is not usu- ally treated in standard textbooks, and the more specialized literature is concentrated on particular types of media and on particular applications rather than on general methodology. In this sense a comprehensive review of general-relativistic ray optics in media would fill a gap in the literature. It is the purpose of this monograph to provide such a review. Actually, this monograph grew out of a more special idea. It was my orig- inal plan to write a review on variational principles for ight rays in general relativistic media, and to give some applications to astronomy and astro-‘VI Preface physics, in particular to the theory of gravitational lensing. However, I soon realized the necessity of precisely formulating the mathematical theory of light rays in general before I could tackle the question of whether these light rays are characterized by a variational principle, ‘The sections on variational principles and on applications are now at the end of Part Il, in which a gen- eral mathematical framework for ray optics is set up. This is written in the language of symplectic geometry, thereby elucidating the well-known analogy between ray optics and the phase-space formulation of classical mechanics. Moreover, I found it desirable to also treat the question of how to derive ray optics as an approximation scheme from Maxwells equations. This is the topic of Part I which serves the purpose of physically motivating the fandamental definitions of Part II. In vacuo, the passage from Maxwell's ‘equations to ray optics is, of course, an elementary textbook matter and the generalization to isotropic and non-dispersive media is quite straightforward. However, for anisotropic and/or dispersive media this passage is more subtle. In Part I two types of media are discussed in detail, viz., an anisotropic one and a dispersive one, and the emphasis is on general methodology. Thave organized the material in such a way that it should be possible to read Part II without having read Part I. This is not recommended, of course, but the reader might wish to do so. Both parts begin with an introductory section containing a brief guide to the literature and a statement of assump- tions and notations used throughout. Whenever the reader feels that a symbol needs explanation or that the underlying assumptions are not clearly stated, he or she should consult the introductory section of the respective part. Also, the index might be of help if problems of that kind occur. Large parts of this monograph present material which, in essence, is not new. However, I hope that the formulation chosen here might give some new insight. As to Part I, our discussion of the passage from Maxwell's equations to ray optics includes several mathematical details which are difficult to find in the literature, although the general features are certainly known to experts. ‘To mention just one example, itis certainly known to experts that in a linear but anisotropic medium on a general-relativistic spacetime the light rays are determined by two “optical Finsler metrics”; to the best of my knowledge, however, a full proof ofthis fact is given here for the first time. As to Part II, the basic formalism is just the 170-year-old Hamiltonian optics, rewritten in modern mathematical terminology and adapted to the framework of general relativity. However, the presentation is based on some general mathematical definitions which have not been used before. This remark applies, in partic- ular, to Definition 5.1.1, which is the definition of what I call “ray-optical structures”. This definition formalizes the widely accepted idea that all of ray optics can be derived from a “dispersion relation”. (The term “ray sys- tem” is sometimes used by Vladimir Arnold and his collaborators in a similar though not quite identical sense.) It also applies, e.g, to Definition 6.4.1, onPreface VI “dilation-invariant” ray optical structures, which characterizes dispersion-free media in a geometric way. On the other hand, I want to direct the reader's attention to the fact that this monograph contains some particular results which, as far as I know, have not been known before. These include, e.g. «© the general redshift formula for light rays in media on a general-relativistic spacetime in Sect. 6.2; ‘* the results on light bundles in isotropic non-dispersive media on a general- relativistic spacetime in Sect. 6.4, in particular the generalized “reciprocity theorem” (Theorem 6.4.3); © Theorem 7.3.1, which can be viewed as a version of Fermat's principle for light rays in (possibly anisotropic and dispersive) media on general- relativistic spacetimes; © Theorem 7.5.4, which generalizes the “Morse index theorem” of Rieman- nian geometry to the case of light rays in stationary media on stationary general-relativistic spacetimes. Some of the questions raised in this monograph remain unanswered, j.e., to some extent this is an interim report on work in progress. In particular, this remark applies to the following two special issues. (a) In Part I we are able to prove that for the linear medium treated in Chap. 2 ray optics is associated ‘with approximate solutions of Maxwell's equations, Le. that ray optics gives ‘a viable approximation scheme for electromagnetic radiation. Unfortunately, ‘we are not able to prove a similar result for the plasma model of Chap. 3. ‘This is a gap which should be filled in the future. (b) In Part II we are able to establish a Morse index theorem for light rays in stationary media, However, itis still an open question whether these results can be generalized to the non-stationary case in which, up to now, a Morse theory exists only for vacuum rays. With Fermat's principle in the form of Theorem 7.3.1 we have a starting point for setting up a Morse theory for light rays In arbitrary (non-stationary) media. This is an interesting problem to be tackled in future work. ‘This monograph in its present form is a slightly revised version of my Ha- bilitation thesis. I would like to use this opportunity to thank the members of the Habilitation Committee, Karl-Eberhard Hellwig, Erwin Sedlmayr, Bernd ‘Wegner, John Beem, Friedrich Wilhelm Heh, and Gernot Neugebauer, for their interest in this work and for several useful comments. In particular, I ‘would like to thank Bernd Wegner for paving the way to having this text published with Springer Verlag. While working at this monograph I have profited from many discussions, in particular with my academic teacher Karl-Bberhard Hellwig and his collab- orators at the Technical University in Berlin, but also with other colleagues. Special thanks are due to Wolfgang Hasse and Marcus Kriele for collabora- tion on various aspects of light propagation in general relativity; to Wolfgang‘VIII Preface Rindler for hospitality at the University of Texas at Dallas and for discus- sions on the fundamentals of general relativity; to John Beem for hospitality at the University of Missouri at Columbia and for discussions on Lorentzian geometry; to Gernot Neugebauer and his collaborators for hospitality at the University of Jena and for discussions on various aspects of general relativity; to Paolo Piccione, Fabio Giannoni, and Antonio Masiello for hospitality dur- ing several visits to Italy and to Brazil and for collaboration on Morse theory; and to Jiirgen Ehlers and Arlie Petters for fruitful discussions on Fermat’s principle and gravitational lensing. Also, I have enjoyed discussions on this subject with students during seminars and classes in Berlin, Osnabrick, and So Paulo. Finally, I am grateful to the Deutsche Forschungsgemeinschaft for spon- soring this work with a Habilitation stipend, and to the Wigner Foundation, to the Deutscher Akademischer Austauschdienst, and to the Fundagéo de ‘Amparo 4 Pesquisa do Estado de So Paulo for financially supporting my visits to Dallas, Columbia, and Séo Paulo. Berlin, August 1999 Volker PerlickContents Part I. From Maxwell’s equations to ray optics 1. Introduction to Part I......... 11 A brief guide to the literature 1.2 Assumptions and notations 2. Light propagation in linear dielectric and permeable media 2.1 Maxell’ equations in nea deletic and penoeabe media 2.2 Approximate-plane-wave families . : 23. Asymptotic solutions of Maxwell's equations -. 2.4 Derivation ofthe eikonal equation and transport equations 2.5 Discussion of the eikonal equation ........+.+++. 2.6 Discussion of transport equations and the introduction of rays 2.7 Ray optics as an approximation scheme SS 3. Light propagation in other kinds of media 8.1 Methodological remarks on dispersive media . 3.2 Light propagation in a non-magnetized plasma .. 19 31 Part I. A mathematical framework for ray optics 4, Introduction to Part II 4.1 A brief guide to the literature 4.2 Assumptions and notations 5. Ray-optical structures on arbitrary manifolds. : 5.1 Definition and basic properties of ray-optical structures. 5.2 Regularity notions for ray-optical structures . 5.3 Symmetries of ray-optical structures ...... 54. Dilationlnvarantrayoptical sructures 5.5 Eikonal equation .... 5.6 Causties........ . 67 222 1008. References. Contents Ray-optical structures on Lorentzian manifolds ........... 111 6.1 The vacuum ray-optical structure. . ut 6.2 Observer fields, frequency, and redshift. 3 6.3 Isotropic ray-optical structures 120 6.4 Light bundles in isotropic media. 123 6.5 Stationary ray-optical structures. . 131 6.6 Stationary ray optics in vacuum and in simple media, 141 Variational principles for rays....... 4g 7.1 The principle of stationary action: The general case. 149 7.2 The principle of stationary action: The strongly regular case . 154 7.3 Fermat’s principle . 156 74 A Hilbert manifold setting for variational problems 165 7.5 A Morse theory for strongly hyperregular ray-optical structures ............. 168 Applications ...... + 183, 8.1 Doppler effect, aberration, and drag effect in isotropic media . 183, 8.2. Light rays in a uniformly accelerated medium on Minkowski space... ...+.sse+2+ 8.3. Light propagation in a plasma on Kerr spacetime . 8.4 Gravitational lensing. fe anPart I From Maxwell’s equations to ray optics1. Introduction to Part I In Part I we recapitulate the general ideas of how to derive the laws of ray optics from Maxwell’s equations. We presuppose a general-relativistic spacetime as background, and we consider media which are general enough to elucidate all relevant features of the method. Chapter 2 treats the case of a linear (not necessarily isotropic) dielectric and permeable medium in full detail. Chapter 3 discusses dispersive media in general and a simple plasma model in particular. In this way the material presented in Part I serves two purposes. First, it motivates our mathematical frame-work for ray optics, to be set up in Part II below. Second, it provides us with physically important examples of ray optical structures to which we shall recur frequently. 1.1 A brief guide to the literature In Part I we have to assume some familiarity on the reader’s side with ‘Maxwell’s equations in matter on a general-relativistic spacetime. Whereas vacuum Maxwell's equations are detailed in any textbook on general relativ- ity, the matter case is not usually treated in extenso. For general aspects of ‘the phenomenology of electromagnetic media in general relativity we refer to Bressan [17] who gives many earlier references. The case of a linear (not necessarily isotropic) dielectric ‘and permeable medium which is at the basis of Chap. 2 is briefly treated by Schmutzer [127], Chap. IV, following an original article by Marx {91}. ‘The general-relativistic plasma model which is at the basis of Chap. 3 is systematically treated in two articles by Breuer and Ehlers (18) (19); for earlier ‘work on the same subject we refer to Madore [90], to Biédk and Hadrava (14), and to Anile and Pantano (5) (6) As an aside it should be mentioned that the phenomenological theory of electromagnetic media can be derived from electron theory by statisti- cal methods, ic., that the macroscopic (phenomenological) Maxwell equa- tions can be derived from a sort of microscopic Maxwell equations. For linear isotropic media in inertial motion on flat spacetime this is a standard text- book matter; the generalization te accelerated media is due to Kaufmann (67). For a general-relativistic plasma, the derivation of phenomenological proper-41, Introduction to Part I ties from the kinetic theory of photons is discussed in the above-mentioned article by Bigék and Hadrava (14 ‘The main topic of Part Iis the derivation of the laws of ray optics from Maxwell's equations. The basic idea is to make an approzimate-plane-wave ansatz for the electromagnetic field and to assume that this ansatz satis- fies Maxwell's equations in an asymptotic sense for high frequencies. This results in a dynamical law for wave surfaces which can be rewritten equiv- alently as a dynamical law for rays. In optics the dynamical law for wave surfaces is usually called the eikonal equation. It is formally analogous to the Hamilton-Jacobi equation of classical mechanics, whereas the dynamical law for rays is formally analogous to Hamilton's equations. Mathematically, this so-called ray method is, of course, not restricted to Maxwell’s equations but applies equally well to other partial differential equations with or without, relevance to physics. In this sense, the ray method has applications not only to optics but also to acoustics and to wave mechanics. In the latter context, the ray method is known as JWKB method, refering to the pioneering work of Jeffreys, Wentzel, Kramers and Brioullin, and is detailed in virtually any textbook on quantum mechanics. In this brief guide to the literature we shall concentrate on the ray method in optics. As to other applications we refer to the comprehensive list of refer- ences given in monographs stich as Keller, Lewis and Seckler {70] or Jeffrey and Kawahara [66]. Purely mathematical aspects of the ray method can be found in textbooks on partial differential equations. Particularly useful for our purposes are, e-g., the books by Chazarain and Piriou [26] and by Egorov and Shubin [36]. Whereas rudiments of the ray method can be traced back to work of Li- ouville and Green around 1830, it was first carried through in the context of optics by Sommerfeld and Runge (132] in the year 1911, following a sug- gestion by Debye. The work of Sommerfeld and Runge was restricted to the vacuum Maxwell equations in an inertial system, and the only goal was to derive the corresponding eikonal equation. Their treatment was generalized ‘and systematized by Luneburg [88] who considered infinite asymptotic series solutions rather than just asymptotic solutions of lowest order as Sommer- feld and Runge did. Later, the method was extended from the vacuum case to the case of light propagation in matter. This was a very active field of reasearch in the 1960s, see, e.g,, Lewis (84), Chen (27] and Kravtsov [75]. All these papers are restricted to special relativity in the sense that they are resupposing a flat spacetime. Nonetheless, the techniques used are of inter- est also in view of general relativity. The reason is that vacuum Maxwell's equations on a general-relativistic spacetime are very similar to Maxwell's ‘equations in an inhomogeneous medium on flat spacetime, at least locally. ‘This was first observed by Plebatiski (120]. Note, however, that global as- ects which do not carry over to general relativity are brought into play whenever temporal Fourier expansions (as eg. by Lewis (84]) and/or spatial1.2 Assumptions and notations 5 Fourier expansions (as e.g. by Chen (27]) are used. A global treatment of the ray method that does carry over to general relativity is possible in terms of ‘the Lagrangian manifold techniques introduced in the 1960s by Maslov and Amold, see Arnold (7), Maslov (94), Duistermaat (31] or Guillemin and Stern- berg 55]. In Part I we are concerned with local questions only. However, we shall touch upon Lagrangian manifold techniques and their relevance for the investigation of caustics in Part II below. Tn general relativity, the passage from Maxwell's equations to ray optics ‘was carried through for the first time by Laue [77] in the year 1920. In this paper, which is the written version of a talk given by Laue at the 86. Natur- forscherversammlung, the author demonstrated how to derive from vacuum ‘Maxwells equations on a curved spacetime the light like geodesic equation for the rays. Laue's treatment followed closely the seminal paper by Sommerfeld ‘and Runge (132). A more systematic general-relativistic treatment of the ray ‘method in optics, including asymptotic solutions of arbitrarily high order, was brought forward much later by Ehlers (38). He considered linear isotropic non- dispersive media on an arbitrary general-relativistic spacetime and derived not only the eikonal equation for the rays but also transport equations of arbi- trary order for the polarization plane along the rays. In particular, his results put earlier findings about light propagation in such media by Gordon [60] and Pham Mau Quan [117] on a mathematically firm basis. At least for the ‘vacuum case, the main results can now be found in many textbooks on gen- ‘eral relativity, see, e.g., Misner, Thorne and Wheeler [98], Straumann {136], or Stephani [133]. The general-relativistic relevance of higher order terms in the asymptotic series expansion was discussed by Dwivedi and Kantowski [2] and by Anile [4]. A goneral-relativistic treatment of the ray method for dispersive media, exemplified with a special plasma model, is due to Breuer and Ehlers [18] [19] who modified and enhanced earlier work by Madore [90], by Bigék and Hadrava (14), and by Anile and Pantano [5] [6] 1.2 Assumptions and notations ‘We assume a general-relativistic spacetime, ie. a four-dimensional O° man: ifold with a metric of Lorentzian signature (-+,-+,+,—-). On this spacetime background we consider Maxwell’s equations, using units making the dielec- tricity and permeability constants of vacuum equal to one, E> = jp = 1 ‘Thereby, in particular, the vacuum velocity of light is set equal to one. We restrict ourselves to the C™ category in the sense that throughout Part I all maps and tensor fields are tacitly assumed to be infinitely often differentiable. We work in local coordinates using standard index notation, Throughout, Einstein's summation convention is in force with latin indices running from 1 to 4 and with greek indices running from 1 to 3.The (covariant) components of the spacetime metric will be denoted by gab. As usual, we define g by gab 9% = 53, where 5g denotes the Kronecker delta, and we use gap (and g*,6 1, Introduction to Part I respectively) to lower (and raise, respectively) indices. With respect toa co- ordinate system x = (z!,2?,2°,z"), partial derivatives 22- will be denoted by 8, for short, whereas 'V_ means covariant derivative with respect to the Levi-Civita connection of our metric. For the sake of brevity, we shall speak of “a tensor field Q*°” if we mean “a tensor field whose contravariant com- ponents in a coordinate system are Q*” etc. Our treatment will be purely local throughout Part I. ‘Therefore, the use of local coordinates and index notation is no restriction whatsoever.2. Light propagation in linear dielectric and permeable media On our spacetime manifold we consider Maxwell's equations in a linear but not necessarily isotropic medium, i. e., in a medium phenomenclogically char- acterized by a dielectricity tensor field and a permeability tensor field. It is our goal to detive and to discuss the laws of ray optics in such a medium, ‘The standard textbook problem of light propagation in vacuo is, of course, included as a special case. ‘The results of this chapter cover a wide range of applications including light propagation in gases (isotropic case) and crystals (anisotropic case) as long as dispersion is ignored. For dispersive media we refer to Chap. 3 below. In view of applications to astrophysics, the isotropic case is more interesting than the (much more complicated) anisotropic case. On the other hand, a thorough treatment of the anisotropic case is highly instructive from ‘a methodological point of view. In particular, it gives us the opportunity to discuss the phenomenon of bitefringence. 2.1 Maxwell’s equations in linear dielectric and permeable media On our spacetime manifold, the source-free Maxwell equations for (macro- scopic) electromagnetic fields in matter can be written in local coordinates m4 TeFea=0 and V'Gr.=0, (2.1) or, using partial rather than covariant derivatives, as f°! O,Fea =0 and 1% 55(nea®! Gey) = 0. (22) In (2.1) and (2.2), nates denotes the totally antisymmetric Levi-Civita tensor field (volume form) of our metric which is defined by the equation masa = +y/[det(Gea)] - (2.3) Here the plus sign is valid if the coordinate system is right-handed and the minus sign is valid if itis left-handed. In other words, we have to choose anpropagation in linoar dielectric and permeable m orientation on the domain of our coordinate system to fix the sign ambiguity of the Levi-Civita tensor field. However, this is irrelevant since Maxwell's ‘equations are invariant under nated + ~Mabed) and so are all the relevant results in Part I. If the reader is not familiar with volume elements he or she may consult, e.g., Wald [146], p. 432. Fay and Gas denote the electromagnetic field strength and the electromag- netic excitation, respectively, both of which are antisymmetric second rank tensor fields. With respect to a reference system, given in terms of a time-like vector field U* with UU, = —1, we can introduce the electric field strength Eq =FapU® (2.4) and the magnetic field strength Ba =~} Mate U? FO (2.5) such that Fay = ~1%a BeUs + By Us — Ea Us» (2.6) Here we have used the familiar property. naboa tt = —55 51 85 — 65 65 5 — 6555 6 + 53 62 SR + 65 5f 8 + 55 54 5k (2.7) of the Levi-Civita tensor field, ef, e.g,, Wald [146], equation (B.2.12) Similarly, we introduce the electric excitation (2.8) and the magnetic excitation He (2.9) such that Gas = - 14a» HeUa + Dye - Dae (2.10) ‘With respect to the reference system used for their definitions, the electric and magnetic field strengths are purely spatial one-forms, and so are the electric and magnetic excitations, E,U*=B,U°=0 and DaU' HaU°=0 (a1) Our terminology of calling E, and Bg the “field strengths” (in german: Feldstirken) and D, and Hq the “excitations” (in german: Erregungen) fol- lows Gustav Mie and Arnold Sommerfeld. This terminology is reasonable2.1 Maxwells equations in linear dicloctric and permeable media 9 since B, and By determine the Lorentz force exerted on a charged test parti- cle whereas, in the presence of field-producing charges and currents, D, and Hg are the fields “excited” by those sources via Maxwell's equations. The traditional terminology of calling Ha the “magnetic field strength” is mis- leading. Moreover, it is highly inconvenient from a relativistic point of view where E, and B,, rather than B and Hg, are united into an antisymmetric second rank tensor field on spacetime. In what follows we consider Maxwell's equations in the form (2.2). As long as only the metric is known, (2.2) gives us eight component equations for twelve unknown functions. (‘The unknown functions are the six independent components of the electromagnetic field strength plus the six independent components of the electromagnetic excitation.) Hence, (2.2) is an underde- termined system of partial differential equations. It must be supplemented by constitutive equations relating the electromagnetic feld strength with the electromagnetic excitation. Thereby the medium is characterized in a phe- nomenological way. In this chapter we consider linear dielectric and permeable ‘media according to the following definition. Definition 2.1.1. A linear dielectric and permeable medium is, by defini- tion, a medium characterized by constitutive equations that take the form Dy =6a°Ey and Ba = jie’ Hy, (2.12) ‘in some reference system U, with second rank tensor fields ¢,* and jg? sat- isfying the following conditions: (a) ao =0. (b) e* a (c) 8 Za Zp > 0 and i*” Z, Zy > 0 for all (Z1, Zz, Zs, Zs) # (0,0,0,0) with U8 Z_ =0. We refer to the distinguished reference system U> as to the rest system, to eq as to the diclectrcity tensor field and to 1,” as to the permeability tensor field of the medium, Condition (a) of Definition 2.1.1 guarantees that the constitutive equa- tions (2.12) are in agreement with Da U® = 0 and B, U® = 0, Conditions (b) and (c) imply that in the rest system of the medium the energy density w=}(D.E*+ Bal"), (2.18) of the electromagnetic field is positive definite. Altogether, conditions (a), (b) and (c) guarantee that the dielectricity and permeability tensor fields are “spatially invertible”. We can, thus, define (y~*),? by the properties Ue (ja =0, orth = ty, 14) (Nat wa? = 55 + Ua UF10 2. Light propagation in linear diloctsc and permeable media, ‘The constitutive equations 2.12) ean then be united in single equation, Gan = (Janu Ne! ter" Us + es UU — ce? UUs) Fog (2-15) ‘The following special case deserves particular interest. Definition 2.1.2. A linear dielectric and permeable medium is called iso- tropic if the dielectricity and permeability tensor fields are of the special form ea? =e (5+U.U*) and pe? = (5 +U,U"), (2.16) ‘with some scalar functions € and jt. (Condition (c) of Definition 2.1.1 then requires © and ys to be strictly positive.) In the isotropic case, (2.15) reduces to + (1 en) (FoaUs — PoaUa)U4) « (2.7) In particular, vacuum can be characterized as a linear isotropic medium with € 1. In this case (and, more generally, in any isotropic medium with eq = 1) U* drops out from (2.17) and the constitutive equations take the form (2.12) in any reference system. This is in agreement with the obvious fact that for vacuum any reference system can be viewed as the rest system. of the medium, ‘We emphasize that our phenomenological constitutive equations are phys- ically reasonable in the rotational as well as in the irrotational case, ie., U® need not be hypersurface-orthogonal. Although this should be clear from the general rules of relativity, there is still a debate on this issue, even in the case of an isotropic medium on flat spacetime, see, e., Pellegrini and Swit [106]. We are now going to analyze the dynamics of electromagnetic fields in ‘linear dielectric and permeable medium. We have already mentioned that ‘Maxwell's equations (2.2) alone give us eight equations for twelve unknown functions. With (2.12) at hand, and assuming that U*, ¢," and p4,? are known, ‘we can eliminate six of the unknown functions. Now (2.2) gives us eight equa- tions for six functions, ie., the system looks overdetermined. However, only six of those eight equations are evolution equations, governing the dynamics of electromagnetic fields, whereas the other two equations are constraints. This is most easily verified in a local coordinate system (21,27,2°,) in which the hypersurfaces 2* = const. are space-like such that 2* can be viewed as a local time function. Owing to the antisymmetry of the Levi-Civita tensor, the a = 4 components of equations (2.2) do not involve any 9, derivative. Hence, these two equations are to be viewed as constraints whereas the te- maining six equations, ie., the a = 1,2,3 components of equations (2.2), are the evolution equations governing the dynamics. Again owing to the anti- symmetry of the Levi-Civita tensor field, the evolution equations preserve the constraints in the following sense. If the constraints are written in the form2.1 Maxwells equations in linear dielectric and permeable media 11 ©; =0 and C2 = 0, then the evolution equations imply that 801 = fi Cr and 04C2 = faC2 with some spacetime functions f, and fa. Hence, if a solution of the evolution equations satisfies the constraints on some initial hypersurface «' = const., then it satisfies the constraints everywhere (on some neighborhood of any point of the initial hypersurface, that is). In other ‘words, locally around any one point all solutions of Maxwell's equations can be found in the following way. Step 1. Choose a space-like hypersurface through that point, Step 2. Choose a local coordinate system such that the chosen hyper- surface is given by the equation z* = const. Step 3. Solve the evolution equations with all initial data that satisty the constraints. In the rest of this section we shall prove that the initial value problem con- sidered in Step 3 is well-posed in the sense that it is characterized by a local ‘existence and uniqueness theorem, provided that the initial hypersurface has been chosen appropriately. Conditions (a), (b) and (c) of Definition 2.1.1 will prove essential for this result. First we introduce special coordinates according to the following defini- tion. Definition 2.1.3. Let U* denote the rest system of a linear dielectric and permeable medium and fix a spacetime point 29. Then a local coordinate sys- tem (2!,22,23,24), defined on a neighborhood of zo, is called adapted to U* near 29 if (a) U* is given by the equation U* = b= 63 , (b) gan =0 for x =1,2,3 at the point zo . For any linear dielectric and permeable medium, itis obvious that adapted coordinates are characterized by the following existence and uniqueness prop- erty. If we choose a spacetime point zo and a hypersurface S that is orthogonal to U® at zo, then there is a coordinate system adapted to U® near zo such that S is represented by the equation «* = const, Another coordinate system (@!,27,2°3,2't) is, again, adapted to U* near zo if and only if it is related to (e4,22,2°, 24) by a coordinate transformation of the special form ar alM(at, 22,29) (2.18) ato al(at2,2504) mn (b) of Definition 2.1.3 makes sure that at the point zp the hypersurface a* = const. intersects the respective integral curve of U® or- thogonally. ‘This implies that, on a sufficiently small neighborhood of zo, all hypersurfaces 2 = const. are space-like. Of course, they cannot be orthogo- nal to U* on a whole neighborhood unless the medium is non-rotating. Hence,12 2. Light propagation in linear dielectric and permeable media in an adapted coordinate system the mixed components ga and gM of the ‘metric need not vanish except at the central point zo. The spatial components give positive definite 3 x 8 matrices (gj) and (gH”) on some neighborhood of 20; at the point 29 these matrices are inverse to each other. The temporal components gs and g* are strictly negative functions on some neighborhood cof zo ; at the point 29 they are inverse to each other. Now we consider a linear dielectric and permeable medium in a coordinate system adapted to its ret system U® near some point 29, Then (2.11) reduces to By=By=0 and Dy=Hy=0 (2.19) owing to condition (a) of Definition 2.1.3. Hence, (2.12) simplifies to Doe =o" Ey and By = to” Hy (2.20) Conditions (b) and (c) of Definition 2.1.1 guarantee that £2? and jo? are positive definite and symmetric with respect to gf". We can, thus, define vo? and wz, which are again positive definite and symmetric with respect to gf", by bu Vo" ve? = OB and 6,7 U9" w,? = 6 (2.21) For the following it will be convenient to introduce the quantities 2, =%p°Be and Y,=tp"De, (2.22) and to use Z1, Z2, Zs,¥i, Ya, Ys for the six independent components of the electromagnetic field. That is to say, we start from Maxwell's equations (2.2) ‘with (2.6) and (2.10); we use part (a) of Definition 2.1.3 and equation (2.19); wwe eliminate B, and H, with the help of (2.20); finally, we express Dy and Bz in terms of Z, and Y, by means of (2.22). After a little bit of algebra, ‘the a = 1,2,3 components of Maxwell’s equations (2.2) give us evolution ‘equations of the form ano for the dynamical variables Zy ‘Ye 2-(2) and v-() 2 2a, %, Here L', L?, L° and L‘ are z-dependent 6 x 6 matrices of the form ) and P= (2 oo) (2.25)2.1 Maxwell's equations in linear dielectric and permeable media 13 where Q is a 3 x 3 matrix with components Qa7 = wy nal vy” = wy” v9 Vy” gp4 (2.26) A? is a 3 x 3 matrix with components (227) (.)T means transposition with respect to 9?” such that, e.g. (QT)x" 9 = Qu? a; (2.28) M is a6 x 6 matrix whose components involve the spacetime metric along with oy” and w,". For the investigation of the evolution equations (2.28) the following two observations are crucial, (@) At the central point 2» of our adapted coordinate system we have ga = 0 and, thus, Q = 0. By continuity, L' is invertible on some neighborhood of zo to which we can restrict our considerations. Hence, (2.23) can be solved for the Ok derivative. (b) Z, L?, L° and L4 are symmetric (=self-adjoint) with respect to the positive definite scalar product: (2) : (2) =4(Zi-Za+¥i-¥s) (2.29) Here the dots on the right-hand side refer to the scalar product defined by a = 9" Ob, (2.30) for any two C*-valued functions a and b on the neighborhood considered, with the overbar denoting complex conjugation. (To be sure, in (2.23) all quantities are real. For later purposes, however, we need the complex version of this scalar product.) ‘These two observations imply that (2.23) satisfies the defining properties of a symmetric hyperbolic system of partial differential equations. By a well-known theorem (see, e.g., Theorem 4.5 in Chazarain and Piriou [26] or Sect. 4.12 in Egorov and Shubin [36)) this guarantees local existence and uniqueness of a solution Z, Y for any initial data Zo, Yo given on our hypersurface 4 = const. (Please recall our stipulation of tacitly working in the C™ cate- gory throughout Part I. Had we restricted ourselves to the analytic category instead, property (a) alone would guarantee local existence and uniqueness ‘of a solution to any initial data, owing to the well-known Cauchy-Kovalevsky theorem.) Moreover, the fact that (2.23) is symmetric hyperbolic implies that solutions Z, Y are bounded in terms of so-called energy inequalities, ee, ¢.6-14 2, Light propagation in linear dielectric and permeable media, ‘Theorem 4.3 in Chazarain and Piriou [26] or Theorem 2.63 in Egorov and Shubin [36]. We are going to employ these facts later. ‘The explicit form of the matrix M in (2.23) will be of no interest for us in the following. What really matters is the structure of the L*, ie., the {information contained in the 6 x 6 matrix L(z,p) = Pa L*(z) (2.31) Here the first argument « = (21,2?,2°, 2) ranges over the coordinate neigh- borhood considered and the second argument p = (p1,p2,Ps,P4) ranges over RS, Tho matrix (2, p) defined by (2.31) is called the principal matriz or the characteristic matriz of the system of differential equations (2.23). Its deter- minant, which gives a homogeneous polynomial of degree six in the pay is called the principal determinant or the characteristic determinant of (2.23). ‘We shall sce later that the laws of ray optics in our medium are coded in the characteristic determinant. ‘The notions of characteristic matrix and characteristic determinant can be introduced for any system of k** order partial differential equations, linear in the highest order derivatives, that gives n equations for n dynamical vati- ables. The characteristic matrix is then formed in a fashion similar to (2.31) from the coefficients of the highest order derivatives. If these coefficients are independent of the unknown functions, Le, ifthe system of differential equa- tions is semi-linear, the characteristic matrix is of the form E(@,P) = Poy ** Pay EY **(z). (2.32) Hence, its determinant gives a homogeneous polynomial of degree nk with respect to the Da. 2.2 Approximate-plane-wave families In the preceding section we have discussed Maxwell’s equations in a linear dielectric and permeable medium. The laws of light propagation in such a ‘medium are determined by the dynamics of wavelike solutions of those eque- tions. In this section we clarify what is meant by the attribute “wavelike”. ‘The following definition is basic. Definition 2.2.1. An approximate-plane-wave family és a one-parameter family of antisymmetric second rank tensor fields of the form Fay(a,2) = Re{e'()/* f.y(a,2)} (2.33) with the following properties. (a) The coordinates x = (z',2?,2°,2*) range over some open subset of the spacetime manifold and the parameter a ranges over the strictly positive real numbers, o € R*.22 Approximateplanesvave families 15 (b) S is @ real-valued function whose gradient has no zeros, i-e., 5(c) = (215(c),25(2), O52), 045(2)) # (0,0,0,0) (2.84) for all x in the neighborhood considered. We refer to S as to the eikonal function of the approzimate-plane-wave family. (c) For each a € R*, fau(a, -) is a compler-valued antisymmetric second rank tensor field, Moreover, fay admits a Taylor expansion of the form Nott favlayz) = > a f5 (2) + O(aN*?) (2.35) for all integers No > —1, where £3(e) = NI lim, Zr fasle2) (2.36) We refer to f%, as to the N** order amplitude of the approzimate-plane- wave family (a) For all x in the neighborhood considered, (S(2)) #0 (237) In (2.33), i denotes, of course, the imaginary unit, i = ~1, and Re denotes the real part of a complex number. ‘We call $ the “eikonal function” because an approximate-plane-wave fam- ily satisfies Maxwell’s equations in an asymptotic sense to be discussed later only if S satisfies a partial differential equation which is known as the eikonal ‘equation. The term “eikonal”, which was introduced in 1895 by Bruns [23] in a more special context, is derived from the greek word eikon which means “image”. This terminology is, indeed, justified since the eikonal equation is the fundamental equation of ray optics; so it governs, in particular, the ray optical laws of image formation. ‘According to our general stipulation that all maps and tensor fields are tacitly assumed to be infinitely often differentiable it goes without saying that fas(0,) is a C™ function of a € R*. A Taylor expansion of the form (2.35) is valid if and only if this function admits a C™° extension into the point a = 0. Note that we do not assume that the O(aN*) term in (2.35) ‘goes to zero for NV -+ co, ie., we do not assume analyticity with respect to «. {It is important to realize that an approximate-plane-wave family cannot ‘converge for « —+ 0. This is an immediate consequence of the following lemma which will often be used in the following. Lemma 2.2.1. Let $ be the eikonal function of an approzimate-plane-wave ‘family, according to Definition 2.2.1 (b). Let u be a complex valued function defined on the same open subset of spacetime as the approzimate-plane-wave family. (As always in Part I, we tacitly assume that w is of class C™ and, thus, continuous). Then linn Re{e'9/*u} exists pointwise only ifu = 0.16 2. Light propagation in linear dielectric and permeable media Proof. If w is different from zero at some point, it is different from zero, by continuity, on a whole neighborhood. For almost all points « of this neigh- borhood, (2.34) implies that $(-) #0 and the limit does not exist. o We are now going to justify the name “approximate-plane-wave family”. (More fully, (2.83) should be called a “locally-approximate-plane-and-mono- chromatic-wave family”. This terminology, however, seems a little bit too cumbersome.) The physical idea behind Definition 2.2.1 becomes clear if we consider the special case that the tensor fields 2a. and fas(a, -) are covari- antly constant (and non-zero, as assured by (2.34) and (2.37)), ie., that the equations V5, = 0 and Vefas(a, +) = 0 are satisfied. Then (2.93) gives a one-parameter family of monochromatic plane waves. With respect to an inertial system (ie., a covariantly constant time-like vector field V* with {445° V° = ~1), the frequency of such a wave is given by w = 2V* 2,8 and the spatial wave covector is given by ka = 20,5 —wgasV°. Hence, the limit + 0 corresponds to infinitely high frequency with respect to all inertial systems V2 with V¢d,5 #0. Now this is a very special case since on a spacetime without symme- try there are no non-zero covariantly constant vector fields. Therefore, as we want to work with ansatz (2.33) on an arbitrary spacetime, we cannot assume that 8,5 and fas(ay, -) are covariantly constant. However, if we restrict our consideration to a sufficiently small neighborhood, 8,5 and fas(a, -) deviate arbitrarily little from being covariantly constant. Similarly, on a sufficiently small neighborhood, any time-like vector field V° with ga» V¢ V? = ~1 devi- ates arbitrarily little from an inertial system. However small this neighbor- hood may be, by choosing a sufficiently small we can have arbitrarily many ‘wave periods in this small spacetime region. ‘This reasoning justifies the terminology introduced in Definition 2.2.1. Please note that (2.34) and (2.37) are essential to guarantee that (2.33) gives ‘an approximately plane and monochromatic wave near each point for a suf- ficiently close to zero. In correspondence with this interpretation we shall refer to the hyper- surfaces S = const. as to the wave surfaces of our approximate-plane-wave family. The alternative terms eikonal surfaces and phase surfaces are also common. Moreover, we call wlay2) = 2 4.5(2) V*(2) (2.38) its frequency function and we call kq(a,2) = 2,5(«) — (a2) gos(z) V*(a) (2.39) its spatial wave covector field with respect to the observer field V9; here all those timelike vector fields with gayV*V? = —1 are admitted for which V9,5 has no zeros.2.3 Asymptotic solutions of Maxwell's equations 17 Ibis worthwile to note that from an approximate-plane-wave family (2.33) wwe can produce non-monochromatic waves by integrating over a with an appropriate density function w, Fas(z) = | Fas(a,2) w(a) dex. (2.40) ‘This can be viewed as a generalized Fourier synthesis. Here we have to assume that a < az, with a sufficiently small to justify the approximate-plane- ‘wave interpretation. Moreover, it is also possible to form superpositions of approximate-plane-wave families with different elkonal functions S. 2.3 Asymptotic solutions of Maxwell's equations ‘To study the dynamics of wavelike electromagnetic fields in our medium wwe have to plug our approximate-plane-wave ansatz (2.33) into Maxwell’s equations, Le., ito (2.2) supplemented with our constitutive equations. Un- fortunately, only in very special cases is it possible to determine the eikonal function $ and the amplitudes /2 in such a way that the resulting equations are exactly satisfied for some a € R*. It is the characteristic feature of the ray method to determine and fj in such a way that Maxwell's equations are satisfied, rather than for some finite value of a, asymptotically for a — 0. In this way the ray method gives us the dynamics of wave surfaces and wave amplitudes in the high frequency limit. To put this rigorously we introduce the following notation. Definition 2.3.1. For N € Z, an approzimate-plane-wave family Fas(ax +) in the sense of Definition 2.2.1 és ealled an N* order asymptotic solution of Mazwell’s equations if dim, (def Fea(as -)) =0, dim, (e185 Gerla, -))) =0- Here Gaj(ay,-) is related to Fas(a, +) by the constitutive equations of the ‘medium, In (2.41), the limits are meant to be performed pointwise with respect to the spacetime coordinates; we shall restrict ourselves to nelghbothoods on which the convergence is uniform. For the evaluation of (2.41), the following ‘two observations are crucial. (@) The metric is independent of a and so are the other tensor fields that center into the constitutive equations, 1e., U%, e4! and y4'. Hence, the special form in which a enters into the approximate-plane-wave ansatz (2.33) together with the linearity of the constitutive equations implies that (2.41) is trivially satisfied for N < 1. (2.41)18 2, Light propagation in linear dielectric and permeable media (b) If (2.41) holds for NV = No, then it holds all the more for N < No. ‘These two observations together suggest that N§ order asymptotic solutions can be found for arbitrarily large No > 0 by first evaluating (2.41) for N= —1 and then proceeding step by step up to N’ = No. We shall see in the following that this inductive procedure gives us dynamical laws for the eikonal function S and, step by step, for the amplitudes {3 up to arbitrarily large order N=Nq, If we want to get dynamical laws for ff for all N € N, we have to assume that our approximate-plane-wave family (2.33) satisfies (2.41) for all N ¢ N or, what is the same, for all N’€ Z. In this case Fas(cy -) is called an infinite asymptotic series solution of Maxwell's equations. Fasla, +) Fig. 2.1. For N > 0, an N* order asymptotic solution Fas(cy, +) of Maccwell's equa- tions approaches the space of exact solutions of Maxwell's equations asymptotically for a.~+ 0, as will be proven in Sect. 2.7. (2.41) does, of course, not imply that the one-parameter family Fao(ay +) converges pointwise (or in any other sense) towards an exact solution of Maxwell's equations for a —> 0. We have already emphasized that for an approximate-plane-wave family the limit lim, Fas(a, -) cannot exist, This raises the question of whether asymptotic solutions can be viewed as approx- imate solutions. This question will be answered in Sect. 2.7 below by proving the following result. Let Fys(a, -) be an approximate-plane-wave family that is an NV" order asymptotic solution of Maxwell's equations in a linear di- electric and permeable medium for some V > 0. Then there exists, locally around any one point, a one-parameter family F3,(a, -) of exact solutions of Maxwell’s equations such that F(a, -) — Fab(a, +) goes to zero in the pointwise sense (and even with respect to some finer norms involving arbi- trarily high derivatives) as a1 for a — 0. In other words, for a sufficiently small the members of our approximate-plane-wave family can be viewed as, arbitrarily good approximations to exact solutions of Maxwell's equations. Figure 2.1 illustrates this situation in the infinite-dimensional space of (C°°)2.4 Derivation of the elkonal equation and transport equations 19 ‘antisymmetric second-rank tensor fields defined on some open spacetime do- main. 2.4 Derivation of the eikonal equation and transport equations In this section we derive, in a linear dielectric and permeable medium, the dynamical equations for wave surfaces and for wave amplitudes in the high frequency limit, We do that locally around any spacetime point x9. As a preparation, we prove the following fact. Proposition 2.4.1. Consider an approsimate-plane-wave family Fau(2, +) that is an N* onder asymptotic solution of Marwell’s equations in a lin- ear dielectric and permeable medium for some N > —1. Then the frequency function (2.38) of Fas(a, +) with respect to the rest system of the medium (V2 =U*) has no zeros. Proof. We introduce, around any spacetime point zo, a coordinate system adapted to U® in the sense of Definition 2.1.3. We are done if we can show that 09 is different from zero at 2, By assumption, our approximate-plane- ‘wave family satisfies (2.41) for N = —1, ie. 1 BS f= 0, (2.42) tet BS med! Boy = 0 (2.43) where g0, is related to 2, by the constitutive equations. (Here we made use of Lemma 2.2.1.) Now let us assume that 45 = 0 at zo. At this point, the a= 4 component of (2.42) implies 9°" 9,508 =0 (2.44) for the magnetic part 0° of f9,, whereas the a = p components of (2.42) imply 8,Se2 -9,5e9 =0 (2.45) for the electric part e2 of f2,. Similarly, (2.43) results in gM” 8,8d2 =0, (2.46) and 8,5 he, — 8,849 =0 (2.47) for the electric part d? and for the magnetic part h? of g%,, Note that 5 is real whereas the amplitudes are complex. (2.45) and (2.48) imply o' &Ba,5 =0. (2.48)20 2 Light propagation in linear dielectric and permeable media Similarly, (2.44) and (2.47) imply oH BOS = Recall that we are at a point where 45 = 0. Thus, con: quires (845,625, 855) # (0,0,0). Hence, by condition (c) of Definition but (2.48) implies that (e2, e8,e8) = (0,0,0) and (2.49) implies that (62, 08, 08) = (0,0, 0). This shows that our hypothesis of 8,5 having a zero gives a contra- diction to (2.37). B ‘To analyze the dynamics of wave surfaces and amplitudes in the high frequency limit near an arbitrary spacetime point to, we introduce near x9 & coordinate system which is adapted to the rest system U of the medium in the sense of Definition 2.1.3. We can then express electromagnetic fields in terms of the dynamical variables Zi, Zz, Zs, Yi, Ya,¥s introduced in (2.22). ‘Then any approximate-plane-wave family takes the form Gea). referer S » nl3}) +orar}, (250) for any integer No > —1. Here the complex amplitudes f24 from (2.35) are ex- pressed in terms of C*-valued functions 2 and y". The following proposition sis necosary ang ufcientconions on the atonal function 5 and on the amplitudes 2, y such that (2.50) is an asymptotic solution of Maxwell's equations. Proposition 2.4.2. Consider, locally around any spacetime point x, a co- ordinate system (2',22,2°,2*) adapted to the rest system U® of a linear dielectric and permeable medium. Then an approzimate-plane-wave family, represented in this coordinate system in the form (2.50), is an asymptotic solution of Marvell’s equations in lowest non-trivial order N = —1 if and only if &4S has no zeros and 8,5 L* (3) = () (2.51) For No > 0, such an approzimate-plane-wave family is an N§* order asymp- totic solution of Maxwell’s equations if and only if, in addition, " oe (184.+M) (jn) = 488 LP ¢ ws) (2.52) for 0 < N < No. Here L* and M denote the same matrices as in the ‘evolution equation (2.23). Proof. In our adapted coordinate system, we decompose the asymptotic ‘Maxwell’s equations (2.41) into constraint part (2 = 4) and evolution part24 Derivation of the eikonal equation and transport equations 21 (a= p). Ifthese equations are satisfied by an approximate plane-wave family for some NV > ~1, Proposition 24.1 implies that 045 has no zeros. Under this condition the evolution part of (2.41) alone already implies the con- straint part of (2.41). his is easy to verify using the fact that, as outlined in Sect. 2.1, the evolution equations preserve the constraints. In other words, we ‘can forget about the constraints and concentrate on evaluating the evolution part of (2.41). According to (2.23), this takes the form bm, ( ak (E8@.+M) ( 3) ) = () (2.53) {in terms of the variables Z1, Z2, Zs, Yi, Y2, ¥s. Hence, our approximate-plane- ‘wave family is an asymptotic solution of Maxwell’s equations to lowest: non- trivial order N = —1 if and only if 45 has no zeros and (2.53) is satisfied for N = 1. By feeding (2.50) into (2.53) for N = ~1 we see that the latter condition is equivalent to (2.51), owing to Lemma 2.2.1. For No > 0, our approximate-plane-wave family is an Nj* order solution if and only if in addition (2.58) is satisfied for all 0 < NV < No. Upon feeding (2.50) into (2.53), itis easy to prove by induction over IV that this is true if and only if (2.52) is satisfied for 0 < N' < No. 0 Condition (d) of Definition 2.2.1 requires that, if (2.50) represents an approximate-plane-wave family, 2° and y? do not vanish simultaneously, Clearly, such a solution 2°,y? of (2.51) exists if and only if det(0.5L*) =0 (2.54) ‘This is a first order partial differential equation for S, homogeneous of degree six with respect to the components of the gradient of S. If S satisfies (2.54) and if 045 has no zeros, S is called a solution of the eikonal equation of the linear dielectric and permeable medium considered. By Proposition 2.4.2, this is a necessary and sufficient condition for $ to be the eikonal function of an approximate-plane-wave family that satisfies Maxwell’s equations asymptot- ically to order N = —1 at least. In the theory of partial differential equations (2.54) is called the characteristic equation of the system of evolution equa- tions (2.23). In the next section we discuss the eikonal equation in our medium in more detail. In particular, we free ourselves from the special coordinates used s0 far. If we have a solution $ of the eikonal equation, Proposition 2.4.2 can be used to construct an asymptotic solution of arbitrarily high order. To that end the amplitudes 2” and y™ have to be determined inductively with the help of (2.51) and (2.52), Clearly, 21 and y+! are not uniquely determined through 2 and y since, for a solution of the eikonal equation, 9,5 L* has a non-trivial kernel. Let Ps(z) denote the 6 x 6 matrix that projects orthogonally onto the kernel of 8,5(z) L(x), where “orthogonally” refers to22 2, Light propagation in linear dielectric and permeable media tthe scalar product (2.29). For any solution S of (2.54) the rank of Ps(z) is bigger than or equal to one. We shall prove Inter that, owing to the special form of the matrices L*(z), the rank of Ps(z) cannot be bigger than two. In general, the rank depends, of course, on 2. Let us write aff aN sf) (a8) (af | =P: and = - . (2.55) (i ) i (*) Ci () wt el ‘This decomposition of the amplitudes 2% and y” implies, via (2.50), a de- composition of Z and Y and thus, via (2.22), a decomposition of the electric and of the magnetic component of our approximate-plane-wave family. In terms of the decomposition (2.55), the inductive scheme for the ampli- tudes is given by the following proposition. Proposition 2.4.8. Let S be a solution of the eikonal equation and fiz an integer No > 0. Then the one-parameter family (2.50) is an. NG order asym- ptotic solution of Maxwell's equations if and only if the amplitudes 2 and o” satisfy (2.58) ond aN ee (1—Ps)(I* 0, + M) (3) 18,8 I (i) (87) i a at PsL*d,( | +PsM i () . () for 0.< N < No. (2.56) is called the 0* order polarization condition, (2.57) is called the (N +1) order polarization condition and (2.58) is called the N* order transport equation. 1 Ps (L*a,+M) (i) (2.58) Proof. (2.56) is obviously equivalent to (2.51). To prove that (2.57) and (2.58) together are equivalent to (2.52), we decompose (2.52) into two equations by applying Ps and 1— Pg respectively. The first, equation gives (2.57), the second equation gives (2.58). This is readily verified with the help of the equations 0,5 L* Ps = 0 and ,5 Ps L* = 0. (The first equation is trivial ‘and the second follows from the fact that 0,5 L* is symmetric with respect to the scalar product (2.29).) 3 Since (2.57) can be solved for z{’*" and yj!*, by this equation the + components of z+ and y+ are algebraically determined through the2.4 Derivation of the eikonal equation and transport equations 23 lower order amplitudes 2” and y%. This gives a restriction on the allowed directions of the electric and magnetic field vectors which justifies the name “polarization condition”. If 2) and yj’ are known, (2.58) gives a system of first order differential equations for zi and yi. Later we shall associate solutions of the eikonal equation with congruences of rays. The name “transport equation” refers to the fact that (2.58) gives us ordinary differential equations (ie., “transport laws") for the components of =’ and y/¥ along each ray, as will be shown in Sect. 2.4 below. In spacetime regions where Ps has constant rank, (2.68) admits a well- posed initial value problem in the following sense. If 1 < rank Pg = k = const, (2.59) ‘we can choose k basis vector fields a1,...,a% (complex six-tuples depending on 2), orthonormal with respect to the scalar product (2.29), such that, k Ps=))0a@aa, (2.60) ant where @ denotes the standard tensor product on C°. Hence, z{! and yf are of the form a & ( 4) = Seles (1) wt)” a with some C-valued functions €¥. Then the NV‘ order transport equation (2.58) gives a system of k differential equations for the k coefficients £1" which is symmetric hyperbolic. (‘This follows from the facts that each matrix L* is symmetric with respect to the scalar product (2.29) and that L* is close to 1L,) Hence, local existence and uniqueness of solutions €N,..,€{" is guaranteed for arbitrary initial values given on a hypersurface x* = const. By solving the transport equations in this way at each level N, we determine that part of the polarization direction which is not fixed already by the polarization condition, and we determine the intensity of our approximate plane wave. ‘Now it is clear how, for a solution S of the eikonal function that satis- fies the rank condition (2.59), the amplitudes 2 and y"’ can be determined inductively to construct an Nj order asymptotic solution of Maxwell's equa- tions. 1. The induction starts with setting 2f = yf = 0. 2. The N** step of the induction, 0 < N < No, is given by the following prescription. With 2)” and yf! known, determine 21 and yi by solving (2.58) with arbitrary initial values. (‘The only restriction on the initial values ig that 2° and y% must not vanish simultaneously.) Then, deter- mine 2!" and y}!*? with the help of (2.57).24 2 Light propagation in linear dielectric and permeable media ‘The other amplitudes (ie, 2, y¥, zf"*4, yf! for N > No +1) and the O(a%*2) term can be chosen arbitrarily. (E.g., they could be set equal to zero.) Then (2.50) gives an approximate-plane-wave family that satisfies Maxwell's equations asymptotically to order No. ‘This construction can be carried through for arbitrarily large No, ic, it ‘can be used to construct (non-convergent) infinite asymptotic series solutions of Maxwell's equations. In the very special case that the induction yields aN y = 0 for some N > 1 we can set 2" and y! equal to zero for M > N to get an approximate-plane-wave family that satisfies Maxwell's aus ‘exactly for all a ¢ R* ‘The results of this section show how to construct, locally around any spacetime point, an approximate-plane-wave family that satisfies Maxwells ‘equations in a linear dielectric and permeable medium asymptotically to some order N > 0. The physical relevance of those one-parameter families is in the fact that they can be interpreted as approzimate solutions of Maxwell's equations as well. This will be proven in Sect. 2.7 below. Already now we emphasize that this is not true for asymptotic solutions of lowest non-trivial order NV = ~1. In other words, ifit is our goal to set up a viable approximation scheme for exact Maxwell fields we have to consider approximate-plane-wave families that satisfy Maxwell’s equations asymptotically to order N = 0 at least. In this order we get polarization conditions that fix 2®, y?, 2}, yi, and ‘we get transport equations for 22 and y}. This N = 0 theory is often called the geometric optics approzimation of Maxwell fields. 2.5 Discussion of the eikonal equation In the preceding section we have derived the eikonal equation of our medium, locally around an arbitrarily chosen point, in a special coordinate system. It is now our goal to analyze the structure of this equation and, in particular, to rewrite the eikonal equation in covariant form. In a coordinate system adapted to the rest system of the medium, the cikonal equation was given by (2.54) supplemented with the condition that 4S has no zeros. Cleatly, the characteristic matrix L(2,p) = p,L°(2) is a real 6x6 matrix, symmetric with respect to the scalar product (2.29). Henoe, it has six real eigenvalues and the characteristic determinant det(pa Z*(z)) is the product of these eigenvalues. If we want to bring the eikonal equation in a more explicit form we have to determine these six eigenvalues. First we reduce this six-dimensional eigenvalue problem to a three- dimensional eigenvalue problem. To that end we introduce, for all z in the spacetime neighborhood considered and for all p = (p1,p2,ps,Pa) € R4, the real 3 x 3 matrix We?) Tray (0292) +94 4°) (202)2.5 Discussion of the elkonal equation 25 which, by (2.25), enters into the characteristic matrix according to noron(38)+V (wey) em ‘The (strictly positive) factor y/—gei(z) was introduced in (2.62) for later convenience. Then the 3 x 3 matrix W/(z,p)"W (c,p) is obviously positive semidefinite and symmetric with respect to the scalar product (2.29). Hence, it has three real eigenvectors 1 (2, p);v2(2P), usp) which are orthonormal with respect to the scalar product (2.29), and the pertaining eigenvalues are real and non-negative. We denote these eigenvalues by hi(z,p)*, ha(z,p)?, ha(z,p)? with ha(z,p) > 0 for A = 1,2,3. Similarly, the 3 x 3 matrix W(c, p) W(2,p)" has three real eigenvectors v1 (1,7); v2(,p), vs (ap) which are orthonormal with respect to the scalar product (2.29), and the pertaining eigenvalues are the same as for W(x, p)7 W(x, p), i. W(2,p)" W(2,p) walt) = halzp)? ua(zsP) » Wer) We ealap)—halaleaer), OD for A=1,2,3, The bases of eigenvectors can be chosen in such a way that W(z,p) wa(z,p) = halz,p) va(zP) » Wen) valzp) = hales0) eater)» tee for A = 1,2,3. (In the non-degenerate case, Le., if the eigenvalues h(x, p)*, ha(x,p)?, hha(z,p)? are mutually different, the eigenvectors wa(z,p) and va(z,p) are unique up to sign and the equations (2.65) are automatically true up to sign.) These equations imply that the characteristic matrix (2.63) a pattie) (2300)) = (r+ V=aaleThatern) ( 24°6?)) for A = 1,2,3. This equation gives us six (real) eigenvalues of the 6 x 6 matrix p,Z* and pertaining eigenvectors in terms of the eigenvalues and ‘eigenvectors of the 3 x 3 matrices W(z,p)? W(z, p) and W(z,p) W(2,p)". 'As the characteristic determinant is the product of these six eigenvalues, the ‘eikonal equation (2.54) takes the form (2.66) 2 TI ((@)? + guahat-,95)?) 267 aa supplemented with the condition that 345 has no zeros. To get a more explicit form of the eikonal equation, we have to calculate the eigenvalues a(x, p)? of26 2, Light propagation in linear dielectric and permeable media ‘the matrix W (2,7) W(c,p). If we insert the general expressions (2.26) and (2.27) for the components of the matrices Q(z) and A?(c) into the definition (2.62) of W(z,p), we find that the components of the matrix (x,p) = W(c,p)" W(a,p) are Re" (ap) = RY," (2) Pa Po (2.68) with 1 RM a) = Fea Mok) wy a) (2) md*() v,"(@)- (2.69) ‘The three eigenvalues h(z,p)?, ha(z,p)? and ha(,p)? of the matrix R(z,p) are then given by hayala,p)? = FR" (2) Pore + 5 RY Ma) RP(2) (2) R4,7(2)) paPoPepay (2.70) Ag(z,p)? =0. ‘The appearance of the square root in (2.5) has the unpleasant consequence that fy and hy might fail to be differentiable at some points even if all input functions are C as tacitly assumed throughout Part I In the following we assume that hy and fg are O® functions at all points with (pi, p2, p3,Pa) # (,0,0,0). “The whole calculation was done around an arbitrarily chosen spacetime 2; in a coordinate system adapted to the rest system U* of the medium. From Sect. 2.1 we lmow that such a coordinate eystom is unique, locally near 2, to within coordinate transformations of the special form (2.18). If we perform such a coordinate change, viewing p = (p1,P2,Ps,ps) &8 canonical momentum coordinates conjugate to x = (z',2%,2°,:*) which transform as Ox? Pe We= Fae Po> en) the components of the matrix F(z, p) = W(z,p)" W(c, p) transform accord- ing to ox i RN) = Fax Bop RMP)» (2.72) fas can be read from (2.68) and (2.68). a is the reason why we intro- duced the factor /=gea in (2.62).) The eigenvalues of the matrix F(z, p) are, thus, invariant under coordinate transformations of the form (2.18), Le. (2's pt)? = ha(z,p)?. In other words, hy and fz are uniquely determined (global and invariant) functions on the cotangent bundle over spacetime. Hence, for A= 1 and A = 2, the function2.5 Discussion of the eikonal equation 27 1 (halen)? -U5(e) U°(e) ps2) (2.78) is a uniquely determined (global and invariant) function on the cotangent bundle over spacetime. We refer to H and H as to the partial Hamiltonians of our linear dielectric and permeable medium. The eikonal equation can then be formulated in the following way. Proposition 2.5.1. A real-valued function S, defined on some open space- time region U, is a solution of the eikonal equation if and only if Hj(z,08(z)) Ha(2,25(2)) =0 (2.74) and 85(z) # (0,0,0,0) for all x € U. Here Hy and Hz denote the partial Hamiltonians introduced in (2.73) Proof. $ is a solution of the eikonal equation near any spacetime point if and only if, in adapted coordinates near this point, (2.67) holds and 243 has no zeros. Since, by (2.6), ig vanishes, this is true if and only if (mt- 5)? + -(045)*) (hal (-,98)? + + £(as)?) =0 (2.75) holds and 845 hes no zeros. From (2.5) we read that /(c,p) and ha(, p) fare non-zero at points (2,p) with pa = 0 but (pi,p2,ps) # (0,0,0). (This follows from the fact that (v,?(z)) and (w,?(z)) are invertible 3 x 3 matrices and that the kernel of the matrix (nod®(z)p,) is exactly one-dimensional if (p1.Pa,Pa) # (0,0,0).) Hence, for a solution of (2.75) the condition 85 7 0 is equivalent to 88 # (0,0,0,0). With the help of (2.78) we can rewste (2.75) in the coordinate invariant form (2.74), ‘We shall refer to the equations Ha(z,88(z)) =0 (2.76) for A = 1 and A = 2 as to the partial eikonal equations. A solution of the eikonal equation has to satisfy at each point at least one of the two partial ‘eikonal equations. In the terminology of classical mechanics, (2.76) is called ‘the Hamilton-Jacobi equation of the Hamiltonian H 4. The set of all (z,p) with p # (0,0,0,0) and Ha(z,p) =0 (277) is called the A-branch of the characteristic variety and the equation (2.77) is called the A-dispersion relation of our medium. The following proposition gives some information on the geometry of the A-branch of the characteristic variety. Proposition 2.5.2. For A = 1 and A = 2, the partial Hamiltonian Ha introduced in (2.73) has the following properties.28 2, Light propagation in linear dielectric and permeable media (a) Ha is homogeneous of degree two with respect to the momentum coordi. nates, Ha(a,tp) =? Ha(e,p) (2.78) for all real nurabers t. (b) Ha satisfies the differential equation Uela) Mae Lota) (2.79) (©) At all points (2,p) with p # (0,0,0,0) but U°(z) pp =0 the partial Hamit- tonian is strictly positive, Halayp) > 0. (2.80) In (b) and (c), U* denotes the rest system of the medium. Proof. (a) is obviously true in the special coordinates where ha(z,p)? is given by (2.5). As a consequence, it is true in any coordinates since the conjugate momenta transform homogeneously according to (2.71). To prove (b), we read from (2.5) that, in the special coordinates used there, the mo- mentum coordinates enter into ha(z,p)? only in terms of the combination 944(z) Pe — 940() pa. Thus, the coordinate invariant differential equation Uy32-(ha(z,p)*) = 0 holds true. ‘To prove (c) it suffices to verify from (2.5) that h(x, p)? is non-zero if, in the coordinates used there, pg = 0 but (p1,p2,P3) # (0,0,0). This follows from the fact that (v,°(x)) and (w,?(2)) are invertible 3 x 3 matrices and that the kernel of the matrix (n,/(z)py) is exactly one-dimensional if (p1,p2,p2) # (0,0,0).. a By differentiating (2.78) with respect to t and setting t = 1 afterwards, part (a) of this proposition implies that H4 satisfies the equations rege — an(a,, st) hee ofan) Hale,p) (282) ‘Thus, the Hamiltonian H, Is similar to the quadratic form of metric tensor, H(c,p) = 49°(2)paps, but with a metric tensor that depends not only on. x but also homogeneously on p. Such generalized metrics are usually called Finsler metrics; we may thus say that each partial Hamiltonian H4 defines & Finsler metric on the cotangent bundle over spacetime. Note, however, that some authors include the assumption of positive-definiteness into the defi nition of the term ‘Finsler metric”, whereas our metric (0H. (zp)/2p.0) ‘cannot be positive definite. This follows from differentiating (2.79) with re- spect to pp which demonstrates that U,Us2?H4(z,p)/@pa0 <0. For litera- ture on Finsler structures we refer to Rund [124] and to Asanov (10),2.5 Discussion of the eikonal equation 29 From part (b) and (¢) of Proposition 2.6.2 we read that Fe es) (0,0,0,0) (2.83) ‘on the A-branch of the characteristic variety. Hence, this branch is a codi- mension-one submanifold of the cotangent bundle which is transverse to the fibers. By patt (a) of Proposition 2.5.2, the intersection of this manifold with ach fiber has a “conic” structure. In general, the union of the 1-branch and of the 2-branch of the charac- teristic variety need not be a manifold. ‘The two branches might intersect or coalesce. It i, of course, also possible that the two branches coincide com- pletely. (This is necessarily true if the medium is isotropic, as we shall verify soon.) Whenever the two branches do not coincide, the medium is called birefringent or double-refractive. ‘The fact that the two branches can intersect or coalesce is related to the following unpleasant feature. Whereas (2.83) guarantees that either partial eikonal equation (2.76) can be solved, locally around any one point, for one of the partial derivatives 0,5, 25, 035, 045, this is not necessarily true for the full eikonal equation (2.74). Hence it is not guaranteed that we can find a hypersurface through each point such that initial data for S on that hypersurface determine a solution of (2.74) uniquely ‘The term “birefringence” refers to the fact that a light wave that enters into such a medium from vacuum will split up, in general, into two waves. In the appproximate-plane-wave setting considered here, one of the two waves has an eikonal function that solves (2.76) with A = 1, the other one with A= 2. In general, a solution of the full eikonal equation (2.74) can satisfy (2.76) with A = 1 at some points and with A = 2 at other points. Moreover, ‘there might be solutions of the full elkonal equation which solve (2.74) with A= Land with A = 2 simultaneously. Ifthe two branches of the characteristic variety coincide, this is true for all solutions of the full eikonal equation. It is worthwile to note that the partial Hamiltonians can be changed according to Ha(#sp) — Hala?) Fa(s,p) Ha(z,p) (2.84) for A = 1,2, where F4 is any real-valued function that has no zeros on the Avbranch of the characteristic variety. Clearly, such a transformation does, not affect the solutions of the partial eikonal equations. In this sense, the dynamics of wave surfaces in our medium determines two equivalence classes of Hamiltonians. A transformation of the form (2.84) does, of course, not preserve the degree-two homogeneity of H4 with respect to the momentum coordinates. Thus, it will lead to a representation in which the Finsler struc- ture is “hidden”. Finally, we illustrate the results of this section by considering two more special kinds of media.30 2, Light propagation in linear diclectric and permeable media, Example 2.5.1. The Isotropic Case If our linear dielectric and permeable medium is isotropic in the sense of Definition 2.1.2, the two branches of the characteristic variety coincide and are given by the mull cone bundle of a Lorentzian metric, In particular, there is no birefringence in an isotropic medium. ‘To verify these well-known facts ‘with the help of our general results, we first observe that, in the isotropic case, (2.21) simplifies to 1 7 ae awl us Upon inserting this into (2.69) and using the identity (2.7) of the Levi-Civita tensor field, the non-zero eigenvalues in (2.5) take the form 5. (2.85) (2) + U(x) U(a) =(a)u(z) ‘Thus, the partial Hamiltonians (2.78) coincide and are given by H(z,p) = Hy(x,p) = Ha(e,p) = 3.99"(2) poo» (2.87) fy (,p)? = halen)” Pam (2.86) where Flo" eu) vows (2.88) are the contravariant components of a Lorentzian metric which is called the optical metric of the isotropic medium. Please note that g%#U,U, = —1 and that 92 U, Xp = 0 implies 92° XqXs= 2 g@ Xq Xp. Both observations together imply that the optical metric is, indeed, of Lorentzian signature (Fett). ‘The strictly positive function n= Va (2.89) is called the index of refraction of the isotropic medium. If m = 1 (and, in particular, in the vacuum case ¢ = j= 1) the optical metric and the space- time metric coincide, o@ = g®. Hence, the eikonal equation in an isotropic ‘medium has exactly the same structure as in vacutum; we just have to replace the spacetime metric with the optical metric. This is a well-known result. It ‘was derived, with increasing mathematical rigor, by Gordon [50], Pham Mau Quan [117] and Ehlers {38} Example 2.5.2. The Uniaxial Case Now we specialize from a general linear dielectric and permeable medium to the case that the permeability tensor field has the same form as for vacuum, 18> = g* + U*U®. Moreover, we assume that the dielectricity tensor field, which can be written in the form e® = e, Xf Xt -+eX$X$-+es X¢ X$ with26 Discussion of transport equations and the introduction of rays 31 Gob XE Xb = bye Gab X$U" = 0 and eg > 0 for o = 1,2,3, has a double eigenvalue, say ¢1 = ¢2. In this case the functions hi(z,p)? and ha(z,p)? of (2.5) are bilinear with respect to the momentum coordinates. An example for such a medium is a uniazial erystal. For the partial Hamiltonians (2.73) we find in this special case after a quick calculation Ha(z,p) = for A= 1and A= 2, where 924 (2) Pa Po (2.90) sth = 2 (gt +ueu) ue", a Axe xt a xgxt) 42 xe xt—ueu. Ps a : (291) Hence, either branch of the characteristic variety is the null cone bundle of ‘a Lorentzian metric. Generalizing the terminology from the isotropic case, ‘the two metrics (2.91) can be called the optical metrics of the medium. The first optical metric does not distinguish a spatial direction, {.., itis of the same kind as the optical metric in an isotropic medium. The second optical ‘atric, however, reflects the fact that the Xz-direction is distinguished in the medium considered. In a situation like this the 1-branch of the characteristic variety is usually called the ordinary branch whereas the 2-branch is called the extraordinary branch, In this terminology solutions of the partial eikonal equation (2.76) with A = 1 are associated with ordinary waves and solutions with A=2 are associated with extraordinary waves. If the eigenvalues €1, €2, €3 of the dielectricity tensor field are mutually different, the two characteristic varieties are no longer the null cone bundles of Lorentzian metrics. An example for such a medium is a biazial crystal. If we want to speak of “optical metrics” in such a medium, we have to understand the term “metric” in the Finslerian sense. Both these optical Finsler metrics display the anisotropy of the medium in a symmetrical way, Le., it is not justified to distinguish one of them by the attribute “ordinary”. For this reason, we prefer to speak of I-waves and 2-waves (rather than of ordinary ‘waves and extraordinary waves) in general anisotropic media. 2.6 Discussion of transport equations and the introduction of rays In this section we associate solutions of the eikonal equation in a linear di- electric and permeable medium with congruences of rays. The guiding idea is to introduce the notion of rays in such a way that the transport equations (2.58) can be reinterpreted as ordinary differential equations along rays.32 2 Light propagation in linear dieloctric and permeable media According to Proposition 2.5.1 the left-hand side of the elkonal equation has a product structure. ‘This suggests to introduce, for solutions S of the eikonal equation, the following terminology. $ is called a solution of multiplic- ‘ty two iff it satisfies both partial eikonal equations (2.76), and it is called a solution of multiplicity one iff it satisfies exactly one of the two partial eikonal equations. The multiplicity can, of course, change from point to point. In the following we consider solutions of the eikonal equation on neighbor- hoods where the multiplicity is constant. Note that, for any given solution, there exists such a neighborhood near almost all spacetime points. We begin ‘our discussion with solutions of multiplicity one, later we consider the some- ‘what more complicated case of solutions of multiplicity two. We introduce the following definition, Definition 2.6.1. Let S be a solution of the eikonal equation according to Proposition 2.5.1. Assume that, on some open spacetime region U, S is of ‘multiplicity one, i.e., that the partial eikonal equation (2.76) is satisfied for A=1, say, but not for A=2 at all points of U. Then the vector field oe, K%(2) = ae as(e)) (2.92) on U is called the transport vector field and its integral curves are called the rays associated with S. Jn the theory of partial differential equations the rays are also known as (b+) characteristic curves. Owing to (2.83) the transport vector field has no zeros, ie., the rays ‘are immersed curves. Changing the partial Hamiltonian acoording to (2.84) corresponds to reparametrizing the rays. Please note that 6q5(t) K*(z) =0. (2.93) ‘This follows from the fact that Hy satisfies equation (2.81) consequence of the homogeneity property established in Proposition 2.5.2 (a). ‘Thus, the transport vector field is tangent to the hypersurfaces $ = const. ‘We want to show now that the transport equations can be viewed as ordinary differential equations along rays. We do that locally around any point 2 of the neighborhood 2 mentioned in Definition 2.6.1. To that end we Introduce a coordinate system adapted to the rest system U/* of the medium near 9. (Please recall Definition 2.1.3.) In such a coordinate system, the partial Hamiltonians (2.73) take the form Halesp) = (balsa) + BS) (halos) - Bir i.e,, the partial eikonal equations (2.76) are equivalent to2.6 Discussion of transport equations and the introduction of rays 33 ha(v,05(2)) + (2.95) Faas fo for A = 1,2. Here and in the following, the upper sign corresponds to negative frequency solutions, &S < 0, and the lower sign corresponds to positive frequency solutions, 0,5 > 0. With (2.95) the partial derivative of His takes the fom OH, Opa (2,05(z)) -—L.). @96) With these informations at hand, we take a closer look at the N** order transport equation, i.e, at equation (2.58) viewed as a differential equation for 2¥’ and yf with 2 and yj assumed known. In the situation of Defini- tion 2.6.1, the projection operator Ps(z) onto the kernel of 9,5(c) L*(2) is given, in terms of the eigenvectors ua and v4 of (2.68), by Po(e)= ( (2, 95(2)) Je ( u:(2,85(2)) , 0,(2,05(2))) © \01(2,08(2)), em) and zY and y¥Y are necessarily of the form x AN) _ neq) ( 1 25(2)) ) ne Ge) EN) (4 u4(c,05(2)) ee with a C-valued function €%. After multiplication with the non-vanishing factor 845(z)/g44(z), the N* order transport equation (2.58) reduces to a differential equation for the function ¢% of the form K*(2) 06% (e) + Se) EM) +8 Here K* is an abbreviation for 5(2) a) ar.) ( u1(%OS(@)) ule) (a FO eu (asia) } 2) (2.99) K%(z) and f(z) and kY(z) are known C-valued functions. Clearly, (2.99) gives an ordinary differential equation for €¥ along each integral curve of the vector field K@. We now show that K° is, indeed, the transport vector field defined through (2.92). We first observe that (2.66) implies (236%) 2 (22h) (a V=Gal@) halesP)) Sas (2.101)34 2. Light propagation in linear dielectric and permeable media for A, B = 1,2. Upon differentiation with respect to pp, (2.101) yields ualP)) phyq) ( “a(aP) Cao) He) (e267) = VGA) (bale) — hae) (ACG? ) ae ( p.)) - vatesp)) "Spe (082) (st V=aa 4 (0) ban (2.102) If we evaluate this equation with A = B = 1 along p = 95(z), we see that the right-hand side of (2.100) coincides with the right-hand side of (2.96) for A= 1. This proves that the vector field K* in (2.99) is, indeed, the transport vector field associated with 8. Now let us investigate to what extent these results carry over to the case of solutions of multiplicity two. In analogy to Definition 2.6.1, we introduce the following notions. Definition 2.6.2. Let S be a solution of the eikonal equation according to Proposition 2.5.1. Assume that, on some open spacetime region U, S is of multiplicity two, i.e., that the partial eikonal equation (2.76) is satisfied for A=1 and A=2 at all points of U. Then for A=1 and A = 2 the vector field aH, K3(a) = Ba («.as(e)) (2.103) onU is called the A-transport vector field and its integral curves are called the A-rays associated with S. We shall also refer to K#(z) and K§(2) as to the partial transport vector fields associated with S. For a solution of multiplicity two, Kf and K may or may not coincide, (If they are collinear, they can be made equal by a transformation of the form (2.84).) If the two branches of the characteristic variety coincide, all solutions are of multiplicity two with Kf = Kg. This is the case for an isotropic medium where, by (2.87), = Ki(o) = ot) a5(2). (2.104) Hence, there is only one congruence of rays associated with each solution of the eikonal equation in an isotropic medium. In the anisotropic case we have to live with the situation that solutions of the eikonal equation might be associated with two different congruences of rays. Clearly, this makes it more complicated to interpret the transport equations as ordinary differential equations along rays. We are now going to work out the details. In the situation of Definition 2.6.2, (2.97) is to be replaced with cosy2.6 Discussion of transport equations and the inroduction of rays 35, rso=$ (GMa (st6%) mw and 2 and yl are of the form 2N(@)\ _ Pa apo ( wa(ZO5(2)) (iS) Lae) (: v4(z,05(2)) eee ‘with two C-valued functions € and €lY to be determined. Upon multipli- cation with the non-vanishing factor 445(z)/g.a(2), the transport equation (2.58) gives a system of two coupled differential equations for éY and él! of the form KG(2) 6% (2) + D> fan(e)6B(2) + kA (2) =0, A=1,2. (2.107) Ba Here 4 is an abbreviation for = 28 ( 6) 9 ( (2, 25(2)) ) ne BA) = Fa) (4 va(e.05(e)) £u(c,05(4)) for A= 1,2 and fap, kif are known C-valued functions. To put the transport equation into the form (2.107), we made use of the fact that, by (2.6), our multiplicity-two solution satisfies (2G2Q)) 7 (S88) = (BQ) Ho MEER) ‘To verify that the K% given by (2.108) are, indeed, the partial transport vector fields defined in (2.103), we consider (2.6) with A = B. This shows ‘that the right-hand side of (2.108) coincides with the right-hand side of (2.96). Ti the two partial transport vector fields coincide, (2.107) gives an ordin differential equation for the tuple (€W,€4") along the integral curves of Kf ‘Kg. In the general case, the situation is more complicated. (2.107) with A gives an ordinary differential equation for €{Y along the integral curves of ‘K¢ that involves &’, and (2.107) with A = 2 gives an ordinary differential ‘equation for é¥ along the integral curves of K$ that involves €) We summarize our discussion in the following way. For a solution of the cikonal equation in a linear dielectric and permeable medium, Definition 2.6.1 gives a transport vector field and, thus, a congruence of rays on open sub- sets on which the multiplicity is one, and Definition 2.6.2 gives two partial transport vector fields and, thus, two congruences of rays on open subsets on (2.109) °36 2, Light propagation in linear dielectric and permeable m which the multiplicity is two. What is left out is the set of all points where ‘the multiplicity changes. By continuous extension into such point we might get pathologies such as bifurcating rays. Any ray is an integral curve of a vector field K given by (2.103) with A=1 and/or A= 2, For any such integral curve s > 2(s) we can define a map s+ p(s) by p(s) = A5(z(s)), thereby getting a solution of Hamilton's equations Ha(2(0),2(°)) (0) = oe (2(s),2(6)) (2110) tats) = ~52A (2(9),96)- We call any immersed curve s ++ 2(s) for which (2.6) is satisfied, with some s+ p(s), an A-ray for short, A = 1, : If the partial Hamiltonian H, is changed into 4 by a transformation of the form (2.84), the A-rays undergo a reparametrization but they are unchanged otherwise. In other words, the A-rays are determined, up to their parametrization, by the A-branch of the characteristic variety. In the uniaxial case discussed in Example 2.6.2, the 1-rays are called ordi- ‘nary rays and the 2-rays are called extraordinary rays . If we solve Hamilton’s equations (2.6) with the partial Hamiltonians given by (2.90), we find that the ordinary and extraordinary rays are the light-like geodesics of the first and the second optical metric (2.91), respectively. In the isotropic case there is only one optical metric and the notions of 1- rays and 2-rays coincide. By solving Hamilton's equations (2.6) with Hy = Hp given by (2.87), we find that the rays are exactly the light-like geodesics of the optical metric. This implies, of course, in particular the familiar textbook result that in vacuum the rays are exactly the light-like geodesics of the spacetime metric. 2.7 Ray optics as an approximation scheme ‘From the preceding sections we know that rays are associated with asymptotic. solutions of Maxwell's equations. We are now going to show that they are associated, moreover, with approzimate solutions of Maxwell's equations. For the physical interpretation of ray optics this is a crucial point. Let us start with a solution S of the eikonal equation which, in a medium. of the kind under consideration, is given by (2.74) with partial Hamiltonians (2.73). As always, we assume that 5 is given on some open neighborhood of spacetime and that its gradient has no zeros. Moreover, we have to assume in2.7 Ray optics as an approximation scheme 37 the following that $ is associated with a unique congruence of rays. In other words, we have to assume that either S is a solution of multiplicity one or that is a solution of multiplicity two for which the two partial transport ‘vector fields coincide, Itis our goal to associate such an eikonal function $ with an approximate solution of Maxwel’s equations. To that end, we fix a spacetime point and, on a neighborhood of that point, a coordinate system adapted to the rest system of the medium in the sense of Definition 2.1.3, We use the inductive ‘method of Sect. 2.5 to construct an V** order asymptotic solution (Flcrn) = nofeserre 5 a (Sule) +010} any ime of the evolution equations (2.23), where V can be chosen as large as we want. "This leaves the O(a"*2) term arbitrary. For any choice of this term, (2.111) is automatically an N* order asymptotic solution of the constraints as well It is not difficult to check that the O(a%*?) term can be chosen in such ‘a way that the constraints are eractly satisfied on the initial hypersurface xt = const. These initial values determine a unique exact solution of the evolution equations (2.23) for each a thereby giving us a one-parameter family of exact solutions that wil be denoted by Z*(a, -), ¥*(a, +). Now the difference AZ(a, +) = Z(a,+)- 2"(a,+), (2.112) AY (a, -)=¥(@,-)-¥"(a,-), satisfies (u#0.+M) (ae 3) = O(a") (2.13) and vanishes on the initial hypersurface. We have already stressed in Sect. 2.1 that the differential operator on the left-hand side of (2.113) is symmetric hyperbolic with respect to the scalar product (2.29). Hence, we have the so-called energy inequalities at our disposal (see, e.g., Theorem 4.3 in Chaz- rain and Piriou [26] or Theorem 2.63 in Egorov and Shubin (36)). As a consequence, (2.113) implies the existence of a constant C' such that the inequality ‘AZ(a,+)\ (AZ(o, 3 ( ave, 3) : ( a 3) < Caren) (2114) holds on an appropriately chosen (relatively compact) neighborhood. ‘The constant C can be written as an integral over this spacetime neighborhood ‘where the integrand involves the (known) tensor fields gab, U% £48 and pj‘. ‘Actually, the energy inequalities allow to estimate AZ and AY not only’ in38 2, Light propagation in linear dielectric and permeable media the pointwise sense as in (2.114) but even in terms of Sobolev norms involving arbitrarily high derivatives. For our purpose, however, (2.114) will do. (2.114) can be rewritten in terms of the field strengths By By B=(B,) ond B= é) (2.115) Ba, Es, rather than in terms of our dynamical variables Z and Y. Since the constitu- tive equations are linear, and since the dielectricity and permeability tensor fields can be uniformly estimated on compact subsets of spacetime, we get an inequality of the form ‘AB(a,-)) . (AB(a,-)) ¢ gaa (ae) (42623) see, ens) where @ is another constant. This shows that for a suficiently small B(a, ) and E(a, -) are arbitrarily close to the exact solution B*(a, -) and E*(ay, -), i.e., that our V* order asymptotic solution is indeed an approximate solution, recall Figure 2.1. The higher NV, the faster AB(a, +) and AE(a, -) converge toto bra In physical terms, the possibility to measure electric and magnetic field strengths is limited by some measuring accuracy 6. If a is so small that the right-hand side of (2.116) is smaller than 62, (2.116) implies that an observer moving along an z*-line cannot distinguish, by way of measurement, the ap- proximate solution from the exact solution. It is important to realize that this is true only for observers moving along an x‘-line (or at a small velocity with respect to the z*-lines). If we exclude the case that approximate solution and exact solution coincide, we can always find observers, moving at a high veloc- ity with respect to the «lines, who measure an arbitrarily large difference between them. This follows immediately from the transformation behavior of electric and magnetic field strength under a Lorentz transformation, given in any textbook on special relativity. In other words, the question of whether or not our N** order asymptotic solution, for some finite value of a can be viewed as a valid approximation for some specific exact solution depends on the observer field with respect to which electric and magnetic field strengths are to be measured. A similar observation, based on a different argument, was brought forward by Mashhoon (92)) who only considered light propagation in vacuum. He came to the conclusion that the equations of general relativistic ray optics have a meaning only in the limit of infinite frequency but not in the sense of a physically reasonable approximation for any finite frequency value. We do not share this radical point of view. Our results show that the ray method dow give a viable approximation schone for light propagation In 8 meds of the kind under consideration in the following sense. Any solution S of the eikonal equation which is associated with a unique congruence of rays2.7 Ray optics as an approximation scheme 39 can be viewed locally as the eikonal function of an approximate-plane-wave family that satisfies Maxwell's equations asymptotically to order N. This was shown in Sect. 2.4 for arbitrary N > 0. Moreover, we can find a one- parameter family of exact solutions of Maxwell's equations such that the difference between asymptotic solution and exact solution goes to zero for @ = 0 like a1, This follows from (2.114) or from the equivalent result (2.116). We just have to keep in mind that the constant C in (2.114) and the ‘constant C'in (2.116) depend on the observer field U®; it is impossible to find ‘error bounds that are valid with respect to all observer fields simultaneously. ‘Having thus associated a congruence of rays with a one-parameter family of exact solutions of Maxwell's equations F3,(a, -), it is natural to ask if ‘the rays are related, at least in the sense of an approximation, to the energy flux of F7,(a, -). After all, the intuitive idea behind ray optics is to view light propagation as a sort of energy transfer along rays. We need some more calculations to prove that this idea is, indeed, correct for media of the kind under consideration. We start again with a solution S of the eikonal equation and assume that it is associated with a unique congruence of rays. We construct, lo- cally around any point as outlined above, an approximate-plane-wave family Fas(ay +) with eikonal function S and a one-parameter family of exact solu- tions F%,(a, +) such that (2.114) holds for N = 0 at least. ‘Then the energy flux of F(a, ~) in the rest system U* of the medium is given by S**(a, +) = U2) T*s*(a,2) (2.117) where T*,2(a, +) is the Minkowski energy-momentum tensor of Fe(0 °), T*s%(ay +) = Fela, -)G°* (a +) ~ 455 Foal)" ‘The component of the energy flux four-vector (2.117) orthogonal to U® gives the familiar Poynting vector, whereas the component parallel to U* gives the energy density of the electromagnetic field. In a coordinate system adapted to US, in the sense of Definition 2.1.3, Fy, and Gz, can be expressed in terms of our dynamical variables Z and Y, asin (2.22). Then (2.117) takes the form V= Baal} S*2(a, +) = 9° (2) nul (2) ¥—¥(2) w4"(2) 25 (ay -)¥e (e+) - 8 9°" (2) (Zola +) Ze(ay -) + ¥5(0, +) ¥e(@)) - (219) 1). (2418) 1 Since (2.114) holds with V = 0, (2.119) can be rewritten in terms of our approximate-plane-wave family Zp(a, +), Yo(a, «) in the form V=H44(e) S°° (a5 +) = 9 (2) n° PC) Y9™(@) woh (z) Za(04 +) Yalan +) — 4 9G 9°" )(Zo(@s-) Zrlas-)+¥ol@s )¥e(@s-)) +O(@) (2.120)40 2. Light propagation in linear dielectric and permeable media Since Zp(a, +) and ¥o(a, -) are given by (2.60) with No = 0, we have Z(a, +) ¥y(os +) = 3 Re: Ste 28 (a) y2(o) + Fla) v9(0)} + O(a). (2.121) Let us denote by < f > (2) the average of a spacetime function f taken over a neighborhood of on which the gradient of S and the amplitudes 28, v), can be viewed as approximately constant. (Please recall our discussion of approximate-plane-wave families in Sect. 2.2.) For a sufficiently small, the first term on the right-hand side of (2.121) gives an average arbitrarily close to zero, We may thus write <2p(ay-)¥ulas+) > $Re{2h wf) (2.122) where © = y means that the difference between x and y can be made arbi- trarily small by choosing a sufficiently small. Similar expressions hold for the averaged products < Z(a, +) Z,(a, +) > and < ¥a(a, +)¥,(ay +) >. With these equations at hand, we can calculate the averaged energy flux from (2.119). If we assume that the background fields (je., the spacetime metric and the tensor fields that characterize the medium) do not vary significantly cover the neighborhood used for the averaging procedure, we find 2 AB < S°*(a,-) > 6 Refs" AM au <5 te ) ee el | (2.123) BGRo{z? Qu") + 54 (29-29 + 4° -9?) where the 3 x 3 matrices Q and A” from (2.26) and (2.27) are used. ‘We shall now show that the right-hand side of (2.123) is, indeed, propor- tional to the transport vector field of our eikonal function. To that end, we recall that Z(a, -), ¥(a, -) is an N" order asymptotic solution of Maxwell’s equations for NV = 0 at least. Thus, 2° and y/ have to satisfy the 0® order polarization condition (2.56). This implies 20) _ garg ( m1(t,05(2)) (8) -#@ (les) oe with a C-valued function €° if $ is of multiplicity one, and 2%(2)\ _ ua(z,08(2)} (98) - Ea (Ea) ren! with C-valued functions €? and €9 if S is of multiplicity two. In the first case the transport vector field is given by (2.100), and (2.124) implies (6) om Ke277 Ray optics as an approximation scheme 41 with some R-valued function u. In the second case the partial transport vector fields are given by (2.108). Since we assume that our eikonal equation is associated with a unique congruence of rays, the two partial transport vector fields coincide, K? = Kg = K®, and (2.126) holds in this case as well. With L* given by (2.25), (2.126) takes the form 2K = u( bg Ree? -A’y?} + 52 Re{2" Qu") +65 (22-27 +9999). (2.127) ‘Comparison of (2.123) and (2.127) shows that 6 vKt (2128) ‘with some R-valued function v. In other words, the averaged energy flux of the exact Maxwell field follows the rays up to terms that can be made arbitrarily small by choosing a sufficiently small. Please note that we have considered the energy flux only in the rest system of the medium. This is important unless in the vacuum case where there is no distinguished rest system of the ‘medium and (2.128) holds for the energy flux with respect to any observer field. ‘With these findings we have completed our discussion of light propagation in a linear dielectric and permeable medium. In particular, we have now es- tablished the missing link between asymptotic solutions and approximate so- lutions. Let us emphasize the main point again. For the mathematical deriva- tion of eikonal equation and transport equations through a mathematically ‘well-defined limit procedure it is necessary to consider approximate-plane- wave families that satisfy Maxwell's equations asymptotically for a -+ 0. From a physical point of view, however, this limit @ -» 0 is a purely formal device. The physical meaning of the method is in the fact that the result- ing approximate-plane-wave families give approximate solutions of Maxwell's equations for (sufficiently small but) finite values of a.3. Light propagation in other kinds of media Tn Chap. 2 we considered the homogeneous Maxwell equations (2.1) supple- mented by linear constitutive equations (2.12). This ansatz does, of course, not cover all sorts of media with relevance to physics. Modifications of the following kind are possible. First, we could replace ansatz, (2.12) with more complicated relations between field strengths and excitations. Second, we could introduce a current, i.e, a source term in the second Maxwell equation (and, similarly, in the first Maxwell equation if the hypothetical existence ‘of magnetic monopoles is to be taken into account). In the latter case, the current must be specified by additional equations. E.¢., we could assume, in analogy to (2.12), a linear relation between 3-current and electric field strength in the rest system of the medium, thereby generalizing Ohm’s law. For any such specification of the medium we can investigate if the resulting system of equations determines reasonable dynamics for the electromagnetic field. Here, a dynamical law is to be viewed as “reasonable” ifit is governed by ‘aset of evolution equations characterized by a local existence and uniqueness theorem. This set of evolution equations might be supplemented by a set of constraints that are preserved by the evolution equations. If the medium under consideration gives rise to a dynamical law of this kind, it is reasonable to proceed along the lines of Chap. 2, .¢., to consider approximate-plane-wave families (2.38) that satisfy evolution equations and constraints asymptotically for « —» 0 to some order NV. The passage to ray optics has been achieved if it is possible to derive, on this assumption, an eikonal equation of the form H(2,95(2)) =0, where H can be chosen as a product, 1 -Hi, with, each partial Hamiltonian H4, A = 1... satisfying condition (2.88) on its characteristic variety. This guarantees that each solution $ of a partial eikonal equation H(z, 85(z)) = 0 is associated with a nowhere vanishing transport vector field (2.103) whose integral curves give a congruence of A-rays, ie., of solutions to Hamilton’s equations (2.6) projected to spacetime. Even if all this works out nicely, it is of course not guaranteed that ray ‘optics in the medium under consideration can be viewed as a valid approx- imation scheme for exact Maxwell fields. This has to be checked, along the lines of Sect. 2.7, in each case individually.44 8, Light propagation in other kinds of media ‘We are now going to discuss the question of if and how such a treat- ment of media more general than considered in Chap. 2 is able to cover the phenomenon of dispersion. 3.1 Methodological remarks on dispersive media ‘The most important motivation to go beyond the kind of media considered in ‘Chap. 2is the following, We have found that the media considered in Chap. 2 are characterized by an eikonal equation of the form H(z, 95(z)) = 0 where ‘the Hamiltonian HT can be chosen homogeneous with respect to the momenta. In other words, if a 4-momentum p = (p1,.--»p4) satisfies the dispersion relation at some spacetime point z, then any multiple tp = (tp1,...,tPa) also satisfies the dispersion relation at this spacetime point. Whenever this homogeneity property is satisfied the medium is called dispersion-free or non- dispersive; otherwise, it is called dispersive. In a non-dispersive medium, a ray is fixed by giving an initial event and an initial direction for the spatial wave covector (with respect to any normalized time-like vector field); in a dispersive ‘medium, one has to give the length of the spatial covector in addition. This definition can be rephrased in terms of phase velocities and group velocities to yield the familiar physics textbook definition of dispersive and non-dispersive media, see Sect. 6.2 in Part IT below. Hence, we have to ask ourselves what sort of modified ansatz for the medium could be able to cover the phenomenon of dispersion. This is an important question not only from a theoretical point of view but also in view ‘of applications to astropyhsics. Dispersion plays a role for light propagation, in planetary and stellar atmospheres and in interstellar plasma clouds. ‘A closer look at the treatment of Chap. 2 shows that the following features are causative for the homogeneity of the eikonal equation. (a) Evolution equations and constraints give a system of linear differential ‘equations for the electromagnetic field strength. (b) ‘The limit a — 0 is taken on a fixed background, ie., neither evolution equations nor constraints involve the parameter a. ‘As a matter of fact, it is easy to check that, whenever (a) and (b) are sat- isfied, the eikonal equation arises in the form H(z,05(z)) = 0 where the Haniltonian H is a homogeneous polynomial with respect to the momenta. (Afterwards, we are free to change the Hamiltonian according to transfor mations of the form given in (2.84) for H = H. This leaves, of course, the homogeneity of the eikonal equation unchanged.) If we want to treat disper- sive media we have, thus, to modify the method of Chap. 2 by violating at least one of the two properties (a) and (b). ‘The most obvious idea to violate property (a) is to modify the linear con- stitutive equations (2.12) by adding terms quadratic in the field strengths.3.1 Methodological remarks on dispersive media 45 This is common practice in ordinary optics where it gives rise to interest- ing effects with relevance to strong electromagnetic wave fields. However, it is quite evident that such non-linear terms are not typically the origin of dispersion in crystals, gases, or plasmas. As a matter of fact, dispersion is, frequently observed in situations where the field strengths are far too weak to make non-linear modifications of the constitutive equations necessary. More- ‘over, there are several technical problems associated with the ray method if ‘evolution equations and/or constraints are non-linear. Contrary to the linear case, our assumption that the approximate-plane-wave family (2.33) satisfies the differential equations asymptotically cannot be evaluated inductively in general. The reason is that even in lowest non-trivial order amplitudes 3 with arbitrarily large N’ may show up, ie., we do not get an eikonal equation for § alone, In other words, in a non-linear medium the propagation of wave surfaces in the high frequency limit. and, thus, the corresponding propaga- tion of rays depends on the amplitudes of the wave fields, In this sense, there is no self-contained theory of ray optics for such media, To be sure, there fare some non-linear equations that do give an eikonal equation for S alone. ‘This is true, in particular, of semi-linear equations, ie., of equations which are linear in the highest order derivatives of the dynamical variables (field strengths) with coefficients independent of these variables. However, for the inductive method of determining the amplitudes f.Y to carry over we need nothing less than linearity. For this reason, only in the linear case is it pos- sible to check, along the lines of Sect. 2.7, whether or not ray optics gives a ‘viable approximation scheme. It is worthwile to mention another problem with non-linear equations. ‘Suppose we know that some approximate-plane-wave family (2.33) stays close to exact solutions, for 0 << ag, within some given error bounds. Then it is still possible that a generalized Fourier integral (2.40), formed with this family over a real interval [a1, a9] C [0, ao}, deviates from all exact solutions by an arbitrarily large amount. In this sense, studying approximate-plane- -wave families of the form (2.33) is of limited usefulness in a non-linear medium since it gives no information on non-monochromatic waves. ‘These arguments show that it is somewhat problematic to apply the ray ‘method to non-linear differential equations. As a matter of fact, the existing literature on this topic is much more ‘heuristic? than in the linear case. A typical reference is the book by Jeffrey and Kawahara [66] where many applications to physics are mentioned. ‘Those applications refer mainly to fluid mechanics where nonlinear effects are more important than in optics. In ‘our context, the following strategy is advisable. When dealing with a medium for electromagnetic fields that gives non-linear evolution equations and/or non-linear constraints, it is reasonable to linearize these equations around ‘a (“background”) solution and to apply the ray method to the linearized ‘equations. The resulting theory is valid for all wave fields which are sufficiently ‘weak such that their selfinteractions, caused by the non-linearities of the46 3. Light propagation in other kinds of media. full equations, can be ignored. Interactions with the background field are, of course, taken into account. Following this line of thought, the only possibility to treat dispersive me- dia is by violating the above-mentioned property (b). At first sight, the idea to smuggle the parameter a into the differential equations seems alien to op- tics. (This is a major difference to the JWKB method in wave mechanics. In ‘the latter case, the role of a is played by Planck's constant f which, evidently, appears in Schréidinger’s equation.) Nonetheless, there is a sound method of achieving this goal. Strictly speaking, this method comes in various different ‘variants. The common feature is that one considers asymptotic behavior of approximate-plane-wave families on a one-parameter family of background geometries, rather than on a fixed background geometry. Circumstances per- mitted, this gives an eikonal equation (and transport equations) in close anal- ogy to the treatment of Chap. 2. The crucial point is that even in the case of linear differential equations the eikonal equation need not be homogeneous ‘with respect to 85, i.., dispersion is not excluded. The physical meaning of ‘an eikonal equation derived that way depends, of course, on the way in which the background geometries depend on the parameter. In the following section ‘we demonstrate the method by way of a special example. 3.2 Light propagation in a non-magnetized plasma In this section we consider a simple plasma model as a medium for electro- ‘magnetic waves and we perform the passage to ray optics in such a way that dispersion is taken into account. Apart from some modifications, our treat- ‘ment follows Breuer and Ehlers {18} (19]. For earlier references on the same subject we refer to Madore {90}, to Bicék and Hadrava (14) and to Anile and Pantano [5] (6) ‘We restrict ourselves to the most simple plasma model, viz., to a two-fluid model with vanishing pressure. Then the dynamical system to be considered is governed by the equations AjeFeg = 0, (31) VoFt = J*-enU® (3.2) mU* VU =e F4,U*, (3.3) Va(nU*) =0, (34) Gav U* US (35) (3.1) and (3.2) are the Maxwell equations for the electromagnetic field strength tensor Fay, where square brackets around indices mean antisym- metrization. In (3.2), the ionic current is denoted by J*, whereas the elec- tronic current is written as the product of electron charge e, electron particle3.2 Light propagation in a non-magnetized plasma 47 density n, and electron 4-velocity U*. In mathematical terms, ¢ is a nega- tive constant, n is a nonnegative scalar function, and U* is a vector field normalized by (3.8) (3.3) is the equation of motion for the electron fluid (Buler equation plus Lorentz force), where m is a positive constant with the meaning of the electron ‘mass, Here we assume, as already mentioned, that the pressure of the electron fluid vanishes. This is a legitimate approximation as long as the plasma is sufficiently cold. (8.4) is the equation of charge conservation of the electron component, Please note that (3.2) already implies conservation of the total charge, Ve(J* +enU*) =0, but not of the electron component alone. ‘We want to view (3.1)-(3.8) as a system of non-linear first order differen- tial equations for F,, n and U* with the metric gay and the ionic current J® assumed known. Viewed in this sense, (9.1)-(8.6) give us 444444141 = 14 component equations for 6+ 144 = 11 unknown functions. In a local coordi- nate system with time-like 2lines and space-like hypersurfaces <* = const. cour 14 equations split up into 11 evolution equations and 3 constraints. It is easy to verify that the evolution equations preserve the constraints. More- over, Breuer and Ehlers (18] were able to show that the system of evolution equations admits a locally well-posed initial value problem, and that the equations (3.1)-(3.5) are linearization stable. The latter property guarantees that solutions of the linearized equations are close to solutions of the full equations, Le., that linearization gives a meaningful approximation. ‘This is of particular relevance for us since, following the strategy out- lined in Sect. 3.1, we are now going to linearize (3.1)-(3.5) around some (“background”) solution. For simplicity we restrict ourselves to the case of a background solution with vanishing electromagnetic field. In other words, our background solution is given by a nonnegative scalar function fi and a vector field U* that satisfy the following set of equations. o=s+ents, (3.8) bv =0, (37) velit") =0, 8) 950° = -1 9) Now we linearize the equations (3.1)-(8.5) around this background solution, i.e., we consider these equations for perturbed fields + Pop, (3.10) fan, (a1) +0", (3.12)48 3, Light propagation in other kinds of media ‘and we drop all terms of second and higher order with respect to the perturba- tions Fap, fi, O*. The resulting equations govern the dynamics of sufficiently weak electromagnetic waves J'5 in our plasma which, acoording to Fg = 0, is assumed non-magnetized. We shall presuppose that the metric gap and the ionic current J* are unperturbed. The first assumption is in agreement with. our general stipulation to work on a fixed metric background, i., to disre- gard the back-reaction, governed by Einstein’s field equations, of matter and electromagnetic fields on the metric. The second assumption means that the ‘effect of the electromagnetic wave on the ions is ignored. This is a reasonable approximation since the inertia of the ions is much bigger than that of the electrons. On these assumptions, the linearized system of equations for the perturbations takes the following form. AaFrg =, (13) vik = of 0* 40nd, (14) mb*Y,0° + m0 v.02 =eF*,0°, (3.15) Va(nt® +009) (3.16) gan 0* =0 (317) With gos, % and U* known, (3.13)-(3.17) is a system of first order linear differential equations for Fas, f and U*. It is our goal to find dynamical equations for Fg, alone, i.e., to eliminate 7 and U*. This is indeed possible provided that the background density 7i has no zeros, n>o, (3.18) in the spacetime region considered. If this condition is satisfied, we can pro- ceed in the following way. From (3.14) we find, with the help of (8.9) and (3.17), ei=-U.VF*, (3.19) eh0? = VihO(52 +090.) . (3.20) Since we can divide by f, (8:20) can be used to eliminate O* from (8:15). ‘This results in the following linear second order differential equation for Fay: 5962 +56.) vival + (v9b"(62 + 0°.) + Veb*) Veh ape (3.21)3.2 Light propagation in a nom-magnetized plasma 49 If we have a solution Fg, of (3.13) and (3.21), we can define ft and U* by (3.19) and (3.20), respectively. It is easy to check that then the fall system of equations (3.13)-(8.17) is satisfied, In other words, we have reduced this system (8.13)-(3.17) to dynamical equations for #4, alone, given by (3.13) and (3.21). ‘To rewrite (3.18) and (8.21) in a more convenient form, we express Fg, in terms of a potential Ag, Puy = Ae4y = Vis » (3.22) and we assume that A, satisfies the Landau gauge condition in the rest system of the background electron fluid, A,t*=0. (8.23) Ik is a standard exercise in Maxwell theory to verify that any antisymmetric tensor field Fy that satisfies (3.13) can be locally represented in this way, and that Aq is (locally) uniquely determined by Pay up to gauge transformations Ag— Ag+ dah (3.24) where h is any spacetime function that is constant along the flow lines of U*. Tn other words, ‘h can be freely prescribed on a hypersurface transverse to ee (3 22), (8.18) is automatically satisfied and (5.21) tals the form Dv Ay =0 (3.25) where the differential operator D*/ is defined by DI Ay = 0°68 +020.) Vo(VIV" — 9! VEVa) Ay + (vob*e2 + bet.) + ¥eb9) (WIV — ov A)Ay ~ ZA GI V*— gB°W)A;- (8.26) (8.25) determines the dynamics of electromagnetic waves in our plasma. (3.25) consists of four component equations, but only three of them are in- dependent since the equation b.D%A;=0 (3.27) is identically satisfied for any A,. By the Landau gauge condition (3.23), Ay has three independent components. Hence, we have as many equations as ‘unknown functions. In this sense, (3.25) gives a determined system of linear third order differential equations for the electromagnetic potential. To make50 3, Light propagation in other kinds of media this explicit, one can choose, on an appropriate open subset of spacetime, an orthonormal tetrad field Fy, Ea, Bs, Es with B} = U%. By (3.23), Ay is of the form Ay = gx A" Ek with some scalar functions A+, A?, A? on that domain. Multiplication of (3.25) with goe BE gives us three equations (numbered by 4 = 1,2,3) for the three functions A*, A?, AS. It is shown in Breuer and Ehlers [18] [19] that this system of linear differential equations admits a local existence and uniqueness theorem for any data A", U°9,4", UU" ,0,4" prescribed on a space-like hypersurface. Viewed in this sense, (3.25) is the system of evolution equations for elec- tromagnetic waves in our plasma. Those evolution equations are of second order in the field strengths, and they are not supplemented by constraints. ‘They are, thus, quite different from the evolution equations (2.23) in a linear dielectric and permeable medium. Unfortunately, (3.25) is not of the kind for which standard theorems guarantee the validity of energy inequalities. With the dynamical law (8.25) at hand, we can now perform the passage to ray optics. Sinoe it is our goal to take dispersion into account, we proceed in a way different from Chap. 2. As outlined in Sect. 3.1, it will be crucial to consider one-parameter families of background fields rather than fixed back- ground fields. The background fields that enter into the differential operator ‘D+ are the metric gap, the electron number density 71 and the electron 4- velocity U*. Let us fix such a set of background fields which have to satisfy (3.6)-(3.9) and (3.18). Further, let us fix a spacetime point and a coordinate system around this point, We assume that the chosen point is represented by the coordinates 29 = (x},23,23,14) and that the considered coordinate domain is star-shaped with respect to 29 in R‘, The latter condition means that for any point z in this domain the straight line between z and zo is com- pletely contained in this domain. Refering to this fixed coordinate system, ‘we define new background fields, depending on a real parameter §, by 4as(8,2) = gaa(0 + Ale —=0)) + (328) (8,2) = R(z0 + Ble —20)) (3.29) (8,2) =0*(e9 + A(x 20)) (8.30) For 0 <6 <1, the new background fields 944(9, -), 2(8, -), and U°(8, -) are ‘well defined on the star-shaped domain considered, and they satisfy again ‘equations (3.6)-(3.9) and condition (3.18). (This observation does not carry over if an electromagnetic background field Fg, # 0 is to be taken into account. For a magnetized Plasma, one: cannot. assume the same $-dependence for all background fields gap, i, bs, and Fas) For 8 — 0, the components of the background fields become constant in ‘the coordinate system under consideration. In this sense, gap(0, -), 71(0, -)3.2 Light propagation in a non-magnetized plasma 51 and U*(0, ) are homogeneous fields. In particular, g5(0,~) is a flat metric and U4(0, ) is covariantly constant, ie., an inertial system, with respect to this metric. For this reason, we shall refer to the limit 8 —» 0 as to the homogeneous background limit. If we replace in (3.2) the original background fields gas, i and U* by gav(8, -), (8, -) and U(8, -), respectively, we get a one-parameter family of differential operators D*/(8, -). It is our plan to enter into the differential equation D*/(8, -)4,(, -) = 0 with an approximate-plane-wave ansatz for the potential A,(8, -). Hence, we consider two-parameter families of the form Ajlas6,2) = (331) gRe{eistotAe 0/2, (ayt9 + Ae —20))} which satisfy the Landau gauge condition 54(6,2) Ap(a, 6,2) = (6.82) ‘We assume that the complex amplitudes are of the form ost Gy(a, +) = > af'(-)aN + O(ao*) (3.33) for all integers No > —1 and that Peal 8,2) = AeAy(a, 8,2) = (3.34) of tS (9,848) ao +e —2)) +0(0)} is an approximate-plane-wave family, in the sense of Sect. 2.2, for any fixed 6 with 0 < 8 < 1. For an approximate plane wave in this family, the frequency function with respect to the background electron rest system (3.30) is then siven by w(a,8,2) = (3.38) £U*(c0 + 62 —z0)) 245 (z0 + (2 —20)) « ‘To perform the passage to ray optics, we have to assume that our approximate- plane-wave family satisfies the dynamical equations asymptotically. Since we have two parameters a and 9 at our disposal, we can consider asymptotic to different kinds of limits. is to keep 6 fixed and to consider the condition Jim, (aeD%8, Aras, ») =0 (3.36)52, Light propagation in other kinds of media for N € Z. This is essentially the same kind of limit as considered in Chap. 2. Tt can be characterized as the high frequency limit on a fized background. In ‘the case at hand, the lowest non-trivial order is NV = —3. We leave it to the reader to compute from (3.36) with IV = —3 that the resulting eikonal equa- tion equals the vacuum eikonal equation in the background metric gas(,-), ie., that the corresponding rays are exactly the light-like geodesics of this background metric. In other words, if the high-frequency limit is taken on fixed background, the plasma has no influence on the rays. In particular, there is no dispersion. (If this kind of limit is to be considered, one can, of course, stick to the case = 1 throughout, ie. there is no need to introduce the parameter @ at all.) Now we want to consider a different kind of limit, namely to let and a .g0 to zero simultaneously with the quotient $ kept fixed. We can then simply put a= and consider the condition Jim, (AD"(a, Aaa, ») =0 (3.37) for N € Z. Keeping 4 fixed implies that the frequency function (3.95) is kept fixed at the point 2. Therefore, this kind of limit can be characterized as the homogeneous background limit with fized frequency at z9. We shall now prove that this limit gives, indeed, a different eikonal equation. To that end, ‘we have to assume that (3.37) holds in lowest non-trivial order which is now given by IV =0. This is true if and only if the equation fa) =0 (338) holds at 29, where Q,/ is an abbreviation for Qs = (3.39) ia5(-0,0's-0,0°0,50!s +61 0'80,5 + 5, Si). ‘Here we have used the equation 4 (20) a9(t0) = 0 (3.40) which follows from the Landau gauge condition (3.32). Since (3.84) is sup- posed to be an approximate-plano-wave family, @) must be non-zero and linearly independent of 8S. The condition that (3.38) admits a solution a of this kind at 2, gives the desired eikonal equation at +» for S. We have, thus, to solve the eigenvalue problem of Q,/ restricted to the orthocomplement of G/., We find that there are three real eigenvalues, viz. a= Ubas(—(b°as)?+ Eh), (3.41) =r =0as(a4sa,8 +i). (3.42)3.2 Light propagation in a non-magnetized plasma 53 Ieither [2,5 = 0 or 0,5 = £/2H Ue, all three eigenvalues coincide and (3.38) is satisfied by any 9. Otherwise, we find A: # Ag = 2g. In the latter case, the eigenspace pertaining to A, is one-dimensional and spanned by 0,5 + 08° O48 whereas the eigenspace pertaining to Az = As is two- dimensional and consists of all X, with Of X; = 0f8Xy =0. Equation (8.38) admits a non-trivial solution a which is perpendicular to U if and only if one of the eigenvalues As, Az, Az {8 zero. From the form of the eigenspaces we see that in any such case a can be chosen linearly independent of 8,5. Hence, the eikonal equation takes the form 2 Aas = 0 which is equivalent to Bas(—( 4,8)" +i) (a#sa5+ i) =0 (3.43) Let us be precise about this result. Our assumption that the asymptotic condition (3.37) holds in lowest non-trivial order requires that satisfies (8.43) at the point zr around which the construction was done. Although we have used a fixed coordinate system around the chosen space- time point to perform the homogeneous background limit, the eikonal equa- tion is a covariant equation (i¢., independent of this coordinate system). If 'S satisfies this covariant equation (3.43) on an open spacetime domain U, it is associated with an asymptotic solution of lowest non-trivial order, in the homogencous-background sense, around any point of U. That is to say, to any such $ we can find a non-trivial amplitude @y(a, -) on U such that the following holds. If we choose any coordinate system around any point of U, thereby defining the one-parameter family of operators D*/(8, -) and the ‘two-parameter family (3.31) of electromagnetic fields, the asymptotic condi- tion (3.37) is satisied for N’ = 0. As a matter of fact, a similar statement is true for any N. However, this more general result does not follow from our reasoning s0 far. ‘Owing to the terms proportional to fi, the eikonal equation (3.43) is not homogeneous with respect to 0S. This indicates dispersion. ‘The product structure of the eikonal equation (3.43) suggests to introduce three partial Hamiltonians Be) m, (3.44) Ha(a,p) =} (-U°@) Oe) pare + SH(2)) , (8.45) Hale,p) = 4 (992) pare + $2) « (3.46) ‘Our assumptions guarantee that each partial Hamiltonian satisfies condition (2.83) on its characteristic variety. We are, of course, free to change each partial Hamiltonian by a transformation of the form (2.84).